Datasets:

---

bigcode

/

jupyter-parsed

like 4

Follow

BigCode 1.34k

[ "CC0-1.0" ]

[ "CC0-1.0" ]

[ "CC0-1.0" ]

[ [ [ "from requests import get\nURL = 'https://dl.fedoraproject.org/pub/alt/iot/32/IoT/x86_64/images/Fedora-IoT-32-20200603.0.x86_64.raw.xz'", "_____no_output_____" ], [ "# This downloads the entire response\nr = get(URL, stream=True)\nfor b in r.iter_content(chunk_size=None):\n print(len(b))", "533830860\n" ], [ "# For comparison: this yeilds new chunks as they arrive\nr = get(URL, stream=True)\nfor b in r.iter_content(chunk_size=2**23):\n print(len(b))", "8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n8388608\n5348556\n" ] ] ]

[ "code" ]

[ [ "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import pandas as pd\r\nimport numpy as np\r\nimport matplotlib.pyplot as plt\r\nimport seaborn as sns\r\n%matplotlib inline\r\nimport warnings\r\nwarnings.filterwarnings(\"ignore\")", "_____no_output_____" ], [ "df = pd.read_csv(\"Data.csv\")\r\ndf.head(5)", "_____no_output_____" ], [ "df['sentiment'] = np.where(df['sentiment'] == \"positive\", 1, 0)\r\ndf.head()", "_____no_output_____" ], [ "df['sentiment'].value_counts().sort_index().plot(kind='bar',color = 'blue')\r\nplt.xlabel('Sentiment')\r\nplt.ylabel('Count')", "_____no_output_____" ], [ "df = df.sample(frac=0.1, random_state=0) \r\ndf.dropna(inplace=True)", "_____no_output_____" ], [ "from sklearn.model_selection import train_test_split\r\nX_train, X_test, y_train, y_test = train_test_split(df['review'], df['sentiment'],test_size=0.1, random_state=0)", "_____no_output_____" ], [ "def cleanText(raw_text, remove_stopwords=False, stemming=False, split_text=False):\r\n text = BeautifulSoup(raw_text, 'html.parser').get_text()\r\n letters_only = re.sub(\"[^a-zA-Z]\", \" \", text)\r\n words = letters_only.lower().split() \r\n \r\n if remove_stopwords:\r\n stops = set(stopwords.words(\"english\"))\r\n words = [w for w in words if not w in stops]\r\n \r\n if stemming==True:\r\n\r\n stemmer = SnowballStemmer('english') \r\n words = [stemmer.stem(w) for w in words]\r\n \r\n if split_text==True:\r\n return (words)\r\n \r\n return( \" \".join(words))", "_____no_output_____" ], [ "import re\r\nimport nltk\r\nfrom nltk.corpus import stopwords \r\nfrom nltk.stem.porter import PorterStemmer\r\nfrom nltk.stem import SnowballStemmer, WordNetLemmatizer\r\nfrom nltk import sent_tokenize, word_tokenize, pos_tag\r\nfrom bs4 import BeautifulSoup \r\nimport logging\r\nfrom wordcloud import WordCloud\r\nfrom gensim.models import word2vec\r\nfrom gensim.models import Word2Vec\r\nfrom gensim.models.keyedvectors import KeyedVectors\r\n\r\nX_train_cleaned = []\r\nX_test_cleaned = []\r\n\r\nfor d in X_train:\r\n X_train_cleaned.append(cleanText(d))\r\nprint('Show a cleaned review in the training set : \\n', X_train_cleaned[10])\r\n \r\nfor d in X_test:\r\n X_test_cleaned.append(cleanText(d))", "Show a cleaned review in the training set : \n the crimson rivers is one of the most over directed over the top over everything mess i ve ever seen come out of france there s nothing worse than a french production trying to out do films made in hollywood and cr is a perfect example of such a wannabe horror action buddy flick i almost stopped it halfway through because i knew it wouldn t amount to anything but french guys trying to show off the film starts off promisingly like some sort of expansive horror film but it quickly shifts genres from horror to action to x files type to buddy flick that in the end cr is all of it and also none of it it s so full of clich s that at one point i thought the whole thing was a comedy the painful dialogue and those silent pauses with fades outs and fades ins just at the right expositionary moments made me groan i thought only films made in hollywood used this hackneyed technique the chase scene with vincent cassel running after the killer is so over directed and over done that it s almost a thing of beauty the climax on top of the mountain with the stupid revelation about the killer s with cassel and reno playing buddies like nolte and murphy in hrs completely derailed what little credibility the film had by then it s difficult to believe that the director of the crimson rivers also directed gothika which though had its share of problems doesn t even come close to the awfulness of this overbaked confused film\n" ] ], [ [ "## CountVectorizer with Mulinomial Naive Bayes (Benchmark Model)", "_____no_output_____" ] ], [ [ "from sklearn.feature_extraction.text import CountVectorizer,TfidfVectorizer\r\nfrom sklearn.naive_bayes import BernoulliNB, MultinomialNB\r\ncountVect = CountVectorizer() \r\nX_train_countVect = countVect.fit_transform(X_train_cleaned)\r\nprint(\"Number of features : %d \\n\" %len(countVect.get_feature_names())) #6378 \r\nprint(\"Show some feature names : \\n\", countVect.get_feature_names()[::1000])\r\n\r\n\r\n# Train MultinomialNB classifier\r\nmnb = MultinomialNB()\r\nmnb.fit(X_train_countVect, y_train)", "Number of features : 36751 \n\nShow some feature names : \n ['aa', 'ameche', 'auggie', 'betrayals', 'bright', 'cathryn', 'clownhouse', 'copying', 'dazzle', 'disarray', 'dvd', 'estimation', 'fighter', 'fusion', 'greenfinch', 'henson', 'imaginings', 'ir', 'kint', 'linklater', 'maropis', 'misik', 'nectar', 'organise', 'performing', 'pre', 'rages', 'reputedly', 'saddled', 'sexiness', 'smith', 'steal', 'swoozie', 'tinfoil', 'unattuned', 'vernacular', 'willed']\n" ], [ "import pickle\r\npickle.dump(countVect,open('countVect_imdb.pkl','wb'))", "_____no_output_____" ], [ "from sklearn import metrics\r\nfrom sklearn.metrics import accuracy_score,roc_auc_score\r\ndef modelEvaluation(predictions):\r\n '''\r\n Print model evaluation to predicted result \r\n '''\r\n print (\"\\nAccuracy on validation set: {:.4f}\".format(accuracy_score(y_test, predictions)))\r\n print(\"\\nAUC score : {:.4f}\".format(roc_auc_score(y_test, predictions)))\r\n print(\"\\nClassification report : \\n\", metrics.classification_report(y_test, predictions))\r\n print(\"\\nConfusion Matrix : \\n\", metrics.confusion_matrix(y_test, predictions))", "_____no_output_____" ], [ "predictions = mnb.predict(countVect.transform(X_test_cleaned))\r\nmodelEvaluation(predictions)", "\nAccuracy on validation set: 0.8140\n\nAUC score : 0.8142\n\nClassification report : \n precision recall f1-score support\n\n 0 0.79 0.86 0.82 249\n 1 0.85 0.77 0.81 251\n\n accuracy 0.81 500\n macro avg 0.82 0.81 0.81 500\nweighted avg 0.82 0.81 0.81 500\n\n\nConfusion Matrix : \n [[214 35]\n [ 58 193]]\n" ], [ "import pickle\r\npickle.dump(mnb,open('Naive_Bayes_model_imdb.pkl','wb'))", "_____no_output_____" ] ], [ [ "# TfidfVectorizer with Logistic Regression", "_____no_output_____" ] ], [ [ "from sklearn.linear_model import LogisticRegression\r\ntfidf = TfidfVectorizer(min_df=5) #minimum document frequency of 5\r\nX_train_tfidf = tfidf.fit_transform(X_train)\r\nprint(\"Number of features : %d \\n\" %len(tfidf.get_feature_names())) #1722\r\nprint(\"Show some feature names : \\n\", tfidf.get_feature_names()[::1000])\r\n\r\n# Logistic Regression\r\nlr = LogisticRegression()\r\nlr.fit(X_train_tfidf, y_train)", "Number of features : 10505 \n\nShow some feature names : \n ['00', 'belonged', 'completion', 'dubious', 'garbage', 'interviewing', 'million', 'plays', 'rough', 'strike', 'vein']\n" ], [ "feature_names = np.array(tfidf.get_feature_names())\r\nsorted_coef_index = lr.coef_[0].argsort()\r\nprint('\\nTop 10 features with smallest coefficients :\\n{}\\n'.format(feature_names[sorted_coef_index[:10]]))\r\nprint('Top 10 features with largest coefficients : \\n{}'.format(feature_names[sorted_coef_index[:-11:-1]]))", "\nTop 10 features with smallest coefficients :\n['bad' 'worst' 'awful' 'no' 'waste' 'poor' 'terrible' 'boring' 'even'\n 'minutes']\n\nTop 10 features with largest coefficients : \n['great' 'and' 'excellent' 'best' 'it' 'wonderful' 'very' 'also' 'well'\n 'love']\n" ], [ "predictions = lr.predict(tfidf.transform(X_test_cleaned))\r\nmodelEvaluation(predictions)", "\nAccuracy on validation set: 0.8500\n\nAUC score : 0.8500\n\nClassification report : \n precision recall f1-score support\n\n 0 0.85 0.85 0.85 249\n 1 0.85 0.85 0.85 251\n\n accuracy 0.85 500\n macro avg 0.85 0.85 0.85 500\nweighted avg 0.85 0.85 0.85 500\n\n\nConfusion Matrix : \n [[211 38]\n [ 37 214]]\n" ], [ "from sklearn.model_selection import GridSearchCV\r\nfrom sklearn import metrics\r\nfrom sklearn.metrics import roc_auc_score, accuracy_score\r\nfrom sklearn.pipeline import Pipeline\r\nestimators = [(\"tfidf\", TfidfVectorizer()), (\"lr\", LogisticRegression())]\r\nmodel = Pipeline(estimators)\r\n\r\n\r\nparams = {\"lr__C\":[0.1, 1, 10], \r\n \"tfidf__min_df\": [1, 3], \r\n \"tfidf__max_features\": [1000, None], \r\n \"tfidf__ngram_range\": [(1,1), (1,2)], \r\n \"tfidf__stop_words\": [None, \"english\"]} \r\n\r\ngrid = GridSearchCV(estimator=model, param_grid=params, scoring=\"accuracy\", n_jobs=-1)\r\ngrid.fit(X_train_cleaned, y_train)\r\nprint(\"The best paramenter set is : \\n\", grid.best_params_)\r\n\r\n\r\n# Evaluate on the validaton set\r\npredictions = grid.predict(X_test_cleaned)\r\nmodelEvaluation(predictions)", "The best paramenter set is : \n {'lr__C': 10, 'tfidf__max_features': None, 'tfidf__min_df': 3, 'tfidf__ngram_range': (1, 2), 'tfidf__stop_words': None}\n\nAccuracy on validation set: 0.8720\n\nAUC score : 0.8720\n\nClassification report : \n precision recall f1-score support\n\n 0 0.87 0.87 0.87 249\n 1 0.87 0.88 0.87 251\n\n accuracy 0.87 500\n macro avg 0.87 0.87 0.87 500\nweighted avg 0.87 0.87 0.87 500\n\n\nConfusion Matrix : \n [[216 33]\n [ 31 220]]\n" ] ], [ [ "# Word2Vec\n<br>\n\n**Step 1 : Parse review text to sentences (Word2Vec model takes a list of sentences as inputs)**\n\n**Step 2 : Create volcabulary list using Word2Vec model.**\n\n**Step 3 : Transform each review into numerical representation by computing average feature vectors of words therein.**\n\n**Step 4 : Fit the average feature vectors to Random Forest Classifier.**", "_____no_output_____" ] ], [ [ "tokenizer = nltk.data.load('tokenizers/punkt/english.pickle')\r\n\r\ndef parseSent(review, tokenizer, remove_stopwords=False):\r\n\r\n raw_sentences = tokenizer.tokenize(review.strip())\r\n sentences = []\r\n for raw_sentence in raw_sentences:\r\n if len(raw_sentence) > 0:\r\n sentences.append(cleanText(raw_sentence, remove_stopwords, split_text=True))\r\n return sentences\r\n\r\n\r\n# Parse each review in the training set into sentences\r\nsentences = []\r\nfor review in X_train_cleaned:\r\n sentences += parseSent(review, tokenizer,remove_stopwords=False)\r\n \r\nprint('%d parsed sentence in the training set\\n' %len(sentences))\r\nprint('Show a parsed sentence in the training set : \\n', sentences[10])", "4500 parsed sentence in the training set\n\nShow a parsed sentence in the training set : \n ['the', 'crimson', 'rivers', 'is', 'one', 'of', 'the', 'most', 'over', 'directed', 'over', 'the', 'top', 'over', 'everything', 'mess', 'i', 've', 'ever', 'seen', 'come', 'out', 'of', 'france', 'there', 's', 'nothing', 'worse', 'than', 'a', 'french', 'production', 'trying', 'to', 'out', 'do', 'films', 'made', 'in', 'hollywood', 'and', 'cr', 'is', 'a', 'perfect', 'example', 'of', 'such', 'a', 'wannabe', 'horror', 'action', 'buddy', 'flick', 'i', 'almost', 'stopped', 'it', 'halfway', 'through', 'because', 'i', 'knew', 'it', 'wouldn', 't', 'amount', 'to', 'anything', 'but', 'french', 'guys', 'trying', 'to', 'show', 'off', 'the', 'film', 'starts', 'off', 'promisingly', 'like', 'some', 'sort', 'of', 'expansive', 'horror', 'film', 'but', 'it', 'quickly', 'shifts', 'genres', 'from', 'horror', 'to', 'action', 'to', 'x', 'files', 'type', 'to', 'buddy', 'flick', 'that', 'in', 'the', 'end', 'cr', 'is', 'all', 'of', 'it', 'and', 'also', 'none', 'of', 'it', 'it', 's', 'so', 'full', 'of', 'clich', 's', 'that', 'at', 'one', 'point', 'i', 'thought', 'the', 'whole', 'thing', 'was', 'a', 'comedy', 'the', 'painful', 'dialogue', 'and', 'those', 'silent', 'pauses', 'with', 'fades', 'outs', 'and', 'fades', 'ins', 'just', 'at', 'the', 'right', 'expositionary', 'moments', 'made', 'me', 'groan', 'i', 'thought', 'only', 'films', 'made', 'in', 'hollywood', 'used', 'this', 'hackneyed', 'technique', 'the', 'chase', 'scene', 'with', 'vincent', 'cassel', 'running', 'after', 'the', 'killer', 'is', 'so', 'over', 'directed', 'and', 'over', 'done', 'that', 'it', 's', 'almost', 'a', 'thing', 'of', 'beauty', 'the', 'climax', 'on', 'top', 'of', 'the', 'mountain', 'with', 'the', 'stupid', 'revelation', 'about', 'the', 'killer', 's', 'with', 'cassel', 'and', 'reno', 'playing', 'buddies', 'like', 'nolte', 'and', 'murphy', 'in', 'hrs', 'completely', 'derailed', 'what', 'little', 'credibility', 'the', 'film', 'had', 'by', 'then', 'it', 's', 'difficult', 'to', 'believe', 'that', 'the', 'director', 'of', 'the', 'crimson', 'rivers', 'also', 'directed', 'gothika', 'which', 'though', 'had', 'its', 'share', 'of', 'problems', 'doesn', 't', 'even', 'come', 'close', 'to', 'the', 'awfulness', 'of', 'this', 'overbaked', 'confused', 'film']\n" ] ], [ [ "## Creating Volcabulary List usinhg Word2Vec Model", "_____no_output_____" ] ], [ [ "from wordcloud import WordCloud\r\nfrom gensim.models import word2vec\r\nfrom gensim.models.keyedvectors import KeyedVectors\r\nnum_features = 300 #embedding dimension \r\nmin_word_count = 10 \r\nnum_workers = 4 \r\ncontext = 10 \r\ndownsampling = 1e-3 \r\n\r\nprint(\"Training Word2Vec model ...\\n\")\r\nw2v = Word2Vec(sentences, workers=num_workers, min_count = min_word_count,\\\r\n window = context, sample = downsampling)\r\nw2v.init_sims(replace=True)\r\nw2v.save(\"w2v_300features_10minwordcounts_10context\") #save trained word2vec model\r\n\r\nprint(\"Number of words in the vocabulary list : %d \\n\" %len(w2v.wv.index2word)) #4016 \r\nprint(\"Show first 10 words in the vocalbulary list vocabulary list: \\n\", w2v.wv.index2word[0:10])", "Training Word2Vec model ...\n\n" ] ], [ [ "## Averaging Feature Vectors", "_____no_output_____" ] ], [ [ "def makeFeatureVec(review, model, num_features):\r\n '''\r\n Transform a review to a feature vector by averaging feature vectors of words \r\n appeared in that review and in the volcabulary list created\r\n '''\r\n featureVec = np.zeros((num_features,),dtype=\"float32\")\r\n nwords = 0.\r\n index2word_set = set(model.wv.index2word) #index2word is the volcabulary list of the Word2Vec model\r\n isZeroVec = True\r\n for word in review:\r\n if word in index2word_set: \r\n nwords = nwords + 1.\r\n featureVec = np.add(featureVec, model[word])\r\n isZeroVec = False\r\n if isZeroVec == False:\r\n featureVec = np.divide(featureVec, nwords)\r\n return featureVec", "_____no_output_____" ], [ "def getAvgFeatureVecs(reviews, model, num_features):\r\n '''\r\n Transform all reviews to feature vectors using makeFeatureVec()\r\n '''\r\n counter = 0\r\n reviewFeatureVecs = np.zeros((len(reviews),num_features),dtype=\"float32\")\r\n for review in reviews:\r\n reviewFeatureVecs[counter] = makeFeatureVec(review, model,num_features)\r\n counter = counter + 1\r\n return reviewFeatureVecs", "_____no_output_____" ], [ "X_train_cleaned = []\r\nfor review in X_train:\r\n X_train_cleaned.append(cleanText(review, remove_stopwords=True, split_text=True))\r\ntrainVector = getAvgFeatureVecs(X_train_cleaned, w2v, num_features)\r\nprint(\"Training set : %d feature vectors with %d dimensions\" %trainVector.shape)\r\n\r\n\r\n# Get feature vectors for validation set\r\nX_test_cleaned = []\r\nfor review in X_test:\r\n X_test_cleaned.append(cleanText(review, remove_stopwords=True, split_text=True))\r\ntestVector = getAvgFeatureVecs(X_test_cleaned, w2v, num_features)\r\nprint(\"Validation set : %d feature vectors with %d dimensions\" %testVector.shape)", "_____no_output_____" ] ], [ [ "# Random Forest Classifer", "_____no_output_____" ] ], [ [ "from sklearn.ensemble import RandomForestClassifier\r\nrf = RandomForestClassifier(n_estimators=1000)\r\nrf.fit(trainVector, y_train)\r\npredictions = rf.predict(testVector)\r\nmodelEvaluation(predictions)", "_____no_output_____" ] ], [ [ "## LSTM\n<br>\n\n**Step 1 : Prepare X_train and X_test to 2D tensor.**\n \n**Step 2 : Train a simple LSTM (embeddign layer => LSTM layer => dense layer).**\n \n**Step 3 : Compile and fit the model using log loss function and ADAM optimizer.**", "_____no_output_____" ] ], [ [ "from keras.preprocessing import sequence\r\nfrom keras.utils import np_utils\r\nfrom keras.models import Sequential\r\nfrom keras.layers.core import Dense, Dropout, Activation, Lambda\r\nfrom keras.layers.embeddings import Embedding\r\nfrom keras.layers.recurrent import LSTM, SimpleRNN, GRU\r\nfrom keras.preprocessing.text import Tokenizer\r\nfrom collections import defaultdict\r\nfrom keras.layers.convolutional import Convolution1D\r\nfrom keras import backend as K\r\nfrom keras.layers.embeddings import Embedding", "Using TensorFlow backend.\nWARNING:root:Limited tf.compat.v2.summary API due to missing TensorBoard installation.\nWARNING:root:Limited tf.compat.v2.summary API due to missing TensorBoard installation.\nWARNING:root:Limited tf.compat.v2.summary API due to missing TensorBoard installation.\nWARNING:root:Limited tf.compat.v2.summary API due to missing TensorBoard installation.\nWARNING:root:Limited tf.compat.v2.summary API due to missing TensorBoard installation.\nWARNING:root:Limited tf.compat.v2.summary API due to missing TensorBoard installation.\nWARNING:root:Limited tf.summary API due to missing TensorBoard installation.\n" ], [ "top_words = 40000 \r\nmaxlen = 200 \r\nbatch_size = 62\r\nnb_classes = 4\r\nnb_epoch = 6\r\n\r\n\r\n# Vectorize X_train and X_test to 2D tensor\r\ntokenizer = Tokenizer(nb_words=top_words) #only consider top 20000 words in the corpse\r\ntokenizer.fit_on_texts(X_train)\r\n# tokenizer.word_index #access word-to-index dictionary of trained tokenizer\r\n\r\nsequences_train = tokenizer.texts_to_sequences(X_train)\r\nsequences_test = tokenizer.texts_to_sequences(X_test)\r\n\r\nX_train_seq = sequence.pad_sequences(sequences_train, maxlen=maxlen)\r\nX_test_seq = sequence.pad_sequences(sequences_test, maxlen=maxlen)\r\n\r\n\r\n# one-hot encoding of y_train and y_test\r\ny_train_seq = np_utils.to_categorical(y_train, nb_classes)\r\ny_test_seq = np_utils.to_categorical(y_test, nb_classes)\r\n\r\nprint('X_train shape:', X_train_seq.shape)\r\nprint(\"========================================\")\r\nprint('X_test shape:', X_test_seq.shape)\r\nprint(\"========================================\")\r\nprint('y_train shape:', y_train_seq.shape)\r\nprint(\"========================================\")\r\nprint('y_test shape:', y_test_seq.shape)\r\nprint(\"========================================\")", "X_train shape: (4500, 200)\n========================================\nX_test shape: (500, 200)\n========================================\ny_train shape: (4500, 4)\n========================================\ny_test shape: (500, 4)\n========================================\n" ], [ "model1 = Sequential()\r\nmodel1.add(Embedding(top_words, 128, dropout=0.2))\r\nmodel1.add(LSTM(128, dropout_W=0.2, dropout_U=0.2)) \r\nmodel1.add(Dense(nb_classes))\r\nmodel1.add(Activation('softmax'))\r\nmodel1.summary()", "Model: \"sequential_1\"\n_________________________________________________________________\nLayer (type) Output Shape Param # \n=================================================================\nembedding_1 (Embedding) (None, None, 128) 5120000 \n_________________________________________________________________\nlstm_1 (LSTM) (None, 128) 131584 \n_________________________________________________________________\ndense_1 (Dense) (None, 4) 516 \n_________________________________________________________________\nactivation_1 (Activation) (None, 4) 0 \n=================================================================\nTotal params: 5,252,100\nTrainable params: 5,252,100\nNon-trainable params: 0\n_________________________________________________________________\n" ], [ "model1.compile(loss='binary_crossentropy',\r\n optimizer='adam',\r\n metrics=['accuracy'])\r\n\r\nmodel1.fit(X_train_seq, y_train_seq, batch_size=batch_size, nb_epoch=nb_epoch, verbose=1)\r\n\r\n# Model evluation\r\nscore = model1.evaluate(X_test_seq, y_test_seq, batch_size=batch_size)\r\nprint('Test loss : {:.4f}'.format(score[0]))\r\nprint('Test accuracy : {:.4f}'.format(score[1]))", "Epoch 1/6\n4500/4500 [==============================] - 22s 5ms/step - loss: 0.3760 - accuracy: 0.7594\nEpoch 2/6\n4500/4500 [==============================] - 24s 5ms/step - loss: 0.2857 - accuracy: 0.8577\nEpoch 3/6\n4500/4500 [==============================] - 24s 5ms/step - loss: 0.1591 - accuracy: 0.9347\nEpoch 4/6\n4500/4500 [==============================] - 24s 5ms/step - loss: 0.0838 - accuracy: 0.9699\nEpoch 5/6\n4500/4500 [==============================] - 24s 5ms/step - loss: 0.0385 - accuracy: 0.9874\nEpoch 6/6\n4500/4500 [==============================] - 24s 5ms/step - loss: 0.0225 - accuracy: 0.9925\n500/500 [==============================] - 1s 1ms/step\nTest loss : 0.4559\nTest accuracy : 0.8750\n" ], [ "len(X_train_seq),len(y_train_seq)", "_____no_output_____" ], [ "print(\"Size of weight matrix in the embedding layer : \", \\\r\n model1.layers[0].get_weights()[0].shape)\r\n\r\n# get weight matrix of the hidden layer\r\nprint(\"Size of weight matrix in the hidden layer : \", \\\r\n model1.layers[1].get_weights()[0].shape)\r\n\r\n# get weight matrix of the output layer\r\nprint(\"Size of weight matrix in the output layer : \", \\\r\n model1.layers[2].get_weights()[0].shape)", "Size of weight matrix in the embedding layer : (40000, 128)\nSize of weight matrix in the hidden layer : (128, 512)\nSize of weight matrix in the output layer : (128, 4)\n" ], [ "import pickle\r\npickle.dump(model1,open('model1.pkl','wb'))", "_____no_output_____" ] ], [ [ "## LSTM with Word2Vec Embedding", "_____no_output_____" ] ], [ [ "2v = Word2Vec.load(\"w2v_300features_10minwordcounts_10context\")\r\n\r\nembedding_matrix = w2v.wv.syn0 \r\nprint(\"Shape of embedding matrix : \", embedding_matrix.shape)", "_____no_output_____" ], [ "top_words = embedding_matrix.shape[0] #4016 \r\nmaxlen = 300 \r\nbatch_size = 62\r\nnb_classes = 4\r\nnb_epoch = 7\r\n\r\n\r\n# Vectorize X_train and X_test to 2D tensor\r\ntokenizer = Tokenizer(nb_words=top_words) #only consider top 20000 words in the corpse\r\ntokenizer.fit_on_texts(X_train)\r\n# tokenizer.word_index #access word-to-index dictionary of trained tokenizer\r\n\r\nsequences_train = tokenizer.texts_to_sequences(X_train)\r\nsequences_test = tokenizer.texts_to_sequences(X_test)\r\n\r\nX_train_seq1 = sequence.pad_sequences(sequences_train, maxlen=maxlen)\r\nX_test_seq1 = sequence.pad_sequences(sequences_test, maxlen=maxlen)\r\n\r\n\r\n# one-hot encoding of y_train and y_test\r\ny_train_seq1 = np_utils.to_categorical(y_train, nb_classes)\r\ny_test_seq1 = np_utils.to_categorical(y_test, nb_classes)\r\n\r\nprint('X_train shape:', X_train_seq1.shape)\r\nprint(\"========================================\")\r\nprint('X_test shape:', X_test_seq1.shape)\r\nprint(\"========================================\")\r\nprint('y_train shape:', y_train_seq1.shape)\r\nprint(\"========================================\")\r\nprint('y_test shape:', y_test_seq1.shape)\r\nprint(\"========================================\")", "_____no_output_____" ], [ "len(X_train_seq1),len(y_train_seq1)", "_____no_output_____" ], [ "embedding_layer = Embedding(embedding_matrix.shape[0], #4016\r\n embedding_matrix.shape[1], #300\r\n weights=[embedding_matrix])\r\n\r\nmodel2 = Sequential()\r\nmodel2.add(embedding_layer)\r\nmodel2.add(LSTM(128, dropout_W=0.2, dropout_U=0.2)) \r\nmodel2.add(Dense(nb_classes))\r\nmodel2.add(Activation('softmax'))\r\nmodel2.summary()", "_____no_output_____" ], [ "model2.compile(loss='binary_crossentropy',\r\n optimizer='adam',\r\n metrics=['accuracy'])\r\n\r\nmodel2.fit(X_train_seq1, y_train_seq1, batch_size=batch_size, nb_epoch=nb_epoch, verbose=1)\r\n\r\n# Model evaluation\r\nscore = model2.evaluate(X_test_seq1, y_test_seq1, batch_size=batch_size)\r\nprint('Test loss : {:.4f}'.format(score[0]))\r\nprint('Test accuracy : {:.4f}'.format(score[1]))", "_____no_output_____" ], [ "print(\"Size of weight matrix in the embedding layer : \", \\\r\n model2.layers[0].get_weights()[0].shape) \r\n\r\nprint(\"Size of weight matrix in the hidden layer : \", \\\r\n model2.layers[1].get_weights()[0].shape) \r\n\r\nprint(\"Size of weight matrix in the output layer : \", \\\r\n model2.layers[2].get_weights()[0].shape) ", "_____no_output_____" ] ] ]

[ "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "print('Hello World')", "Hello World\n" ], [ "import numpy", "_____no_output_____" ], [ "print(numpy.pi)", "3.141592653589793\n" ], [ "import numpy as np", "_____no_output_____" ], [ "print(np.pi)", "3.141592653589793\n" ], [ "import numpy as np", "_____no_output_____" ], [ "import numpy as np\nimport matplotlib.pyplot as plt", "_____no_output_____" ], [ "%pinfo print", "_____no_output_____" ], [ "%matplotlib inline", "_____no_output_____" ], [ "plt.style.use('../../solving_pde_mooc/notebooks/styles/mainstyle.use')", "_____no_output_____" ], [ "#create an empty python list that will create values of delta for all k\ndelta_list = []\ndelta_list_2 = []\n\n#we have to loop for k=1,2,3,4,5,6,7,8,9\n\n#we will use the range function range(min,max,step)\n#this creates values of min, min+1*step, min+2*step, ..., max-step\n#as you can see, max value is not (actually never!) included in the list\n\n#min and step are optional and if you don't input them, then default values will be used\n#the default are min=0,step=1\n\n#range(5) will give 0,1,2,3,4 since it takes min=0, step=1 (default)\n#range(1,5) will give 1,2,3,4 since it takes min=1\n#range(2,5) will give 2,3,4\n#range(2,8,2) will give 2,4,6\n\n#once we have create the list of k, we have to create the delta_list\n#we can create the delta_list by adding/appending values to the end of the list each time\n#we can do that by using the append function as done below\nfor k in range(1,10):\n delta_list.append(2**(-k))\n \nprint('\\n')\nprint(delta_list)\n\n#another way to create a list is to direcly add values to it\ndelta_list_2 = [2**(-k) for k in range(1,10)]\n\nprint('\\n')\nprint(delta_list_2)\n\n#here a list is being created using the values and fuctions that we want\n#but to do operations on it, we need to create a numerical array out of it\n\n#np(Numpy) helps us create an array out of it as shown below\ndelta = np.array(delta_list_2)\n\nprint('\\n')\nprint(delta)", "\n\n[0.5, 0.25, 0.125, 0.0625, 0.03125, 0.015625, 0.0078125, 0.00390625, 0.001953125]\n\n\n[0.5, 0.25, 0.125, 0.0625, 0.03125, 0.015625, 0.0078125, 0.00390625, 0.001953125]\n\n\n[0.5 0.25 0.125 0.0625 0.03125 0.015625\n 0.0078125 0.00390625 0.00195312]\n" ], [ "%%timeit\n\n#method 1 - using append function\n\n#create an empty list first\ndelta_list = []\n\n#append values to the list\nfor k in range(1,10):\n delta_list.append(2**(-k))\n\n#assign the list to an array using numpy\ndelta = np.array(delta_list)\n", "2.71 µs ± 14.4 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)\n" ], [ "#%%timeit\n\n#method 2 - directly adding values to the list\n\n#create an empty list first\ndelta_list = []\n\n#directly add values to the list\ndelta_list = [2**(-k) for k in range(1,10)]\n\n#assign the list to an array using numpy\ndelta = np.array(delta_list)\n\nprint(delta_list[8])", "0.001953125\n" ], [ "#%%timeit\n\n#method 3 - directly create a numpy array\n\n#first create a numpy array with all zeros\ndelta_array = np.zeros(9)\n\nprint('\\n')\nprint(delta_array)\n\n#this will create an array of 9 elements - indexing as 0,1,2,...,8\n#numpy array starts indexing from 0 and goest to N-1\n# var =np.zeros(9) \n#this will mean that the first element in the array will be var[0]\n#last element in the array will be var(8)\n\nprint('\\n')\nprint(len(delta))\nprint('\\n')\n\n#fill the array\n#using len function, since i will start from zero\n#also, numpy array starts indexing from 0 to N-1\nfor i in range(len(delta_array)):\n print(i)\n delta_array[i]=2**(-i-1)\n \nprint('\\n')\nprint(delta_array[0])\nprint('\\n')\nprint(delta_array[8])", "\n\n[0. 0. 0. 0. 0. 0. 0. 0. 0.]\n\n\n9\n\n\n0\n1\n2\n3\n4\n5\n6\n7\n8\n\n\n0.5\n\n\n0.001953125\n" ], [ "#the method I am most used to\n\ndelta_array = np.zeros(9)\n\nfor i in range(len(delta_array)):\n delta_array[i]=2**(-i-1)\n \nprint(delta_array)\nprint('\\n')\nprint(delta_array[8])", "[0.5 0.25 0.125 0.0625 0.03125 0.015625\n 0.0078125 0.00390625 0.00195312]\n\n\n0.001953125\n" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Implementing TF-IDF\n------------------------------------\n\nHere we implement TF-IDF, (Text Frequency - Inverse Document Frequency) for the spam-ham text data.\n\nWe will use a hybrid approach of encoding the texts with sci-kit learn's TFIDF vectorizer. Then we will use the regular TensorFlow logistic algorithm outline.\n\nCreating the TF-IDF vectors requires us to load all the text into memory and count the occurrences of each word before we can start training our model. Because of this, it is not implemented fully in Tensorflow, so we will use Scikit-learn for creating our TF-IDF embedding, but use Tensorflow to fit the logistic model.\n\nWe start by loading the necessary libraries.", "_____no_output_____" ] ], [ [ "import tensorflow as tf\nimport matplotlib.pyplot as plt\nimport csv\nimport numpy as np\nimport os\nimport string\nimport requests\nimport io\nimport nltk\nfrom zipfile import ZipFile\nfrom sklearn.feature_extraction.text import TfidfVectorizer\nfrom tensorflow.python.framework import ops\nops.reset_default_graph()", "_____no_output_____" ] ], [ [ "Start a computational graph session.", "_____no_output_____" ] ], [ [ "sess = tf.Session()", "_____no_output_____" ] ], [ [ "We set two parameters, `batch_size` and `max_features`. `batch_size` is the size of the batch we will train our logistic model on, and `max_features` is the maximum number of tf-idf textual words we will use in our logistic regression.", "_____no_output_____" ] ], [ [ "batch_size = 200\nmax_features = 1000", "_____no_output_____" ] ], [ [ "Check if data was downloaded, otherwise download it and save for future use", "_____no_output_____" ] ], [ [ "save_file_name = 'temp_spam_data.csv'\nif os.path.isfile(save_file_name):\n text_data = []\n with open(save_file_name, 'r') as temp_output_file:\n reader = csv.reader(temp_output_file)\n for row in reader:\n text_data.append(row)\nelse:\n zip_url = 'http://archive.ics.uci.edu/ml/machine-learning-databases/00228/smsspamcollection.zip'\n r = requests.get(zip_url)\n z = ZipFile(io.BytesIO(r.content))\n file = z.read('SMSSpamCollection')\n # Format Data\n text_data = file.decode()\n text_data = text_data.encode('ascii',errors='ignore')\n text_data = text_data.decode().split('\\n')\n text_data = [x.split('\\t') for x in text_data if len(x)>=1]\n \n # And write to csv\n with open(save_file_name, 'w') as temp_output_file:\n writer = csv.writer(temp_output_file)\n writer.writerows(text_data)", "_____no_output_____" ] ], [ [ "We now clean our texts. This will decrease our vocabulary size by converting everything to lower case, removing punctuation and getting rid of numbers.", "_____no_output_____" ] ], [ [ "texts = [x[1] for x in text_data]\ntarget = [x[0] for x in text_data]\n\n# Relabel 'spam' as 1, 'ham' as 0\ntarget = [1. if x=='spam' else 0. for x in target]\n\n# Normalize text\n# Lower case\ntexts = [x.lower() for x in texts]\n\n# Remove punctuation\ntexts = [''.join(c for c in x if c not in string.punctuation) for x in texts]\n\n# Remove numbers\ntexts = [''.join(c for c in x if c not in '0123456789') for x in texts]\n\n# Trim extra whitespace\ntexts = [' '.join(x.split()) for x in texts]", "_____no_output_____" ] ], [ [ "Define tokenizer function and create the TF-IDF vectors with SciKit-Learn.", "_____no_output_____" ] ], [ [ "import nltk\nnltk.download('punkt')", "[nltk_data] Downloading package punkt to /home/jovyan/nltk_data...\n[nltk_data] Unzipping tokenizers/punkt.zip.\n" ], [ "def tokenizer(text):\n words = nltk.word_tokenize(text)\n return words\n\n# Create TF-IDF of texts\ntfidf = TfidfVectorizer(tokenizer=tokenizer, stop_words='english', max_features=max_features)\nsparse_tfidf_texts = tfidf.fit_transform(texts)", "/srv/venv/lib/python3.6/site-packages/sklearn/feature_extraction/text.py:1089: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n if hasattr(X, 'dtype') and np.issubdtype(X.dtype, np.float):\n" ] ], [ [ "Split up data set into train/test.", "_____no_output_____" ] ], [ [ "train_indices = np.random.choice(sparse_tfidf_texts.shape[0], round(0.8*sparse_tfidf_texts.shape[0]), replace=False)\ntest_indices = np.array(list(set(range(sparse_tfidf_texts.shape[0])) - set(train_indices)))\ntexts_train = sparse_tfidf_texts[train_indices]\ntexts_test = sparse_tfidf_texts[test_indices]\ntarget_train = np.array([x for ix, x in enumerate(target) if ix in train_indices])\ntarget_test = np.array([x for ix, x in enumerate(target) if ix in test_indices])", "_____no_output_____" ] ], [ [ "Now we create the variables and placeholders necessary for logistic regression. After which, we declare our logistic regression operation. Remember that the sigmoid part of the logistic regression will be in the loss function.", "_____no_output_____" ] ], [ [ "# Create variables for logistic regression\nA = tf.Variable(tf.random_normal(shape=[max_features,1]))\nb = tf.Variable(tf.random_normal(shape=[1,1]))\n\n# Initialize placeholders\nx_data = tf.placeholder(shape=[None, max_features], dtype=tf.float32)\ny_target = tf.placeholder(shape=[None, 1], dtype=tf.float32)\n\n# Declare logistic model (sigmoid in loss function)\nmodel_output = tf.add(tf.matmul(x_data, A), b)", "_____no_output_____" ] ], [ [ "Next, we declare the loss function (which has the sigmoid in it), and the prediction function. The prediction function will have to have a sigmoid inside of it because it is not in the model output.", "_____no_output_____" ] ], [ [ "# Declare loss function (Cross Entropy loss)\nloss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=model_output, labels=y_target))\n\n# Prediction\nprediction = tf.round(tf.sigmoid(model_output))\npredictions_correct = tf.cast(tf.equal(prediction, y_target), tf.float32)\naccuracy = tf.reduce_mean(predictions_correct)", "_____no_output_____" ] ], [ [ "Now we create the optimization function and initialize the model variables.", "_____no_output_____" ] ], [ [ "# Declare optimizer\nmy_opt = tf.train.GradientDescentOptimizer(0.0025)\ntrain_step = my_opt.minimize(loss)\n\n# Intitialize Variables\ninit = tf.global_variables_initializer()\nsess.run(init)", "_____no_output_____" ] ], [ [ "Finally, we perform our logisitic regression on the 1000 TF-IDF features.", "_____no_output_____" ] ], [ [ "train_loss = []\ntest_loss = []\ntrain_acc = []\ntest_acc = []\ni_data = []\nfor i in range(10000):\n rand_index = np.random.choice(texts_train.shape[0], size=batch_size)\n rand_x = texts_train[rand_index].todense()\n rand_y = np.transpose([target_train[rand_index]])\n sess.run(train_step, feed_dict={x_data: rand_x, y_target: rand_y})\n \n # Only record loss and accuracy every 100 generations\n if (i+1)%100==0:\n i_data.append(i+1)\n train_loss_temp = sess.run(loss, feed_dict={x_data: rand_x, y_target: rand_y})\n train_loss.append(train_loss_temp)\n \n test_loss_temp = sess.run(loss, feed_dict={x_data: texts_test.todense(), y_target: np.transpose([target_test])})\n test_loss.append(test_loss_temp)\n \n train_acc_temp = sess.run(accuracy, feed_dict={x_data: rand_x, y_target: rand_y})\n train_acc.append(train_acc_temp)\n \n test_acc_temp = sess.run(accuracy, feed_dict={x_data: texts_test.todense(), y_target: np.transpose([target_test])})\n test_acc.append(test_acc_temp)\n if (i+1)%500==0:\n acc_and_loss = [i+1, train_loss_temp, test_loss_temp, train_acc_temp, test_acc_temp]\n acc_and_loss = [np.round(x,2) for x in acc_and_loss]\n print('Generation # {}. Train Loss (Test Loss): {:.2f} ({:.2f}). Train Acc (Test Acc): {:.2f} ({:.2f})'.format(*acc_and_loss))", "Generation # 500. Train Loss (Test Loss): 0.92 (0.93). Train Acc (Test Acc): 0.39 (0.40)\nGeneration # 1000. Train Loss (Test Loss): 0.71 (0.74). Train Acc (Test Acc): 0.56 (0.56)\nGeneration # 1500. Train Loss (Test Loss): 0.58 (0.62). Train Acc (Test Acc): 0.66 (0.66)\nGeneration # 2000. Train Loss (Test Loss): 0.59 (0.56). Train Acc (Test Acc): 0.67 (0.74)\nGeneration # 2500. Train Loss (Test Loss): 0.58 (0.52). Train Acc (Test Acc): 0.74 (0.77)\nGeneration # 3000. Train Loss (Test Loss): 0.55 (0.49). Train Acc (Test Acc): 0.76 (0.79)\nGeneration # 3500. Train Loss (Test Loss): 0.47 (0.47). Train Acc (Test Acc): 0.80 (0.81)\nGeneration # 4000. Train Loss (Test Loss): 0.47 (0.46). Train Acc (Test Acc): 0.81 (0.83)\nGeneration # 4500. Train Loss (Test Loss): 0.44 (0.45). Train Acc (Test Acc): 0.84 (0.83)\nGeneration # 5000. Train Loss (Test Loss): 0.47 (0.45). Train Acc (Test Acc): 0.82 (0.84)\nGeneration # 5500. Train Loss (Test Loss): 0.46 (0.44). Train Acc (Test Acc): 0.84 (0.84)\nGeneration # 6000. Train Loss (Test Loss): 0.47 (0.44). Train Acc (Test Acc): 0.82 (0.85)\nGeneration # 6500. Train Loss (Test Loss): 0.46 (0.44). Train Acc (Test Acc): 0.84 (0.85)\nGeneration # 7000. Train Loss (Test Loss): 0.45 (0.44). Train Acc (Test Acc): 0.86 (0.85)\nGeneration # 7500. Train Loss (Test Loss): 0.48 (0.44). Train Acc (Test Acc): 0.84 (0.85)\nGeneration # 8000. Train Loss (Test Loss): 0.37 (0.44). Train Acc (Test Acc): 0.88 (0.85)\nGeneration # 8500. Train Loss (Test Loss): 0.42 (0.44). Train Acc (Test Acc): 0.88 (0.85)\nGeneration # 9000. Train Loss (Test Loss): 0.38 (0.44). Train Acc (Test Acc): 0.89 (0.85)\nGeneration # 9500. Train Loss (Test Loss): 0.49 (0.44). Train Acc (Test Acc): 0.81 (0.85)\nGeneration # 10000. Train Loss (Test Loss): 0.50 (0.44). Train Acc (Test Acc): 0.84 (0.85)\n" ] ], [ [ "Here is matplotlib code to plot the loss and accuracies.", "_____no_output_____" ] ], [ [ "# Plot loss over time\nplt.plot(i_data, train_loss, 'k-', label='Train Loss')\nplt.plot(i_data, test_loss, 'r--', label='Test Loss', linewidth=4)\nplt.title('Cross Entropy Loss per Generation')\nplt.xlabel('Generation')\nplt.ylabel('Cross Entropy Loss')\nplt.legend(loc='upper right')\nplt.show()\n\n# Plot train and test accuracy\nplt.plot(i_data, train_acc, 'k-', label='Train Set Accuracy')\nplt.plot(i_data, test_acc, 'r--', label='Test Set Accuracy', linewidth=4)\nplt.title('Train and Test Accuracy')\nplt.xlabel('Generation')\nplt.ylabel('Accuracy')\nplt.legend(loc='lower right')\nplt.show()", "_____no_output_____" ], [ "test complete; Gopal", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Table of Contents\n <p><div class=\"lev1 toc-item\"><a href=\"#Texte-d'oral-de-modélisation---Agrégation-Option-Informatique\" data-toc-modified-id=\"Texte-d'oral-de-modélisation---Agrégation-Option-Informatique-1\"><span class=\"toc-item-num\">1&nbsp;&nbsp;</span>Texte d'oral de modélisation - Agrégation Option Informatique</a></div><div class=\"lev2 toc-item\"><a href=\"#Préparation-à-l'agrégation---ENS-de-Rennes,-2016-17\" data-toc-modified-id=\"Préparation-à-l'agrégation---ENS-de-Rennes,-2016-17-11\"><span class=\"toc-item-num\">1.1&nbsp;&nbsp;</span>Préparation à l'agrégation - ENS de Rennes, 2016-17</a></div><div class=\"lev2 toc-item\"><a href=\"#À-propos-de-ce-document\" data-toc-modified-id=\"À-propos-de-ce-document-12\"><span class=\"toc-item-num\">1.2&nbsp;&nbsp;</span>À propos de ce document</a></div><div class=\"lev2 toc-item\"><a href=\"#Implémentation\" data-toc-modified-id=\"Implémentation-13\"><span class=\"toc-item-num\">1.3&nbsp;&nbsp;</span>Implémentation</a></div><div class=\"lev3 toc-item\"><a href=\"#Une-bonne-structure-de-donnée-pour-des-intervalles-et-des-graphes-d'intervales\" data-toc-modified-id=\"Une-bonne-structure-de-donnée-pour-des-intervalles-et-des-graphes-d'intervales-131\"><span class=\"toc-item-num\">1.3.1&nbsp;&nbsp;</span>Une bonne structure de donnée pour des intervalles et des graphes d'intervales</a></div><div class=\"lev3 toc-item\"><a href=\"#Algorithme-de-coloriage-de-graphe-d'intervalles\" data-toc-modified-id=\"Algorithme-de-coloriage-de-graphe-d'intervalles-132\"><span class=\"toc-item-num\">1.3.2&nbsp;&nbsp;</span>Algorithme de coloriage de graphe d'intervalles</a></div><div class=\"lev3 toc-item\"><a href=\"#Algorithme-pour-calculer-le-stable-maximum-d'un-graphe-d'intervalles\" data-toc-modified-id=\"Algorithme-pour-calculer-le-stable-maximum-d'un-graphe-d'intervalles-133\"><span class=\"toc-item-num\">1.3.3&nbsp;&nbsp;</span>Algorithme pour calculer le <em>stable maximum</em> d'un graphe d'intervalles</a></div><div class=\"lev2 toc-item\"><a href=\"#Exemples\" data-toc-modified-id=\"Exemples-14\"><span class=\"toc-item-num\">1.4&nbsp;&nbsp;</span>Exemples</a></div><div class=\"lev3 toc-item\"><a href=\"#Qui-a-tué-le-Duc-de-Densmore-?\" data-toc-modified-id=\"Qui-a-tué-le-Duc-de-Densmore-?-141\"><span class=\"toc-item-num\">1.4.1&nbsp;&nbsp;</span>Qui a tué le Duc de Densmore ?</a></div><div class=\"lev4 toc-item\"><a href=\"#Comment-résoudre-ce-problème-?\" data-toc-modified-id=\"Comment-résoudre-ce-problème-?-1411\"><span class=\"toc-item-num\">1.4.1.1&nbsp;&nbsp;</span>Comment résoudre ce problème ?</a></div><div class=\"lev4 toc-item\"><a href=\"#Solution\" data-toc-modified-id=\"Solution-1412\"><span class=\"toc-item-num\">1.4.1.2&nbsp;&nbsp;</span>Solution</a></div><div class=\"lev3 toc-item\"><a href=\"#Le-problème-des-frigos\" data-toc-modified-id=\"Le-problème-des-frigos-142\"><span class=\"toc-item-num\">1.4.2&nbsp;&nbsp;</span>Le problème des frigos</a></div><div class=\"lev3 toc-item\"><a href=\"#Le-problème-du-CSA\" data-toc-modified-id=\"Le-problème-du-CSA-143\"><span class=\"toc-item-num\">1.4.3&nbsp;&nbsp;</span>Le problème du CSA</a></div><div class=\"lev3 toc-item\"><a href=\"#Le-problème-du-wagon-restaurant\" data-toc-modified-id=\"Le-problème-du-wagon-restaurant-144\"><span class=\"toc-item-num\">1.4.4&nbsp;&nbsp;</span>Le problème du wagon restaurant</a></div><div class=\"lev4 toc-item\"><a href=\"#Solution-via-l'algorithme-de-coloriage-de-graphe-d'intervalles\" data-toc-modified-id=\"Solution-via-l'algorithme-de-coloriage-de-graphe-d'intervalles-1441\"><span class=\"toc-item-num\">1.4.4.1&nbsp;&nbsp;</span>Solution via l'algorithme de coloriage de graphe d'intervalles</a></div><div class=\"lev2 toc-item\"><a href=\"#Bonus-?\" data-toc-modified-id=\"Bonus-?-15\"><span class=\"toc-item-num\">1.5&nbsp;&nbsp;</span>Bonus ?</a></div><div class=\"lev3 toc-item\"><a href=\"#Visualisation-des-graphes-définis-dans-les-exemples\" data-toc-modified-id=\"Visualisation-des-graphes-définis-dans-les-exemples-151\"><span class=\"toc-item-num\">1.5.1&nbsp;&nbsp;</span>Visualisation des graphes définis dans les exemples</a></div><div class=\"lev2 toc-item\"><a href=\"#Conclusion\" data-toc-modified-id=\"Conclusion-16\"><span class=\"toc-item-num\">1.6&nbsp;&nbsp;</span>Conclusion</a></div>", "_____no_output_____" ], [ "# Texte d'oral de modélisation - Agrégation Option Informatique\n## Préparation à l'agrégation - ENS de Rennes, 2016-17\n- *Date* : 3 avril 2017\n- *Auteur* : [Lilian Besson](https://GitHub.com/Naereen/notebooks/)\n- *Texte*: Annale 2006, \"Crime Parfait\"", "_____no_output_____" ], [ "## À propos de ce document\n- Ceci est une *proposition* de correction, partielle et probablement non-optimale, pour la partie implémentation d'un [texte d'annale de l'agrégation de mathématiques, option informatique](http://Agreg.org/Textes/).\n- Ce document est un [notebook Jupyter](https://www.Jupyter.org/), et [est open-source sous Licence MIT sur GitHub](https://github.com/Naereen/notebooks/tree/master/agreg/), comme les autres solutions de textes de modélisation que [j](https://GitHub.com/Naereen)'ai écrite cette année.\n- L'implémentation sera faite en OCaml, version 4+ :", "_____no_output_____" ] ], [ [ "Sys.command \"ocaml -version\";;", "The OCaml toplevel, version 4.04.2\n" ] ], [ [ "----\n## Implémentation\nLa question d'implémentation était la question 2) en page 7.\n\n> « Proposer une structure de donnée adaptée pour représenter un graphe d'intervalles dont une représentation sous forme de famille d’intervalles est connue.\n> Implémenter de manière efficace l’algorithme de coloriage de graphes d'intervalles et illustrer cet algorithme sur une application bien choisie citée dans le texte. »\n\nNous allons donc d'abord définir une structure de donnée pour une famille d'intervalles ainsi que pour un graphe d'intervalle, ainsi qu'une fonction convertissant l'un en l'autre.\n\nCela permettra de facilement définr les différents exemples du texte, et de les résoudre.", "_____no_output_____" ], [ "### Une bonne structure de donnée pour des intervalles et des graphes d'intervales\n\n- Pour des **intervalles** à valeurs réelles, on se restreint par convénience à des valeurs entières.", "_____no_output_____" ] ], [ [ "type intervalle = (int * int);;\ntype intervalles = intervalle list;;", "_____no_output_____" ] ], [ [ "- Pour des **graphes d'intervalles**, on utilise une simple représentation sous forme de liste d'adjacence, plus facile à mettre en place en OCaml qu'une représentation sous forme de matrice. Ici, tous nos graphes ont pour sommets $0 \\dots n - 1$.", "_____no_output_____" ] ], [ [ "type sommet = int;;\ntype voisins = sommet list;;\ntype graphe_intervalle = voisins list;;", "_____no_output_____" ] ], [ [ "> *Note:* j'ai préféré garder une structure très simple, pour les intervalles, les graphes d'intervalles et les coloriages, mais on perd un peu en lisibilité dans la fonction coloriage.\n> \n> Implicitement, dès qu'une liste d'intervalles est fixée, de taille $n$, ils sont numérotés de $0$ à $n-1$. Le graphe `g` aura pour sommet $0 \\dots n-1$, et le coloriage sera un simple tableau de couleurs `c` (i.e., d'entiers), donnant en `c[i]` la couleur de l'intervalle numéro `i`.\n>\n> Une solution plus intelligente aurait été d'utiliser des tables d'association, cf. le module [Map](http://caml.inria.fr/pub/docs/manual-ocaml/libref/Map.html) de OCaml, et le code proposé par Julien durant son oral.", "_____no_output_____" ], [ "- On peut rapidement écrire une fonction qui va convertir une liste d'intervalle (`intervalles`) en un graphe d'intervalle. On crée les sommets du graphes, via `index_intvls` qui associe un intervalle à son indice, et ensuite on ajoute les arêtes au graphe selon les contraintes définissant un graphe d'intervalle :\n\n $$ \\forall I, I' \\in V, (I,I') \\in E \\Leftrightarrow I \\neq I' \\;\\text{and}\\; I \\cap I' \\neq \\emptyset $$\n \n Donc avec des intervales $I = [x,y]$ et $I' = [a,b]$, cela donne :\n\n $$ \\forall I = [x,y], I' = [a,b] \\in V, (I,I') \\in E \\Leftrightarrow (x,y) \\neq (a,b) \\;\\text{and}\\; \\neg (b < x \\;\\text{or}\\; y < a) $$", "_____no_output_____" ] ], [ [ "let graphe_depuis_intervalles (intvls : intervalles) : graphe_intervalle =\n let n = List.length intvls in (* Nomber de sommet *)\n let array_intvls = Array.of_list intvls in (* Tableau des intervalles *)\n let index_intvls = Array.to_list (\n Array.init n (fun i -> (\n array_intvls.(i), i) (* Associe un intervalle à son indice *)\n )\n ) in\n let gr = List.map (fun (a, b) -> (* Pour chaque intervalle [a, b] *)\n List.filter (fun (x, y) -> (* On ajoute [x, y] s'il intersecte [a, b] *)\n (x, y) <> (a, b) (* Intervalle différent *)\n && not ( (b < x) || (y < a) ) (* pas x---y a---b ni a---b x---y *)\n ) intvls\n ) intvls in\n (* On transforme la liste de liste d'intervalles en une liste de liste d'entiers *)\n List.map (fun voisins ->\n List.map (fun sommet -> (* Grace au tableau index_intvls *)\n List.assoc sommet index_intvls\n ) voisins\n ) gr\n;;", "_____no_output_____" ] ], [ [ "### Algorithme de coloriage de graphe d'intervalles\n\n> Étant donné un graphe $G = (V, E)$, on cherche un entier $n$ minimal et une fonction $c : V \\to \\{1, \\cdots, n\\}$ telle que si $(v_1 , v_2) \\in E$, alors $c(v_1) \\neq c(v_2)$.\n\nOn suit les indications de l'énoncé pour implémenter facilement cet algorithme.\n\n> Une *heuristique* simple pour résoudre ce problème consiste à appliquer l’algorithme glouton suivant :\n> - tant qu'il reste reste des sommets non coloriés,\n> + en choisir un\n> + et le colorier avec le plus petit entier qui n’apparait pas dans les voisins déjà coloriés.\n\n\n> En choisissant bien le nouveau sommet à colorier à chaque fois, cette heuristique se révelle optimale pour les graphes d’intervalles.\n\nOn peut d'abord définir un type de donnée pour un coloriage, sous la forme d'une liste de couple d'intervalle et de couleur.\nAinsi, `List.assoc` peut être utilisée pour donner le coloriage de chaque intervalle.", "_____no_output_____" ] ], [ [ "type couleur = int;;\ntype coloriage = (intervalle * couleur) list;;\n\nlet coloriage_depuis_couleurs (intvl : intervalles) (c : couleur array) : coloriage =\n Array.to_list (Array.init (Array.length c) (fun i -> (List.nth intvl i), c.(i)));;\n\nlet quelle_couleur (intvl : intervalle) (colors : coloriage) = \n List.assoc intvl colors\n;;", "_____no_output_____" ] ], [ [ "Ensuite, l'ordre partiel $\\prec_i$ sur les intervalles est défini comme ça :\n\n$$ I = (a,b) \\prec_i J=(x, y) \\Longleftrightarrow a < x $$", "_____no_output_____" ] ], [ [ "let ordre_partiel ((a, _) : intervalle) ((x, _) : intervalle) =\n a < x\n;;", "_____no_output_____" ] ], [ [ "On a ensuite besoin d'une fonction qui va calculer l'inf de $\\mathbb{N} \\setminus \\{x : x \\in \\mathrm{valeurs} \\}$:", "_____no_output_____" ] ], [ [ "let inf_N_minus valeurs =\n let res = ref 0 in (* Très important d'utiliser une référence ! *)\n while List.mem !res valeurs do\n incr res;\n done;\n !res\n;;", "_____no_output_____" ] ], [ [ "On vérifie rapidement sur deux exemples :", "_____no_output_____" ] ], [ [ "inf_N_minus [0; 1; 3];; (* 2 *)\ninf_N_minus [0; 1; 2; 3; 4; 5; 6; 10];; (* 7 *)", "_____no_output_____" ] ], [ [ "Enfin, on a besoin d'une fonction pour trouver l'intervalle $I \\in V$, minimal pour $\\prec_i$, tel que $c(I) = +\\infty$.", "_____no_output_____" ] ], [ [ "let trouve_min_interval intvl (c : coloriage) (inf : couleur) =\n let colorie inter = quelle_couleur inter c in\n (* D'abord on extraie {I : c(I) = +oo} *)\n let intvl2 = List.filter (fun i -> (colorie i) = inf) intvl in\n (* Puis on parcourt la liste et on garde le plus petit pour l'ordre *)\n let i0 = ref 0 in\n for j = 1 to (List.length intvl2) - 1 do\n if ordre_partiel (List.nth intvl2 j) (List.nth intvl2 !i0) then\n i0 := j;\n done;\n List.nth intvl2 !i0;\n;;", "_____no_output_____" ] ], [ [ "Et donc tout cela permet de finir l'algorithme, tel que décrit dans le texte :\n\n<img style=\"width:65%;\" alt=\"images/algorithme_coloriage.png\" src=\"images/algorithme_coloriage.png\">", "_____no_output_____" ] ], [ [ "let coloriage_intervalles (intvl : intervalles) : coloriage =\n let n = List.length intvl in (* Nombre d'intervalles *)\n let array_intvls = Array.of_list intvl in (* Tableau des intervalles *)\n let index_intvls = Array.to_list (\n Array.init n (fun i -> (\n array_intvls.(i), i) (* Associe un intervalle à son indice *)\n )\n ) in\n let gr = graphe_depuis_intervalles intvl in\n let inf = n + 10000 in (* Grande valeur, pour +oo *)\n let c = Array.make n inf in (* Liste des couleurs, c(I) = +oo pour tout I *)\n let maxarray = Array.fold_left max (-inf - 10000) in (* Initialisé à -oo *)\n while maxarray c = inf do (* Il reste un I in V tel que c(I) = +oo *)\n begin (* C'est la partie pas élégante *)\n (* On récupère le coloriage depuis la liste de couleurs actuelle *)\n let coloriage = (coloriage_depuis_couleurs intvl c) in\n (* Puis la fonction [colorie] pour associer une couleur à un intervalle *)\n let colorie inter = quelle_couleur inter coloriage in\n (* On choisit un I, minimal pour ordre_partiel, tel que c(I) = +oo *)\n let inter = trouve_min_interval intvl coloriage inf in\n (* On trouve son indice *)\n let i = List.assoc inter index_intvls in\n (* On trouve les voisins de i dans le graphe *)\n let adj_de_i = List.nth gr i in\n (* Puis les voisins de I en tant qu'intervalles *)\n let adj_de_I = List.map (fun j -> List.nth intvl j) adj_de_i in\n (* Puis on récupère leurs couleurs *)\n let valeurs = List.map colorie adj_de_I in\n (* c(I) = inf(N - {c(J) : J adjacent a I} ) *)\n c.(i) <- inf_N_minus valeurs;\n end;\n done;\n coloriage_depuis_couleurs intvl c;\n;;", "_____no_output_____" ] ], [ [ "Une fois qu'on a un coloriage, à valeurs dans $0,\\dots,k$ on récupère le nombre de couleurs comme $1 + \\max c$, i.e., $k+1$.", "_____no_output_____" ] ], [ [ "let max_valeurs = List.fold_left max 0;;", "_____no_output_____" ], [ "let nombre_chromatique (colorg : coloriage) : int =\n 1 + max_valeurs (List.map snd colorg)\n;;", "_____no_output_____" ] ], [ [ "### Algorithme pour calculer le *stable maximum* d'un graphe d'intervalles\nOn répond ici à la question 7.\n\n> « Proposer un algorithme efficace pour construire un stable maximum (i.e., un ensemble de sommets indépendants) d'un graphe d’intervalles dont on connaı̂t une représentation sous forme d'intervalles.\n> On pourra chercher à quelle condition l'intervalle dont l'extrémité droite est la plus à gauche appartient à un stable maximum. »", "_____no_output_____" ], [ "**FIXME, je ne l'ai pas encore fait.**", "_____no_output_____" ], [ "----\n## Exemples\nOn traite ici l'exemple introductif, ainsi que les trois autres exemples proposés.", "_____no_output_____" ], [ "### Qui a tué le Duc de Densmore ?\n\n> On ne rappelle pas le problème, mais les données :\n\n> - Ann dit avoir vu Betty, Cynthia, Emily, Felicia et Georgia.\n- Betty dit avoir vu Ann, Cynthia et Helen.\n- Cynthia dit avoir vu Ann, Betty, Diana, Emily et Helen.\n- Diana dit avoir vu Cynthia et Emily.\n- Emily dit avoir vu Ann, Cynthia, Diana et Felicia.\n- Felicia dit avoir vu Ann et Emily.\n- Georgia dit avoir vu Ann et Helen.\n- Helen dit avoir vu Betty, Cynthia et Georgia.\n\nTranscrit sous forme de graphe, cela donne :", "_____no_output_____" ] ], [ [ "(* On définit des entiers, c'est plus simple *)\nlet ann = 0\nand betty = 1\nand cynthia = 2\nand diana = 3\nand emily = 4\nand felicia = 5\nand georgia = 6\nand helen = 7;;\n\nlet graphe_densmore = [\n [betty; cynthia; emily; felicia; georgia]; (* Ann *)\n [ann; cynthia; helen]; (* Betty *)\n [ann; betty; diana; emily; helen]; (* Cynthia *)\n [cynthia; emily]; (* Diana *)\n [ann; cynthia; diana; felicia]; (* Emily *)\n [ann; emily]; (* Felicia *)\n [ann; helen]; (* Georgia *)\n [betty; cynthia; georgia] (* Helen *)\n];;", "_____no_output_____" ] ], [ [ "\n> Figure 1. Graphe d'intervalle pour le problème de l'assassinat du duc de Densmore.", "_____no_output_____" ], [ "Avec les prénoms plutôt que des numéros, cela donne :", "_____no_output_____" ], [ "\n> Figure 2. Graphe d'intervalle pour le problème de l'assassinat du duc de Densmore.", "_____no_output_____" ], [ "#### Comment résoudre ce problème ?\n> Il faut utiliser la caractérisation du théorème 2 du texte, et la définition des graphes parfaits.\n\n- Définition + Théorème 2 (point 1) :\n\nOn sait qu'un graphe d'intervalle est parfait, et donc tous ses graphes induits le sont aussi.\nLa caractérisation via les cordes sur les cycles de taille $\\geq 4$ permet de dire qu'un quadrilatère (cycle de taille $4$) n'est pas un graphe d'intervalle.\nDonc un graphe qui contient un graphe induit étant un quadrilatère ne peut être un graphe d'intervalle.\n\nAinsi, sur cet exemple, comme on a deux quadrilatères $A B H G$ et $A G H C$, on en déduit que $A$, $G$, ou $H$ ont menti.\n\n- Théorème 2 (point 2) :\n\nEnsuite, si on enlève $G$ ou $H$, le graphe ne devient pas un graphe d'intervalle, par les considérations suivantes, parce que son complémentaire n'est pas un graphe de comparaison.\n\nEn effet, par exemple si on enlève $G$, $A$ et $H$ et $D$ forment une clique dans le complémentaire $\\overline{G}$ de $G$, et l'irréflexivité d'une éventuelle relation $R$ rend cela impossible. Pareil si on enlève $H$, avec $G$ et $B$ et $D$ qui formet une clique dans $\\overline{G}$.\n\nPar contre, si on enlève $A$, le graphe devient triangulé (et de comparaison, mais c'est plus dur à voir !).\n\nDonc seule $A$ reste comme potentielle menteuse.", "_____no_output_____" ], [ "> « Mais... Ça semble difficile de programmer une résolution automatique de ce problème ? »\n\nEn fait, il suffit d'écrire une fonction de vérification qu'un graphe est un graphe d'intervalle, puis on essaie d'enlever chaque sommet, tant que le graphe n'est pas un graphe d'intervalle.\n\nSi le graphe devient valide en enlevant un seul sommet, et qu'il n'y en a qu'un seul qui fonctionne, alors il y a un(e) seul(e) menteur(se) dans le graphe, et donc un(e) seul(e) coupable !", "_____no_output_____" ], [ "#### Solution\n\nC'est donc $A$, i.e., Ann l'unique menteuse et donc la coupable.\n\n> Ce n'est pas grave de ne pas avoir réussi à répondre durant l'oral !\n> Au contraire, vous avez le droit de vous détacher du problème initial du texte !", "_____no_output_____" ], [ "> Une solution bien expliquée peut être trouvée dans [cette vidéo](https://youtu.be/ZGhSyVvOelg) :", "_____no_output_____" ], [ "<iframe width=\"640\" height=\"360\" src=\"https://www.youtube.com/embed/ZGhSyVvOelg\" frameborder=\"1\" allowfullscreen></iframe>", "_____no_output_____" ], [ "### Le problème des frigos\n> Dans un grand hopital, les réductions de financement public poussent le gestionnaire du service d'immunologie à faire des économies sur le nombre de frigos à acheter pour stocker les vaccins. A peu de chose près, il lui faut stocker les vaccins suivants :\n\n> | Numéro | Nom du vaccin | Température de conservation\n| :-----: | :------------ | -------------------------: |\n| 0 | Rougeole-Rubéole-Oreillons (RRO) | $4 \\cdots 12$ °C\n| 1 | BCG | $8 \\cdots 15$ °C\n| 2 | Di-Te-Per | $0 \\cdots 20$ °C\n| 3 | Anti-polio | $2 \\cdots 3$ °C\n| 4 | Anti-hépatite B | $-3 \\cdots 6$ °C\n| 5 | Anti-amarile | $-10 \\cdots 10$ °C\n| 6 | Variole | $6 \\cdots 20$ °C\n| 7 | Varicelle | $-5 \\cdots 2$ °C\n| 8 | Antihaemophilus | $-2 \\cdots 8$ °C\n\n> Combien le gestionaire doit-il acheter de frigos, et sur quelles températures doit-il les régler ?", "_____no_output_____" ] ], [ [ "let vaccins : intervalles = [\n (4, 12);\n (8, 15);\n (0, 20);\n (2, 3);\n (-3, 6);\n (-10, 10);\n (6, 20);\n (-5, 2);\n (-2, 8)\n]", "_____no_output_____" ] ], [ [ "Qu'on peut visualiser sous forme de graphe facilement :", "_____no_output_____" ] ], [ [ "let graphe_vaccins = graphe_depuis_intervalles vaccins;;", "_____no_output_____" ] ], [ [ "\n> Figure 3. Graphe d'intervalle pour le problème des frigos et des vaccins.", "_____no_output_____" ], [ "Avec des intervalles au lieu de numéro :", "_____no_output_____" ], [ "\n> Figure 4. Graphe d'intervalle pour le problème des frigos et des vaccins.", "_____no_output_____" ], [ "On peut récupérer une coloriage minimal pour ce graphe :", "_____no_output_____" ] ], [ [ "coloriage_intervalles vaccins;;", "_____no_output_____" ] ], [ [ "La couleur la plus grande est `5`, donc le nombre chromatique de ce graphe est `6`.", "_____no_output_____" ] ], [ [ "nombre_chromatique (coloriage_intervalles vaccins);;", "_____no_output_____" ] ], [ [ "Par contre, la solution au problème des frigos et des vaccins réside dans le nombre de couverture de cliques, $k(G)$, pas dans le nombre chromatique $\\chi(G)$.\n\nOn peut le résoudre en répondant à la question 7, qui demandait de mettre au point un algorithme pour construire un *stable maximum* pour un graphe d'intervalle.", "_____no_output_____" ], [ "### Le problème du CSA\n\n> Le Conseil Supérieur de l’Audiovisuel doit attribuer de nouvelles bandes de fréquences d’émission pour la stéréophonie numérique sous-terraine (SNS).\n> Cette technologie de pointe étant encore à l'état expérimental, les appareils capables d'émettre ne peuvent utiliser que les bandes de fréquences FM suivantes :\n\n> | Bandes de fréquence | Intervalle (kHz) |\n| :-----------------: | ---------: |\n| 0 | $32 \\cdots 36$ |\n| 1 | $24 \\cdots 30$ |\n| 2 | $28 \\cdots 33$ |\n| 3 | $22 \\cdots 26$ |\n| 4 | $20 \\cdots 25$ |\n| 5 | $30 \\cdots 33$ |\n| 6 | $31 \\cdots 34$ |\n| 7 | $27 \\cdots 31$ |\n\n> Quelles bandes de fréquences doit-on retenir pour permettre à le plus d'appareils possibles d'être utilisés, sachant que deux appareils dont les bandes de fréquences s'intersectent pleinement (pas juste sur les extrémités) sont incompatibles.", "_____no_output_____" ] ], [ [ "let csa : intervalles = [\n (32, 36);\n (24, 30);\n (28, 33);\n (22, 26);\n (20, 25);\n (30, 33);\n (31, 34);\n (27, 31)\n];;", "_____no_output_____" ], [ "let graphe_csa = graphe_depuis_intervalles csa;;", "_____no_output_____" ] ], [ [ "\n> Figure 5. Graphe d'intervalle pour le problème du CSA.", "_____no_output_____" ], [ "Avec des intervalles au lieu de numéro :", "_____no_output_____" ], [ "\n> Figure 6. Graphe d'intervalle pour le problème du CSA.", "_____no_output_____" ], [ "On peut récupérer une coloriage minimal pour ce graphe :", "_____no_output_____" ] ], [ [ "coloriage_intervalles csa;;", "_____no_output_____" ] ], [ [ "La couleur la plus grande est `3`, donc le nombre chromatique de ce graphe est `4`.", "_____no_output_____" ] ], [ [ "nombre_chromatique (coloriage_intervalles csa);;", "_____no_output_____" ] ], [ [ "Par contre, la solution au problème CSA réside dans le nombre de couverture de cliques, $k(G)$, pas dans le nombre chromatique $\\chi(G)$.\n\nOn peut le résoudre en répondant à la question 7, qui demandait de mettre au point un algorithme pour construire un *stable maximum* pour un graphe d'intervalle.", "_____no_output_____" ], [ "### Le problème du wagon restaurant\n\n> Le chef de train de l'Orient Express doit aménager le wagon restaurant avant le départ du train. Ce wagon est assez petit et doit être le moins encombré de tables possibles, mais il faut prévoir suffisemment de tables pour accueillir toutes personnes qui ont réservé :\n\n> | Numéro | Personnage(s) | Heures de dîner | En secondes |\n| :----------------- | --------- | :---------: | :---------: |\n| 0 | Le baron et la baronne Von Haussplatz | 19h30 .. 20h14 | $1170 \\cdots 1214$\n| 1 | Le général Cook | 20h30 .. 21h59 | $1230 \\cdots 1319$\n| 2 | Les époux Steinberg | 19h .. 19h59 | $1140 \\cdots 1199$\n| 3 | La duchesse de Colombart | 20h15 .. 20h59 | $1215 \\cdots 1259$\n| 4 | Le marquis de Carquamba | 21h .. 21h59 | $1260 \\cdots 1319$\n| 5 | La Vociafiore | 19h15 .. 20h29 | $1155 \\cdots 1229$\n| 6 | Le colonel Ferdinand | 20h .. 20h59 | $1200 \\cdots 1259$\n\n\n> Combien de tables le chef de train doit-il prévoir ?", "_____no_output_____" ] ], [ [ "let restaurant = [\n (1170, 1214);\n (1230, 1319);\n (1140, 1199);\n (1215, 1259);\n (1260, 1319);\n (1155, 1229);\n (1200, 1259)\n];;", "_____no_output_____" ], [ "let graphe_restaurant = graphe_depuis_intervalles restaurant;;", "_____no_output_____" ] ], [ [ "\n> Figure 7. Graphe d'intervalle pour le problème du wagon restaurant.", "_____no_output_____" ], [ "Avec des intervalles au lieu de numéro :", "_____no_output_____" ], [ "\n> Figure 8. Graphe d'intervalle pour le problème du wagon restaurant.", "_____no_output_____" ] ], [ [ "coloriage_intervalles restaurant;;", "_____no_output_____" ] ], [ [ "La couleur la plus grande est `2`, donc le nombre chromatique de ce graphe est `3`.", "_____no_output_____" ] ], [ [ "nombre_chromatique (coloriage_intervalles restaurant);;", "_____no_output_____" ] ], [ [ "#### Solution via l'algorithme de coloriage de graphe d'intervalles\nPour ce problème là, la solution est effectivement donnée par le nombre chromatique.\n\nLa couleur sera le numéro de table pour chaque passagers (ou couple de passagers), et donc le nombre minimal de table à installer dans le wagon restaurant est exactement le nombre chromatique.\n\nUne solution peut être la suivante, avec **3 tables** :\n\n| Numéro | Personnage(s) | Heures de dîner | Numéro de table |\n| :----------------- | --------- | :---------: | :---------: |\n| 0 | Le baron et la baronne Von Haussplatz | 19h30 .. 20h14 | 2\n| 1 | Le général Cook | 20h30 .. 21h59 | 1\n| 2 | Les époux Steinberg | 19h .. 19h59 | 0\n| 3 | La duchesse de Colombart | 20h15 .. 20h59 | 2\n| 4 | Le marquis de Carquamba | 21h .. 21h59 | 0\n| 5 | La Vociafiore | 19h15 .. 20h29 | 1\n| 6 | Le colonel Ferdinand | 20h .. 20h59 | 0\n\nOn vérifie manuellement que la solution convient.\nChaque passager devra quitter sa tableau à la minute près par contre !", "_____no_output_____" ], [ "On peut afficher la solution avec un graphe colorié.\nLa table `0` sera <span style=\"color:red;\">rouge</span>, `1` sera <span style=\"color:blue;\">bleu</span> et `2` sera <span style=\"color:yellow;\">jaune</span> :", "_____no_output_____" ], [ "\n> Figure 9. Solution pour le problème du wagon restaurant.", "_____no_output_____" ], [ "----\n## Bonus ?", "_____no_output_____" ], [ "### Visualisation des graphes définis dans les exemples\n\n- J'utilise une petite fonction facile à écrire, qui convertit un graphe (`int list list`) en une chaîne de caractère au format [DOT Graph](http://www.graphviz.org/doc/info/lang.html).\n- Ensuite, un appel `dot -Tpng ...` en ligne de commande convertit ce graphe en une image, que j'inclus ensuite manuellement.", "_____no_output_____" ] ], [ [ "(** Transforme un [graph] en une chaîne représentant un graphe décrit par le langage DOT,\n voir http://en.wikipedia.org/wiki/DOT_language pour plus de détails sur ce langage.\n\n @param graphname Donne le nom du graphe tel que précisé pour DOT\n @param directed Vrai si le graphe doit être dirigé (c'est le cas ici) faux sinon. Change le style des arêtes ([->] ou [--])\n @param verb Affiche tout dans le terminal.\n @param onetoone Si on veut afficher le graphe en mode carré (échelle 1:1). Parfois bizarre, parfois génial.\n*)\nlet graph_to_dotgraph ?(graphname = \"graphname\") ?(directed = false) ?(verb = false) ?(onetoone = false) (glist : int list list) =\n let res = ref \"\" in\n let log s =\n if verb then print_string s; (* Si [verb] affiche dans le terminal le résultat du graphe. *)\n res := !res ^ s\n in\n log (if directed then \"digraph \" else \"graph \");\n log graphname; log \" {\";\n if onetoone then\n log \"\\n size=\\\"1,1\\\";\";\n let g = Array.of_list (List.map Array.of_list glist) in\n (* On affiche directement les arc, un à un. *)\n for i = 0 to (Array.length g) - 1 do\n for j = 0 to (Array.length g.(i)) - 1 do\n if i < g.(i).(j) then\n log (\"\\n \\\"\"\n ^ (string_of_int i) ^ \"\\\" \"\n ^ (if directed then \"->\" else \"--\")\n ^ \" \\\"\" ^ (string_of_int g.(i).(j)) ^ \"\\\"\"\n );\n done;\n done;\n log \"\\n}\\n// generated by OCaml with the function graphe_to_dotgraph.\";\n!res;;", "_____no_output_____" ], [ "(** Fonction ecrire_sortie : plus pratique que output. *)\nlet ecrire_sortie monoutchanel machaine =\n output monoutchanel machaine 0 (String.length machaine);\n flush monoutchanel;;\n\n(** Fonction ecrire_dans_fichier : pour écrire la chaine dans le fichier à l'adresse renseignée. *)\nlet ecrire_dans_fichier ~chaine ~adresse =\n let mon_out_channel = open_out adresse in\n ecrire_sortie mon_out_channel chaine;\n close_out mon_out_channel;;\n", "_____no_output_____" ], [ "let s_graphe_densmore = graph_to_dotgraph ~graphname:\"densmore\" ~directed:false ~verb:false graphe_densmore;;\nlet s_graphe_vaccins = graph_to_dotgraph ~graphname:\"vaccins\" ~directed:false ~verb:false graphe_vaccins;;\nlet s_graphe_csa = graph_to_dotgraph ~graphname:\"csa\" ~directed:false ~verb:false graphe_csa;;\nlet s_graphe_restaurant = graph_to_dotgraph ~graphname:\"restaurant\" ~directed:false ~verb:false graphe_restaurant;;", "_____no_output_____" ], [ "ecrire_dans_fichier ~chaine:s_graphe_densmore ~adresse:\"/tmp/densmore.dot\" ;;\n(* Sys.command \"fdp -Tpng /tmp/densmore.dot > images/densmore.png\";; *)\n\necrire_dans_fichier ~chaine:s_graphe_vaccins ~adresse:\"/tmp/vaccins.dot\" ;;\n(* Sys.command \"fdp -Tpng /tmp/vaccins.dot > images/vaccins.png\";; *)\n\necrire_dans_fichier ~chaine:s_graphe_csa ~adresse:\"/tmp/csa.dot\" ;;\n(* Sys.command \"fdp -Tpng /tmp/csa.dot > images/csa.png\";; *)\n\necrire_dans_fichier ~chaine:s_graphe_restaurant ~adresse:\"/tmp/restaurant.dot\" ;;\n(* Sys.command \"fdp -Tpng /tmp/restaurant.dot > images/restaurant.png\";; *)", "_____no_output_____" ] ], [ [ "On pourrait étendre cette fonction pour qu'elle prenne les intervalles initiaux, pour afficher des bonnes étiquettes et pas des entiers, et un coloriage pour colorer directement les noeuds, mais ça prend du temps pour pas grand chose.", "_____no_output_____" ], [ "----\n## Conclusion\n\nVoilà pour la question obligatoire de programmation, sur l'algorithme de coloriage.\n\n- on a décomposé le problème en sous-fonctions,\n- on a fait des exemples et *on les garde* dans ce qu'on présente au jury,\n- on a testé la fonction exigée sur de petits exemples et sur un exemple de taille réelle (venant du texte)\n\nEt on a pas essayé de faire *un peu plus*.\nAvec plus de temps, on aurait aussi pu écrire un algorithme pour calculer le stable maximum (ensemble de sommets indépendants de taille maximale).\n\n> Bien-sûr, ce petit notebook ne se prétend pas être une solution optimale, ni exhaustive.", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown", "markdown", "markdown" ], [ "code", "code", "code", "code" ], [ "markdown", "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import torch\nfrom torch.autograd import grad\nimport torch.nn as nn\nfrom numpy import genfromtxt\nimport torch.optim as optim\nimport matplotlib.pyplot as plt\nimport torch.nn.functional as F\nimport math\n\ntuberculosis_data = genfromtxt('tuberculosis.csv', delimiter=',') #in the form of [t, S,L,I,T]\n\ntorch.manual_seed(1234)", "_____no_output_____" ], [ "%%time\n\nPATH = 'tuberculosis' \n\nclass DINN(nn.Module):\n def __init__(self, t, S_data, L_data, I_data, T_data):\n super(DINN, self).__init__()\n self.t = torch.tensor(t, requires_grad=True)\n self.t_float = self.t.float()\n self.t_batch = torch.reshape(self.t_float, (len(self.t),1)) #reshape for batch \n self.S = torch.tensor(S_data)\n self.L = torch.tensor(L_data)\n self.I = torch.tensor(I_data)\n self.T = torch.tensor(T_data) \n self.N = torch.tensor(1001)\n\n self.losses = [] #keep the losses\n self.save = 2 #which file to save to\n\n #learnable parameters\n self.delta_tilda = torch.nn.Parameter(torch.rand(1, requires_grad=True)) #torch.tensor(500)\n self.beta_tilda = torch.nn.Parameter(torch.rand(1, requires_grad=True)) #torch.tensor(13)\n self.c_tilda = torch.nn.Parameter(torch.rand(1, requires_grad=True)) #torch.tensor(1)\n self.mu_tilda = torch.nn.Parameter(torch.rand(1, requires_grad=True)) #torch.tensor(0.143)\n self.k_tilda = torch.nn.Parameter(torch.rand(1, requires_grad=True)) #torch.tensor(0.5)\n self.r_1_tilda = torch.nn.Parameter(torch.rand(1, requires_grad=True)) #torch.tensor(2)\n self.r_2_tilda = torch.nn.Parameter(torch.rand(1, requires_grad=True)) #torch.tensor(1)\n self.beta_prime_tilda = torch.nn.Parameter(torch.rand(1, requires_grad=True)) #torch.tensor(13)\n self.d_tilda = torch.nn.Parameter(torch.rand(1, requires_grad=True)) #torch.tensor(0)\n\n #matrices (x4 for S, L, I, T) for the gradients\n self.m1 = torch.zeros((len(self.t), 4)); self.m1[:, 0] = 1\n self.m2 = torch.zeros((len(self.t), 4)); self.m2[:, 1] = 1\n self.m3 = torch.zeros((len(self.t), 4)); self.m3[:, 2] = 1\n self.m4 = torch.zeros((len(self.t), 4)); self.m4[:, 3] = 1\n\n #values for norm\n self.S_max = max(self.S)\n self.S_min = min(self.S)\n self.L_max = max(self.L)\n self.L_min = min(self.L)\n self.I_max = max(self.I)\n self.I_min = min(self.I)\n self.T_max = max(self.T)\n self.T_min = min(self.T)\n\n #normalize \n self.S_hat = (self.S - self.S_min) / (self.S_max - self.S_min)\n self.L_hat = (self.L - self.L_min) / (self.L_max - self.L_min)\n self.I_hat = (self.I - self.I_min) / (self.I_max - self.I_min)\n self.T_hat = (self.T - self.T_min) / (self.T_max - self.T_min)\n\n #NN\n self.net_tuberculosis = self.Net_tuberculosis()\n self.params = list(self.net_tuberculosis.parameters())\n self.params.extend(list([self.delta_tilda ,self.beta_tilda ,self.c_tilda ,self.mu_tilda ,self.k_tilda ,self.r_1_tilda ,self.r_2_tilda ,self.beta_prime_tilda ,self.d_tilda]))\n\n \n #force parameters to be in a range\n @property\n def delta(self):\n return torch.tanh(self.delta_tilda) * 20 + 500 #self.delta_tilda \n @property\n def beta(self):\n return torch.tanh(self.beta_tilda) * 3 + 12 #self.beta_tilda\n @property\n def c(self):\n return torch.tanh(self.c_tilda) * 2 + 1 #self.c_tilda\n @property\n def mu(self):\n return torch.tanh(self.mu_tilda) * 0.1 + 0.2 #self.mu_tilda\n @property\n def k(self):\n return torch.tanh(self.k_tilda) * 0.5 + 0.5 #self.k_tilda\n @property\n def r_1(self):\n return torch.tanh(self.r_1_tilda) + 2 #self.r_1_tilda\n @property\n def r_2(self):\n return torch.tanh(self.r_2_tilda) * 2 + 1 #self.r_2_tilda\n @property\n def beta_prime(self):\n return torch.tanh(self.beta_prime_tilda) * 3 + 12 #self.beta_prime_tilda\n @property\n def d(self):\n return torch.tanh(self.d_tilda) * 0.4 #self.d_tilda\n\n #nets\n class Net_tuberculosis(nn.Module): # input = [t]\n def __init__(self):\n super(DINN.Net_tuberculosis, self).__init__()\n self.fc1=nn.Linear(1, 20) #takes 100 t's\n self.fc2=nn.Linear(20, 20)\n self.fc3=nn.Linear(20, 20)\n self.fc4=nn.Linear(20, 20)\n self.fc5=nn.Linear(20, 20)\n self.fc6=nn.Linear(20, 20)\n self.fc7=nn.Linear(20, 20)\n self.fc8=nn.Linear(20, 20)\n self.out=nn.Linear(20, 4) #outputs S, L, I, T\n\n def forward(self, t):\n tuberculosis=F.relu(self.fc1(t))\n tuberculosis=F.relu(self.fc2(tuberculosis))\n tuberculosis=F.relu(self.fc3(tuberculosis))\n tuberculosis=F.relu(self.fc4(tuberculosis))\n tuberculosis=F.relu(self.fc5(tuberculosis))\n tuberculosis=F.relu(self.fc6(tuberculosis))\n tuberculosis=F.relu(self.fc7(tuberculosis))\n tuberculosis=F.relu(self.fc8(tuberculosis))\n tuberculosis=self.out(tuberculosis)\n return tuberculosis \n\n def net_f(self, t_batch): \n\n tuberculosis_hat = self.net_tuberculosis(t_batch)\n\n S_hat, L_hat, I_hat, T_hat = tuberculosis_hat[:,0], tuberculosis_hat[:,1], tuberculosis_hat[:,2], tuberculosis_hat[:,3]\n\n #S_hat\n tuberculosis_hat.backward(self.m1, retain_graph=True)\n S_hat_t = self.t.grad.clone()\n self.t.grad.zero_()\n\n #L_hat\n tuberculosis_hat.backward(self.m2, retain_graph=True)\n L_hat_t = self.t.grad.clone()\n self.t.grad.zero_()\n\n #I_hat\n tuberculosis_hat.backward(self.m3, retain_graph=True)\n I_hat_t = self.t.grad.clone()\n self.t.grad.zero_()\n \n #T_hat\n tuberculosis_hat.backward(self.m4, retain_graph=True)\n T_hat_t = self.t.grad.clone()\n self.t.grad.zero_()\n\n #unnormalize\n S = self.S_min + (self.S_max - self.S_min) * S_hat\n L = self.L_min + (self.L_max - self.L_min) * L_hat\n I = self.I_min + (self.I_max - self.I_min) * I_hat\n T = self.T_min + (self.T_max - self.T_min) * T_hat\n \n #equations\n f1_hat = S_hat_t - (self.delta - self.beta * self.c * S * I / self.N - self.mu * S) / (self.S_max - self.S_min) \n f2_hat = L_hat_t - (self.beta * self.c * S * I / self.N - (self.mu + self.k + self.r_1) * L + self.beta_prime * self.c * T * 1/self.N) / (self.L_max - self.L_min) \n f3_hat = I_hat_t - (self.k*L - (self.mu + self.d) * I - self.r_2 * I) / (self.I_max - self.I_min) \n f4_hat = T_hat_t - (self.r_1 * L + self.r_2 * I - self.beta_prime * self.c * T * 1/self.N - self.mu*T) / (self.T_max - self.T_min) \n\n return f1_hat, f2_hat, f3_hat, f4_hat, S_hat, L_hat, I_hat, T_hat\n \n def load(self):\n # Load checkpoint\n try:\n checkpoint = torch.load(PATH + str(self.save)+'.pt') \n print('\\nloading pre-trained model...')\n self.load_state_dict(checkpoint['model'])\n self.optimizer.load_state_dict(checkpoint['optimizer_state_dict'])\n self.scheduler.load_state_dict(checkpoint['scheduler'])\n epoch = checkpoint['epoch']\n loss = checkpoint['loss']\n self.losses = checkpoint['losses']\n print('loaded previous loss: ', loss)\n except RuntimeError :\n print('changed the architecture, ignore')\n pass\n except FileNotFoundError:\n pass\n\n def train(self, n_epochs):\n #try loading\n self.load()\n\n #train\n print('\\nstarting training...\\n')\n \n for epoch in range(n_epochs):\n #lists to hold the output (maintain only the final epoch)\n S_pred_list = []\n L_pred_list = []\n I_pred_list = []\n T_pred_list = []\n\n f1_hat, f2_hat, f3_hat, f4_hat, S_hat_pred, L_hat_pred, I_hat_pred, T_hat_pred = self.net_f(self.t_batch)\n self.optimizer.zero_grad()\n\n S_pred_list.append(self.S_min + (self.S_max - self.S_min) * S_hat_pred)\n L_pred_list.append(self.L_min + (self.L_max - self.L_min) * L_hat_pred)\n I_pred_list.append(self.I_min + (self.I_max - self.I_min) * I_hat_pred)\n T_pred_list.append(self.T_min + (self.T_max - self.T_min) * T_hat_pred)\n\n loss = (\n torch.mean(torch.square(self.S_hat - S_hat_pred)) + torch.mean(torch.square(self.L_hat - L_hat_pred)) + \n torch.mean(torch.square(self.I_hat - I_hat_pred)) + torch.mean(torch.square(self.T_hat - T_hat_pred))+\n torch.mean(torch.square(f1_hat)) + torch.mean(torch.square(f2_hat)) +\n torch.mean(torch.square(f3_hat)) + torch.mean(torch.square(f4_hat))\n )\n\n loss.backward()\n\n self.optimizer.step()\n self.scheduler.step() \n # self.scheduler.step(loss) \n\n self.losses.append(loss.item())\n\n if epoch % 1000 == 0: \n print('\\nEpoch ', epoch)\n\n #loss + model parameters update\n if epoch % 4000 == 9999:\n #checkpoint save\n print('\\nSaving model... Loss is: ', loss)\n torch.save({\n 'epoch': epoch,\n 'model': self.state_dict(),\n 'optimizer_state_dict': self.optimizer.state_dict(),\n 'scheduler': self.scheduler.state_dict(),\n 'loss': loss,\n 'losses': self.losses,\n }, PATH + str(self.save)+'.pt')\n if self.save % 2 > 0: #its on 3\n self.save = 2 #change to 2\n else: #its on 2\n self.save = 3 #change to 3\n\n print('epoch: ', epoch)\n print('#################################')\n \n #plot\n plt.plot(self.losses, color = 'teal')\n plt.xlabel('Epochs')\n plt.ylabel('Loss')\n return S_pred_list, L_pred_list, I_pred_list, T_pred_list", "_____no_output_____" ], [ "%%time\n\ndinn = DINN(tuberculosis_data[0], tuberculosis_data[1], tuberculosis_data[2], tuberculosis_data[3], tuberculosis_data[4])\n\nlearning_rate = 1e-3\noptimizer = optim.Adam(dinn.params, lr = learning_rate)\ndinn.optimizer = optimizer\n\nscheduler = torch.optim.lr_scheduler.CyclicLR(dinn.optimizer, base_lr=1e-7, max_lr=1e-3, step_size_up=1000, mode=\"exp_range\", gamma=0.85, cycle_momentum=False)\n\ndinn.scheduler = scheduler\n\ntry: \n S_pred_list, L_pred_list, I_pred_list, T_pred_list = dinn.train(1) #train\nexcept EOFError:\n if dinn.save == 2:\n dinn.save = 3\n S_pred_list, L_pred_list, I_pred_list, T_pred_list = dinn.train(1) #train\n elif dinn.save == 3:\n dinn.save = 2\n S_pred_list, L_pred_list, I_pred_list, T_pred_list = dinn.train(1) #train", "_____no_output_____" ], [ "plt.plot(dinn.losses[3000000:], color = 'teal')\nplt.xlabel('Epochs')\nplt.ylabel('Loss')", "_____no_output_____" ], [ "fig = plt.figure(figsize=(12,12))\n\nax = fig.add_subplot(111, facecolor='#dddddd', axisbelow=True)\nax.set_facecolor('xkcd:white')\n\nax.scatter(tuberculosis_data[0], tuberculosis_data[1], color = 'pink', alpha=0.5, lw=2, label='S Data', s=20)\nax.plot(tuberculosis_data[0], S_pred_list[0].detach().numpy(), 'navy', alpha=0.9, lw=2, label='S Prediction', linestyle='dashed')\n\nax.scatter(tuberculosis_data[0], tuberculosis_data[2], color = 'violet', alpha=0.5, lw=2, label='L Data', s=20)\nax.plot(tuberculosis_data[0], L_pred_list[0].detach().numpy(), 'dodgerblue', alpha=0.9, lw=2, label='L Prediction', linestyle='dashed')\n\nax.scatter(tuberculosis_data[0], tuberculosis_data[3], color = 'darkgreen', alpha=0.5, lw=2, label='I Data', s=20)\nax.plot(tuberculosis_data[0], I_pred_list[0].detach().numpy(), 'gold', alpha=0.9, lw=2, label='I Prediction', linestyle='dashed')\n\nax.scatter(tuberculosis_data[0], tuberculosis_data[4], color = 'red', alpha=0.5, lw=2, label='T Data', s=20)\nax.plot(tuberculosis_data[0], T_pred_list[0].detach().numpy(), 'blue', alpha=0.9, lw=2, label='T Prediction', linestyle='dashed')\n\n\nax.set_xlabel('Time /days',size = 20)\nax.set_ylabel('Number',size = 20)\n#ax.set_ylim([-1,50])\nax.yaxis.set_tick_params(length=0)\nax.xaxis.set_tick_params(length=0)\nplt.xticks(size = 20)\nplt.yticks(size = 20)\n# ax.grid(b=True, which='major', c='black', lw=0.2, ls='-')\nlegend = ax.legend(prop={'size':20})\nlegend.get_frame().set_alpha(0.5)\nfor spine in ('top', 'right', 'bottom', 'left'):\n ax.spines[spine].set_visible(False)\nplt.savefig('tuberculosis.pdf')\nplt.show()", "_____no_output_____" ], [ "#vaccination! \n\nimport numpy as np\nfrom scipy.integrate import odeint\nimport matplotlib.pyplot as plt\n\n# Initial conditions\nS0 = 1000\nL0 = 0\nI0 = 1\nT0 = 0\nN = 1001 #S0 + L0 + I0 + T0\n\n# A grid of time points (in days)\nt = np.linspace(0, 40, 50) \n\n#parameters\ndelta = dinn.delta\nprint(delta)\nbeta = dinn.beta\nprint(beta)\nc = dinn.c\nprint(c)\nmu = dinn.mu\nprint(mu)\nk = dinn.k\nprint(k)\nr_1 = dinn.r_1\nprint(r_1)\nr_2 = dinn.r_2\nprint(r_2)\nbeta_prime = dinn.beta_prime\nprint(beta_prime)\nd = dinn.d\nprint(d)\n\n# The SIR model differential equations.\ndef deriv(y, t, N, delta ,beta ,c ,mu ,k ,r_1 ,r_2 ,beta_prime,d ):\n S, L, I, T= y\n\n dSdt = delta - beta * c * S * I / N - mu * S\n dLdt = beta * c * S * I / N - (mu + k + r_1) * L + beta_prime * c * T * 1/N\n dIdt = k*L - (mu + d) * I - r_2 * I\n dTdt = r_1 * L + r_2 * I - beta_prime * c * T * 1/N - mu*T\n\n return dSdt, dLdt, dIdt, dTdt\n\n\n# Initial conditions vector\ny0 = S0, L0, I0, T0\n# Integrate the SIR equations over the time grid, t.\nret = odeint(deriv, y0, t, args=(N, delta ,beta ,c ,mu ,k ,r_1 ,r_2 ,beta_prime,d ))\nS, L, I, T = ret.T\n\n# Plot the data on two separate curves for S(t), I(t)\nfig = plt.figure(figsize=(12,12))\n\nax = fig.add_subplot(111, facecolor='#dddddd', axisbelow=True)\nax.set_facecolor('xkcd:white')\n\nax.plot(t, S, 'violet', alpha=0.5, lw=2, label='S_pred', linestyle='dashed')\nax.plot(tuberculosis_data[0], tuberculosis_data[1], 'grey', alpha=0.5, lw=2, label='S')\n\nax.plot(t, L, 'darkgreen', alpha=0.5, lw=2, label='L_pred', linestyle='dashed')\nax.plot(tuberculosis_data[0], tuberculosis_data[2], 'purple', alpha=0.5, lw=2, label='L')\n\nax.plot(t, I, 'blue', alpha=0.5, lw=2, label='I_pred', linestyle='dashed')\nax.plot(tuberculosis_data[0], tuberculosis_data[3], 'teal', alpha=0.5, lw=2, label='I')\n\nax.plot(t, T, 'black', alpha=0.5, lw=2, label='T_pred', linestyle='dashed')\nax.plot(tuberculosis_data[0], tuberculosis_data[4], 'red', alpha=0.5, lw=2, label='T')\n\nax.set_xlabel('Time /days',size = 20)\nax.set_ylabel('Number',size = 20)\n#ax.set_ylim([-1,50])\nax.yaxis.set_tick_params(length=0)\nax.xaxis.set_tick_params(length=0)\nplt.xticks(size = 20)\nplt.yticks(size = 20)\nax.grid(b=True, which='major', c='black', lw=0.2, ls='-')\nlegend = ax.legend(prop={'size':20})\nlegend.get_frame().set_alpha(0.5)\nfor spine in ('top', 'right', 'bottom', 'left'):\n ax.spines[spine].set_visible(False)\nplt.show()", "_____no_output_____" ], [ "#calculate relative MSE loss\nimport math\n\nS_total_loss = 0\nS_den = 0\nL_total_loss = 0\nL_den = 0\nI_total_loss = 0\nI_den = 0\nT_total_loss = 0\nT_den = 0\nfor timestep in range(len(t)):\n S_value = tuberculosis_data[1][timestep] - S[timestep]\n S_total_loss += S_value**2\n S_den += (tuberculosis_data[1][timestep])**2\n\n L_value = tuberculosis_data[2][timestep] - L[timestep]\n L_total_loss += L_value**2\n L_den += (tuberculosis_data[2][timestep])**2\n\n I_value = tuberculosis_data[3][timestep] - I[timestep]\n I_total_loss += I_value**2\n I_den += (tuberculosis_data[3][timestep])**2\n T_value = tuberculosis_data[4][timestep] - T[timestep]\n T_total_loss += T_value**2\n T_den += (tuberculosis_data[4][timestep])**2\n\nS_total_loss = math.sqrt(S_total_loss/S_den)\nL_total_loss = math.sqrt(L_total_loss/L_den)\nI_total_loss = math.sqrt(I_total_loss/I_den)\nT_total_loss = math.sqrt(T_total_loss/T_den)\n\nprint('S_total_loss: ', S_total_loss)\nprint('I_total_loss: ', L_total_loss)\nprint('S_total_loss: ', I_total_loss)\nprint('I_total_loss: ', T_total_loss)", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "empty" ] ] ]

[ "empty" ]

[ [ "empty" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "#IMPORT SEMUA LIBARARY", "_____no_output_____" ], [ "#IMPORT LIBRARY PANDAS\nimport pandas as pd\n#IMPORT LIBRARY UNTUK POSTGRE\nfrom sqlalchemy import create_engine\nimport psycopg2\n#IMPORT LIBRARY CHART\nfrom matplotlib import pyplot as plt\nfrom matplotlib import style\n#IMPORT LIBRARY BASE PATH\nimport os\nimport io\n#IMPORT LIBARARY PDF\nfrom fpdf import FPDF\n#IMPORT LIBARARY CHART KE BASE64\nimport base64\n#IMPORT LIBARARY EXCEL\nimport xlsxwriter ", "_____no_output_____" ], [ "#FUNGSI UNTUK MENGUPLOAD DATA DARI CSV KE POSTGRESQL", "_____no_output_____" ], [ "def uploadToPSQL(columns, table, filePath, engine):\n #FUNGSI UNTUK MEMBACA CSV\n df = pd.read_csv(\n os.path.abspath(filePath),\n names=columns,\n keep_default_na=False\n )\n #APABILA ADA FIELD KOSONG DISINI DIFILTER\n df.fillna('')\n #MENGHAPUS COLUMN YANG TIDAK DIGUNAKAN\n del df['kategori']\n del df['jenis']\n del df['pengiriman']\n del df['satuan']\n \n #MEMINDAHKAN DATA DARI CSV KE POSTGRESQL\n df.to_sql(\n table, \n engine,\n if_exists='replace'\n )\n \n #DIHITUNG APABILA DATA YANG DIUPLOAD BERHASIL, MAKA AKAN MENGEMBALIKAN KELUARAN TRUE(BENAR) DAN SEBALIKNYA\n if len(df) == 0:\n return False\n else:\n return True", "_____no_output_____" ], [ "#FUNGSI UNTUK MEMBUAT CHART, DATA YANG DIAMBIL DARI DATABASE DENGAN MENGGUNAKAN ORDER DARI TANGGAL DAN JUGA LIMIT\n#DISINI JUGA MEMANGGIL FUNGSI MAKEEXCEL DAN MAKEPDF", "_____no_output_____" ], [ "def makeChart(host, username, password, db, port, table, judul, columns, filePath, name, subjudul, limit, negara, basePath):\n #TEST KONEKSI DATABASE\n try:\n #KONEKSI KE DATABASE\n connection = psycopg2.connect(user=username,password=password,host=host,port=port,database=db)\n cursor = connection.cursor()\n #MENGAMBL DATA DARI TABLE YANG DIDEFINISIKAN DIBAWAH, DAN DIORDER DARI TANGGAL TERAKHIR\n #BISA DITAMBAHKAN LIMIT SUPAYA DATA YANG DIAMBIL TIDAK TERLALU BANYAK DAN BERAT\n postgreSQL_select_Query = \"SELECT * FROM \"+table+\" ORDER BY tanggal ASC LIMIT \" + str(limit)\n \n cursor.execute(postgreSQL_select_Query)\n mobile_records = cursor.fetchall() \n uid = []\n lengthx = []\n lengthy = []\n #MELAKUKAN LOOPING ATAU PERULANGAN DARI DATA YANG SUDAH DIAMBIL\n #KEMUDIAN DATA TERSEBUT DITEMPELKAN KE VARIABLE DIATAS INI\n for row in mobile_records:\n uid.append(row[0])\n lengthx.append(row[1])\n if row[2] == \"\":\n lengthy.append(float(0))\n else:\n lengthy.append(float(row[2]))\n\n #FUNGSI UNTUK MEMBUAT CHART\n #bar\n style.use('ggplot')\n \n fig, ax = plt.subplots()\n #MASUKAN DATA ID DARI DATABASE, DAN JUGA DATA TANGGAL\n ax.bar(uid, lengthy, align='center')\n #UNTUK JUDUL CHARTNYA\n ax.set_title(judul)\n ax.set_ylabel('Total')\n ax.set_xlabel('Tanggal')\n \n ax.set_xticks(uid)\n #TOTAL DATA YANG DIAMBIL DARI DATABASE, DIMASUKAN DISINI\n ax.set_xticklabels((lengthx))\n b = io.BytesIO()\n #CHART DISIMPAN KE FORMAT PNG\n plt.savefig(b, format='png', bbox_inches=\"tight\")\n #CHART YANG SUDAH DIJADIKAN PNG, DISINI DICONVERT KE BASE64\n barChart = base64.b64encode(b.getvalue()).decode(\"utf-8\").replace(\"\\n\", \"\")\n #CHART DITAMPILKAN\n plt.show()\n \n #line\n #MASUKAN DATA DARI DATABASE\n plt.plot(lengthx, lengthy)\n plt.xlabel('Tanggal')\n plt.ylabel('Total')\n #UNTUK JUDUL CHARTNYA\n plt.title(judul)\n plt.grid(True)\n l = io.BytesIO()\n #CHART DISIMPAN KE FORMAT PNG\n plt.savefig(l, format='png', bbox_inches=\"tight\")\n #CHART YANG SUDAH DIJADIKAN PNG, DISINI DICONVERT KE BASE64\n lineChart = base64.b64encode(l.getvalue()).decode(\"utf-8\").replace(\"\\n\", \"\")\n #CHART DITAMPILKAN\n plt.show()\n \n #pie\n #UNTUK JUDUL CHARTNYA\n plt.title(judul)\n #MASUKAN DATA DARI DATABASE\n plt.pie(lengthy, labels=lengthx, autopct='%1.1f%%', \n shadow=True, startangle=180)\n \n plt.axis('equal')\n p = io.BytesIO()\n #CHART DISIMPAN KE FORMAT PNG\n plt.savefig(p, format='png', bbox_inches=\"tight\")\n #CHART YANG SUDAH DIJADIKAN PNG, DISINI DICONVERT KE BASE64\n pieChart = base64.b64encode(p.getvalue()).decode(\"utf-8\").replace(\"\\n\", \"\")\n #CHART DITAMPILKAN\n plt.show()\n \n #MENGAMBIL DATA DARI CSV YANG DIGUNAKAN SEBAGAI HEADER DARI TABLE UNTUK EXCEL DAN JUGA PDF\n header = pd.read_csv(\n os.path.abspath(filePath),\n names=columns,\n keep_default_na=False\n )\n #MENGHAPUS COLUMN YANG TIDAK DIGUNAKAN\n header.fillna('')\n del header['tanggal']\n del header['total']\n #MEMANGGIL FUNGSI EXCEL\n makeExcel(mobile_records, header, name, limit, basePath)\n #MEMANGGIL FUNGSI PDF\n makePDF(mobile_records, header, judul, barChart, lineChart, pieChart, name, subjudul, limit, basePath) \n \n #JIKA GAGAL KONEKSI KE DATABASE, MASUK KESINI UNTUK MENAMPILKAN ERRORNYA\n except (Exception, psycopg2.Error) as error :\n print (error)\n\n #KONEKSI DITUTUP\n finally:\n if(connection):\n cursor.close()\n connection.close()", "_____no_output_____" ], [ "#FUNGSI MAKEEXCEL GUNANYA UNTUK MEMBUAT DATA YANG BERASAL DARI DATABASE DIJADIKAN FORMAT EXCEL TABLE F2\n#PLUGIN YANG DIGUNAKAN ADALAH XLSXWRITER", "_____no_output_____" ], [ "def makeExcel(datarow, dataheader, name, limit, basePath):\n #MEMBUAT FILE EXCEL\n workbook = xlsxwriter.Workbook(basePath+'jupyter/BLOOMBERG/SektorHargaInflasi/excel/'+name+'.xlsx')\n #MENAMBAHKAN WORKSHEET PADA FILE EXCEL TERSEBUT\n worksheet = workbook.add_worksheet('sheet1')\n #SETINGAN AGAR DIBERIKAN BORDER DAN FONT MENJADI BOLD\n row1 = workbook.add_format({'border': 2, 'bold': 1})\n row2 = workbook.add_format({'border': 2})\n #MENJADIKAN DATA MENJADI ARRAY\n data=list(datarow)\n isihead=list(dataheader.values)\n header = []\n body = []\n \n #LOOPING ATAU PERULANGAN, KEMUDIAN DATA DITAMPUNG PADA VARIABLE DIATAS\n for rowhead in dataheader:\n header.append(str(rowhead))\n \n for rowhead2 in datarow:\n header.append(str(rowhead2[1]))\n \n for rowbody in isihead[1]:\n body.append(str(rowbody))\n \n for rowbody2 in data:\n body.append(str(rowbody2[2]))\n \n #MEMASUKAN DATA DARI VARIABLE DIATAS KE DALAM COLUMN DAN ROW EXCEL\n for col_num, data in enumerate(header):\n worksheet.write(0, col_num, data, row1)\n \n for col_num, data in enumerate(body):\n worksheet.write(1, col_num, data, row2)\n \n #FILE EXCEL DITUTUP\n workbook.close()", "_____no_output_____" ], [ "#FUNGSI UNTUK MEMBUAT PDF YANG DATANYA BERASAL DARI DATABASE DIJADIKAN FORMAT EXCEL TABLE F2\n#PLUGIN YANG DIGUNAKAN ADALAH FPDF", "_____no_output_____" ], [ "def makePDF(datarow, dataheader, judul, bar, line, pie, name, subjudul, lengthPDF, basePath):\n #FUNGSI UNTUK MENGATUR UKURAN KERTAS, DISINI MENGGUNAKAN UKURAN A4 DENGAN POSISI LANDSCAPE\n pdf = FPDF('L', 'mm', [210,297])\n #MENAMBAHKAN HALAMAN PADA PDF\n pdf.add_page()\n #PENGATURAN UNTUK JARAK PADDING DAN JUGA UKURAN FONT\n pdf.set_font('helvetica', 'B', 20.0)\n pdf.set_xy(145.0, 15.0)\n #MEMASUKAN JUDUL KE DALAM PDF\n pdf.cell(ln=0, h=2.0, align='C', w=10.0, txt=judul, border=0)\n \n #PENGATURAN UNTUK UKURAN FONT DAN JUGA JARAK PADDING\n pdf.set_font('arial', '', 14.0)\n pdf.set_xy(145.0, 25.0)\n #MEMASUKAN SUB JUDUL KE PDF\n pdf.cell(ln=0, h=2.0, align='C', w=10.0, txt=subjudul, border=0)\n #MEMBUAT GARIS DI BAWAH SUB JUDUL\n pdf.line(10.0, 30.0, 287.0, 30.0)\n pdf.set_font('times', '', 10.0)\n pdf.set_xy(17.0, 37.0)\n \n #PENGATURAN UNTUK UKURAN FONT DAN JUGA JARAK PADDING\n pdf.set_font('Times','',10.0) \n #MENGAMBIL DATA HEADER PDF YANG SEBELUMNYA SUDAH DIDEFINISIKAN DIATAS\n datahead=list(dataheader.values)\n pdf.set_font('Times','B',12.0) \n pdf.ln(0.5)\n \n th1 = pdf.font_size\n \n #MEMBUAT TABLE PADA PDF, DAN MENAMPILKAN DATA DARI VARIABLE YANG SUDAH DIKIRIM\n pdf.cell(100, 2*th1, \"Kategori\", border=1, align='C')\n pdf.cell(177, 2*th1, datahead[0][0], border=1, align='C')\n pdf.ln(2*th1)\n pdf.cell(100, 2*th1, \"Jenis\", border=1, align='C')\n pdf.cell(177, 2*th1, datahead[0][1], border=1, align='C')\n pdf.ln(2*th1)\n pdf.cell(100, 2*th1, \"Pengiriman\", border=1, align='C')\n pdf.cell(177, 2*th1, datahead[0][2], border=1, align='C')\n pdf.ln(2*th1)\n pdf.cell(100, 2*th1, \"Satuan\", border=1, align='C')\n pdf.cell(177, 2*th1, datahead[0][3], border=1, align='C')\n pdf.ln(2*th1)\n \n #PENGATURAN PADDING\n pdf.set_xy(17.0, 75.0)\n \n #PENGATURAN UNTUK UKURAN FONT DAN JUGA JARAK PADDING\n pdf.set_font('Times','B',11.0) \n data=list(datarow)\n epw = pdf.w - 2*pdf.l_margin\n col_width = epw/(lengthPDF+1)\n \n #PENGATURAN UNTUK JARAK PADDING\n pdf.ln(0.5)\n th = pdf.font_size\n \n #MEMASUKAN DATA HEADER YANG DIKIRIM DARI VARIABLE DIATAS KE DALAM PDF\n pdf.cell(50, 2*th, str(\"Negara\"), border=1, align='C')\n for row in data:\n pdf.cell(40, 2*th, str(row[1]), border=1, align='C')\n pdf.ln(2*th)\n \n #MEMASUKAN DATA ISI YANG DIKIRIM DARI VARIABLE DIATAS KE DALAM PDF\n pdf.set_font('Times','B',10.0)\n pdf.set_font('Arial','',9)\n pdf.cell(50, 2*th, negara, border=1, align='C')\n for row in data:\n pdf.cell(40, 2*th, str(row[2]), border=1, align='C')\n pdf.ln(2*th)\n \n #MENGAMBIL DATA CHART, KEMUDIAN CHART TERSEBUT DIJADIKAN PNG DAN DISIMPAN PADA DIRECTORY DIBAWAH INI\n #BAR CHART\n bardata = base64.b64decode(bar)\n barname = basePath+'jupyter/BLOOMBERG/SektorHargaInflasi/img/'+name+'-bar.png'\n with open(barname, 'wb') as f:\n f.write(bardata)\n \n #LINE CHART\n linedata = base64.b64decode(line)\n linename = basePath+'jupyter/BLOOMBERG/SektorHargaInflasi/img/'+name+'-line.png'\n with open(linename, 'wb') as f:\n f.write(linedata)\n \n #PIE CHART\n piedata = base64.b64decode(pie)\n piename = basePath+'jupyter/BLOOMBERG/SektorHargaInflasi/img/'+name+'-pie.png'\n with open(piename, 'wb') as f:\n f.write(piedata)\n \n #PENGATURAN UNTUK UKURAN FONT DAN JUGA JARAK PADDING\n pdf.set_xy(17.0, 75.0)\n col = pdf.w - 2*pdf.l_margin\n widthcol = col/3\n #MEMANGGIL DATA GAMBAR DARI DIREKTORY DIATAS\n pdf.image(barname, link='', type='',x=8, y=100, w=widthcol)\n pdf.set_xy(17.0, 75.0)\n col = pdf.w - 2*pdf.l_margin\n pdf.image(linename, link='', type='',x=103, y=100, w=widthcol)\n pdf.set_xy(17.0, 75.0)\n col = pdf.w - 2*pdf.l_margin\n pdf.image(piename, link='', type='',x=195, y=100, w=widthcol)\n pdf.ln(2*th)\n \n #MEMBUAT FILE PDF\n pdf.output(basePath+'jupyter/BLOOMBERG/SektorHargaInflasi/pdf/'+name+'.pdf', 'F')", "_____no_output_____" ], [ "#DISINI TEMPAT AWAL UNTUK MENDEFINISIKAN VARIABEL VARIABEL SEBELUM NANTINYA DIKIRIM KE FUNGSI\n#PERTAMA MANGGIL FUNGSI UPLOADTOPSQL DULU, KALAU SUKSES BARU MANGGIL FUNGSI MAKECHART\n#DAN DI MAKECHART MANGGIL FUNGSI MAKEEXCEL DAN MAKEPDF", "_____no_output_____" ], [ "#DEFINISIKAN COLUMN BERDASARKAN FIELD CSV\ncolumns = [\n \"kategori\",\n \"jenis\",\n \"tanggal\",\n \"total\",\n \"pengiriman\",\n \"satuan\",\n]\n\n#UNTUK NAMA FILE\nname = \"SektorHargaInflasi3_6\"\n#VARIABLE UNTUK KONEKSI KE DATABASE\nhost = \"localhost\"\nusername = \"postgres\"\npassword = \"1234567890\"\nport = \"5432\"\ndatabase = \"bloomberg_SektorHargaInflasi\"\ntable = name.lower()\n#JUDUL PADA PDF DAN EXCEL\njudul = \"Data Sektor Harga Inflasi\"\nsubjudul = \"Badan Perencanaan Pembangunan Nasional\"\n#LIMIT DATA UNTUK SELECT DI DATABASE\nlimitdata = int(8)\n#NAMA NEGARA UNTUK DITAMPILKAN DI EXCEL DAN PDF\nnegara = \"Indonesia\"\n#BASE PATH DIRECTORY\nbasePath = 'C:/Users/ASUS/Documents/bappenas/'\n#FILE CSV\nfilePath = basePath+ 'data mentah/BLOOMBERG/SektorHargaInflasi/' +name+'.csv';\n#KONEKSI KE DATABASE\nengine = create_engine('postgresql://'+username+':'+password+'@'+host+':'+port+'/'+database)\n\n#MEMANGGIL FUNGSI UPLOAD TO PSQL\ncheckUpload = uploadToPSQL(columns, table, filePath, engine)\n#MENGECEK FUNGSI DARI UPLOAD PSQL, JIKA BERHASIL LANJUT MEMBUAT FUNGSI CHART, JIKA GAGAL AKAN MENAMPILKAN PESAN ERROR\nif checkUpload == True:\n makeChart(host, username, password, database, port, table, judul, columns, filePath, name, subjudul, limitdata, negara, basePath)\nelse:\n print(\"Error When Upload CSV\")", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import pytest\nfrom scipy.stats import zscore\nfrom mne.preprocessing import create_ecg_epochs\nfrom sklearn.model_selection import train_test_split\n%run parameters.py\n%run Utility_Functions.ipynb", "_____no_output_____" ], [ "%matplotlib qt5\ndata = np.load('All_Subject_IR_Index_'+str(epoch_length)+'.npy')\nprint(data.shape)\nsb.set()\ndef ir_plot(data):\n for i, subject in enumerate(subjects):\n temp = []\n for j, trial in enumerate(trials):\n temp.append(data[i][j][:])\n plt.subplot(3,6,i+1)\n plt.boxplot(temp, showfliers=False)\n plt.tight_layout()\n \nir_plot(data)\n\n\ndata = np.load('All_Subject_IR_Index_'+str(epoch_length)+'.npy')\nprint(data.shape)\n\nplt.figure()\ndef ir_plot(data):\n for i, subject in enumerate(subjects):\n temp = []\n for j, trial in enumerate(trials):\n if trial=='HighFine' or trial=='LowFine':\n temp.append(data[i][j][:])\n plt.subplot(3,6,i+1)\n for element in temp:\n plt.plot(element)\n plt.tight_layout()\n \ndef min_max(data):\n data -= data.min()\n# data /= data.ptp()\n \n return data\n\nplt.figure()\ndef ir_plot(data):\n for i, subject in enumerate(subjects):\n temp = []\n for j, trial in enumerate(trials):\n if trial=='HighFine' or trial=='LowFine':\n temp.append(data[i][j][:])\n temp_z = zscore(np.vstack((np.expand_dims(temp[0], axis=1),np.expand_dims(temp[1], axis=1))))\n plt.plot(temp_z[0:len(temp[0])], 'r')\n plt.plot(temp_z[len(temp[0]):], 'b')\n plt.tight_layout()\n \ndef test_epoch_length(subjects, trials):\n s = []\n for subject in subjects:\n for trial in trials:\n read_eeg_path = '../Cleaned Data/' + subject + '/EEG/'\n read_force_path = '../Cleaned Data/' + subject + '/Force/'\n cleaned_eeg = mne.read_epochs(read_eeg_path + subject + '_' + trial + '_' + str(epoch_length) \n + '_cleaned_epo.fif', verbose=False)\n cleaned_force = mne.read_epochs(read_force_path + subject + '_' + trial + '_' + str(epoch_length) \n + '_cleaned_epo.fif', verbose=False)\n eeg = cleaned_eeg.get_data()\n force = cleaned_force.get_data()\n \n # Check whether eeg and force data are same\n assert eeg.shape[0]==force.shape[0]\n s.append(subject)\n \n # Check whether all subjects were tested\n assert len(s)==len(subjects), 'Huston! We have got a problem!'\n \n return 'Reached moon!'\n\n\ndef test_data():\n x = np.load('PSD_X_Data_' + str(epoch_length) + '.npy')\n y = np.load('IR_Y_Data_' + str(epoch_length) + '.npy')\n \n assert x.shape[0]==y.shape[0], \"Houston we've got a problem!\"\n \n \ndef test_psd_image():\n x = np.load('PSD_X_Data_' + str(epoch_length) +'.npy')\n \n plt.imshow(x[5,:,0].reshape(image_size, image_size))\n \ndef test_x_y_length():\n x = np.load('X.npy')\n y = np.load('Y.npy')\n assert x.shape[0]==y.shape[0], 'Huston! We have got a problem!'\n \n return 'Reached moon!'", "_____no_output_____" ], [ "def test_x_y_length():\n x = np.load('X.npy')\n y = np.load('Y.npy')\n print(sum(y)/len(y))\n assert x.shape[0]==y.shape[0], 'Huston! We have got a problem!'\n \n return 'Reached moon!'\n\ntest_x_y_length()", "[0.26972963 0.4975338 0.23273657]\n" ], [ "x = np.load('X.npy')\ny = np.load('Y.npy')\n\nprint(x.shape)\n\nx_normal = x[np.argmax(y, axis=1)==1,:,:]\ny_normal = y[np.argmax(y, axis=1)==1]\n\nprint(np.argmax(y, axis=1)==0)\n\nx_low = x[np.argmax(y, axis=1)==0,:,:]\ny_low = y[np.argmax(y, axis=1)==0]\n\nprint(x_low.shape)\n\nx_high = x[np.argmax(y, axis=1)==2,:,:]\ny_high = y[np.argmax(y, axis=1)==2]\n\nx_normal, x_test, y_normal, y_test = train_test_split(x_normal, y_normal, test_size = 0.50)\n\nx_balanced = np.vstack((x_low, x_normal, x_high))", "(10948, 20, 256)\n[ True True True ... False False False]\n(2953, 20, 256)\n" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Produit matriciel avec une matrice creuse\n\nLes dictionnaires sont une façon assez de représenter les matrices creuses en ne conservant que les coefficients non nuls. Comment écrire alors le produit matriciel ?", "_____no_output_____" ] ], [ [ "from jyquickhelper import add_notebook_menu\nadd_notebook_menu()", "_____no_output_____" ] ], [ [ "## Matrice creuse et dictionnaire\n\nUne [matrice creuse](https://fr.wikipedia.org/wiki/Matrice_creuse) ou [sparse matrix](https://en.wikipedia.org/wiki/Sparse_matrix) est constituée majoritairement de 0. On utilise un dictionnaire avec les coefficients non nuls. La fonction suivante pour créer une matrice aléatoire.", "_____no_output_____" ] ], [ [ "import random\n\ndef random_matrix(n, m, ratio=0.1):\n mat = {}\n nb = min(n * m, int(ratio * n * m + 0.5))\n while len(mat) < nb:\n i = random.randint(0, n-1)\n j = random.randint(0, m-1)\n mat[i, j] = 1\n return mat\n\nmat = random_matrix(3, 3, ratio=0.5)\nmat", "_____no_output_____" ] ], [ [ "## Calcul de la dimension\n\nPour obtenir la dimension de la matrice, il faut parcourir toutes les clés du dictionnaire.", "_____no_output_____" ] ], [ [ "def dimension(mat):\n maxi, maxj = 0, 0\n for k in mat:\n maxi = max(maxi, k[0])\n maxj = max(maxj, k[1])\n return maxi + 1, maxj + 1\n\ndimension(mat)", "_____no_output_____" ] ], [ [ "Cette fonction possède l'inconvénient de retourner une valeur fausse si la matrice ne possède aucun coefficient non nul sur la dernière ligne ou la dernière colonne. Cela peut être embarrassant, tout dépend de l'usage.", "_____no_output_____" ], [ "## Produit matriciel classique\n\nOn implémente le produit matriciel classique, à trois boucles.", "_____no_output_____" ] ], [ [ "def produit_classique(m1, m2):\n dim1 = dimension(m1)\n dim2 = dimension(m2)\n if dim1[1] != dim2[0]:\n raise Exception(\"Impossible de multiplier {0}, {1}\".format(\n dim1, dim2))\n res = {}\n for i in range(dim1[0]):\n for j in range(dim2[1]):\n s = 0\n for k in range(dim1[1]):\n s += m1.get((i, k), 0) * m2.get((k, j), 0)\n if s != 0: # Pour éviter de garder les coefficients non nuls.\n res[i, j] = s \n return res\n\nsimple = {(0, 1): 1, (1, 0): 1}\nproduit_classique(simple, simple)", "_____no_output_____" ] ], [ [ "Sur la matrice aléatoire...", "_____no_output_____" ] ], [ [ "produit_classique(mat, mat)", "_____no_output_____" ] ], [ [ "## Produit matriciel plus élégant\n\nA-t-on vraiment besoin de s'enquérir des dimensions de la matrice pour faire le produit matriciel ? Ne peut-on pas tout simplement faire une boucle sur les coefficients non nul ?", "_____no_output_____" ] ], [ [ "def produit_elegant(m1, m2):\n res = {}\n for (i, k1), v1 in m1.items():\n if v1 == 0:\n continue\n for (k2, j), v2 in m2.items(): \n if v2 == 0:\n continue\n if k1 == k2:\n if (i, j) in res:\n res[i, j] += v1 * v2\n else :\n res[i, j] = v1 * v2\n return res\n\nproduit_elegant(simple, simple)", "_____no_output_____" ], [ "produit_elegant(mat, mat)", "_____no_output_____" ] ], [ [ "## Mesure du temps\n\nA priori, la seconde méthode est plus rapide puisque son coût est proportionnel au produit du nombre de coefficients non nuls dans les deux matrices. Vérifions.", "_____no_output_____" ] ], [ [ "bigmat = random_matrix(100, 100)\n%timeit produit_classique(bigmat, bigmat)", "658 ms ± 81.3 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n" ], [ "%timeit produit_elegant(bigmat, bigmat)", "157 ms ± 9.33 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)\n" ] ], [ [ "C'est beaucoup mieux. Mais peut-on encore faire mieux ?", "_____no_output_____" ], [ "## Dictionnaires de dictionnaires\n\nCa sonne un peu comme [mille millions de mille sabords](https://fr.wikipedia.org/wiki/Vocabulaire_du_capitaine_Haddock) mais le dictionnaire que nous avons créé a pour clé un couple de coordonnées et valeur des coefficients. La fonction ``produit_elegant`` fait plein d'itérations inutiles en quelque sorte puisque les coefficients sont nuls. Peut-on éviter ça ?\n\nEt si on utilisait des dictionnaires de dictionnaires : ``{ ligne : { colonne : valeur } }``.", "_____no_output_____" ] ], [ [ "def matrice_dicodico(mat):\n res = {}\n for (i, j), v in mat.items():\n if i not in res:\n res[i] = {j: v}\n else:\n res[i][j] = v\n return res\n\nmatrice_dicodico(simple)", "_____no_output_____" ] ], [ [ "Peut-on adapter le calcul matriciel élégant ? Il reste à associer les indices de colonnes de la première avec les indices de lignes de la seconde. Cela pose problème en l'état quand les indices de colonnes sont inaccessibles sans connaître les indices de lignes d'abord à moins d'échanger l'ordre pour la seconde matrice.", "_____no_output_____" ] ], [ [ "def matrice_dicodico_lc(mat, ligne=True):\n res = {}\n if ligne:\n for (i, j), v in mat.items():\n if i not in res:\n res[i] = {j: v}\n else:\n res[i][j] = v\n else:\n for (j, i), v in mat.items():\n if i not in res:\n res[i] = {j: v}\n else:\n res[i][j] = v\n return res\n\nmatrice_dicodico_lc(simple, ligne=False)", "_____no_output_____" ] ], [ [ "Maintenant qu'on a fait ça, on peut songer au produit matriciel.", "_____no_output_____" ] ], [ [ "def produit_elegant_rapide(m1, m2):\n res = {}\n for k, vs in m1.items():\n if k in m2:\n for i, v1 in vs.items():\n for j, v2 in m2[k].items():\n if (i, j) in res:\n res[i, j] += v1 * v2\n else :\n res[i, j] = v1 * v2\n\n return res\n\nm1 = matrice_dicodico_lc(simple, ligne=False)\nm2 = matrice_dicodico_lc(simple)\nproduit_elegant_rapide(m1, m2)", "_____no_output_____" ], [ "m1 = matrice_dicodico_lc(mat, ligne=False)\nm2 = matrice_dicodico_lc(mat)\nproduit_elegant_rapide(m1, m2)", "_____no_output_____" ] ], [ [ "On mesure le temps avec une grande matrice.", "_____no_output_____" ] ], [ [ "m1 = matrice_dicodico_lc(bigmat, ligne=False)\nm2 = matrice_dicodico_lc(bigmat)\n%timeit produit_elegant_rapide(m1, m2)", "6.46 ms ± 348 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)\n" ] ], [ [ "Beaucoup plus rapide, il n'y a plus besoin de tester les coefficients non nuls. La comparaison n'est pas très juste néanmoins car il faut transformer les deux matrices avant de faire le calcul. Et si on l'incluait ?", "_____no_output_____" ] ], [ [ "def produit_elegant_rapide_transformation(mat1, mat2):\n m1 = matrice_dicodico_lc(mat1, ligne=False)\n m2 = matrice_dicodico_lc(mat2)\n return produit_elegant_rapide(m1, m2)\n\nproduit_elegant_rapide_transformation(simple, simple)", "_____no_output_____" ], [ "%timeit produit_elegant_rapide_transformation(bigmat, bigmat)", "7.17 ms ± 635 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)\n" ] ], [ [ "Finalement ça vaut le coup... mais est-ce le cas pour toutes les matrices.", "_____no_output_____" ] ], [ [ "%matplotlib inline", "_____no_output_____" ], [ "import time\nmesures = []\n\nfor ratio in [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.99]:\n big = random_matrix(100, 100, ratio=ratio)\n \n t1 = time.perf_counter()\n produit_elegant_rapide_transformation(big, big)\n t2 = time.perf_counter()\n dt = (t2 - t1)\n obs = {\"dicodico\": dt, \"ratio\": ratio}\n \n if ratio <= 0.3:\n # après c'est trop lent\n t1 = time.perf_counter()\n produit_elegant(big, big)\n t2 = time.perf_counter()\n dt = (t2 - t1)\n obs[\"dico\"] = dt\n \n t1 = time.perf_counter()\n produit_classique(big, big)\n t2 = time.perf_counter()\n dt = (t2 - t1)\n obs[\"classique\"] = dt\n\n mesures.append(obs)\n print(obs) ", "{'dicodico': 0.005037441000240506, 'ratio': 0.1, 'dico': 0.14238126200052648, 'classique': 0.6123743010002727}\n{'dicodico': 0.026836189999812632, 'ratio': 0.2, 'dico': 0.6422706439998365, 'classique': 0.6319077640000614}\n{'dicodico': 0.05713747299978422, 'ratio': 0.3, 'dico': 1.5467935550004768, 'classique': 0.6079049600002691}\n{'dicodico': 0.11274368399972445, 'ratio': 0.4, 'classique': 0.6851242430002458}\n{'dicodico': 0.16862485899946478, 'ratio': 0.5, 'classique': 0.6875519010000062}\n{'dicodico': 0.22460795999995753, 'ratio': 0.6, 'classique': 0.7007410579999487}\n{'dicodico': 0.3225403609994828, 'ratio': 0.7, 'classique': 0.6795763820000502}\n{'dicodico': 0.40570493699942745, 'ratio': 0.8, 'classique': 0.6769405260001804}\n{'dicodico': 0.5422258439994039, 'ratio': 0.9, 'classique': 0.6723052600000301}\n{'dicodico': 0.6474834919999921, 'ratio': 0.99, 'classique': 0.6958359640002527}\n" ], [ "from pandas import DataFrame\ndf = DataFrame(mesures)\nax = df.plot(x=\"ratio\", y=\"dicodico\", label=\"dico dico\")\ndf.plot(x=\"ratio\", y=\"dico\", label=\"dico\", ax=ax)\ndf.plot(x=\"ratio\", y=\"classique\", label=\"classique\", ax=ax)\nax.legend();", "_____no_output_____" ] ], [ [ "Cette dernière version est efficace.", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/mohd-faizy/03_TensorFlow_In-Practice/blob/master/03_callbacks.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ], [ "# __Callbacks API__\n\nA __callback__ is an object that can perform actions at various stages of training (e.g. at the start or end of an epoch, before or after a single batch, etc).\n\n_You can use callbacks to:_\n\n- Write TensorBoard logs after every batch of training to monitor your metrics.\n- Periodically save your model to disk.\n- Do early stopping.\n- Get a view on internal states and statistics of a model during training...and more\n", "_____no_output_____" ], [ "## __Usage of callbacks via the built-in `fit()` loop__\n\nYou can pass a list of callbacks (as the keyword argument callbacks) to the `.fit()` method of a model:\n\n```\nmy_callbacks = [\n tf.keras.callbacks.EarlyStopping(patience=2),\n tf.keras.callbacks.ModelCheckpoint(filepath='model.{epoch:02d}-{val_loss:.2f}.h5'),\n tf.keras.callbacks.TensorBoard(log_dir='./logs'),\n]\nmodel.fit(dataset, epochs=10, callbacks=my_callbacks)\n\n```\nThe relevant methods of the callbacks will then be called at each stage of the training.", "_____no_output_____" ], [ "__Using custom callbacks__\n\nCreating new callbacks is a simple and powerful way to customize a training loop. Learn more about creating new callbacks in the guide [__Writing your own Callbacks__](https://keras.io/guides/writing_your_own_callbacks/), and refer to the documentation for the [__base Callback class__](https://keras.io/api/callbacks/base_callback/).", "_____no_output_____" ], [ "__Available callbacks__\n\n```\n- Base Callback class\n- ModelCheckpoint\n- TensorBoard\n- EarlyStopping\n- LearningRateScheduler\n- ReduceLROnPlateau\n- RemoteMonitor\n- LambdaCallback\n- TerminateOnNaN\n- CSVLogger\n- ProgbarLogger\n```", "_____no_output_____" ], [ "> [__Writing your own callbacks__](https://www.tensorflow.org/guide/keras/custom_callback)", "_____no_output_____" ] ], [ [ "import tensorflow as tf\n\n# Defining the callback class\nclass myCallback(tf.keras.callbacks.Callback):\n def on_epoch_end(self, epoch, logs={}):\n if(logs.get('accuracy')>0.6):\n print(\"\\nReached 60% accuracy so cancelling training!\")\n self.model.stop_training = True\n\nmnist = tf.keras.datasets.fashion_mnist\n\n(x_train, y_train),(x_test, y_test) = mnist.load_data()\nx_train, x_test = x_train / 255.0, x_test / 255.0\n\ncallbacks = myCallback()\n\nmodel = tf.keras.models.Sequential([\n tf.keras.layers.Flatten(input_shape=(28, 28)),\n tf.keras.layers.Dense(512, activation=tf.nn.relu),\n tf.keras.layers.Dense(10, activation=tf.nn.softmax)\n])\nmodel.compile(optimizer=tf.optimizers.Adam(),\n loss='sparse_categorical_crossentropy',\n metrics=['accuracy'])\n\nmodel.fit(x_train, y_train, epochs=10, callbacks=[callbacks])", "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-labels-idx1-ubyte.gz\n32768/29515 [=================================] - 0s 0us/step\nDownloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-images-idx3-ubyte.gz\n26427392/26421880 [==============================] - 0s 0us/step\nDownloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-labels-idx1-ubyte.gz\n8192/5148 [===============================================] - 0s 0us/step\nDownloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-images-idx3-ubyte.gz\n4423680/4422102 [==============================] - 0s 0us/step\nEpoch 1/10\n1866/1875 [============================>.] - ETA: 0s - loss: 0.4722 - accuracy: 0.8303\nReached 60% accuracy so cancelling training!\n1875/1875 [==============================] - 7s 4ms/step - loss: 0.4723 - accuracy: 0.8302\n" ] ] ]

[ "markdown", "code" ]

[ [ "markdown", "markdown", "markdown", "markdown", "markdown", "markdown" ], [ "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "from PIL import Image\nimport numpy as np\nimport os\nimport cv2\nimport keras\nfrom keras.utils import np_utils\nfrom keras.models import Sequential\nfrom keras.layers import Conv2D,MaxPooling2D,Dense,Flatten,Dropout\nimport pandas as pd\nimport sys\nimport tensorflow as tf\n%matplotlib inline\nimport matplotlib.pyplot as plt\nimport plotly.express as px", "_____no_output_____" ], [ "def readData(filepath, label):\n cells = []\n labels = []\n file = os.listdir(filepath)\n for img in file:\n try:\n image = cv2.imread(filepath + img)\n image_from_array = Image.fromarray(image, 'RGB')\n size_image = image_from_array.resize((50, 50))\n cells.append(np.array(size_image))\n labels.append(label)\n except AttributeError as e:\n print('Skipping file: ', img, e)\n print(len(cells), ' Data Points Read!')\n return np.array(cells), np.array(labels)", "_____no_output_____" ], [ "def genesis_train(file):\n \n print('Reading Training Data')\n \n ParasitizedCells, ParasitizedLabels = readData(file + '/Parasitized/', 1)\n UninfectedCells, UninfectedLabels = readData(file + '/Uninfected/', 0)\n Cells = np.concatenate((ParasitizedCells, UninfectedCells))\n Labels = np.concatenate((ParasitizedLabels, UninfectedLabels))\n \n print('Reading Testing Data')\n \n TestParasitizedCells, TestParasitizedLabels = readData('./input/fed/test/Parasitized/', 1)\n TestUninfectedCells, TestUninfectedLabels = readData('./input/fed/test/Uninfected/', 0)\n TestCells = np.concatenate((TestParasitizedCells, TestUninfectedCells))\n TestLabels = np.concatenate((TestParasitizedLabels, TestUninfectedLabels))\n \n s = np.arange(Cells.shape[0])\n np.random.shuffle(s)\n Cells = Cells[s]\n Labels = Labels[s]\n \n sTest = np.arange(TestCells.shape[0])\n np.random.shuffle(sTest)\n TestCells = TestCells[sTest]\n TestLabels = TestLabels[sTest]\n \n num_classes=len(np.unique(Labels))\n len_data=len(Cells)\n print(len_data, ' Data Points')\n \n (x_train,x_test)=Cells, TestCells\n (y_train,y_test)=Labels, TestLabels\n \n # Since we're working on image data, we normalize data by divinding 255.\n x_train = x_train.astype('float32')/255 \n x_test = x_test.astype('float32')/255\n train_len=len(x_train)\n test_len=len(x_test)\n \n #Doing One hot encoding as classifier has multiple classes\n y_train=keras.utils.to_categorical(y_train,num_classes)\n y_test=keras.utils.to_categorical(y_test,num_classes)\n \n #creating sequential model\n model=Sequential()\n model.add(Conv2D(filters=16,kernel_size=2,padding=\"same\",activation=\"relu\",input_shape=(50,50,3)))\n model.add(MaxPooling2D(pool_size=2))\n model.add(Conv2D(filters=32,kernel_size=2,padding=\"same\",activation=\"relu\"))\n model.add(MaxPooling2D(pool_size=2))\n model.add(Conv2D(filters=64,kernel_size=2,padding=\"same\",activation=\"relu\"))\n model.add(MaxPooling2D(pool_size=2))\n model.add(Dropout(0.2))\n model.add(Flatten())\n model.add(Dense(500,activation=\"relu\"))\n model.add(Dropout(0.2))\n model.add(Dense(2,activation=\"softmax\"))#2 represent output layer neurons \n# model.summary()\n\n # compile the model with loss as categorical_crossentropy and using adam optimizer\n model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])\n \n #Fit the model with min batch size as 50[can tune batch size to some factor of 2^power ] \n model.fit(x_train, y_train, batch_size=100, epochs=5, verbose=1)\n \n scores = model.evaluate(x_test, y_test)\n print(\"Loss: \", scores[0]) #Loss\n print(\"Accuracy: \", scores[1]) #Accuracy\n\n #Saving Model\n model.save(\"./output.h5\")\n return len_data, scores[1]", "_____no_output_____" ], [ "def update_train(file, d):\n \n print('Reading Training Data')\n \n ParasitizedCells, ParasitizedLabels = readData(file + '/Parasitized/', 1)\n UninfectedCells, UninfectedLabels = readData(file + '/Uninfected/', 0)\n Cells = np.concatenate((ParasitizedCells, UninfectedCells))\n Labels = np.concatenate((ParasitizedLabels, UninfectedLabels))\n \n print('Reading Testing Data')\n \n TestParasitizedCells, TestParasitizedLabels = readData('./input/fed/test/Parasitized/', 1)\n TestUninfectedCells, TestUninfectedLabels = readData('./input/fed/test/Uninfected/', 0)\n TestCells = np.concatenate((TestParasitizedCells, TestUninfectedCells))\n TestLabels = np.concatenate((TestParasitizedLabels, TestUninfectedLabels))\n \n s = np.arange(Cells.shape[0])\n np.random.shuffle(s)\n Cells = Cells[s]\n Labels = Labels[s]\n \n sTest = np.arange(TestCells.shape[0])\n np.random.shuffle(sTest)\n TestCells = TestCells[sTest]\n TestLabels = TestLabels[sTest]\n \n num_classes=len(np.unique(Labels))\n len_data=len(Cells)\n print(len_data, ' Data Points')\n \n (x_train,x_test)=Cells, TestCells\n (y_train,y_test)=Labels, TestLabels\n \n # Since we're working on image data, we normalize data by divinding 255.\n x_train = x_train.astype('float32')/255 \n x_test = x_test.astype('float32')/255\n train_len=len(x_train)\n test_len=len(x_test)\n \n #Doing One hot encoding as classifier has multiple classes\n y_train=keras.utils.to_categorical(y_train,num_classes)\n y_test=keras.utils.to_categorical(y_test,num_classes)\n \n #creating sequential model\n model=Sequential()\n model.add(Conv2D(filters=16,kernel_size=2,padding=\"same\",activation=\"relu\",input_shape=(50,50,3)))\n model.add(MaxPooling2D(pool_size=2))\n model.add(Conv2D(filters=32,kernel_size=2,padding=\"same\",activation=\"relu\"))\n model.add(MaxPooling2D(pool_size=2))\n model.add(Conv2D(filters=64,kernel_size=2,padding=\"same\",activation=\"relu\"))\n model.add(MaxPooling2D(pool_size=2))\n model.add(Dropout(0.2))\n model.add(Flatten())\n model.add(Dense(500,activation=\"relu\"))\n model.add(Dropout(0.2))\n model.add(Dense(2,activation=\"softmax\"))#2 represent output layer neurons \n # model.summary()\n\n model.load_weights(\"./output.h5\")\n \n # compile the model with loss as categorical_crossentropy and using adam optimizer\n model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])\n \n #Fit the model with min batch size as 50[can tune batch size to some factor of 2^power ] \n model.fit(x_train, y_train, batch_size=100, epochs=5, verbose=1)\n \n \n scores = model.evaluate(x_test, y_test)\n print(\"Loss: \", scores[0]) #Loss\n print(\"Accuracy: \", scores[1]) #Accuracy\n\n #Saving Model\n model.save(\"./weights/\" + str(d) + \".h5\")\n return len_data, scores[1]", "_____no_output_____" ], [ "FLAccuracy = {}\n# FLAccuracy['Complete Dataset'] = genesis_train('./input/cell_images')\nFLAccuracy['Genesis'] = genesis_train('./input/fed/genesis')\nFLAccuracy['d1'] = update_train('./input/fed/d1', 'd1')\nFLAccuracy['d2'] = update_train('./input/fed/d2', 'd2')\nFLAccuracy['d3'] = update_train('./input/fed/d3', 'd3')\nFLAccuracy['d4'] = update_train('./input/fed/d4', 'd4')\nFLAccuracy['d5'] = update_train('./input/fed/d5', 'd5')\nFLAccuracy['d6'] = update_train('./input/fed/d6', 'd6')\nFLAccuracy['d7'] = update_train('./input/fed/d7', 'd7')\nFLAccuracy['d8'] = update_train('./input/fed/d8', 'd8')\nFLAccuracy['d9'] = update_train('./input/fed/d9', 'd9')\nFLAccuracy['d10'] = update_train('./input/fed/d10', 'd10')\nFLAccuracy['d11'] = update_train('./input/fed/d11', 'd11')\nFLAccuracy['d12'] = update_train('./input/fed/d12', 'd12')\nFLAccuracy['d13'] = update_train('./input/fed/d13', 'd13')\nFLAccuracy['d14'] = update_train('./input/fed/d14', 'd14')\nFLAccuracy['d15'] = update_train('./input/fed/d15', 'd15')\nFLAccuracy['d16'] = update_train('./input/fed/d16', 'd16')\nFLAccuracy['d17'] = update_train('./input/fed/d17', 'd17')\nFLAccuracy['d18'] = update_train('./input/fed/d18', 'd18')\nFLAccuracy['d19'] = update_train('./input/fed/d19', 'd19')\nFLAccuracy['d20'] = update_train('./input/fed/d20', 'd20')", "Reading Training Data\n686 Data Points Read!\n696 Data Points Read!\nReading Testing Data\n2740 Data Points Read!\n2783 Data Points Read!\n1382 Data Points\nEpoch 1/5\n14/14 [==============================] - 2s 135ms/step - loss: 0.6728 - accuracy: 0.6027\nEpoch 2/5\n14/14 [==============================] - 2s 150ms/step - loss: 0.6296 - accuracy: 0.6469\nEpoch 3/5\n14/14 [==============================] - 2s 147ms/step - loss: 0.5789 - accuracy: 0.7142\nEpoch 4/5\n14/14 [==============================] - 2s 149ms/step - loss: 0.5440 - accuracy: 0.7315\nEpoch 5/5\n14/14 [==============================] - 2s 129ms/step - loss: 0.4903 - accuracy: 0.7721\n173/173 [==============================] - 3s 16ms/step - loss: 0.4749 - accuracy: 0.7900\nLoss: 0.47488918900489807\nAccuracy: 0.7899692058563232\nReading Training Data\n528 Data Points Read!\n533 Data Points Read!\nReading Testing Data\n2740 Data Points Read!\n2783 Data Points Read!\n1061 Data Points\nEpoch 1/5\n11/11 [==============================] - 1s 91ms/step - loss: 0.4673 - accuracy: 0.7879\nEpoch 2/5\n11/11 [==============================] - 1s 90ms/step - loss: 0.4264 - accuracy: 0.8162\nEpoch 3/5\n11/11 [==============================] - 1s 96ms/step - loss: 0.3726 - accuracy: 0.8530\nEpoch 4/5\n11/11 [==============================] - 1s 96ms/step - loss: 0.3147 - accuracy: 0.8699\nEpoch 5/5\n11/11 [==============================] - 1s 86ms/step - loss: 0.2670 - accuracy: 0.9057\n173/173 [==============================] - 2s 11ms/step - loss: 0.3025 - accuracy: 0.8749\nLoss: 0.3024612367153168\nAccuracy: 0.8748868107795715\nReading Training Data\n522 Data Points Read!\n528 Data Points Read!\nReading Testing Data\n2740 Data Points Read!\n2783 Data Points Read!\n1050 Data Points\nEpoch 1/5\n11/11 [==============================] - 2s 167ms/step - loss: 0.5450 - accuracy: 0.7448\nEpoch 2/5\n11/11 [==============================] - 1s 100ms/step - loss: 0.4732 - accuracy: 0.7914\nEpoch 3/5\n11/11 [==============================] - 1s 126ms/step - loss: 0.4347 - accuracy: 0.8105\nEpoch 4/5\n11/11 [==============================] - 1s 91ms/step - loss: 0.3879 - accuracy: 0.8419\nEpoch 5/5\n11/11 [==============================] - 2s 153ms/step - loss: 0.3488 - accuracy: 0.8724\n173/173 [==============================] - 3s 15ms/step - loss: 0.3182 - accuracy: 0.8803\nLoss: 0.31824126839637756\nAccuracy: 0.8803186416625977\nReading Training Data\n692 Data Points Read!\n655 Data Points Read!\nReading Testing Data\n2740 Data Points Read!\n2783 Data Points Read!\n1347 Data Points\nEpoch 1/5\n14/14 [==============================] - 2s 122ms/step - loss: 0.5021 - accuracy: 0.7602\nEpoch 2/5\n14/14 [==============================] - 1s 102ms/step - loss: 0.4287 - accuracy: 0.8092\nEpoch 3/5\n14/14 [==============================] - 1s 100ms/step - loss: 0.3923 - accuracy: 0.8448\nEpoch 4/5\n14/14 [==============================] - 1s 102ms/step - loss: 0.3223 - accuracy: 0.8753\nEpoch 5/5\n14/14 [==============================] - 1s 102ms/step - loss: 0.2833 - accuracy: 0.8886\n173/173 [==============================] - 2s 12ms/step - loss: 0.2785 - accuracy: 0.8839\nLoss: 0.27846673130989075\nAccuracy: 0.8839398622512817\nReading Training Data\n448 Data Points Read!\n410 Data Points Read!\nReading Testing Data\n2740 Data Points Read!\n2783 Data Points Read!\n858 Data Points\nEpoch 1/5\n9/9 [==============================] - 1s 92ms/step - loss: 0.5274 - accuracy: 0.7541\nEpoch 2/5\n9/9 [==============================] - 1s 93ms/step - loss: 0.4711 - accuracy: 0.8019\nEpoch 3/5\n9/9 [==============================] - 1s 92ms/step - loss: 0.4223 - accuracy: 0.8135\nEpoch 4/5\n9/9 [==============================] - 1s 92ms/step - loss: 0.3792 - accuracy: 0.8415\nEpoch 5/5\n9/9 [==============================] - 1s 94ms/step - loss: 0.3516 - accuracy: 0.8520\n173/173 [==============================] - 2s 13ms/step - loss: 0.3922 - accuracy: 0.8479\nLoss: 0.392183780670166\nAccuracy: 0.8479087352752686\nReading Training Data\n838 Data Points Read!\n838 Data Points Read!\nReading Testing Data\n2740 Data Points Read!\n2783 Data Points Read!\n1676 Data Points\nEpoch 1/5\n17/17 [==============================] - 2s 102ms/step - loss: 0.5155 - accuracy: 0.7572\nEpoch 2/5\n17/17 [==============================] - 2s 107ms/step - loss: 0.4178 - accuracy: 0.8204\nEpoch 3/5\n17/17 [==============================] - 2s 103ms/step - loss: 0.3393 - accuracy: 0.8646\nEpoch 4/5\n17/17 [==============================] - 2s 104ms/step - loss: 0.2733 - accuracy: 0.9004\nEpoch 5/5\n17/17 [==============================] - 2s 110ms/step - loss: 0.2165 - accuracy: 0.9189\n173/173 [==============================] - 2s 12ms/step - loss: 0.2611 - accuracy: 0.9015\nLoss: 0.26110416650772095\nAccuracy: 0.901502788066864\nReading Training Data\n599 Data Points Read!\n567 Data Points Read!\nReading Testing Data\n2740 Data Points Read!\n2783 Data Points Read!\n1166 Data Points\nEpoch 1/5\n12/12 [==============================] - 1s 103ms/step - loss: 0.6068 - accuracy: 0.6930\nEpoch 2/5\n12/12 [==============================] - 1s 114ms/step - loss: 0.4938 - accuracy: 0.7804\nEpoch 3/5\n12/12 [==============================] - 1s 104ms/step - loss: 0.4533 - accuracy: 0.7985\nEpoch 4/5\n12/12 [==============================] - 2s 137ms/step - loss: 0.4060 - accuracy: 0.8233\nEpoch 5/5\n12/12 [==============================] - 1s 121ms/step - loss: 0.3403 - accuracy: 0.8739\n173/173 [==============================] - 3s 15ms/step - loss: 0.3295 - accuracy: 0.8725\nLoss: 0.3295148015022278\nAccuracy: 0.8725330233573914\nReading Training Data\n418 Data Points Read!\n395 Data Points Read!\nReading Testing Data\n2740 Data Points Read!\n2783 Data Points Read!\n813 Data Points\nEpoch 1/5\n9/9 [==============================] - 1s 99ms/step - loss: 0.5564 - accuracy: 0.7306\nEpoch 2/5\n9/9 [==============================] - 1s 108ms/step - loss: 0.4825 - accuracy: 0.7958\nEpoch 3/5\n9/9 [==============================] - 1s 116ms/step - loss: 0.4358 - accuracy: 0.8266\nEpoch 4/5\n9/9 [==============================] - 1s 161ms/step - loss: 0.3995 - accuracy: 0.8426\nEpoch 5/5\n9/9 [==============================] - 1s 130ms/step - loss: 0.3645 - accuracy: 0.8610\n173/173 [==============================] - 2s 12ms/step - loss: 0.3681 - accuracy: 0.8647\nLoss: 0.3681491017341614\nAccuracy: 0.8647474050521851\nReading Training Data\n716 Data Points Read!\n729 Data Points Read!\nReading Testing Data\n2740 Data Points Read!\n2783 Data Points Read!\n1445 Data Points\nEpoch 1/5\n15/15 [==============================] - 1s 96ms/step - loss: 0.5056 - accuracy: 0.7661\nEpoch 2/5\n15/15 [==============================] - 1s 96ms/step - loss: 0.4308 - accuracy: 0.8187\nEpoch 3/5\n15/15 [==============================] - 1s 93ms/step - loss: 0.3585 - accuracy: 0.8526\nEpoch 4/5\n15/15 [==============================] - 1s 95ms/step - loss: 0.2907 - accuracy: 0.8858\nEpoch 5/5\n15/15 [==============================] - 1s 96ms/step - loss: 0.2397 - accuracy: 0.9038\n173/173 [==============================] - 2s 11ms/step - loss: 0.2601 - accuracy: 0.8972\nLoss: 0.26010987162590027\nAccuracy: 0.8971573710441589\nReading Training Data\n530 Data Points Read!\n572 Data Points Read!\nReading Testing Data\n2740 Data Points Read!\n2783 Data Points Read!\n1102 Data Points\nEpoch 1/5\n12/12 [==============================] - 1s 85ms/step - loss: 0.5228 - accuracy: 0.7486\nEpoch 2/5\n12/12 [==============================] - 1s 86ms/step - loss: 0.4758 - accuracy: 0.7976\nEpoch 3/5\n12/12 [==============================] - 1s 89ms/step - loss: 0.4397 - accuracy: 0.8022\nEpoch 4/5\n12/12 [==============================] - 1s 85ms/step - loss: 0.3850 - accuracy: 0.8348\nEpoch 5/5\n12/12 [==============================] - 1s 85ms/step - loss: 0.3386 - accuracy: 0.8548\n173/173 [==============================] - 2s 11ms/step - loss: 0.2926 - accuracy: 0.8914\nLoss: 0.2926398813724518\nAccuracy: 0.8913633823394775\nReading Training Data\n695 Data Points Read!\n701 Data Points Read!\nReading Testing Data\n2740 Data Points Read!\n2783 Data Points Read!\n1396 Data Points\nEpoch 1/5\n14/14 [==============================] - 1s 88ms/step - loss: 0.4831 - accuracy: 0.7636\nEpoch 2/5\n" ], [ "FLAccuracy", "_____no_output_____" ], [ "FLAccuracyDF = pd.DataFrame.from_dict(FLAccuracy, orient='index', columns=['DataSize', 'Accuracy'])\nFLAccuracyDF", "_____no_output_____" ], [ "FLAccuracyDF.index", "_____no_output_____" ], [ "n = 0\nfor w in FLAccuracy:\n if 'Complete' in w:\n continue\n n += FLAccuracy[w][0]\nprint('Total number of data points in this round: ', n)", "Total number of data points in this round: 22035\n" ], [ "FLAccuracyDF['Weightage'] = FLAccuracyDF['DataSize'].apply(lambda x: x/n)", "_____no_output_____" ], [ "FLAccuracyDF", "_____no_output_____" ], [ "def scale(weight, scaler):\n scaledWeights = []\n for i in range(len(weight)):\n scaledWeights.append(scaler * weight[i])\n return scaledWeights\n\ndef getScaledWeight(d, scaler):\n \n #creating sequential model\n model=Sequential()\n model.add(Conv2D(filters=16,kernel_size=2,padding=\"same\",activation=\"relu\",input_shape=(50,50,3)))\n model.add(MaxPooling2D(pool_size=2))\n model.add(Conv2D(filters=32,kernel_size=2,padding=\"same\",activation=\"relu\"))\n model.add(MaxPooling2D(pool_size=2))\n model.add(Conv2D(filters=64,kernel_size=2,padding=\"same\",activation=\"relu\"))\n model.add(MaxPooling2D(pool_size=2))\n model.add(Dropout(0.2))\n model.add(Flatten())\n model.add(Dense(500,activation=\"relu\"))\n model.add(Dropout(0.2))\n model.add(Dense(2,activation=\"softmax\"))#2 represent output layer neurons \n \n fpath = \"./weights/\"+d+\".h5\"\n model.load_weights(fpath)\n weight = model.get_weights()\n scaledWeight = scale(weight, scaler)\n\n return scaledWeight", "_____no_output_____" ], [ "def avgWeights(scaledWeights):\n avg = list()\n for weight_list_tuple in zip(*scaledWeights):\n layer_mean = tf.math.reduce_sum(weight_list_tuple, axis=0)\n avg.append(layer_mean)\n return avg\n\ndef FedAvg(models):\n \n scaledWeights = []\n for m in models:\n scaledWeights.append(getScaledWeight(m, FLAccuracyDF.loc[m]['Weightage']))\n avgWeight = avgWeights(scaledWeights)\n return avgWeight", "_____no_output_____" ], [ "models = ['d1', 'd2', 'd3', 'd4', 'd5', 'd6', 'd7', 'd8', 'd9', 'd10', 'd11', 'd12', 'd13', 'd14', 'd15', 'd16', 'd17', 'd18', 'd19', 'd20']\navgWeight = FedAvg(models)\nprint(avgWeight)", "[<tf.Tensor: shape=(2, 2, 3, 16), dtype=float32, numpy=\narray([[[[ 6.29709959e-02, 3.05600781e-02, 4.73186262e-02,\n -4.84639332e-02, -1.09765619e-01, -2.40126792e-02,\n -2.00001881e-01, 4.64794785e-03, -9.27647501e-02,\n -1.66123241e-01, 2.24855661e-01, -7.29905367e-02,\n 1.65846452e-01, -1.77608833e-01, 8.64765197e-02,\n -2.02627137e-01],\n [ 1.22623229e-02, -2.69816726e-01, -8.74007195e-02,\n -2.52788186e-01, -1.04290135e-01, 1.57856569e-01,\n -2.61925399e-01, -1.27280667e-01, 1.26418844e-01,\n -1.46524444e-01, -2.43654959e-02, 2.42929533e-01,\n -2.60935754e-01, 2.79806722e-02, 1.43053783e-02,\n -1.92507386e-01],\n [-1.32187217e-01, -1.47404790e-01, -3.16053033e-02,\n 1.14077598e-01, -1.20645165e-01, 1.56122614e-02,\n 2.14044124e-01, 1.66435450e-01, -1.01623848e-01,\n -2.04821333e-01, -9.87461675e-03, -2.57626027e-01,\n -2.19773769e-01, -1.49482757e-01, -1.49648502e-01,\n 4.13678773e-02]],\n\n [[ 4.09911200e-02, -2.55751640e-01, 5.51730059e-02,\n 1.51951760e-01, -2.08946750e-01, -1.78974658e-01,\n -2.33739480e-01, -1.88674644e-01, -2.48390481e-01,\n -8.12981203e-02, -1.72898650e-01, -5.53919859e-02,\n 1.58572555e-01, -1.22390337e-01, 2.73622870e-02,\n 4.19585668e-02],\n [-8.34954083e-02, 2.52027541e-01, -1.80260316e-01,\n -2.27775335e-01, -1.51616588e-01, 2.35146865e-01,\n -2.38066748e-01, 1.14227705e-01, 2.09067702e-01,\n 4.44908515e-02, 9.89529341e-02, -1.02538820e-02,\n 1.72871754e-01, 1.98662151e-02, -6.45739734e-02,\n 8.96787941e-02],\n [ 1.29124120e-01, -2.28320614e-01, 1.61692813e-01,\n 6.46706596e-02, 2.05930516e-01, 9.05648619e-02,\n -2.87533049e-02, 6.75788745e-02, 1.68647338e-02,\n 4.18692678e-02, 2.26347685e-01, 1.76918477e-01,\n 1.28755435e-01, -1.25659034e-01, -1.95176736e-01,\n 2.46681631e-01]]],\n\n\n [[[ 6.76263869e-02, -1.64116889e-01, -4.63423915e-02,\n 1.08722940e-01, -2.42184967e-01, 1.78869694e-01,\n 1.28165051e-01, -1.95564687e-01, 1.54643655e-01,\n -8.88377205e-02, 8.41335952e-02, -2.23800123e-01,\n -1.60318166e-01, 1.83068842e-01, -2.22699732e-01,\n 5.19412458e-02],\n [-1.74145699e-01, 1.64810389e-01, 3.44512388e-02,\n -1.51834458e-01, 2.44864017e-01, -2.07519397e-01,\n -4.64312620e-02, -9.52167138e-02, 1.13456339e-01,\n 2.46316984e-01, -1.51859090e-01, 2.41635039e-01,\n 1.27022326e-01, 2.29096413e-01, -1.57743827e-01,\n 1.86367273e-01],\n [ 2.29843587e-01, -2.43908539e-01, -2.01413512e-01,\n 9.97636020e-02, 2.12529823e-01, 6.52621761e-02,\n 2.25567147e-01, -1.94455549e-01, 5.37216440e-02,\n 9.14809331e-02, -1.86621517e-01, 2.08363146e-01,\n 1.42783493e-01, -1.71516404e-01, 2.87856683e-02,\n 7.70495385e-02]],\n\n [[ 1.47195205e-01, 9.29638669e-02, -2.24783435e-01,\n 1.57111585e-01, 2.19253033e-01, -1.44216925e-01,\n 1.88499019e-01, 2.31585249e-01, 1.90144151e-01,\n -9.07958075e-02, -4.38553058e-02, 1.75386280e-01,\n -2.41612554e-01, -1.50471807e-01, 1.24751553e-01,\n -1.57731324e-01],\n [ 1.61211178e-01, 1.91906020e-01, -1.63448617e-01,\n -1.06442757e-01, -7.65680149e-02, -5.57944737e-02,\n -1.51635736e-01, 9.22414038e-05, 2.16817468e-01,\n -1.81188583e-01, 1.44259129e-02, -8.36728215e-02,\n 1.69681653e-01, 1.00190096e-01, 1.78319663e-01,\n -1.59951687e-01],\n [-4.09052819e-02, -1.97755881e-02, -1.84581861e-01,\n -1.45668117e-02, -1.44612893e-01, 2.52977610e-01,\n 1.50122315e-01, -2.20931277e-01, 3.89067531e-02,\n 2.37456664e-01, 1.47347704e-01, -4.30123396e-02,\n -4.51557077e-02, -5.94768040e-02, -2.90106405e-02,\n -2.41003498e-01]]]], dtype=float32)>, <tf.Tensor: shape=(16,), dtype=float32, numpy=\narray([-0.01007514, -0.001635 , -0.01025345, -0.00173561, 0.03268854,\n -0.00208687, 0.00159445, -0.01452221, -0.00939484, -0.00810049,\n 0.00187787, -0.00531907, -0.01053663, -0.01091921, -0.01560304,\n -0.00814741], dtype=float32)>, <tf.Tensor: shape=(2, 2, 16, 32), dtype=float32, numpy=\narray([[[[-0.09537445, 0.0238525 , 0.08094858, ..., -0.11627594,\n 0.01361733, -0.00270094],\n [ 0.03979139, 0.12243818, -0.10753282, ..., -0.02761981,\n 0.00115342, -0.11250883],\n [-0.00319301, 0.03747182, -0.12272758, ..., -0.17002101,\n -0.09323299, -0.09280318],\n ...,\n [ 0.01198404, 0.18341562, -0.08970661, ..., 0.07275134,\n 0.16204083, -0.11643461],\n [ 0.11521284, 0.09957907, -0.06785449, ..., -0.16049376,\n -0.03478961, 0.05059588],\n [-0.10904927, 0.05084139, -0.11017702, ..., -0.13262197,\n -0.09998756, 0.07306522]],\n\n [[ 0.14296961, 0.14634573, -0.07494281, ..., 0.04981283,\n -0.02226478, -0.00955457],\n [-0.18278444, -0.12257586, 0.1105509 , ..., 0.06898673,\n -0.14074963, 0.12880448],\n [ 0.05062443, 0.09011479, -0.01778495, ..., 0.01384717,\n 0.05328264, 0.01165935],\n ...,\n [ 0.11653248, 0.15053472, 0.02755488, ..., -0.00180184,\n -0.11342274, 0.02041092],\n [ 0.11759499, 0.10058165, 0.1134244 , ..., 0.01960301,\n -0.04442984, -0.04641576],\n [-0.01923554, -0.1412403 , -0.0296952 , ..., 0.01371953,\n 0.06250753, 0.09004614]]],\n\n\n [[[ 0.14302328, -0.03723493, 0.05975141, ..., -0.08695951,\n -0.03743268, -0.16613564],\n [ 0.04879532, 0.01573316, 0.03783104, ..., -0.07207087,\n 0.138336 , 0.06230988],\n [-0.17695434, -0.14341961, 0.1074056 , ..., -0.0374577 ,\n -0.04587403, -0.06571089],\n ...,\n [ 0.08705768, -0.09674174, 0.15470633, ..., 0.07526341,\n 0.14084874, -0.07903387],\n [ 0.11766817, -0.07612219, -0.03450292, ..., -0.17648341,\n -0.08644895, -0.1104345 ],\n [ 0.09279143, -0.14395934, -0.12362929, ..., -0.08168439,\n -0.01325884, 0.06796352]],\n\n [[-0.10889354, 0.09163773, 0.064237 , ..., -0.11828446,\n -0.03442774, 0.11232536],\n [ 0.1001339 , 0.06223914, -0.13083588, ..., 0.05516066,\n -0.17094746, 0.06631851],\n [ 0.13773926, 0.10093588, -0.13178204, ..., -0.05654637,\n -0.06447498, 0.10750612],\n ...,\n [ 0.06421222, 0.1465723 , -0.04136471, ..., 0.05052067,\n -0.06008358, -0.12787679],\n [ 0.04828766, 0.05224356, -0.16164221, ..., 0.10612825,\n -0.11255513, 0.04035288],\n [-0.16791548, 0.06760396, -0.13105269, ..., 0.06637175,\n 0.03395404, 0.00674641]]]], dtype=float32)>, <tf.Tensor: shape=(32,), dtype=float32, numpy=\narray([ 7.8103901e-03, -2.5253366e-03, 9.4068382e-05, -1.0054309e-02,\n 5.6395228e-03, 2.2935962e-02, 3.6243699e-04, -6.3295038e-03,\n 3.1995773e-02, 1.3141254e-02, 8.1295166e-03, -1.7962465e-03,\n -5.8354164e-04, -7.3266337e-03, -8.7775812e-03, 7.3068980e-03,\n -1.2754799e-03, -8.6430637e-03, -1.2573479e-02, -5.5672540e-03,\n -1.1904450e-02, -4.4461302e-03, -1.9995815e-03, -3.0116516e-03,\n -1.4999485e-02, -6.2715444e-03, 1.2082155e-02, 9.9959513e-03,\n -3.4943717e-03, 4.8271101e-03, -4.3909899e-03, -7.2158724e-03],\n dtype=float32)>, <tf.Tensor: shape=(2, 2, 32, 64), dtype=float32, numpy=\narray([[[[-0.05097382, -0.103765 , 0.05987458, ..., -0.05191575,\n -0.00475351, 0.11364381],\n [-0.00168266, -0.03842274, -0.07855407, ..., -0.0831281 ,\n -0.0727298 , 0.05130759],\n [-0.05923396, -0.01078768, 0.02821486, ..., 0.02914551,\n -0.06302543, -0.08857528],\n ...,\n [-0.04730515, -0.01813016, 0.05795706, ..., -0.10225762,\n 0.01166681, 0.09502172],\n [-0.09534874, -0.05579872, -0.04130651, ..., -0.13656466,\n 0.09485541, -0.11062077],\n [-0.05683818, -0.04426903, -0.03588507, ..., -0.09614032,\n 0.03641687, -0.04682203]],\n\n [[ 0.16425094, 0.00555054, 0.00494457, ..., 0.00898783,\n 0.10889069, 0.08391576],\n [-0.02203994, -0.05366965, -0.11370585, ..., 0.02346958,\n -0.05212817, 0.04323069],\n [-0.04689383, -0.03736854, -0.01105933, ..., 0.05111279,\n -0.10411535, -0.07766384],\n ...,\n [ 0.00388702, -0.05325527, -0.08057582, ..., -0.09036542,\n 0.08295781, 0.02695396],\n [-0.06440552, -0.01092236, -0.04536002, ..., -0.07186652,\n 0.0181105 , 0.08185989],\n [-0.0419886 , -0.03008262, -0.05637935, ..., 0.04211152,\n 0.02888373, -0.06597874]]],\n\n\n [[[-0.03833229, -0.10916876, 0.1284822 , ..., -0.05868948,\n -0.12631544, 0.12959538],\n [ 0.04788401, 0.03303488, 0.06145496, ..., -0.09839627,\n 0.03922866, 0.09826408],\n [-0.07018621, 0.10417556, -0.02953644, ..., 0.05394573,\n 0.0176542 , -0.01720765],\n ...,\n [ 0.05617737, -0.05888639, -0.09679458, ..., -0.0468474 ,\n 0.08244231, 0.10589295],\n [ 0.0106239 , -0.01670274, -0.12942371, ..., -0.0671926 ,\n -0.02569613, -0.07168184],\n [ 0.08028605, 0.07828661, -0.0697 , ..., -0.11891251,\n 0.01813116, 0.09038625]],\n\n [[ 0.08789235, -0.01428154, 0.06888534, ..., 0.09440541,\n -0.03437117, -0.10806488],\n [-0.06862493, 0.04934879, -0.04362699, ..., -0.07955035,\n 0.05497896, 0.09841695],\n [-0.10580587, -0.00517587, 0.0524519 , ..., -0.00403526,\n 0.02082359, 0.03606828],\n ...,\n [ 0.1386118 , -0.05597519, -0.08895131, ..., -0.04376757,\n -0.04135654, 0.1048066 ],\n [ 0.1007139 , 0.05583981, -0.03825749, ..., -0.11412786,\n 0.01438947, -0.04390105],\n [ 0.00362133, 0.09658727, -0.03246778, ..., 0.05341358,\n 0.02820618, -0.01911995]]]], dtype=float32)>, <tf.Tensor: shape=(64,), dtype=float32, numpy=\narray([-8.1379199e-03, 6.1860224e-03, 1.6091123e-02, -6.1926767e-03,\n -6.7583038e-03, -9.4345007e-03, -6.0022073e-03, 3.3236535e-03,\n 1.1789045e-02, -8.9633754e-03, 6.6024857e-03, -2.8087604e-03,\n -5.3437511e-03, -5.8863903e-03, -1.5286321e-02, -8.4907245e-03,\n -4.6209171e-03, -1.0798765e-02, 1.3004292e-02, 7.3197032e-03,\n -2.3899684e-03, -5.9073414e-03, 1.2828208e-03, 8.9652007e-03,\n -1.0949758e-02, 1.0419855e-03, -3.2625971e-03, 3.0798630e-03,\n -2.5548907e-03, -9.3173067e-04, -9.3546556e-03, 1.8120963e-03,\n -3.8350022e-03, -1.4084974e-02, -3.6110198e-03, -4.5017223e-03,\n -2.4619487e-03, 1.8209267e-02, -2.8989587e-03, 1.6913021e-03,\n -5.3030080e-03, -5.9248605e-03, 1.1082625e-03, -3.8173990e-03,\n 1.8847305e-03, -8.2327640e-03, -1.7164005e-03, 2.6117486e-05,\n -4.1099940e-03, 1.6918827e-02, -5.7169464e-03, -2.2128168e-03,\n 4.6205113e-04, -5.6640753e-03, -2.1636956e-03, -4.5968518e-03,\n -1.8414368e-03, -1.2497818e-02, -7.9843001e-03, 2.1397090e-03,\n -3.7410008e-03, 3.0753077e-03, -9.8327342e-03, -6.4184298e-03],\n dtype=float32)>, <tf.Tensor: shape=(2304, 500), dtype=float32, numpy=\narray([[-1.5044750e-02, -1.9083844e-02, 4.4547915e-02, ...,\n 2.8617123e-02, -3.0377548e-02, 4.0829502e-05],\n [ 3.3865925e-02, -2.2863064e-02, -1.0030516e-02, ...,\n 1.5757322e-02, -1.9825639e-02, -1.3909874e-02],\n [-1.5962871e-02, 1.8267710e-02, 1.4149060e-03, ...,\n -4.8543759e-02, 2.1052580e-02, -1.3631902e-02],\n ...,\n [-1.1880083e-02, -9.0174275e-03, -1.0868900e-02, ...,\n 6.0880934e-03, -3.2359600e-02, -1.5150492e-02],\n [-1.4395309e-02, 1.9456392e-02, 1.5710445e-02, ...,\n 5.6687263e-03, 8.9815492e-03, 5.4982271e-02],\n [ 1.5056844e-02, -3.3522487e-02, -1.8322496e-02, ...,\n -3.1811804e-02, 2.7293323e-02, 3.3484075e-02]], dtype=float32)>, <tf.Tensor: shape=(500,), dtype=float32, numpy=\narray([-5.53416880e-03, 1.12668620e-02, -2.51924153e-03, -4.17344924e-03,\n 9.82279144e-03, -1.77115912e-03, -6.46361941e-03, -5.59850084e-03,\n 1.04833283e-02, -3.00975540e-03, -8.56306928e-04, -5.20168711e-03,\n 8.98524467e-03, 1.28810275e-02, -4.06666519e-03, -3.82398628e-03,\n -5.54959150e-03, -3.64668365e-03, -7.92631507e-03, -6.75905682e-03,\n -6.28078589e-03, -3.44249629e-03, 8.80345237e-03, 9.11579560e-03,\n -3.39526846e-03, -8.37900490e-03, -3.30694043e-03, -5.70433028e-03,\n -6.35400740e-03, -6.88481797e-03, -1.85096229e-03, -5.58825349e-03,\n 1.12006404e-02, -5.44621237e-03, -2.14726408e-03, -4.09880653e-03,\n -4.19273693e-03, -5.61053818e-03, -4.17135097e-03, -4.33228025e-03,\n 9.04962793e-03, 1.30762206e-02, -5.75472182e-03, 1.13392835e-02,\n -5.61683020e-03, -4.13291575e-03, -3.34750861e-03, -4.06149169e-03,\n -3.66749568e-03, -3.24321515e-03, 9.33618098e-03, -7.22316233e-03,\n 9.70353745e-03, -3.68067622e-03, -6.16257172e-03, 8.20684154e-03,\n -4.89435019e-03, -2.14906642e-04, -4.98367054e-03, 8.90939869e-03,\n 1.21129975e-02, -5.61448559e-03, 1.06549365e-02, -5.29230293e-03,\n -7.12551922e-03, -3.38179525e-03, -1.93966873e-04, 1.35435369e-02,\n -2.14829901e-03, -5.71408169e-03, -5.53259440e-03, 1.09508727e-02,\n 7.46000465e-03, 8.89100228e-03, 1.11365514e-02, 1.13110561e-02,\n -3.79776186e-03, -2.44536810e-03, 8.80534574e-03, 6.33697840e-04,\n 1.35701690e-02, -3.75823537e-03, 8.84520821e-03, 1.01940446e-02,\n 1.18791331e-02, -4.45917575e-03, -5.21652494e-03, 1.00889653e-02,\n 1.04786623e-02, -1.21432624e-03, -4.65361541e-03, -4.73555177e-03,\n -2.50295829e-03, -6.23297552e-03, 1.23668527e-02, -5.57112740e-03,\n -5.67774940e-03, -8.34269915e-03, 1.95718429e-04, -7.68144522e-03,\n 1.63959491e-03, -7.55467964e-03, -6.72129123e-03, -5.64150047e-03,\n -4.72860225e-03, -2.58226250e-03, 1.13667669e-02, -4.72749490e-03,\n -6.87813736e-04, -7.16909440e-03, 1.38499383e-02, -4.85397922e-03,\n -4.04685503e-03, -8.88858456e-03, -3.92845739e-03, -5.49211912e-03,\n 8.65745731e-03, -6.40754169e-03, 8.97939131e-03, -7.40244985e-03,\n -4.83700447e-03, -7.29304156e-04, -3.94110195e-03, -6.02743588e-04,\n 1.11664683e-02, -5.32353530e-03, -2.82435399e-03, -3.30526684e-03,\n -6.55194139e-03, -6.93755038e-03, -1.95045010e-04, -4.21771314e-03,\n -2.50204583e-03, -3.01466952e-03, -6.63121603e-03, -4.80043842e-03,\n -7.46097416e-03, 1.32271247e-02, -5.97341685e-03, -5.82827302e-03,\n -3.68164736e-03, -6.31515449e-03, 9.53376200e-03, -5.07155526e-03,\n -2.95582996e-03, 1.04220407e-02, 1.18372804e-02, -2.46381853e-03,\n -4.80389362e-03, -5.61798224e-03, -5.24300430e-03, 1.54450033e-02,\n 1.32732317e-02, -4.70723212e-03, 1.08032888e-02, -6.82017487e-03,\n -6.28149649e-03, -2.42553814e-03, -8.88851658e-03, -5.68734854e-03,\n -4.40417882e-03, -4.07085195e-03, 1.10611124e-02, 8.81560147e-03,\n 1.23010194e-02, 1.01457806e-02, -2.99730501e-03, 1.13221137e-02,\n -5.19194733e-03, -4.76593524e-03, -3.49803339e-03, -5.18953474e-03,\n 1.03672050e-04, -2.19414127e-03, -6.27203798e-03, -4.62104147e-03,\n 9.67853703e-03, 1.19852079e-02, -4.77695186e-03, -3.02255899e-03,\n -3.44580598e-03, -5.03656501e-03, 1.16737988e-02, -1.01109478e-03,\n 1.50007810e-02, 1.32154869e-02, -5.56759816e-03, -5.62080368e-03,\n 1.20012723e-02, -3.59317590e-03, -4.95124375e-03, 1.28721436e-02,\n 8.88990797e-03, -6.95590302e-03, -3.73836188e-03, -9.03419685e-03,\n -7.40891788e-03, -4.48821840e-04, -3.66688869e-03, -3.19493329e-03,\n 1.01427659e-02, 1.11271814e-02, 1.10702636e-02, -2.11332203e-03,\n 2.97213136e-03, 1.19747408e-02, -5.24236402e-03, -1.82243134e-03,\n -5.73743833e-03, -6.53023412e-03, 8.42849817e-03, -1.75264548e-03,\n -3.80321289e-03, 1.21859685e-02, -5.28530916e-03, 8.93038046e-03,\n -1.85946876e-03, -4.21086838e-03, -4.17282851e-03, 9.28098708e-03,\n 1.21705001e-02, -3.06760427e-03, 9.03287902e-03, 1.28004504e-02,\n 5.14049828e-03, -5.60991513e-03, -5.60030155e-03, 1.09166838e-02,\n -1.18235371e-03, -1.28952414e-03, -4.58185887e-03, -5.13173454e-03,\n -3.54259741e-03, 1.11334594e-02, -3.29254218e-03, -2.63589993e-03,\n -5.60598960e-03, -8.67103878e-03, -2.93730665e-03, 2.26680830e-04,\n -5.56087401e-03, -3.64702521e-03, -1.47829112e-03, 1.20302821e-02,\n -3.81234381e-03, -7.11362157e-03, 1.12991966e-02, -9.81083792e-03,\n -3.74554598e-04, -7.74306990e-03, 1.20938942e-03, -7.87460152e-03,\n -4.29497054e-03, -3.82512342e-03, -5.61306300e-03, -5.59601514e-03,\n -4.65118355e-04, 1.30068157e-02, -6.96778996e-04, -7.90145155e-03,\n -4.68531577e-03, -6.41325163e-03, -9.63451690e-04, -1.03702885e-03,\n -2.45038699e-03, -9.89039708e-03, 1.02759171e-02, -5.18792914e-03,\n 3.90443206e-03, -5.46140177e-03, -5.85449254e-03, -7.75552588e-03,\n -2.65215989e-03, 2.57740874e-04, -4.21085767e-03, 9.33959056e-03,\n -8.25493410e-03, 8.85015074e-03, 1.23032620e-02, -2.43170653e-03,\n -4.67009097e-03, 9.55339614e-03, -6.68688398e-03, 1.02293398e-02,\n -2.21140636e-03, -2.83968076e-03, -6.47431705e-03, -1.17618497e-03,\n -2.06996733e-03, -3.74172023e-03, 8.99831671e-03, 1.34283733e-02,\n -2.95580830e-03, -2.78668827e-03, -5.36573119e-03, 1.10625606e-02,\n 1.20213265e-02, 1.29257431e-02, 1.02709904e-02, 1.46592576e-02,\n -2.87306751e-03, -2.57530925e-03, -1.21975131e-03, -1.98024209e-03,\n -3.65329836e-03, -3.85023374e-03, 1.08012799e-02, 9.48995803e-05,\n -5.53057902e-03, 1.33074366e-03, 8.31840653e-03, -5.04636159e-03,\n -2.35966896e-03, 1.21791437e-02, -8.25849071e-04, 3.88702890e-03,\n 1.00835385e-02, -2.26927036e-03, 1.11112688e-02, 1.07091153e-02,\n -4.75936569e-03, -5.16986288e-03, -1.54961017e-04, -4.29618731e-03,\n -2.62062997e-03, -6.55703200e-03, 1.31747909e-02, 9.15726181e-03,\n -6.96338667e-03, -6.86828885e-03, -2.18273280e-03, -1.80473831e-03,\n -5.26898634e-03, -5.49727399e-03, 1.12049617e-02, -6.60270732e-03,\n -5.09557826e-03, 9.42141470e-03, -6.21194812e-03, -5.80852665e-03,\n -2.03451491e-03, -1.75885821e-03, 1.20100761e-02, 9.89157520e-03,\n 1.17626321e-02, 1.30208796e-02, -9.12684627e-05, 1.14509659e-02,\n 9.74610634e-03, -2.73874239e-03, -4.58180439e-03, -4.26019076e-03,\n -3.16335331e-03, 2.72267079e-03, -5.57976356e-03, -3.09378584e-03,\n -3.35071399e-03, -4.05666325e-03, 1.09317005e-02, -5.98766655e-03,\n -1.52193604e-03, -4.38027596e-03, -6.51899609e-04, -3.89595074e-03,\n -3.28053813e-03, 1.11519285e-02, -2.93060159e-03, 9.72441025e-03,\n -8.85107648e-03, 1.08384751e-02, 1.17983613e-02, -6.72848150e-03,\n 1.06212078e-02, -5.12874452e-03, -3.57506657e-03, -5.36638824e-03,\n -4.11110371e-03, -5.96203050e-03, -2.85185175e-03, -7.19294325e-03,\n -7.81475846e-03, 9.83515382e-03, -9.24775959e-04, -7.53887743e-03,\n -3.22614377e-03, 1.22755012e-02, 1.05221616e-02, -9.61643644e-04,\n 1.17060253e-02, -3.14895273e-03, 1.15561383e-02, -4.81133629e-03,\n -2.31146999e-03, 7.01590907e-03, -3.40830768e-04, -6.03314582e-03,\n 9.08868946e-03, -3.73258116e-03, -1.99783803e-03, -2.79753609e-03,\n -8.24323739e-04, 1.19561637e-02, -3.95200495e-03, -1.26268726e-03,\n -1.55346445e-03, -6.04792219e-03, 7.40239490e-03, -2.95300642e-03,\n 9.86943301e-03, 1.24818468e-02, -2.34776968e-03, -6.97968761e-03,\n -2.50496319e-03, 6.33505266e-03, -2.07575876e-03, -1.63455214e-03,\n 1.00206258e-02, -2.48630135e-03, -3.58426711e-03, -3.98963783e-03,\n 1.01473331e-02, 1.10968156e-02, -4.35665529e-03, 1.21512776e-02,\n -6.98138727e-03, -9.01398889e-04, -5.60927298e-03, -1.45288149e-03,\n -3.87207884e-03, -2.94308970e-03, 1.10045495e-02, -3.27208196e-03,\n 7.61404494e-03, -5.47929248e-03, -3.01020127e-03, -9.55906138e-03,\n -5.53280767e-03, -7.28675572e-04, 5.55282598e-03, -1.76397408e-03,\n 6.81685284e-03, 1.01069128e-02, -5.46161924e-03, -1.28993904e-03,\n -1.05295400e-03, -3.52117140e-03, -5.76895103e-03, 2.91664759e-03,\n -1.51595997e-03, 7.78692542e-03, -1.12719163e-02, -3.46895959e-03,\n 1.00675672e-02, -8.27851531e-04, -5.40378690e-03, -5.57581801e-03,\n -7.43054925e-03, -3.99085227e-03, -2.47309078e-03, -3.02337669e-03,\n -2.71165813e-03, -3.62162798e-04, -5.38837304e-03, -4.36798111e-03,\n -1.64058746e-03, -7.32781878e-03, 1.01796910e-02, -5.91279287e-03,\n 9.76460241e-03, -2.69726431e-03, -5.62081765e-03, -2.54856143e-03,\n -4.46600746e-03, -4.22631390e-03, -9.64728184e-04, 1.30091254e-02,\n 1.24156959e-02, -2.78786267e-03, 9.98625811e-03, 1.08450977e-02,\n 8.82346649e-03, 1.14490669e-02, -5.76793309e-03, 9.75369290e-03,\n -4.55533341e-03, 2.60111806e-03, 1.26153072e-02, -3.95245338e-03,\n 2.82460329e-04, -1.97257055e-03, 9.04539507e-03, -5.60731255e-03,\n -3.03759193e-03, -6.28925627e-03, -4.76786727e-03, 1.24086421e-02,\n -2.92672310e-03, -6.27733255e-03, -4.93518915e-03, -4.00497485e-03],\n dtype=float32)>, <tf.Tensor: shape=(500, 2), dtype=float32, numpy=\narray([[ 0.05311327, 0.09514155],\n [ 0.09232233, -0.02851534],\n [-0.08964266, 0.04173108],\n [ 0.08367732, -0.0923907 ],\n [ 0.07473528, -0.08274475],\n [-0.0627539 , 0.09992003],\n [-0.0911191 , -0.07911095],\n [-0.0498883 , -0.09269869],\n [ 0.07364915, -0.06293585],\n [ 0.04505345, 0.08532324],\n [ 0.03102643, 0.07411335],\n [ 0.03223044, -0.05786802],\n [ 0.10995429, -0.01073887],\n [ 0.10785273, 0.00285203],\n [ 0.04749208, 0.05066685],\n [ 0.0555957 , 0.09664138],\n [-0.00818688, 0.0174538 ],\n [-0.08163548, -0.05619446],\n [ 0.02267669, -0.04380619],\n [-0.01865696, 0.0669779 ],\n [ 0.05982472, 0.05374481],\n [ 0.05908397, -0.05340671],\n [ 0.04229654, -0.0705682 ],\n [ 0.12624112, -0.03905471],\n [-0.08749062, -0.01966353],\n [ 0.02720188, 0.02096189],\n [ 0.06169368, 0.08842259],\n [ 0.01870921, -0.08792486],\n [-0.05300749, -0.06052515],\n [ 0.05659251, -0.05551624],\n [-0.00416558, 0.06536365],\n [-0.01034767, 0.05329547],\n [ 0.07533468, -0.05162623],\n [ 0.03910884, 0.06408792],\n [-0.02865157, 0.03982799],\n [-0.00720344, 0.03931104],\n [ 0.03563602, 0.00198104],\n [ 0.0744648 , -0.02432105],\n [-0.09741265, 0.04142449],\n [-0.02363664, -0.05258123],\n [ 0.12679069, -0.04927159],\n [-0.02962878, -0.12274925],\n [-0.08189599, -0.02983104],\n [ 0.08228916, -0.12541004],\n [-0.01287102, 0.03614439],\n [-0.016284 , 0.06752599],\n [ 0.01914192, 0.04822027],\n [ 0.09245484, -0.08001135],\n [-0.04274287, -0.00859715],\n [ 0.05542322, 0.00793769],\n [ 0.03698973, -0.11074325],\n [ 0.05847845, -0.01122535],\n [ 0.05873084, -0.06315666],\n [-0.01872551, 0.0537422 ],\n [ 0.00085652, -0.0669762 ],\n [ 0.09326971, -0.04454394],\n [ 0.03200875, 0.03614467],\n [ 0.04222843, 0.09592429],\n [-0.0204555 , 0.00569754],\n [ 0.05601981, -0.12443047],\n [ 0.00828085, -0.09333592],\n [-0.05703217, -0.02206847],\n [ 0.07337331, -0.01721186],\n [-0.10884454, -0.00482925],\n [ 0.0913316 , 0.08112097],\n [-0.07608335, 0.00085231],\n [-0.09819711, -0.05532588],\n [ 0.07308225, 0.00092411],\n [-0.05119678, 0.05604082],\n [ 0.07414087, 0.10505526],\n [ 0.01682709, 0.01961546],\n [ 0.08240177, -0.10032496],\n [ 0.04001589, -0.08403035],\n [ 0.12239482, 0.06419824],\n [ 0.1324751 , -0.01727679],\n [-0.02381351, -0.10001826],\n [-0.06213731, 0.07949966],\n [-0.04335392, 0.01332419],\n [ 0.04419218, -0.06350309],\n [-0.04222876, 0.10688141],\n [ 0.07051704, -0.01126709],\n [-0.05057632, 0.00396998],\n [ 0.13414493, -0.005617 ],\n [ 0.12455928, 0.00826711],\n [ 0.12764968, -0.07120249],\n [-0.05176536, 0.07963897],\n [ 0.08401894, 0.03100447],\n [ 0.08777595, -0.04021717],\n [ 0.08009473, -0.10042297],\n [ 0.05165022, 0.10122991],\n [-0.02704764, -0.05683092],\n [-0.07406341, -0.05763072],\n [-0.07700881, 0.11013646],\n [ 0.08613678, 0.01541759],\n [ 0.10774246, -0.0432528 ],\n [-0.01273213, 0.0070588 ],\n [-0.01267115, 0.09487985],\n [-0.01893573, -0.01735934],\n [ 0.07963271, 0.09316237],\n [-0.0849181 , 0.09335002],\n [-0.07189753, 0.06319249],\n [-0.01437461, -0.07791349],\n [-0.07779831, -0.03652881],\n [-0.04608301, 0.05867525],\n [-0.08924111, 0.09480068],\n [-0.10260601, -0.02373477],\n [ 0.11865773, -0.01351118],\n [ 0.0479991 , -0.04756196],\n [-0.00401142, 0.09675419],\n [ 0.00615512, -0.02635094],\n [ 0.03999988, -0.05102275],\n [ 0.07417063, 0.09568918],\n [-0.10557904, -0.06227076],\n [ 0.00559694, -0.06304982],\n [ 0.06008749, 0.0375339 ],\n [-0.06326835, 0.02782372],\n [ 0.12035009, -0.09589054],\n [ 0.00579257, 0.08006887],\n [-0.01508083, -0.09803493],\n [ 0.03642597, -0.00953708],\n [ 0.00085658, 0.05022113],\n [-0.0886467 , -0.03710546],\n [-0.03850847, 0.05948083],\n [-0.10435861, 0.10078683],\n [ 0.10006507, 0.07087514],\n [ 0.00495187, 0.01239195],\n [-0.0808455 , 0.05980099],\n [-0.07445095, 0.10324261],\n [ 0.0441043 , 0.0076606 ],\n [-0.07697976, 0.07268078],\n [-0.07653762, -0.01193937],\n [-0.09638125, -0.07475944],\n [-0.06650464, -0.03050037],\n [-0.05667973, 0.08996452],\n [ 0.05176068, -0.0103077 ],\n [ 0.03983513, 0.04569188],\n [-0.01864344, 0.08519334],\n [ 0.0079924 , -0.1436909 ],\n [-0.06566814, -0.07663368],\n [-0.00131069, -0.07904355],\n [ 0.00564205, 0.04069302],\n [-0.01365841, -0.06190141],\n [ 0.09065706, -0.04532593],\n [-0.07778468, 0.06040839],\n [ 0.07657637, 0.00619171],\n [ 0.12044221, -0.05069051],\n [ 0.10355301, -0.01523891],\n [-0.05932459, -0.02066591],\n [-0.04601621, 0.05558309],\n [-0.02698389, 0.07053442],\n [ 0.07882285, -0.04121957],\n [ 0.06135241, -0.0304644 ],\n [ 0.09189945, 0.00352685],\n [ 0.07324959, 0.00517631],\n [ 0.00789318, -0.08145713],\n [ 0.02677604, 0.06527786],\n [-0.06949059, -0.07066917],\n [-0.06701694, 0.00092053],\n [-0.00497779, -0.07953849],\n [-0.0283235 , 0.03393211],\n [-0.00527409, 0.03929835],\n [-0.03482571, 0.04214784],\n [ 0.13532552, -0.0626914 ],\n [ 0.04993704, -0.08098684],\n [ 0.07999712, -0.06564064],\n [ 0.07224462, 0.00101267],\n [-0.08284859, -0.05882983],\n [ 0.0772713 , -0.02075527],\n [ 0.09509362, -0.04933422],\n [ 0.00703417, -0.05456049],\n [ 0.06934486, 0.10816067],\n [ 0.02769655, 0.03253336],\n [-0.02737729, 0.08668435],\n [-0.04420803, 0.08174775],\n [ 0.01564788, 0.04019317],\n [-0.07019652, -0.04801892],\n [ 0.12309071, -0.02950791],\n [ 0.12868382, -0.12376357],\n [-0.0832961 , -0.07362361],\n [-0.05780158, -0.06723551],\n [-0.04076702, 0.08101598],\n [-0.08256809, -0.05024244],\n [ 0.02858803, -0.08623272],\n [ 0.05343185, 0.06022738],\n [ 0.04897041, -0.12582912],\n [ 0.07588698, -0.04684179],\n [-0.09575359, -0.05143378],\n [-0.00919765, 0.08940119],\n [ 0.04028715, -0.05920773],\n [-0.08991487, -0.06903067],\n [-0.10565867, -0.07443611],\n [ 0.05306325, -0.04204487],\n [ 0.11833616, -0.04841996],\n [-0.00116225, -0.01993424],\n [ 0.03919767, -0.02504915],\n [-0.09598167, -0.09569343],\n [-0.05948884, 0.0847825 ],\n [-0.07438079, 0.0150445 ],\n [ 0.00054377, 0.04857477],\n [ 0.05947721, -0.00051871],\n [-0.00574352, -0.08031838],\n [ 0.03730416, -0.0494786 ],\n [-0.00285258, -0.10425571],\n [-0.06427557, 0.08026918],\n [ 0.09416965, 0.03169006],\n [ 0.09222102, -0.04166833],\n [ 0.00919792, 0.0091698 ],\n [-0.04828366, -0.06778532],\n [-0.00445858, -0.00255824],\n [ 0.0004533 , -0.01015858],\n [ 0.10160092, -0.04120492],\n [-0.03873373, -0.00724124],\n [ 0.05889171, -0.07536945],\n [ 0.00992343, -0.12579918],\n [ 0.09181193, 0.07637873],\n [ 0.12409923, -0.00582372],\n [-0.02855872, 0.08005448],\n [ 0.00997379, 0.0599444 ],\n [-0.01220272, 0.04842511],\n [ 0.12037931, 0.0039286 ],\n [ 0.06772764, -0.13433436],\n [-0.08197935, -0.06232351],\n [ 0.06975443, -0.06791086],\n [ 0.06771931, -0.0870792 ],\n [ 0.10175464, -0.05467491],\n [-0.08869459, -0.03781105],\n [ 0.03419706, -0.02091313],\n [ 0.03717361, -0.05176476],\n [ 0.03977473, 0.0802637 ],\n [ 0.0786541 , -0.07467738],\n [ 0.06886227, 0.06881069],\n [-0.0448051 , -0.04030641],\n [ 0.03404806, 0.10317541],\n [ 0.07326619, -0.06823327],\n [-0.03777633, 0.10320984],\n [-0.10982945, 0.09765613],\n [ 0.09580883, 0.00511308],\n [ 0.00609537, -0.02871921],\n [ 0.01074211, 0.0428532 ],\n [ 0.05898644, -0.04757763],\n [ 0.08945356, -0.04275674],\n [-0.0443637 , -0.00082853],\n [-0.05223919, 0.0191266 ],\n [ 0.10952595, -0.02627584],\n [-0.09769403, 0.04058051],\n [ 0.09491251, -0.07337349],\n [ 0.09357981, -0.07139876],\n [-0.08782497, -0.08310542],\n [-0.02649409, -0.00852232],\n [-0.03283405, -0.02455583],\n [ 0.06076952, 0.01649073],\n [-0.04753735, -0.04494003],\n [ 0.03921098, -0.09155199],\n [ 0.0690461 , 0.07946862],\n [-0.09138182, 0.00620584],\n [ 0.07899081, 0.0901005 ],\n [-0.01421051, 0.06142241],\n [ 0.01492564, -0.09382761],\n [-0.09812864, 0.06189889],\n [-0.03061358, -0.01388041],\n [-0.08795238, 0.06047294],\n [ 0.07104512, -0.05074662],\n [-0.01304952, -0.085006 ],\n [-0.0283862 , 0.06999809],\n [ 0.01572364, 0.07186005],\n [ 0.02568827, -0.00206702],\n [ 0.08745052, -0.05517298],\n [ 0.05046008, -0.0602025 ],\n [-0.03640335, -0.10797785],\n [ 0.08553269, -0.02292269],\n [-0.03778361, 0.06580836],\n [-0.06589452, -0.06769627],\n [-0.03918728, 0.01647712],\n [-0.0237859 , 0.09259149],\n [ 0.0724097 , 0.07180366],\n [ 0.1075801 , -0.02824615],\n [ 0.0331415 , 0.01933192],\n [ 0.00501336, -0.07279122],\n [-0.00366991, -0.10399261],\n [-0.07857811, 0.0787955 ],\n [-0.07753023, 0.05651722],\n [ 0.07203721, -0.12172461],\n [ 0.08254173, -0.0240918 ],\n [ 0.01095617, -0.0724567 ],\n [-0.0530875 , 0.03137068],\n [-0.0066072 , 0.06545337],\n [ 0.07457578, -0.0442367 ],\n [ 0.05588755, 0.10217611],\n [-0.02981724, 0.04941597],\n [ 0.05253782, -0.01042885],\n [ 0.10137051, -0.09172978],\n [ 0.08790664, 0.02524709],\n [-0.04589577, 0.07517756],\n [ 0.08342946, 0.09162924],\n [-0.08593765, 0.06982628],\n [ 0.10655744, -0.05617983],\n [ 0.06092152, -0.12207716],\n [ 0.13419878, -0.02252891],\n [ 0.09834116, 0.03958232],\n [ 0.08437857, 0.01460592],\n [ 0.0879554 , -0.01700424],\n [ 0.00131226, 0.09671631],\n [-0.07696921, -0.00924925],\n [-0.00744633, 0.06004588],\n [ 0.05903571, 0.07920177],\n [-0.06260628, -0.09104627],\n [-0.01939411, -0.11509704],\n [-0.08755022, -0.04919781],\n [ 0.08280303, 0.10103364],\n [-0.09067247, 0.010084 ],\n [ 0.06435605, -0.10556137],\n [ 0.09159714, -0.06761235],\n [-0.03296948, 0.08572628],\n [ 0.11521567, 0.05843401],\n [-0.11013817, 0.04792359],\n [ 0.08529602, -0.09699252],\n [ 0.09589008, -0.07699554],\n [-0.04429969, -0.01039457],\n [-0.04172556, -0.1274764 ],\n [ 0.01776977, -0.09997707],\n [ 0.03752356, -0.09220051],\n [-0.0466691 , -0.06307165],\n [-0.10223605, 0.06639349],\n [ 0.0373287 , 0.06885269],\n [-0.1011653 , 0.01160435],\n [ 0.08323063, 0.030745 ],\n [ 0.05227034, -0.04873146],\n [ 0.02358136, -0.07874562],\n [ 0.06542907, 0.04180999],\n [ 0.05650794, -0.09472515],\n [-0.00625252, 0.05431107],\n [ 0.00902724, 0.05882132],\n [ 0.08372131, 0.10223008],\n [ 0.02161542, 0.02079195],\n [ 0.05030798, -0.11809568],\n [ 0.07213019, 0.02946462],\n [-0.02988404, 0.09049877],\n [ 0.09042504, -0.07592832],\n [-0.09396608, -0.0944846 ],\n [-0.02128671, 0.00555106],\n [ 0.0881349 , -0.01951319],\n [ 0.09815294, 0.02091831],\n [ 0.06768715, -0.00866784],\n [ 0.03008059, -0.03210432],\n [ 0.07697138, -0.10169055],\n [ 0.01389083, -0.1269208 ],\n [-0.10234262, -0.05354223],\n [ 0.10600115, 0.018374 ],\n [ 0.00389425, -0.09662258],\n [-0.06471124, 0.07431647],\n [ 0.08016909, 0.10446185],\n [ 0.05589481, 0.0695638 ],\n [-0.05375014, -0.03225301],\n [ 0.0376894 , 0.00021482],\n [-0.07243066, -0.03724802],\n [-0.08446896, -0.00804608],\n [-0.01091738, 0.03005325],\n [ 0.00425963, 0.09999651],\n [ 0.0634597 , -0.01699882],\n [ 0.0536304 , -0.02552321],\n [-0.09320226, 0.12028917],\n [ 0.01233296, 0.02663163],\n [-0.07040945, 0.07650659],\n [ 0.04895047, -0.04399431],\n [-0.02182859, 0.06619221],\n [ 0.13631342, -0.0910524 ],\n [-0.02438866, 0.04361154],\n [ 0.0196631 , -0.0924203 ],\n [-0.02257976, 0.02791588],\n [ 0.10096157, -0.10916837],\n [ 0.0721221 , -0.13175704],\n [-0.03534862, -0.08653457],\n [ 0.09404437, -0.08704901],\n [ 0.09220192, -0.0581664 ],\n [ 0.0577516 , 0.05761899],\n [ 0.02303884, 0.07061604],\n [-0.07656746, 0.04161585],\n [ 0.06183554, 0.00855878],\n [-0.08804803, -0.03012522],\n [-0.06404053, 0.09230365],\n [ 0.05869053, -0.08241767],\n [ 0.05102699, -0.05720156],\n [-0.07484718, -0.02034107],\n [-0.05531253, -0.0525405 ],\n [ 0.06994882, 0.1004664 ],\n [ 0.12005012, -0.03424909],\n [ 0.03128378, -0.06865919],\n [-0.04526852, 0.09185734],\n [ 0.15100084, -0.01347744],\n [ 0.00988858, -0.08292833],\n [ 0.03963636, -0.10915481],\n [-0.03419256, -0.01783324],\n [-0.08656282, -0.03493471],\n [-0.05074825, -0.09571009],\n [-0.03827269, 0.07175969],\n [ 0.07545518, 0.01717809],\n [ 0.07358796, 0.02868232],\n [ 0.06777788, 0.02454507],\n [-0.08414111, -0.00173827],\n [-0.09479126, 0.07037915],\n [ 0.0302516 , 0.10241633],\n [ 0.01278188, -0.11881058],\n [ 0.00672977, 0.05497678],\n [ 0.01997549, 0.10038829],\n [ 0.03824695, 0.10304374],\n [ 0.06208556, 0.0946699 ],\n [ 0.12233598, -0.03543108],\n [ 0.06428068, 0.07997499],\n [-0.00985603, -0.08305734],\n [ 0.02025353, -0.09543061],\n [-0.09335214, -0.04045701],\n [ 0.05653603, -0.03955551],\n [-0.0212149 , 0.05950537],\n [ 0.05479232, -0.08865862],\n [-0.06271227, -0.00616521],\n [-0.08322059, -0.09444351],\n [ 0.06355092, -0.115833 ],\n [-0.06922431, 0.06530183],\n [ 0.05917469, 0.07849282],\n [ 0.01401479, 0.01351963],\n [ 0.03810755, -0.02678735],\n [ 0.04086413, -0.04378407],\n [-0.08662243, -0.05540965],\n [ 0.06556071, -0.00274828],\n [ 0.09472376, -0.03822417],\n [-0.03384451, 0.01003726],\n [ 0.05466221, -0.06395304],\n [-0.04413132, -0.00833847],\n [-0.10400607, 0.11038367],\n [ 0.05504902, 0.03310199],\n [ 0.0670617 , -0.06417592],\n [ 0.05163915, 0.07122194],\n [ 0.0510257 , -0.05159095],\n [-0.07860655, -0.0385831 ],\n [-0.0425971 , 0.08931515],\n [ 0.06477237, 0.07649194],\n [ 0.0130143 , -0.00678981],\n [ 0.09306058, -0.01254047],\n [ 0.05911999, -0.02476157],\n [-0.05037018, 0.03968317],\n [ 0.09468324, -0.09633217],\n [ 0.09046121, -0.04414934],\n [-0.07896487, -0.01889215],\n [-0.05126528, 0.0645128 ],\n [-0.02817298, -0.01976723],\n [-0.06963074, 0.05060264],\n [ 0.04382402, 0.01743763],\n [ 0.02111326, -0.056915 ],\n [-0.07937434, 0.01911365],\n [ 0.0320007 , -0.0966574 ],\n [ 0.08609178, 0.08823472],\n [-0.05968918, 0.07774922],\n [ 0.13485806, -0.06266385],\n [ 0.02274571, 0.06322207],\n [-0.0048017 , 0.04118194],\n [-0.06662469, -0.07862572],\n [-0.0447496 , 0.06557685],\n [ 0.03626973, -0.0363848 ],\n [-0.0801399 , -0.04140452],\n [-0.0892067 , 0.04529453],\n [-0.04640176, 0.05373997],\n [-0.06597373, 0.09341846],\n [ 0.03246468, -0.06266692],\n [ 0.03325224, 0.04605464],\n [-0.088351 , 0.04260119],\n [-0.09440482, -0.05242742],\n [ 0.12047318, -0.04484326],\n [ 0.0587764 , 0.05491043],\n [ 0.11781491, -0.0247802 ],\n [-0.05252584, 0.0948424 ],\n [-0.09270789, 0.00290092],\n [ 0.02953261, 0.06384847],\n [-0.04930701, -0.00180705],\n [-0.01726307, -0.00485664],\n [-0.00343406, 0.05071843],\n [ 0.08734091, -0.04523054],\n [-0.00492472, -0.10022837],\n [-0.08574443, -0.05268416],\n [-0.00830413, -0.11258492],\n [ 0.12223374, -0.12711501],\n [ 0.05255788, -0.02906406],\n [-0.02451387, -0.10607663],\n [ 0.05887705, 0.1014686 ],\n [ 0.10687933, -0.11090266],\n [ 0.01595546, -0.09647645],\n [ 0.08522485, -0.09018736],\n [ 0.05575566, -0.08034538],\n [-0.01402826, 0.09128823],\n [-0.01914284, 0.07313996],\n [ 0.01201154, 0.0353253 ],\n [ 0.02630525, -0.10031029],\n [-0.00714251, 0.02835639],\n [-0.04130204, 0.04700005],\n [ 0.00765558, 0.08880088],\n [-0.07652104, 0.10528626],\n [-0.03428807, -0.1265489 ],\n [-0.08762529, 0.0153307 ],\n [ 0.01514399, -0.04275274],\n [ 0.08736248, 0.04509316],\n [-0.11950286, -0.00858769]], dtype=float32)>, <tf.Tensor: shape=(2,), dtype=float32, numpy=array([ 0.00663124, -0.00663124], dtype=float32)>]\n" ], [ "def testNewGlobal(weight):\n \n print('Reading Testing Data')\n \n TestParasitizedCells, TestParasitizedLabels = readData('./input/fed/test/Parasitized/', 1)\n TestUninfectedCells, TestUninfectedLabels = readData('./input/fed/test/Uninfected/', 0)\n TestCells = np.concatenate((TestParasitizedCells, TestUninfectedCells))\n TestLabels = np.concatenate((TestParasitizedLabels, TestUninfectedLabels))\n \n \n sTest = np.arange(TestCells.shape[0])\n np.random.shuffle(sTest)\n TestCells = TestCells[sTest]\n TestLabels = TestLabels[sTest]\n \n num_classes=len(np.unique(TestLabels))\n \n (x_test) = TestCells\n (y_test) = TestLabels\n \n # Since we're working on image data, we normalize data by divinding 255.\n x_test = x_test.astype('float32')/255\n test_len=len(x_test)\n \n #Doing One hot encoding as classifier has multiple classes\n y_test=keras.utils.to_categorical(y_test,num_classes)\n \n #creating sequential model\n model=Sequential()\n model.add(Conv2D(filters=16,kernel_size=2,padding=\"same\",activation=\"relu\",input_shape=(50,50,3)))\n model.add(MaxPooling2D(pool_size=2))\n model.add(Conv2D(filters=32,kernel_size=2,padding=\"same\",activation=\"relu\"))\n model.add(MaxPooling2D(pool_size=2))\n model.add(Conv2D(filters=64,kernel_size=2,padding=\"same\",activation=\"relu\"))\n model.add(MaxPooling2D(pool_size=2))\n model.add(Dropout(0.2))\n model.add(Flatten())\n model.add(Dense(500,activation=\"relu\"))\n model.add(Dropout(0.2))\n model.add(Dense(2,activation=\"softmax\"))#2 represent output layer neurons \n# model.summary()\n\n model.set_weights(weight)\n\n # compile the model with loss as categorical_crossentropy and using adam optimizer\n model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])\n \n scores = model.evaluate(x_test, y_test)\n print(\"Loss: \", scores[0]) #Loss\n print(\"Accuracy: \", scores[1]) #Accuracy\n\n #Saving Model\n model.save(\"./output.h5\")\n return scores[1]", "_____no_output_____" ], [ "testNewGlobal(avgWeight)", "Reading Testing Data\n2740 Data Points Read!\n2783 Data Points Read!\n173/173 [==============================] - 2s 13ms/step - loss: 0.3326 - accuracy: 0.8702\nLoss: 0.3326050043106079\nAccuracy: 0.8701792359352112\n" ], [ "FLAccuracyDF", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Loan predictions\n\n## Problem Statement\n\nWe want to automate the loan eligibility process based on customer details that are provided as online application forms are being filled. You can find the dataset [here](https://drive.google.com/file/d/1h_jl9xqqqHflI5PsuiQd_soNYxzFfjKw/view?usp=sharing). These details concern the customer's Gender, Marital Status, Education, Number of Dependents, Income, Loan Amount, Credit History and other things as well. \n\n|Variable| Description|\n|: ------------- |:-------------|\n|Loan_ID| Unique Loan ID|\n|Gender| Male/ Female|\n|Married| Applicant married (Y/N)|\n|Dependents| Number of dependents|\n|Education| Applicant Education (Graduate/ Under Graduate)|\n|Self_Employed| Self employed (Y/N)|\n|ApplicantIncome| Applicant income|\n|CoapplicantIncome| Coapplicant income|\n|LoanAmount| Loan amount in thousands|\n|Loan_Amount_Term| Term of loan in months|\n|Credit_History| credit history meets guidelines|\n|Property_Area| Urban/ Semi Urban/ Rural|\n|Loan_Status| Loan approved (Y/N)\n\n\n\n### Explore the problem in following stages:\n\n1. Hypothesis Generation – understanding the problem better by brainstorming possible factors that can impact the outcome\n2. Data Exploration – looking at categorical and continuous feature summaries and making inferences about the data.\n3. Data Cleaning – imputing missing values in the data and checking for outliers\n4. Feature Engineering – modifying existing variables and creating new ones for analysis\n5. Model Building – making predictive models on the data", "_____no_output_____" ], [ "## 1. Hypothesis Generation\n\nGenerating a hypothesis is a major step in the process of analyzing data. This involves understanding the problem and formulating a meaningful hypothesis about what could potentially have a good impact on the outcome. This is done BEFORE looking at the data, and we end up creating a laundry list of the different analyses which we can potentially perform if data is available.\n\n#### Possible hypotheses\nWhich applicants are more likely to get a loan\n\n1. Applicants having a credit history \n2. Applicants with higher applicant and co-applicant incomes\n3. Applicants with higher education level\n4. Properties in urban areas with high growth perspectives\n\nDo more brainstorming and create some hypotheses of your own. Remember that the data might not be sufficient to test all of these, but forming these enables a better understanding of the problem.", "_____no_output_____" ] ], [ [ "import pandas as pd\nimport numpy as np\nfrom matplotlib import pyplot as plt", "_____no_output_____" ], [ "df = pd.read_csv('data.csv')", "_____no_output_____" ], [ "df.head(10)", "_____no_output_____" ], [ "df.shape", "_____no_output_____" ] ], [ [ "## 2. Data Exploration\nLet's do some basic data exploration here and come up with some inferences about the data. Go ahead and try to figure out some irregularities and address them in the next section. ", "_____no_output_____" ], [ "One of the key challenges in any data set are missing values. Lets start by checking which columns contain missing values.", "_____no_output_____" ] ], [ [ "df.isnull().sum()", "_____no_output_____" ] ], [ [ "Look at some basic statistics for numerical variables.", "_____no_output_____" ] ], [ [ "df.dtypes", "_____no_output_____" ], [ "df.nunique()", "_____no_output_____" ] ], [ [ "1. How many applicants have a `Credit_History`? (`Credit_History` has value 1 for those who have a credit history and 0 otherwise)\n2. Is the `ApplicantIncome` distribution in line with your expectation? Similarly, what about `CoapplicantIncome`?\n3. Tip: Can you see a possible skewness in the data by comparing the mean to the median, i.e. the 50% figure of a feature.\n\n", "_____no_output_____" ], [ "Let's discuss nominal (categorical) variable. Look at the number of unique values in each of them.", "_____no_output_____" ], [ "Explore further using the frequency of different categories in each nominal variable. Exclude the ID obvious reasons.", "_____no_output_____" ], [ "### Distribution analysis\n\nStudy distribution of various variables. Plot the histogram of ApplicantIncome, try different number of bins.\n\n", "_____no_output_____" ], [ "\nLook at box plots to understand the distributions. ", "_____no_output_____" ], [ "Look at the distribution of income segregated by `Education`", "_____no_output_____" ], [ "Look at the histogram and boxplot of LoanAmount", "_____no_output_____" ], [ "There might be some extreme values. Both `ApplicantIncome` and `LoanAmount` require some amount of data munging. `LoanAmount` has missing and well as extreme values values, while `ApplicantIncome` has a few extreme values, which demand deeper understanding. ", "_____no_output_____" ], [ "### Categorical variable analysis\n\nTry to understand categorical variables in more details using `pandas.DataFrame.pivot_table` and some visualizations.", "_____no_output_____" ], [ "## 3. Data Cleaning\n\nThis step typically involves imputing missing values and treating outliers. ", "_____no_output_____" ], [ "### Imputing Missing Values\n\nMissing values may not always be NaNs. For instance, the `Loan_Amount_Term` might be 0, which does not make sense.\n\n", "_____no_output_____" ], [ "Impute missing values for all columns. Use the values which you find most meaningful (mean, mode, median, zero.... maybe different mean values for different groups)", "_____no_output_____" ] ], [ [ "from sklearn.impute import SimpleImputer\nfrom sklearn.preprocessing import StandardScaler\nfrom sklearn.pipeline import Pipeline\nfrom sklearn.preprocessing import FunctionTransformer\nfrom sklearn.metrics import accuracy_score\nfrom sklearn.compose import ColumnTransformer\nfrom sklearn.preprocessing import OneHotEncoder\nfrom sklearn.model_selection import GridSearchCV\nfrom sklearn.svm import SVC\nfrom sklearn.linear_model import LogisticRegression\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.compose import make_column_selector\nimport pickle\n", "_____no_output_____" ], [ "ohe=OneHotEncoder(drop='first', sparse=False)", "_____no_output_____" ], [ "X= df.drop(columns=['Loan_Status', 'Loan_ID'])\ny=df['Loan_Status']\n#y=ohe.fit_transform(df['Loan_Status'].to_numpy().reshape(-1, 1))", "_____no_output_____" ], [ "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3,\n random_state=0)", "_____no_output_____" ], [ "column_trans = ColumnTransformer(\n[('imp_most_frequent', SimpleImputer(strategy='most_frequent'), ['Gender', 'Married', \n 'Dependents', 'Self_Employed',\n 'Property_Area',\n 'Education',\n 'Credit_History']),\n ('imp_median', SimpleImputer(strategy='median'), ['LoanAmount', 'Loan_Amount_Term']),\n\n('scaling', StandardScaler(), make_column_selector(dtype_include=np.number))\n \n]\n)\n\ncolumn_enc = ColumnTransformer([('one_hot_enc', OneHotEncoder(handle_unknown='ignore'), \n [0,\n 1,\n 2,\n 3,\n 4,\n 10])])\n ", "_____no_output_____" ], [ "pipeline = Pipeline(steps=[('column_transf', column_trans),\n ('column_enc', column_enc),\n ('classifier', SVC(random_state = 17))])", "_____no_output_____" ], [ "# Find the best hyperparameters using GridSearchCV on the train set\nparam_grid = [\n {'classifier':(SVC(random_state = 17),),\n 'classifier__C': [0.5, 1, 2, 4], \n 'classifier__kernel': ['linear', 'poly', 'rbf', 'sigmoid'],\n 'classifier__degree': [2, 3],\n 'classifier__gamma':['scale', 'auto']},\n \n {'classifier':(LogisticRegression(random_state = 17),),\n 'classifier__penalty':['l1', 'l2', 'elasticnet'],\n 'classifier__C': [0.5, 1, 2, 4],\n 'classifier__solver': ['newton-cg', 'lbfgs', 'liblinear', 'sag', 'saga']\n }]\n\ngrid = GridSearchCV(pipeline, param_grid=param_grid, verbose=2, n_jobs=5, scoring='accuracy')\ngrid.fit(X_train, y_train)", "Fitting 5 folds for each of 124 candidates, totalling 620 fits\n" ], [ "grid.best_params_", "_____no_output_____" ], [ "grid.best_score_", "_____no_output_____" ], [ "y_pred=grid.predict(X_test)", "_____no_output_____" ], [ "accuracy_score(y_test, y_pred)", "_____no_output_____" ], [ "pipeline = Pipeline(steps=[('column_transf', column_trans),\n ('column_enc', column_enc),\n ('classifier', grid.best_params_['classifier'])])", "_____no_output_____" ], [ "pipeline.fit(X_train, y_train)", "_____no_output_____" ], [ "with open('myfile.pickle', 'wb') as file_handle:\n pickle.dump(pipeline, file_handle)", "_____no_output_____" ] ], [ [ "### Extreme values\nTry a log transformation to get rid of the extreme values in `LoanAmount`. Plot the histogram before and after the transformation", "_____no_output_____" ], [ "Combine both incomes as total income and take a log transformation of the same.", "_____no_output_____" ], [ "## 4. Building a Predictive Model", "_____no_output_____" ], [ "Try paramater grid search to improve the results", "_____no_output_____" ], [ "## 5. Using Pipeline\nIf you didn't use pipelines before, transform your data prep, feat. engineering and modeling steps into Pipeline. It will be helpful for deployment.\n\nThe goal here is to create the pipeline that will take one row of our dataset and predict the probability of being granted a loan.\n\n`pipeline.predict(x)`", "_____no_output_____" ], [ "## 6. Deploy your model to cloud and test it with PostMan, BASH or Python", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown", "markdown" ], [ "code", "code", "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown", "markdown", "markdown", "markdown", "markdown", "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Looking at the randomness (or otherwise) of mouse behaviour\n### Also, the randomness (or otherwise) of trial types to know when best to start looking at 'full task' behaviour", "_____no_output_____" ] ], [ [ "# Import libraries\nimport matplotlib.pyplot as plt\n%matplotlib inline\nimport pandas as pd\nimport seaborn as sns \nimport random\nimport copy\nimport numpy as np\nfrom scipy.signal import resample\nfrom scipy.stats import zscore\nfrom scipy import interp\nfrom sklearn.linear_model import LogisticRegression\nfrom sklearn.metrics import confusion_matrix\nfrom sklearn.metrics import accuracy_score\nfrom sklearn import metrics\nfrom sklearn import cross_validation", "_____no_output_____" ], [ "# Load data\n# data loading function\ndef data_load_and_parse(mouse_name):\n tt = pd.read_csv('~/work/whiskfree/data/trialtype_' + mouse_name + '.csv',header=None)\n ch = pd.read_csv('~/work/whiskfree/data/choice_' + mouse_name + '.csv',header=None)\n sess = pd.read_csv('~/work/whiskfree/data/session_' + mouse_name + '.csv',header=None)\n AB = pd.read_csv('~/work/whiskfree/data/AB_' + mouse_name + '.csv',header=None)\n \n clean1 = np.nan_to_num(tt) !=0\n clean2 = np.nan_to_num(ch) !=0\n clean = clean1&clean2\n tt_c = tt[clean].values\n\n ch_c = ch[clean].values\n \n s_c = sess[clean].values\n \n ab_c = AB[clean].values\n \n return tt_c, ch_c, clean, s_c, ab_c", "_____no_output_____" ], [ "mouse_name = '36_r'\ntt, ch, clean, sess, AB = data_load_and_parse(mouse_name)\n\n\n# work out AB/ON trials\nAB_pol = np.nan_to_num(AB) !=0\nON_pol = np.nan_to_num(AB) ==0\ncm_AB = confusion_matrix(tt[AB_pol],ch[AB_pol])\ncm_ON = confusion_matrix(tt[ON_pol],ch[ON_pol])\nprint(cm_AB)\nprint(cm_ON)\nprint(accuracy_score(tt[AB_pol],ch[AB_pol]))\nprint(accuracy_score(tt[ON_pol],ch[ON_pol]))", "[[293 42 47]\n [ 63 213 42]\n [ 85 25 220]]\n[[180 23 28]\n [ 41 139 26]\n [ 52 47 198]]\n0.704854368932\n0.704359673025\n" ], [ "# Format TT/ choice data and plot\nfig, ax = plt.subplots(2,1,figsize=(20,5))\n_ = ax[0].plot(tt[ON_pol][:100],label='TT ON')\n_ = ax[0].plot(ch[ON_pol][:100],label='Ch ON')\nax[0].legend()\n_ = ax[1].plot(tt[AB_pol][:100],label='TT AB')\n_ = ax[1].plot(ch[AB_pol][:100],label='Ch AB')\nax[1].legend()", "_____no_output_____" ], [ "# Measure randomness and plot that\n# First plot cumsum of trial types. Periods of bias (of choice 1 and 3, anyway) will be seen as deviations from the mean line\nplt.plot(np.cumsum(tt[AB_pol][:100]),label='Cumsum TT AB')\nplt.plot(np.cumsum(ch[AB_pol][:100]),label='Cumsum Ch AB')\nplt.plot([0,99],[0,np.sum(tt[AB_pol][:100])],label='Mean cumsum')\nplt.legend()", "_____no_output_____" ], [ "# How about looking at the distribution of individual states, pairs, triples. \n# Compare to random sequence (with no conditions)\nP_i = np.zeros(3)\nP_i[0] = len(tt[tt[AB_pol]==1])\nP_i[1] = len(tt[tt[AB_pol]==2])\nP_i[2] = len(tt[tt[AB_pol]==3])\nwith sns.axes_style(\"white\"):\n _ = plt.imshow(np.expand_dims(P_i/sum(P_i),axis=0),interpolation='none')\n for j in range(0,3):\n plt.text(j, 0, P_i[j]/sum(P_i), va='center', ha='center',bbox=dict(facecolor='white',edgecolor='white', alpha=0.5))\n\n# _ = ax[1].bar([0,1,2],P_i/sum(P_i))\n\n", "/Users/mathew/miniconda/envs/graph_tool/lib/python3.4/site-packages/ipykernel/__main__.py:4: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 2050 but corresponding boolean dimension is 1261\n/Users/mathew/miniconda/envs/graph_tool/lib/python3.4/site-packages/ipykernel/__main__.py:5: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 2050 but corresponding boolean dimension is 1261\n/Users/mathew/miniconda/envs/graph_tool/lib/python3.4/site-packages/ipykernel/__main__.py:6: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 2050 but corresponding boolean dimension is 1261\n" ], [ "# Pairs and triples (in dumb O(n) format)\nP_ij = np.zeros([3,3])\nP_ijk = np.zeros([3,3,3])\nfor i in range(len(tt[AB_pol]) - 2):\n #p_i = tt[AB_pol][i]\n #p_j = tt[AB_pol][i+1]\n #p_k = tt[AB_pol][i+2]\n p_i = ch[AB_pol][i]\n p_j = ch[AB_pol][i+1]\n p_k = ch[AB_pol][i+2]\n P_ij[p_i-1,p_j-1] += 1\n P_ijk[p_i-1,p_j-1,[p_k-1]] += 1\n \ncmap = sns.diverging_palette(220,10, l=70, as_cmap=True, center=\"dark\") # blue to green via black\n\nwith sns.axes_style(\"white\"):\n plt.imshow(P_ij/np.sum(P_ij),interpolation='none',cmap=cmap)\n \nfor i in range(0,3):\n for j in range(0,3):\n plt.text(j, i, \"{0:.2f}\".format(P_ij[i,j]/np.sum(P_ij)*9), va='center', ha='center',bbox=dict(facecolor='white',edgecolor='white', alpha=0.5))\n\n#plt.savefig('figs/graphs/state_transition_matrix_AB'+ mouse_name +'.png')\nplt.savefig('figs/graphs/choice_state_transition_matrix_AB'+ mouse_name +'.png')", "_____no_output_____" ], [ "# Plot P(state) for all 27 triple states\nplt.plot(P_ijk_ON.flatten()/np.sum(P_ijk_ON))\nplt.plot([0,26],[1/27,1/27],'--')\n1/27", "_____no_output_____" ], [ "import graph_tool.all as gt\n", "/Users/mathew/miniconda/envs/graph_tool/lib/python3.4/site-packages/graph_tool/draw/cairo_draw.py:1468: RuntimeWarning: Error importing Gtk module: No module named 'gi'; GTK+ drawing will not work.\n warnings.warn(msg, RuntimeWarning)\n" ], [ "# Transition probabilities between individual states, pairs, triples\ng = gt.Graph()\ng.add_edge_list(np.transpose(P_ij.nonzero()))\nwith sns.axes_style(\"white\"):\n\tplt.imshow(P_ij,interpolation='none')", "_____no_output_____" ], [ "g = gt.Graph(directed = True)\ng.add_vertex(len(P_ij))\nedge_weights = g.new_edge_property('double')\nedge_labels = g.new_edge_property('string')\nfor i in range(P_ij.shape[0]):\n for j in range(P_ij.shape[1]):\n e = g.add_edge(i, j)\n edge_weights[e] = P_ij[i,j]\n edge_labels[e] = str(P_ij[i,j])\n ", "_____no_output_____" ], [ "# Fancy drawing code where node colour/size is degree. Edge colour/size is betweenness\ndeg = g.degree_property_map(\"in\")\n# deg.a = 4 * (np.sqrt(deg.a) * 0.5 + 0.4)\ndeg.a = deg.a*20\nprint(deg.a)\newidth = edge_weights.a / 10\n#ebet.a /= ebet.a.max() / 10.\n#print(ebet.a)\npos = gt.sfdp_layout(g)\n#control = g.new_edge_property(\"vector<double>\")\n#for e in g.edges():\n# d = np.sqrt(sum((pos[e.source()].a - pos[e.target()].a) ** 2))\n# print(d)\n# control[e] = [10, d, 10,d] #[0.3, d, 0.7, d]\n \n \n\ncmap = sns.cubehelix_palette(as_cmap=True) # cubehelix \ncmap = sns.diverging_palette(220,10, l=70, as_cmap=True, center=\"dark\") # blue to green via black\n# gt.graph_draw(g, pos=pos, vertex_size=deg, vertex_fill_color=deg, vorder=deg,\n# edge_color=ebet, eorder=eorder, edge_pen_width=ebet,\n# edge_control_points=control) # some curvy edges\n # output=\"graph-draw.pdf\")\ngt.graph_draw(g, pos=pos, vertex_size=deg, vertex_color=deg, vertex_fill_color=deg, edge_color=edge_weights, edge_text=edge_labels,\n vcmap=cmap,ecmap=cmap, vertex_text=g.vertex_index, vertex_font_size=18,fit_view=0.5)\n #vcmap=plt.cm.Pastel1,ecmap=plt.cm.Pastel1 )\n # edge_control_points=control) # some curvy edges\n # output=\"graph-draw.pdf\")", "[60 60 60]\n" ], [ "# Same as g but normalised so total trials/9 = 1\ng_n = gt.Graph(directed = True)\n\nedge_weights_n = g_n.new_edge_property('double')\nedge_labels_n = g_n.new_edge_property('string')\nnode_size_n = g_n.new_vertex_property('double')\ng_n.add_vertex(len(P_ij))\n\nP_ij_n = P_ij /(P_ij.sum()/9)\nfor i in range(P_ij.shape[0]):\n #v = g_n.add_vertex()\n node_size_n[i] = 3* sum(P_ij)[i] / np.sum(P_ij)\n for j in range(P_ij.shape[1]):\n e = g_n.add_edge(i, j)\n edge_weights_n[e] = P_ij_n[i,j]\n edge_labels_n[e] = \"{0:.2f}\".format(P_ij_n[i,j])", "_____no_output_____" ], [ "# Minimal drawing code, but with scaled colours/weights for network properties\n# Line width changes on each loop ATM. Needs fixing..\npos = gt.sfdp_layout(g_n) \n\n#deg_n = g_n.degree_property_map(\"in\")\n# deg.a = 4 * (np.sqrt(deg.a) * 0.5 + 0.4)\n#deg_n.a = deg_n.a*20\nn_size = copy.copy(node_size_n)\nn_size.a = 50* n_size.a/ max(n_size.a)\n\nedge_w = copy.copy(edge_weights_n)\nedge_w.a = edge_w.a*10\n\ncmap = sns.cubehelix_palette(as_cmap=True) # cubehelix \ncmap = sns.diverging_palette(220,10, l=70, as_cmap=True, center=\"dark\") # blue to green via black\n\ngt.graph_draw(g_n, pos=pos, vertex_color = n_size, vertex_fill_color = n_size, \n vertex_size = n_size,\n edge_pen_width=edge_w, edge_color=edge_weights_n, \n edge_text=edge_labels_n,\n vcmap=cmap,ecmap=cmap, \n vertex_text=g_n.vertex_index, \n vertex_font_size=18,\n output_size=(600,600), fit_view=0.4,\n output=\"figs/graphs/choice_1st_order_transition_AB.pdf\")\n #vcmap=plt.cm.Pastel1,ecmap=plt.cm.Pastel1 )\n # edge_control_points=control) # some curvy edges\n # output=\"graph-draw.pdf\")", "_____no_output_____" ], [ "current_palette = sns.color_palette(\"cubehelix\")\ncurrent_palette = sns.diverging_palette(220,10, l=50, n=7, center=\"dark\")\nsns.palplot(current_palette)\n\n", "_____no_output_____" ], [ "# Now write a loop to construct a tree-type graph\n# Same as g but normalised so total trials/9 = 1\nt = gt.Graph(directed = False)\n\nP_ij_n = P_ij /(P_ij.sum()/9)\nP_ijk_n = P_ijk /(P_ijk.sum()/27)\n\nedge_weights_t = t.new_edge_property('double')\nedge_labels_t = t.new_edge_property('string')\nnode_labels_t = t.new_vertex_property('string')\nnode_size = t.new_vertex_property('double')\nh = t.add_vertex()\nnode_labels_t[h] = \"0\"\n\nfor i in range(P_ij.shape[0]):\n v = t.add_vertex()\n node_labels_t[v] = str(i)\n e = t.add_edge(h,v)\n node_size[v] = sum(P_ij_n)[i] *10\n \n for j in range(P_ij.shape[1]):\n v2 = t.add_vertex()\n node_labels_t[v2] = str(i) + \"-\" + str(j)\n e = t.add_edge(v,v2)\n \n edge_weights_t[e] = P_ij_n[i,j]*10\n edge_labels_t[e] = \"{0:.2f}\".format(P_ij_n[i,j])\n node_size[v2] = P_ij_n[i,j]*20\n \n for k in range(P_ijk.shape[2]):\n v3 = t.add_vertex()\n node_labels_t[v3] = str(i) + \"-\" + str(j) + \"-\" + str(k)\n e2 = t.add_edge(v2,v3)\n \n edge_weights_t[e2] = P_ijk_n[i,j,k]*10\n edge_labels_t[e2] = \"{0:.2f}\".format(P_ijk_n[i,j,k])\n node_size[v3] = P_ijk_n[i,j,k]*20\n", "_____no_output_____" ], [ "#pos = gt.sfdp_layout(t) \n#pos = gt.fruchterman_reingold_layout(t)\npos = gt.radial_tree_layout(t,t.vertex(0))\ncmap = sns.diverging_palette(220,10, l=70, as_cmap=True, center=\"dark\") # blue to green via black\n\ngt.graph_draw(t,pos=pos,vertex_size=node_size,edge_pen_width=edge_weights_t,\n vertex_text = node_labels_t, edge_text=edge_labels_t,\n ecmap=cmap, edge_color = edge_weights_t,\n vcmap=cmap, vertex_color = node_size,vertex_fill_color = node_size,\n output_size=(1000, 1000), fit_view=0.8,\n output=\"figs/graphs/choice_3_step_statespace_AB.pdf\")", "_____no_output_____" ], [ "\"{0:.2f}\".format(P_ijk[1,1,1])", "_____no_output_____" ], [ "\"{0:.2f}\".format(P_ijk[1,1,1])", "_____no_output_____" ], [ "len(P_ij)", "_____no_output_____" ] ], [ [ "# Repeat the trick for ON policy trials", "_____no_output_____" ] ], [ [ "# P_ijk_ON\nP_ij_ON = np.zeros([3,3])\nP_ijk_ON = np.zeros([3,3,3])\nfor i in range(len(tt[AB_pol]) - 2):\n# p_i = tt[ON_pol][i]\n# p_j = tt[ON_pol][i+1]\n# p_k = tt[ON_pol][i+2]\n p_i = ch[AB_pol][i]\n p_j = ch[AB_pol][i+1]\n p_k = ch[AB_pol][i+2]\n P_ij_ON[p_i-1,p_j-1] += 1\n P_ijk_ON[p_i-1,p_j-1,[p_k-1]] += 1\n\n# Make graph\nt_ON = gt.Graph(directed = False)\n\nP_ij_ON = P_ij_ON /(P_ij_ON.sum()/9)\nP_ijk_ON = P_ijk_ON /(P_ijk_ON.sum()/27)\n\nedge_weights_tON = t_ON.new_edge_property('double')\nedge_labels_tON = t_ON.new_edge_property('string')\nnode_labels_tON = t_ON.new_vertex_property('string')\nnode_size_ON = t_ON.new_vertex_property('double')\nh = t_ON.add_vertex()\nnode_labels_tON[h] = \"0\"\n\nfor i in range(P_ij_ON.shape[0]):\n v = t_ON.add_vertex()\n node_labels_tON[v] = str(i)\n e = t_ON.add_edge(h,v)\n node_size_ON[v] = sum(P_ij_ON)[i] *10\n \n for j in range(P_ij_ON.shape[1]):\n v2 = t_ON.add_vertex()\n node_labels_tON[v2] = str(i) + \"-\" + str(j)\n e = t_ON.add_edge(v,v2)\n \n edge_weights_tON[e] = P_ij_ON[i,j]*10\n edge_labels_tON[e] = \"{0:.2f}\".format(P_ij_ON[i,j])\n node_size_ON[v2] = P_ij_ON[i,j]*20\n \n for k in range(P_ijk_ON.shape[2]):\n v3 = t_ON.add_vertex()\n node_labels_tON[v3] = str(i) + \"-\" + str(j) + \"-\" + str(k)\n e2 = t_ON.add_edge(v2,v3)\n \n edge_weights_tON[e2] = P_ijk_ON[i,j,k]*10\n edge_labels_tON[e2] = \"{0:.2f}\".format(P_ijk_ON[i,j,k])\n node_size_ON[v3] = P_ijk_ON[i,j,k]*20\n\n# Plot graph\npos = gt.radial_tree_layout(t_ON,t_ON.vertex(0))\ncmap = sns.diverging_palette(220,10, l=70, as_cmap=True, center=\"dark\") # blue to green via black\n\ngt.graph_draw(t_ON,pos=pos,vertex_size=node_size_ON,edge_pen_width=edge_weights_tON,\n vertex_text = node_labels_tON, edge_text=edge_labels_tON,\n ecmap=cmap, edge_color = edge_weights_tON,\n vcmap=cmap, vertex_color = node_size_ON,\n vertex_fill_color = node_size_ON,\n output_size=(1000, 1000), fit_view=0.8)\n# output=\"figs/graphs/choice_3_step_statespace_AB_\"+ mouse_name +\".pdf\")\n", "_____no_output_____" ], [ "# image of ON trials transition matrix\ncmap = sns.diverging_palette(220,10, l=70, as_cmap=True, center=\"dark\") # blue to green via black\n\nwith sns.axes_style(\"white\"):\n plt.imshow(P_ij_ON/np.sum(P_ij_ON),interpolation='none',cmap=cmap)\n \nfor i in range(0,3):\n for j in range(0,3):\n plt.text(j, i, \"{0:.2f}\".format(P_ij_ON[i,j]), va='center', ha='center',bbox=dict(facecolor='white',edgecolor='white', alpha=0.5))\n\nylabels = ['Anterior','Posterior','No Go']\nplt.xticks([0,1,2],ylabels)\nplt.yticks([0,1,2],ylabels)\n# plt.set_yticks([0,1,2])\n# plt.set_yticklabels(ylabels)\n \n# plt.savefig('figs/graphs/choice_state_transition_matrix_AB_'+ mouse_name +'.png')", "_____no_output_____" ], [ "# Just plot P(state)\nplt.figure(figsize=(16,2))\nax1 = plt.subplot2grid((1,4),(0,0))\nax1.plot(P_ij_ON.flatten()/np.sum(P_ij_ON) * 9)\nax1.plot([0,8],[1,1],'--')\n\nstate_names = np.empty([3,3],dtype=object)\nfor i in range(0,3):\n for j in range(0,3):\n state_names[i,j] = str(i) + \"-\" + str(j)\n \nax1.set_xticks(range(0,9))\nax1.set_xticklabels(state_names.flatten(),rotation=45)\n\nax2 = plt.subplot2grid((1,4),(0,1),colspan=3)\nax2.plot(P_ijk_ON.flatten()/np.sum(P_ijk_ON) * 27)\nax2.plot([0,26],[1,1],'--')\n\nstate_names = np.empty([3,3,3],dtype=object)\nfor i in range(0,3):\n for j in range(0,3):\n for k in range(0,3):\n state_names[i,j,k] = str(i) + \"-\" + str(j) + \"-\" + str(k)\n \n_ = ax2.set_xticks(range(0,27))\n_ = ax2.set_xticklabels(state_names.flatten(),rotation=45)\n\nplt.tight_layout()\n\nplt.savefig('figs/graphs/CH_state_prob_AB_'+ mouse_name +'.png')", "_____no_output_____" ], [ "from scipy.stats import chisquare\n# chisquare(P_ij_ON.flatten())\nchisquare?", "_____no_output_____" ], [ "# First order transition graph\ng_ON = gt.Graph(directed = True)\n\nedge_weights_ON = g_ON.new_edge_property('double')\nedge_labels_ON = g_ON.new_edge_property('string')\nnode_size_ON = g_ON.new_vertex_property('double')\ng_ON.add_vertex(len(P_ij_ON))\n\nfor i in range(P_ij_ON.shape[0]):\n #v = g_n.add_vertex()\n node_size_ON[i] = 3* sum(P_ij_ON)[i] / np.sum(P_ij_ON)\n for j in range(P_ij_ON.shape[1]):\n e = g_ON.add_edge(i, j)\n edge_weights_ON[e] = P_ij_ON[i,j]\n edge_labels_ON[e] = \"{0:.2f}\".format(P_ij_ON[i,j])\n \n# Plot graph\npos = gt.sfdp_layout(g_ON) \nn_size = copy.copy(node_size_ON)\nn_size.a = 50* n_size.a/ max(n_size.a)\n\nedge_w = copy.copy(edge_weights_ON)\nedge_w.a = edge_w.a*10\n\ncmap = sns.cubehelix_palette(as_cmap=True) # cubehelix \ncmap = sns.diverging_palette(220,10, l=70, as_cmap=True, center=\"dark\") # blue to red via black\n\ngt.graph_draw(g_ON, pos=pos, vertex_color = n_size, vertex_fill_color = n_size, \n vertex_size = n_size,\n edge_pen_width=edge_w, edge_color=edge_w, \n edge_text=edge_labels_ON,\n vcmap=cmap,ecmap=cmap, \n vertex_text=g_ON.vertex_index, \n vertex_font_size=18,\n output_size=(800, 800), fit_view=0.45,\n output=\"figs/graphs/choice_1st_order_transition_ON\"+ mouse_name +\".pdf\")\n", "_____no_output_____" ] ], [ [ "# Finally, transition probabilities for choices - do they follow the trial types?\n## (Actually, let's just re-run the code from above changing tt to ch)", "_____no_output_____" ], [ "# Now, let's use graphs to visualise confusion matrices", "_____no_output_____" ] ], [ [ "cm_AB = confusion_matrix(tt[AB_pol],ch[AB_pol])\ncm_ON = confusion_matrix(tt[ON_pol],ch[ON_pol])\nprint(cm_AB)\nprint(cm_ON)\nprint(accuracy_score(tt[AB_pol],ch[AB_pol]))\nprint(accuracy_score(tt[ON_pol],ch[ON_pol]))", "[[401 32 42]\n [ 18 291 43]\n [ 91 63 280]]\n[[199 14 27]\n [ 10 196 15]\n [ 27 54 247]]\n0.770816812054\n0.813688212928\n" ], [ "cmap = sns.diverging_palette(220,10, l=70, as_cmap=True, center=\"dark\") # blue to red via black\nwith sns.axes_style(\"white\"):\n fig, ax = plt.subplots(1,2)\n ax[0].imshow(cm_ON/np.sum(cm_ON),interpolation='none',cmap=cmap)\n ax[1].imshow(cm_AB/np.sum(cm_AB),interpolation='none',cmap=cmap)\n \nfor i in range(0,3):\n for j in range(0,3):\n ax[0].text(j, i, \"{0:.2f}\".format(cm_ON[i,j]), va='center', ha='center',bbox=dict(facecolor='white',edgecolor='white', alpha=0.5))\n ax[1].text(j, i, \"{0:.2f}\".format(cm_AB[i,j]), va='center', ha='center',bbox=dict(facecolor='white',edgecolor='white', alpha=0.5))\n\nax[0].set_title('Mouse ON')\nax[1].set_title('Mouse AB')\n# plt.savefig('figs/graphs/confusion_matrix_AB_'+ mouse_name +'.png')", "_____no_output_____" ] ], [ [ "# Should also look at patterns in licking wrt correct/incorrect ", "_____no_output_____" ] ], [ [ "for v in g.vertices():\n print(v)\nfor e in g.edges():\n print(e)", "_____no_output_____" ], [ "19.19 - 9.92\n", "_____no_output_____" ], [ "# gt.graph_draw(g,output_size=(400,400),fit_view=True,output='simple_graph.pdf')\ngt.graph_draw(g2,output_size=(400,400),fit_view=True)", "_____no_output_____" ], [ "deg.", "_____no_output_____" ], [ "# Stats...", "_____no_output_____" ], [ "len(tt[tt[AB_pol]])", "_____no_output_____" ], [ "gt.graph_draw?", "_____no_output_____" ] ], [ [ "## Load and plot protraction/retraction trial data for one mouse", "_____no_output_____" ] ], [ [ "# quick load and classification of pro/ret data\ntt = pd.read_csv('~/work/whiskfree/data/tt_36_subset_sorted.csv',header=None)\nch = pd.read_csv('~/work/whiskfree/data/ch_36_subset_sorted.csv',header=None)\nproret = pd.read_csv('~/work/whiskfree/data/proret_36_subset_sorted.csv',header=None)\n\ntt = tt.values.reshape(-1,1)\nch = ch.values.reshape(-1,1)\nproret = proret.values.reshape(-1,1)", "_____no_output_____" ], [ "cm = confusion_matrix(tt,ch)\nprint(cm)", "[[231 41 37]\n [ 41 183 33]\n [ 75 18 172]]\n" ], [ "cm_tt_t = confusion_matrix(tt,proret)\ncm_ch_t = confusion_matrix(ch,proret)\n\nprint(cm_tt_t)\nprint(cm_ch_t)\nplt.imshow(cm_tt_t,interpolation='none')", "[[198 28 83]\n [ 81 110 66]\n [ 9 151 105]]\n[[181 56 110]\n [ 60 131 51]\n [ 47 102 93]]\n" ], [ "with sns.axes_style(\"white\"):\n fig, ax = plt.subplots(1,2,figsize=(10,6))\n ax[0].imshow(cm_tt_t/np.sum(cm_tt_t),interpolation='none')\n ax[1].imshow(cm_ch_t/np.sum(cm_ch_t),interpolation='none')\n \nfor i in range(0,3):\n for j in range(0,3):\n ax[0].text(j, i, \"{0:.2f}\".format(cm_tt_t[i,j]), va='center', ha='center',bbox=dict(facecolor='white',edgecolor='white', alpha=0.5))\n ax[1].text(j, i, \"{0:.2f}\".format(cm_ch_t[i,j]), va='center', ha='center',bbox=dict(facecolor='white',edgecolor='white', alpha=0.5))\n\n \nxlabels = ['Retraction','Protraction','No Touch']\nylabels = ['Posterior','Anterior','No Go']\nax[0].set_title('Trialtype | touch type' + '. ' + str(int(100 * accuracy_score(tt,proret))) + '%')\nax[1].set_title('Choice | touch type' + '. ' + str(int(100 * accuracy_score(ch,proret))) + '%')\n\nax[0].set_ylabel('Trial type')\nax[1].set_ylabel('Choice')\n \nfor i in range(0,2):\n ax[i].set_xlabel('Touch type')\n ax[i].set_xticks([0,1,2])\n ax[i].set_xticklabels(xlabels)\n ax[i].set_yticks([0,1,2])\n ax[i].set_yticklabels(ylabels)\n \nplt.tight_layout()\n \n# plt.savefig('../figs/classification/pro_ret/310816/touchtype_confmatrix_both_32.png')\nplt.savefig('../figs/classification/pro_ret/36/touchtype_confmatrix_both_36.png')", "_____no_output_____" ], [ "lr_tt = LogisticRegression(solver='lbfgs',multi_class='multinomial')\nlr_tt.fit(proret,tt)\nc_tt = lr_tt.predict(proret)\nprint('TT prediction accuracy =',accuracy_score(tt,c_tt))\nlr_ch = LogisticRegression(solver='lbfgs',multi_class='multinomial')\nlr_ch.fit(proret,ch)\nc_ch = lr_ch.predict(proret)\nprint('Choice prediction accuracy =',accuracy_score(ch,c_ch))\nprint('Mouse prediction accuracy =',accuracy_score(tt,ch))", "TT prediction accuracy = 0.398315282792\nChoice prediction accuracy = 0.397111913357\nMouse prediction accuracy = 0.705174488568\n" ], [ "print(confusion_matrix(ch,c_ch))\nprint(confusion_matrix(tt,c_tt))\n", "[[237 0 110]\n [191 0 51]\n [149 0 93]]\n[[226 0 83]\n [191 0 66]\n [160 0 105]]\n" ], [ "print(accuracy_score(ch,proret))\nprint(accuracy_score(tt,proret))", "0.487364620939\n0.496991576414\n" ], [ "plt.plot(c_ch)", "_____no_output_____" ], [ "# Confusion matrix predicting trial type based on protraction/retraction\ncm = confusion_matrix(tt,c_tt)\ncm_m = confusion_matrix(tt,ch)\n\n# xlabels = ['Retraction','Protraction','No Touch']\nylabels = ['Posterior','Anterior','No Go']\nwith sns.axes_style(\"white\"):\n fig, ax = plt.subplots(1,2,figsize=(10,6))\n ax[0].imshow(cm,interpolation='none')\n for i in range(0,3):\n for j in range(0,3):\n ax[0].text(j, i, \"{0:.2f}\".format(cm[i,j]), va='center', ha='center',bbox=dict(facecolor='white',edgecolor='white', alpha=0.5))\n\n \n ax[0].set_title('Logistic Regression - TT' + '. ' + str(int(100 * accuracy_score(tt,c_tt))) + '%')\n\n ax[1].imshow(cm_m,interpolation='none')\n for i in range(0,3):\n for j in range(0,3):\n ax[1].text(j, i, \"{0:.2f}\".format(cm_m[i,j]), va='center', ha='center',bbox=dict(facecolor='white',edgecolor='white', alpha=0.5))\n\n ax[1].set_title('Mouse' + '. ' + str(int(100 * accuracy_score(tt,ch))) + '%')\n\n for i in range(0,2):\n ax[i].set_ylabel('True label')\n ax[i].set_xlabel('Predicted label')\n ax[i].set_xticks([0,1,2])\n ax[i].set_xticklabels(ylabels)\n ax[i].set_yticks([0,1,2])\n ax[i].set_yticklabels(ylabels)\n \nplt.tight_layout()\n\n# plt.savefig('../figs/classification/pro_ret/310816/LR_confmatrix_TT_32.png')\nplt.savefig('../figs/classification/pro_ret/36/LR_confmatrix_TT_36.png')", "_____no_output_____" ], [ "# Confusion matrix predicting choice based on protraction/retraction\ncm_ch = confusion_matrix(ch,c_ch)\ncm_m = confusion_matrix(ch,tt)\n\n# xlabels = ['Retraction','Protraction','No Touch']\nylabels = ['Posterior','Anterior','No Go']\nwith sns.axes_style(\"white\"):\n fig, ax = plt.subplots(1,2,figsize=(10,6))\n ax[0].imshow(cm_ch,interpolation='none')\n for i in range(0,3):\n for j in range(0,3):\n ax[0].text(j, i, \"{0:.2f}\".format(cm_ch[i,j]), va='center', ha='center',bbox=dict(facecolor='white',edgecolor='white', alpha=0.5))\n\n \n ax[0].set_title('Logistic Regression - Ch' + '. ' + str(int(100 * accuracy_score(ch,c_ch))) + '%')\n \n ax[1].imshow(cm_m,interpolation='none')\n for i in range(0,3):\n for j in range(0,3):\n ax[1].text(j, i, \"{0:.2f}\".format(cm_m[i,j]), va='center', ha='center',bbox=dict(facecolor='white',edgecolor='white', alpha=0.5))\n\n ax[1].set_title('Mouse' + '. ' + str(int(100 * accuracy_score(ch,tt))) + '%')\n\n for i in range(0,2):\n ax[i].set_ylabel('True label')\n ax[i].set_xlabel('Predicted label')\n ax[i].set_xticks([0,1,2])\n ax[i].set_xticklabels(ylabels)\n ax[i].set_yticks([0,1,2])\n ax[i].set_yticklabels(ylabels)\n \nplt.tight_layout()\n\n# plt.savefig('../figs/classification/pro_ret/310816/LR_confmatrix_Ch_32.png')\nplt.savefig('../figs/classification/pro_ret/36/LR_confmatrix_Ch_36.png')", "_____no_output_____" ], [ "# Correct/incorrect\ncorrect = tt==ch\nerrors = tt!=ch\n\ncm_c = confusion_matrix(ch[correct],proret[correct])\ncm_ic = confusion_matrix(ch[errors],proret[errors])\n\nxlabels = ['Retraction','Protraction','No Touch']\nylabels = ['Posterior','Anterior','No Go']\nwith sns.axes_style(\"white\"):\n fig, ax = plt.subplots(1,2,figsize=(10,6))\n ax[0].imshow(cm_c,interpolation='none')\n for i in range(0,3):\n for j in range(0,3):\n ax[0].text(j, i, \"{0:.2f}\".format(cm_c[i,j]), va='center', ha='center',bbox=dict(facecolor='white',edgecolor='white', alpha=0.5))\n\n \n ax[0].set_title('Correct choice | touch type')\n \n ax[1].imshow(cm_ic,interpolation='none')\n for i in range(0,3):\n for j in range(0,3):\n ax[1].text(j, i, \"{0:.2f}\".format(cm_ic[i,j]), va='center', ha='center',bbox=dict(facecolor='white',edgecolor='white', alpha=0.5))\n\n ax[1].set_title('Incorrect choice | touch type')\n \n for i in range(0,2):\n ax[i].set_ylabel('Choice')\n ax[i].set_xlabel('Touch Type')\n ax[i].set_xticks([0,1,2])\n ax[i].set_xticklabels(xlabels)\n ax[i].set_yticks([0,1,2])\n ax[i].set_yticklabels(ylabels)\n \nplt.tight_layout()\n\n# plt.savefig('../figs/classification/pro_ret/310816/Correct_incorrect_confmatrix_Ch_32.png')\nplt.savefig('../figs/classification/pro_ret/36/Correct_incorrect_confmatrix_Ch_36.png')", "_____no_output_____" ], [ "# Try graph of trialtype/choice/touchtype plots\n# P_ijk_ON\n\n# import graph_tool.all as gt\n\ncm_3 = np.zeros([3,3,3])\nfor i in range(len(tt) - 2):\n \n cm_3[tt[i]-1,proret[i]-1 ,ch[i]-1] += 1\n\n# Make graph\ncm_G = gt.Graph(directed = False)\n\n# trialtypes = ['P','A','NG']\n# touchtypes = ['Ret','Pro','NT']\n# choices = ['P','A','NG']\ntrialtypes = ['Posterior','Anterior','No Go']\ntouchtypes = ['Retraction','Protraction','No Touch']\nchoices = ['Posterior','Anterior','No Go']\n\nedge_weights_cm_G = cm_G.new_edge_property('double')\nedge_labels_cm_G = cm_G.new_edge_property('string')\nnode_labels_cm_G = cm_G.new_vertex_property('string')\nnode_size_cm_G = cm_G.new_vertex_property('double')\nh = cm_G.add_vertex()\nnode_labels_cm_G[h] = \"0\"\n\nfor i in range(cm_3.shape[0]):\n v = cm_G.add_vertex()\n node_labels_cm_G[v] = trialtypes[i]\n e = cm_G.add_edge(h,v)\n node_size_cm_G[v] = np.sum(cm_3[i]) / 4\n \n for j in range(cm_3.shape[1]):\n v2 = cm_G.add_vertex()\n node_labels_cm_G[v2] = touchtypes[j]\n e = cm_G.add_edge(v,v2)\n \n edge_weights_cm_G[e] = np.sum(cm_3[i,j]) /4\n edge_labels_cm_G[e] = str(int(np.sum(cm_3[i,j])))\n node_size_cm_G[v2] = np.sum(cm_3[i,j]) /4\n \n for k in range(cm_3.shape[2]):\n v3 = cm_G.add_vertex()\n node_labels_cm_G[v3] = choices[k]\n e2 = cm_G.add_edge(v2,v3)\n \n edge_weights_cm_G[e2] = int(cm_3[i,j,k])/4\n edge_labels_cm_G[e2] = str(int(cm_3[i,j,k]))\n node_size_cm_G[v3] = int(cm_3[i,j,k])/2\n\n# Plot graph\npos = gt.radial_tree_layout(cm_G,cm_G.vertex(0))\n# cmap = sns.diverging_palette(220,10, l=70, as_cmap=True, center=\"dark\") # blue to green via black\ncmap =plt.get_cmap('Greys')\ngt.graph_draw(cm_G,pos=pos,vertex_size=node_size_cm_G,edge_pen_width=edge_weights_cm_G,\n vertex_text = node_labels_cm_G, #vertex_text_position = 'centered',\n edge_text=edge_labels_cm_G,\n vertex_font_size = 22, vertex_font_family = 'sansserif',\n edge_font_size = 24, edge_font_family = 'sansserif',\n ecmap=cmap, vcmap=cmap, \n edge_color = edge_weights_cm_G,\n vertex_color = node_size_cm_G,\n vertex_fill_color = node_size_cm_G,\n output_size=(1500, 1500), fit_view=0.8,\n# output=\"../figs/classification/pro_ret/310816/tt_touch_ch_graph_BW_\"+ mouse_name +\".pdf\")\n \n output=\"../figs/classification/pro_ret/36/tt_touch_ch_graph_BW_\"+ mouse_name +\".pdf\")\n\n", "_____no_output_____" ], [ "np.sum(cm_3)", "_____no_output_____" ], [ "error_matrix", "_____no_output_____" ], [ "choice_matrix", "_____no_output_____" ], [ "with sns.axes_style(\"white\"):\n cmap = sns.diverging_palette(220,10, l=70, as_cmap=True, center=\"dark\") # blue to green via black\n fig, ax = plt.subplots(1,2)\n ax[0].imshow(error_matrix,interpolation='none',cmap=cmap)\n for i in range(0,3):\n for j in range(0,3):\n ax[0].text(j, i, \"{0:.2f}\".format(error_matrix[i,j]), va='center', ha='center',bbox=dict(facecolor='white',edgecolor='white', alpha=0.5))\n\n \n ax[0].set_title('Error matrix') # + '. ' + str(int(100 * accuracy_score(tt,c_tt))) + '%')\n ax[0].set_ylabel('Trial type')\n ax[0].set_xlabel('Touch type')\n ax[1].imshow(choice_matrix,interpolation='none',cmap=cmap)\n for i in range(0,3):\n for j in range(0,3):\n ax[1].text(j, i, \"{0:.2f}\".format(choice_matrix[i,j]), va='center', ha='center',bbox=dict(facecolor='white',edgecolor='white', alpha=0.5))\n\n ax[1].set_title('Choice matrix') # + '. ' + str(int(100 * accuracy_score(tt,ch))) + '%')\n ax[1].set_ylabel('Choice')\n ax[1].set_xlabel('Touch type')\n# plt.savefig('figs/graphs/pro_ret_confmatrix_TT_32_full.png')", "_____no_output_____" ], [ "plt.plot(c_ch)", "_____no_output_____" ], [ "print(confusion_matrix(ch,proret))\nprint(confusion_matrix(tt,proret))\n", "[[164 54 77]\n [ 86 241 25]\n [ 21 114 133]]\n[[189 15 62]\n [ 80 236 25]\n [ 2 158 148]]\n" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/haroldosfilho/Python/blob/master/TRABALHO1GRAFOS.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ], [ "bibliotecas utilizadas", "_____no_output_____" ], [ "Aperte Play para inicializar as bibliotecas\n", "_____no_output_____" ] ], [ [ "import networkx as nx\nimport matplotlib.pyplot as plt\nimport numpy as np\n", "_____no_output_____" ] ], [ [ "Entre com o número de vértces do seu grafo e digite enter.\n", "_____no_output_____" ] ], [ [ "n = input(\"entre com o numero de vertices:\" )", "entre com o numero de vertices:5\n" ] ], [ [ "aperte o play para tranformar num número inteiro a sua entrada.", "_____no_output_____" ] ], [ [ "num=int(str(n))\nprint(num)", "5\n" ] ], [ [ "aperte o play para gerar a lista dos vértices do seu Grafo", "_____no_output_____" ] ], [ [ "G = nx.path_graph(num)\nlist(G.nodes)\n", "_____no_output_____" ], [ "m = int(input(\"Entre com o número de arestas : \"))", "Entre com o número de arestas : 7\n" ] ], [ [ "Digite as suas arestas, quais vértices estão conectados, aperte enter após cada aresta informada.\n", "_____no_output_____" ] ], [ [ "# creating an empty list\nlst = []\n# iterating till the range\nfor i in range(0, m):\n ele = str(input())\n lst.append(ele) # adding the element\n \nprint(lst)", "01\n12\n13\n23\n24\n34\n02\n['01', '12', '13', '23', '24', '34', '02']\n" ] ], [ [ "aperte play para gerar uma representação no plano do seu Grafo.", "_____no_output_____" ] ], [ [ "G = nx.Graph(lst)\nopts = { \"with_labels\": True, \"node_color\": 'y' }\nnx.draw(G, **opts)\n", "_____no_output_____" ] ], [ [ "aperte o play para gerar os elementos da sua matriz de adjacência do seu Grafo\n", "_____no_output_____" ] ], [ [ "A = nx.adjacency_matrix(G)\nprint(A)", " (0, 1)\t1\n (0, 2)\t1\n (1, 0)\t1\n (1, 2)\t1\n (1, 3)\t1\n (2, 0)\t1\n (2, 1)\t1\n (2, 3)\t1\n (2, 4)\t1\n (3, 1)\t1\n (3, 2)\t1\n (3, 4)\t1\n (4, 2)\t1\n (4, 3)\t1\n" ] ], [ [ "Agora basta apertar o play e a sua matriz de adjacência está pronta!", "_____no_output_____" ] ], [ [ "A = nx.adjacency_matrix(G).toarray()\n\nprint(A)", "[[0 1 1 0 0]\n [1 0 1 1 0]\n [1 1 0 1 1]\n [0 1 1 0 1]\n [0 0 1 1 0]]\n" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# ClarityViz\n\n## Pipeline: .img -> histogram .nii -> graph represented as csv -> graph as graphml -> plotly\n\n### To run:\n\n### Step 1:\n\nFirst, run the following. This takes the .img, generates the localeq histogram as an nii file, gets the nodes and edges as a csv and converts the csv into a graphml\n\n", "_____no_output_____" ] ], [ [ "python runclarityviz.py --token Fear199Coronal --file-type img --source-directory /cis/project/clarity/data/clarity/isoCoronal", "_____no_output_____" ] ], [ [ "### Step 2: \nThen run this. This just converts the graphml into a plotly", "_____no_output_____" ] ], [ [ "python runclarityviz.py --token Fear199Coronal --plotly yes", "_____no_output_____" ] ], [ [ "## Results", "_____no_output_____" ] ], [ [ "Starting pipeline for Fear199.img\nGenerating Histogram...\nFINISHED GENERATING HISTOGRAM\nLoading: Fear199/Fear199localeq.nii\nImage Loaded: Fear199/Fear199localeq.nii\nFINISHED LOADING NII\nCoverting to points...\ntoken=Fear199\ntotal=600735744\nmax=255.000000\nthreshold=0.300000\nsample=0.500000\n(This will take couple minutes)\nAbove threshold=461409948\nSamples=230718301\nFinished\nFINISHED GETTING POINTS", "_____no_output_____" ], [ "~/clarityviztesting/Fear199Coronal$ ls\nFear199Coronal.csv\t Fear199Coronal.graphml Fear199Coronal.nodes.csv\nFear199Coronal.edges.csv Fear199Coronallocaleq.nii Fear199Coronalplotly.html", "_____no_output_____" ] ], [ [ "# Code", "_____no_output_____" ], [ "## runclarityviz.py:", "_____no_output_____" ] ], [ [ "from clarityviz import clarityviz\nimport ...\n\ndef get_args():\n parser = argparse.ArgumentParser(description=\"Description\")\n\n parser.add_argument(\"--token\", type=str, required=True, help=\"The token.\")\n parser.add_argument(\"--file-type\", type=str, required=False, help=\"The file type.\")\n parser.add_argument(\"--source-directory\", type=str, required=False,\n help=\"Optional setting of the source directory.\")\n parser.add_argument(\"--plotly\", type=str, required=False, help=\"Optional method to generate the plotly graphs.\")\n parser.add_argument(\"--generate-nii-from-csv\", type=str, required=False, help=\"script to generate nii\")\n\n args = parser.parse_args()\n\n return args\n\n\ndef main():\n print('ayyooooo')\n args = get_args()\n\n if args.plotly == 'yes':\n ## Type in the path to your csv file here\n thedata = np.genfromtxt(args.token + '/' + args.token + '.csv',\n delimiter=',', dtype='int', usecols = (0,1,2), names=['a','b','c'])\n\n trace1 = go.Scatter3d(\n x = thedata['a'],\n y = thedata['b'],\n z = thedata['c'],\n mode='markers',\n marker=dict(\n size=1.2,\n color='purple', # set color to an array/list of desired values\n colorscale='Viridis', # choose a colorscale\n opacity=0.15\n )\n )\n\n data = [trace1]\n layout = go.Layout(\n margin=dict(\n l=0,\n r=0,\n b=0,\n t=0\n )\n )\n\n fig = go.Figure(data=data, layout=layout)\n print args.token + \"plotly\"\n plotly.offline.plot(fig, filename= args.token + \"/\" + args.token + \"plotly.html\")\n else:\n print('Starting pipeline for %s' % (args.token + '.' + args.file_type))\n if args.source_directory == None:\n c = clarityviz(args.token)\n else:\n c = clarityviz(args.token, args.source_directory)\n\n if args.file_type == 'img':\n #c.loadEqImg()\n c.generateHistogram()\n print('FINISHED GENERATING HISTOGRAM')\n c.loadNiiImg()\n print('FINISHED LOADING NII')\n elif args.file_type == 'nii':\n c.loadNiiImg()\n print('FINISHED LOADING NII')\n\n c.imgToPoints(0.3, 0.5)\n print(\"FINISHED GETTING POINTS\")\n\n c.savePoints()\n\n c.plot3d()\n print(\"FINISHED PLOT3D\")\n\n c.graphmlconvert()\n print(\"FINISHED GRAPHMLCONVERT\")\n\nif __name__ == \"__main__\":\n main()", "_____no_output_____" ] ], [ [ "## clarityviz.py", "_____no_output_____" ] ], [ [ "def generateHistogram(self):\n print('Generating Histogram...')\n if self._source_directory == None:\n path = self._token + '.img'\n else:\n path = self._source_directory + \"/\" + self._token + \".img\"\n\n im = nib.load(path)\n\n im = im.get_data()\n img = im[:,:,:]\n\n shape = im.shape\n #affine = im.get_affine()\n\n x_value = shape[0]\n y_value = shape[1]\n z_value = shape[2]\n\n #####################################################\n\n imgflat = img.reshape(-1)\n\n #img_grey = np.array(imgflat * 255, dtype = np.uint8)\n\n #img_eq = exposure.equalize_hist(img_grey)\n\n #new_img = img_eq.reshape(x_value, y_value, z_value)\n #globaleq = nib.Nifti1Image(new_img, np.eye(4))\n\n ######################################################\n\n #clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))\n clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))\n\n img_grey = np.array(imgflat * 255, dtype = np.uint8)\n #threshed = cv2.adaptiveThreshold(img_grey, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 3, 0)\n\n cl1 = clahe.apply(img_grey)\n\n #cv2.imwrite('clahe_2.jpg',cl1)\n #cv2.startWindowThread()\n #cv2.namedWindow(\"adaptive\")\n #cv2.imshow(\"adaptive\", cl1)\n #cv2.imshow(\"adaptive\", threshed)\n #plt.imshow(threshed)\n\n localimgflat = cl1 #cl1.reshape(-1)\n\n newer_img = localimgflat.reshape(x_value, y_value, z_value)\n localeq = nib.Nifti1Image(newer_img, np.eye(4))\n nib.save(localeq, self._token + '/' + self._token + 'localeq.nii')", "_____no_output_____" ], [ "def loadGeneratedNii(self, path=None, info=False):\n path = self._token + '/' + self._token + 'localeq.nii'\n print(\"Loading: %s\"%(path))\n\n #pathname = path+self._token+\".nii\"\n img = nib.load(path)\n if info:\n print(img)\n #self._img = img.get_data()[:,:,:,0]\n self._img = img.get_data()\n self._shape = self._img.shape\n self._max = np.max(self._img)\n print(\"Image Loaded: %s\"%(path))\n return self", "_____no_output_____" ], [ "def imgToPoints(self, threshold=0.1, sample=0.5, optimize=True):\n \"\"\"Method to extract points data from the img file.\"\"\"\n if not 0 <= threshold < 1:\n raise ValueError(\"Threshold should be within [0,1).\")\n if not 0 < sample <= 1:\n raise ValueError(\"Sample rate should be within (0,1].\")\n if self._img is None:\n raise ValueError(\"Img haven't loaded, please call loadImg() first.\")\n\n total = self._shape[0]*self._shape[1]*self._shape[2]\n print(\"Coverting to points...\\ntoken=%s\\ntotal=%d\\nmax=%f\\nthreshold=%f\\nsample=%f\"\\\n %(self._token,total,self._max,threshold,sample))\n print(\"(This will take couple minutes)\")\n # threshold\n filt = self._img > threshold * self._max\n x, y, z = np.where(filt)\n v = self._img[filt]\n if optimize:\n self.discardImg()\n v = np.int16(255*(np.float32(v)/np.float32(self._max)))\n l = v.shape\n print(\"Above threshold=%d\"%(l))\n # sample\n if sample < 1.0:\n filt = np.random.random(size=l) < sample\n x = x[filt]\n y = y[filt]\n z = z[filt]\n v = v[filt]\n self._points = np.vstack([x,y,z,v])\n self._points = np.transpose(self._points)\n print(\"Samples=%d\"%(self._points.shape[0]))\n print(\"Finished\")\n return self", "_____no_output_____" ], [ "def plot3d(self, infile = None):\n \"\"\"Method for plotting the Nodes and Edges\"\"\"\n filename = \"\"\n points_file = None\n if infile == None:\n points_file = self._points\n filename = self._token\n else:\n self.loadInitCsv(infile)\n infile = self._infile\n filename = self._filename\n\n # points is an array of arrays\n points = self._points\n outpath = self._token + '/'\n nodename = outpath + filename + '.nodes.csv'\n edgename = outpath + filename + '.edges.csv'\n\n with open(nodename, 'w') as nodefile:\n with open(edgename, 'w') as edgefile:\n for ind in range(len(points)):\n #temp = points[ind].strip().split(',')\n temp = points[ind]\n x = temp[0]\n y = temp[1]\n z = temp[2]\n v = temp[3]\n radius = 18\n nodefile.write(\"s\" + str(ind + 1) + \",\" + str(x) + \",\" + str(y) + \",\" + str(z) + \"\\n\")\n for index in range(ind + 1, len(points)):\n tmp = points[index]\n distance = math.sqrt(math.pow(int(x) - int(tmp[0]), 2) + math.pow(int(y) - int(tmp[1]), 2) + math.pow(int(z) - int(tmp[2]), 2))\n if distance < radius:\n edgefile.write(\"s\" + str(ind + 1) + \",\" + \"s\" + str(index + 1) + \"\\n\")\n self._nodefile = nodefile\n self._edgefile = edgefile", "_____no_output_____" ], [ " def graphmlconvert(self, nodefilename = None, edgefilename = None):\n \"\"\"Method for extracting the data to a graphml file, based on the node and edge files\"\"\"\n nodefile = None\n edgefile = None\n\n # If no nodefilename was entered, used the Clarity object's nodefile\n if nodefilename == None:\n #nodefile = self._nodefile\n #nodefile = open(self._nodefile, 'r')\n\n self.loadNodeCsv(self._token + \"/\" + self._token + \".nodes.csv\")\n nodefile = self._nodefile\n else:\n self.loadNodeCsv(nodefilename)\n nodefile = self._nodefile\n\n # If no edgefilename was entered, used the Clarity object's edgefile\n if edgefilename == None:\n #edgefile = self._edgefile\n #edgefile = open(self._edgefile, 'r')\n\n self.loadEdgeCsv(self._token + \"/\" + self._token + \".edges.csv\")\n edgefile = self._edgefile\n else:\n self.loadEdgeCsv(edgefilename)\n edgefile = self._edgefile\n\n # Start writing to the output graphml file\n path = self._token + \"/\" + self._token + \".graphml\"\n with open(path, 'w') as outfile:\n outfile.write(\"<?xml version=\\\"1.0\\\" encoding=\\\"UTF-8\\\"?>\\n\")\n outfile.write(\"<graphml xmlns=\\\"http://graphml.graphdrawing.org/xmlns\\\"\\n\")\n outfile.write(\" xmlns:xsi=\\\"http://www.w3.org/2001/XMLSchema-instance\\\"\\n\")\n outfile.write(\" xsi:schemaLocation=\\\"http://graphml.graphdrawing.org/xmlns\\n\")\n outfile.write(\" http://graphml.graphdrawing.org/xmlns/1.0/graphml.xsd\\\">\\n\")\n\n outfile.write(\" <key id=\\\"d0\\\" for=\\\"node\\\" attr.name=\\\"attr\\\" attr.type=\\\"string\\\"/>\\n\")\n outfile.write(\" <key id=\\\"e_weight\\\" for=\\\"edge\\\" attr.name=\\\"weight\\\" attr.type=\\\"double\\\"/>\\n\")\n outfile.write(\" <graph id=\\\"G\\\" edgedefault=\\\"undirected\\\">\\n\")\n\n for line in nodefile:\n if len(line) == 0:\n continue\n line = line.strip().split(',')\n outfile.write(\" <node id=\\\"\" + line[0] + \"\\\">\\n\")\n outfile.write(\" <data key=\\\"d0\\\">[\" + line[1] + \", \" + line[2] + \", \" + line[3] +\"]</data>\\n\")\n outfile.write(\" </node>\\n\")\n \n for line in edgefile:\n if len(line) == 0:\n continue\n line = line.strip().split(',')\n outfile.write(\" <edge source=\\\"\" + line[0] + \"\\\" target=\\\"\" + line[1] + \"\\\">\\n\")\n outfile.write(\" <data key=\\\"e_weight\\\">1</data>\\n\")\n outfile.write(\" </edge>\\n\")\n\n outfile.write(\" </graph>\\n</graphml>\")", "_____no_output_____" ], [ "def graphmlToPlotly(self, path):\n ## Type in the path to your csv file here\n thedata = np.genfromtxt('../data/points/localeq.csv', delimiter=',', dtype='int', usecols = (0,1,2), names=['a','b','c'])\n\n trace1 = go.Scatter3d(\n x = thedata['a'],\n y = thedata['b'],\n z = thedata['c'],\n mode='markers',\n marker=dict(\n size=1.2,\n color='purple', # set color to an array/list of desired values\n colorscale='Viridis', # choose a colorscale\n opacity=0.15\n )\n )\n\n data = [trace1]\n layout = go.Layout(\n margin=dict(\n l=0,\n r=0,\n b=0,\n t=0\n )\n )\n\n fig = go.Figure(data=data, layout=layout)\n print \"localeq\"\n plotly.offline.plot(fig, filename= \"localeq\")", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "{\n \"nodes\": [\n {\n \"op\": \"null\", \n \"name\": \"data\", \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv0_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(8L, 3L, 3L, 3L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv0_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(3, 3)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"8\", \n \"num_group\": \"1\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(2, 2)\"\n }, \n \"inputs\": [[0, 0, 0], [1, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm0_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(8L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm0_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(8L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm0_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(8L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm0_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(8L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm0_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[2, 0, 0], [3, 0, 0], [4, 0, 0], [5, 0, 1], [6, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu0_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[7, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv1_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(8L, 1L, 3L, 3L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv1_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(3, 3)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"8\", \n \"num_group\": \"8\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[8, 0, 0], [9, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm1_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(8L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm1_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(8L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm1_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(8L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm1_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(8L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm1_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[10, 0, 0], [11, 0, 0], [12, 0, 0], [13, 0, 1], [14, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu1_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[15, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv2_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(16L, 8L, 1L, 1L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv2_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(1, 1)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"16\", \n \"num_group\": \"1\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[16, 0, 0], [17, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm2_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(16L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm2_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(16L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm2_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(16L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm2_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(16L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm2_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[18, 0, 0], [19, 0, 0], [20, 0, 0], [21, 0, 1], [22, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu2_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[23, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv3_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(16L, 1L, 3L, 3L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv3_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(3, 3)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"16\", \n \"num_group\": \"16\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(2, 2)\"\n }, \n \"inputs\": [[24, 0, 0], [25, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm3_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(16L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm3_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(16L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm3_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(16L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm3_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(16L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm3_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[26, 0, 0], [27, 0, 0], [28, 0, 0], [29, 0, 1], [30, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu3_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[31, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv4_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L, 16L, 1L, 1L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv4_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(1, 1)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"32\", \n \"num_group\": \"1\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[32, 0, 0], [33, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm4_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm4_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm4_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm4_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm4_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[34, 0, 0], [35, 0, 0], [36, 0, 0], [37, 0, 1], [38, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu4_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[39, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv5_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L, 1L, 3L, 3L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv5_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(3, 3)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"32\", \n \"num_group\": \"32\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[40, 0, 0], [41, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm5_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm5_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm5_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm5_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm5_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[42, 0, 0], [43, 0, 0], [44, 0, 0], [45, 0, 1], [46, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu5_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[47, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv6_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L, 32L, 1L, 1L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv6_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(1, 1)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"32\", \n \"num_group\": \"1\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[48, 0, 0], [49, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm6_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm6_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm6_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm6_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm6_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[50, 0, 0], [51, 0, 0], [52, 0, 0], [53, 0, 1], [54, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu6_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[55, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv7_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L, 1L, 3L, 3L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv7_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(3, 3)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"32\", \n \"num_group\": \"32\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(2, 2)\"\n }, \n \"inputs\": [[56, 0, 0], [57, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm7_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm7_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm7_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm7_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(32L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm7_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[58, 0, 0], [59, 0, 0], [60, 0, 0], [61, 0, 1], [62, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu7_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[63, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv8_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L, 32L, 1L, 1L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv8_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(1, 1)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"64\", \n \"num_group\": \"1\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[64, 0, 0], [65, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm8_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm8_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm8_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm8_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm8_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[66, 0, 0], [67, 0, 0], [68, 0, 0], [69, 0, 1], [70, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu8_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[71, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv9_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L, 1L, 3L, 3L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv9_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(3, 3)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"64\", \n \"num_group\": \"64\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[72, 0, 0], [73, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm9_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm9_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm9_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm9_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm9_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[74, 0, 0], [75, 0, 0], [76, 0, 0], [77, 0, 1], [78, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu9_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[79, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv10_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L, 64L, 1L, 1L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv10_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(1, 1)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"64\", \n \"num_group\": \"1\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[80, 0, 0], [81, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm10_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm10_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm10_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm10_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm10_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[82, 0, 0], [83, 0, 0], [84, 0, 0], [85, 0, 1], [86, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu10_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[87, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv11_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L, 1L, 3L, 3L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv11_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(3, 3)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"64\", \n \"num_group\": \"64\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(2, 2)\"\n }, \n \"inputs\": [[88, 0, 0], [89, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm11_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm11_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm11_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm11_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(64L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm11_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[90, 0, 0], [91, 0, 0], [92, 0, 0], [93, 0, 1], [94, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu11_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[95, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv12_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L, 64L, 1L, 1L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv12_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(1, 1)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"128\", \n \"num_group\": \"1\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[96, 0, 0], [97, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm12_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm12_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm12_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm12_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm12_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[98, 0, 0], [99, 0, 0], [100, 0, 0], [101, 0, 1], [102, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu12_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[103, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv13_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L, 1L, 3L, 3L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv13_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(3, 3)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"128\", \n \"num_group\": \"128\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[104, 0, 0], [105, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm13_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm13_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm13_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm13_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm13_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[106, 0, 0], [107, 0, 0], [108, 0, 0], [109, 0, 1], [110, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu13_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[111, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv14_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L, 128L, 1L, 1L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv14_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(1, 1)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"128\", \n \"num_group\": \"1\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[112, 0, 0], [113, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm14_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm14_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm14_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm14_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm14_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[114, 0, 0], [115, 0, 0], [116, 0, 0], [117, 0, 1], [118, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu14_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[119, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv15_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L, 1L, 3L, 3L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv15_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(3, 3)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"128\", \n \"num_group\": \"128\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[120, 0, 0], [121, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm15_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm15_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm15_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm15_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm15_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[122, 0, 0], [123, 0, 0], [124, 0, 0], [125, 0, 1], [126, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu15_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[127, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv16_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L, 128L, 1L, 1L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv16_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(1, 1)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"128\", \n \"num_group\": \"1\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[128, 0, 0], [129, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm16_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm16_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm16_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm16_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm16_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[130, 0, 0], [131, 0, 0], [132, 0, 0], [133, 0, 1], [134, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu16_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[135, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv17_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L, 1L, 3L, 3L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv17_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(3, 3)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"128\", \n \"num_group\": \"128\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[136, 0, 0], [137, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm17_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm17_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm17_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm17_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm17_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[138, 0, 0], [139, 0, 0], [140, 0, 0], [141, 0, 1], [142, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu17_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[143, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv18_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L, 128L, 1L, 1L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv18_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(1, 1)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"128\", \n \"num_group\": \"1\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[144, 0, 0], [145, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm18_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm18_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm18_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm18_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm18_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[146, 0, 0], [147, 0, 0], [148, 0, 0], [149, 0, 1], [150, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu18_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[151, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv19_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L, 1L, 3L, 3L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv19_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(3, 3)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"128\", \n \"num_group\": \"128\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[152, 0, 0], [153, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm19_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm19_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm19_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm19_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm19_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[154, 0, 0], [155, 0, 0], [156, 0, 0], [157, 0, 1], [158, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu19_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[159, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv20_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L, 128L, 1L, 1L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv20_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(1, 1)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"128\", \n \"num_group\": \"1\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[160, 0, 0], [161, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm20_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm20_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm20_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm20_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm20_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[162, 0, 0], [163, 0, 0], [164, 0, 0], [165, 0, 1], [166, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu20_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[167, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv21_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L, 1L, 3L, 3L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv21_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(3, 3)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"128\", \n \"num_group\": \"128\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[168, 0, 0], [169, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm21_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm21_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm21_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm21_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm21_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[170, 0, 0], [171, 0, 0], [172, 0, 0], [173, 0, 1], [174, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu21_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[175, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv22_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L, 128L, 1L, 1L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv22_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(1, 1)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"128\", \n \"num_group\": \"1\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[176, 0, 0], [177, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm22_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm22_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm22_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm22_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm22_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[178, 0, 0], [179, 0, 0], [180, 0, 0], [181, 0, 1], [182, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu22_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[183, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv23_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L, 1L, 3L, 3L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv23_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(3, 3)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"128\", \n \"num_group\": \"128\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(2, 2)\"\n }, \n \"inputs\": [[184, 0, 0], [185, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm23_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm23_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm23_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm23_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(128L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm23_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[186, 0, 0], [187, 0, 0], [188, 0, 0], [189, 0, 1], [190, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu23_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[191, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv24_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(256L, 128L, 1L, 1L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv24_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(1, 1)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"256\", \n \"num_group\": \"1\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[192, 0, 0], [193, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm24_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(256L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm24_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(256L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm24_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(256L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm24_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(256L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm24_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[194, 0, 0], [195, 0, 0], [196, 0, 0], [197, 0, 1], [198, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu24_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[199, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv25_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(256L, 1L, 3L, 3L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv25_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(3, 3)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"256\", \n \"num_group\": \"256\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[200, 0, 0], [201, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm25_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(256L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm25_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(256L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm25_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(256L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm25_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(256L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm25_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[202, 0, 0], [203, 0, 0], [204, 0, 0], [205, 0, 1], [206, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu25_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[207, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_conv26_weight\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(256L, 256L, 1L, 1L)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"mobilenet0_conv26_fwd\", \n \"attrs\": {\n \"dilate\": \"(1, 1)\", \n \"kernel\": \"(1, 1)\", \n \"layout\": \"NCHW\", \n \"no_bias\": \"True\", \n \"num_filter\": \"256\", \n \"num_group\": \"1\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[208, 0, 0], [209, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm26_gamma\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(256L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm26_beta\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(256L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm26_running_mean\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"zeros\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(256L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"mobilenet0_batchnorm26_running_var\", \n \"attrs\": {\n \"__dtype__\": \"0\", \n \"__init__\": \"ones\", \n \"__lr_mult__\": \"1.0\", \n \"__shape__\": \"(256L,)\", \n \"__storage_type__\": \"0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"mobilenet0_batchnorm26_fwd\", \n \"attrs\": {\n \"axis\": \"1\", \n \"eps\": \"1e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\", \n \"use_global_stats\": \"False\"\n }, \n \"inputs\": [[210, 0, 0], [211, 0, 0], [212, 0, 0], [213, 0, 1], [214, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"mobilenet0_relu26_fwd\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[215, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_lateral_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_lateral_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c3_lateral\", \n \"attrs\": {\n \"kernel\": \"(1, 1)\", \n \"num_filter\": \"64\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[216, 0, 0], [217, 0, 0], [218, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_lateral_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_lateral_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_lateral_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_lateral_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c3_lateral_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[219, 0, 0], [220, 0, 0], [221, 0, 0], [222, 0, 1], [223, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"rf_c3_lateral_relu\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[224, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_conv1_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_conv1_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c3_det_conv1\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"32\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[225, 0, 0], [226, 0, 0], [227, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_conv1_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_conv1_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_conv1_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_conv1_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c3_det_conv1_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[228, 0, 0], [229, 0, 0], [230, 0, 0], [231, 0, 1], [232, 0, 1]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv1_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv1_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c3_det_context_conv1\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"16\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[225, 0, 0], [234, 0, 0], [235, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv1_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv1_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv1_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv1_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c3_det_context_conv1_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[236, 0, 0], [237, 0, 0], [238, 0, 0], [239, 0, 1], [240, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"rf_c3_det_context_conv1_relu\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[241, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv2_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv2_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c3_det_context_conv2\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"16\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[242, 0, 0], [243, 0, 0], [244, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv2_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv2_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv2_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv2_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c3_det_context_conv2_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[245, 0, 0], [246, 0, 0], [247, 0, 0], [248, 0, 1], [249, 0, 1]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv3_1_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv3_1_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c3_det_context_conv3_1\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"16\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[242, 0, 0], [251, 0, 0], [252, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv3_1_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv3_1_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv3_1_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv3_1_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c3_det_context_conv3_1_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[253, 0, 0], [254, 0, 0], [255, 0, 0], [256, 0, 1], [257, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"rf_c3_det_context_conv3_1_relu\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[258, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv3_2_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv3_2_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c3_det_context_conv3_2\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"16\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[259, 0, 0], [260, 0, 0], [261, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv3_2_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv3_2_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv3_2_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c3_det_context_conv3_2_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c3_det_context_conv3_2_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[262, 0, 0], [263, 0, 0], [264, 0, 0], [265, 0, 1], [266, 0, 1]]\n }, \n {\n \"op\": \"Concat\", \n \"name\": \"rf_c3_det_concat\", \n \"attrs\": {\n \"dim\": \"1\", \n \"num_args\": \"3\"\n }, \n \"inputs\": [[233, 0, 0], [250, 0, 0], [267, 0, 0]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"rf_c3_det_concat_relu\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[268, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_cls_score_stride32_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_cls_score_stride32_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"face_rpn_cls_score_stride32\", \n \"attrs\": {\n \"kernel\": \"(1, 1)\", \n \"num_filter\": \"4\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[269, 0, 0], [270, 0, 0], [271, 0, 0]]\n }, \n {\n \"op\": \"Reshape\", \n \"name\": \"face_rpn_cls_score_reshape_stride32\", \n \"attrs\": {\"shape\": \"(0, 2, -1, 0)\"}, \n \"inputs\": [[272, 0, 0]]\n }, \n {\n \"op\": \"SoftmaxActivation\", \n \"name\": \"face_rpn_cls_prob_stride32\", \n \"attrs\": {\"mode\": \"channel\"}, \n \"inputs\": [[273, 0, 0]]\n }, \n {\n \"op\": \"Reshape\", \n \"name\": \"face_rpn_cls_prob_reshape_stride32\", \n \"attrs\": {\"shape\": \"(0, 4, -1, 0)\"}, \n \"inputs\": [[274, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_bbox_pred_stride32_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_bbox_pred_stride32_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"face_rpn_bbox_pred_stride32\", \n \"attrs\": {\n \"kernel\": \"(1, 1)\", \n \"num_filter\": \"8\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[269, 0, 0], [276, 0, 0], [277, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_landmark_pred_stride32_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_landmark_pred_stride32_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"face_rpn_landmark_pred_stride32\", \n \"attrs\": {\n \"kernel\": \"(1, 1)\", \n \"num_filter\": \"20\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[269, 0, 0], [279, 0, 0], [280, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_lateral_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_lateral_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c2_lateral\", \n \"attrs\": {\n \"kernel\": \"(1, 1)\", \n \"num_filter\": \"64\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[184, 0, 0], [282, 0, 0], [283, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_lateral_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_lateral_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_lateral_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_lateral_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c2_lateral_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[284, 0, 0], [285, 0, 0], [286, 0, 0], [287, 0, 1], [288, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"rf_c2_lateral_relu\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[289, 0, 0]]\n }, \n {\n \"op\": \"UpSampling\", \n \"name\": \"rf_c3_upsampling\", \n \"attrs\": {\n \"num_args\": \"1\", \n \"sample_type\": \"nearest\", \n \"scale\": \"2\", \n \"workspace\": \"512\"\n }, \n \"inputs\": [[225, 0, 0]]\n }, \n {\n \"op\": \"Crop\", \n \"name\": \"crop0\", \n \"attrs\": {\"num_args\": \"2\"}, \n \"inputs\": [[291, 0, 0], [290, 0, 0]]\n }, \n {\n \"op\": \"elemwise_add\", \n \"name\": \"plus0\", \n \"inputs\": [[290, 0, 0], [292, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_aggr_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_aggr_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c2_aggr\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"64\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[293, 0, 0], [294, 0, 0], [295, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_aggr_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_aggr_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_aggr_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_aggr_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c2_aggr_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[296, 0, 0], [297, 0, 0], [298, 0, 0], [299, 0, 1], [300, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"rf_c2_aggr_relu\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[301, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_conv1_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_conv1_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c2_det_conv1\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"32\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[302, 0, 0], [303, 0, 0], [304, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_conv1_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_conv1_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_conv1_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_conv1_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c2_det_conv1_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[305, 0, 0], [306, 0, 0], [307, 0, 0], [308, 0, 1], [309, 0, 1]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv1_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv1_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c2_det_context_conv1\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"16\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[302, 0, 0], [311, 0, 0], [312, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv1_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv1_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv1_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv1_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c2_det_context_conv1_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[313, 0, 0], [314, 0, 0], [315, 0, 0], [316, 0, 1], [317, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"rf_c2_det_context_conv1_relu\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[318, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv2_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv2_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c2_det_context_conv2\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"16\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[319, 0, 0], [320, 0, 0], [321, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv2_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv2_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv2_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv2_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c2_det_context_conv2_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[322, 0, 0], [323, 0, 0], [324, 0, 0], [325, 0, 1], [326, 0, 1]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv3_1_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv3_1_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c2_det_context_conv3_1\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"16\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[319, 0, 0], [328, 0, 0], [329, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv3_1_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv3_1_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv3_1_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv3_1_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c2_det_context_conv3_1_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[330, 0, 0], [331, 0, 0], [332, 0, 0], [333, 0, 1], [334, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"rf_c2_det_context_conv3_1_relu\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[335, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv3_2_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv3_2_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c2_det_context_conv3_2\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"16\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[336, 0, 0], [337, 0, 0], [338, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv3_2_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv3_2_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv3_2_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c2_det_context_conv3_2_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c2_det_context_conv3_2_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[339, 0, 0], [340, 0, 0], [341, 0, 0], [342, 0, 1], [343, 0, 1]]\n }, \n {\n \"op\": \"Concat\", \n \"name\": \"rf_c2_det_concat\", \n \"attrs\": {\n \"dim\": \"1\", \n \"num_args\": \"3\"\n }, \n \"inputs\": [[310, 0, 0], [327, 0, 0], [344, 0, 0]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"rf_c2_det_concat_relu\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[345, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_cls_score_stride16_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_cls_score_stride16_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"face_rpn_cls_score_stride16\", \n \"attrs\": {\n \"kernel\": \"(1, 1)\", \n \"num_filter\": \"4\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[346, 0, 0], [347, 0, 0], [348, 0, 0]]\n }, \n {\n \"op\": \"Reshape\", \n \"name\": \"face_rpn_cls_score_reshape_stride16\", \n \"attrs\": {\"shape\": \"(0, 2, -1, 0)\"}, \n \"inputs\": [[349, 0, 0]]\n }, \n {\n \"op\": \"SoftmaxActivation\", \n \"name\": \"face_rpn_cls_prob_stride16\", \n \"attrs\": {\"mode\": \"channel\"}, \n \"inputs\": [[350, 0, 0]]\n }, \n {\n \"op\": \"Reshape\", \n \"name\": \"face_rpn_cls_prob_reshape_stride16\", \n \"attrs\": {\"shape\": \"(0, 4, -1, 0)\"}, \n \"inputs\": [[351, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_bbox_pred_stride16_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_bbox_pred_stride16_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"face_rpn_bbox_pred_stride16\", \n \"attrs\": {\n \"kernel\": \"(1, 1)\", \n \"num_filter\": \"8\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[346, 0, 0], [353, 0, 0], [354, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_landmark_pred_stride16_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_landmark_pred_stride16_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"face_rpn_landmark_pred_stride16\", \n \"attrs\": {\n \"kernel\": \"(1, 1)\", \n \"num_filter\": \"20\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[346, 0, 0], [356, 0, 0], [357, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_red_conv_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_red_conv_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c1_red_conv\", \n \"attrs\": {\n \"kernel\": \"(1, 1)\", \n \"num_filter\": \"64\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[88, 0, 0], [359, 0, 0], [360, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_red_conv_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_red_conv_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_red_conv_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_red_conv_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c1_red_conv_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[361, 0, 0], [362, 0, 0], [363, 0, 0], [364, 0, 1], [365, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"rf_c1_red_conv_relu\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[366, 0, 0]]\n }, \n {\n \"op\": \"UpSampling\", \n \"name\": \"rf_c2_upsampling\", \n \"attrs\": {\n \"num_args\": \"1\", \n \"sample_type\": \"nearest\", \n \"scale\": \"2\", \n \"workspace\": \"512\"\n }, \n \"inputs\": [[302, 0, 0]]\n }, \n {\n \"op\": \"Crop\", \n \"name\": \"crop1\", \n \"attrs\": {\"num_args\": \"2\"}, \n \"inputs\": [[368, 0, 0], [367, 0, 0]]\n }, \n {\n \"op\": \"elemwise_add\", \n \"name\": \"plus1\", \n \"inputs\": [[367, 0, 0], [369, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_aggr_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_aggr_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c1_aggr\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"64\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[370, 0, 0], [371, 0, 0], [372, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_aggr_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_aggr_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_aggr_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_aggr_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c1_aggr_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[373, 0, 0], [374, 0, 0], [375, 0, 0], [376, 0, 1], [377, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"rf_c1_aggr_relu\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[378, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_conv1_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_conv1_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c1_det_conv1\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"32\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[379, 0, 0], [380, 0, 0], [381, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_conv1_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_conv1_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_conv1_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_conv1_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c1_det_conv1_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[382, 0, 0], [383, 0, 0], [384, 0, 0], [385, 0, 1], [386, 0, 1]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv1_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv1_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c1_det_context_conv1\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"16\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[379, 0, 0], [388, 0, 0], [389, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv1_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv1_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv1_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv1_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c1_det_context_conv1_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[390, 0, 0], [391, 0, 0], [392, 0, 0], [393, 0, 1], [394, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"rf_c1_det_context_conv1_relu\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[395, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv2_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv2_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c1_det_context_conv2\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"16\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[396, 0, 0], [397, 0, 0], [398, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv2_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv2_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv2_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv2_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c1_det_context_conv2_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[399, 0, 0], [400, 0, 0], [401, 0, 0], [402, 0, 1], [403, 0, 1]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv3_1_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv3_1_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c1_det_context_conv3_1\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"16\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[396, 0, 0], [405, 0, 0], [406, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv3_1_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv3_1_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv3_1_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv3_1_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c1_det_context_conv3_1_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[407, 0, 0], [408, 0, 0], [409, 0, 0], [410, 0, 1], [411, 0, 1]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"rf_c1_det_context_conv3_1_relu\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[412, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv3_2_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv3_2_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"rf_c1_det_context_conv3_2\", \n \"attrs\": {\n \"kernel\": \"(3, 3)\", \n \"num_filter\": \"16\", \n \"pad\": \"(1, 1)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[413, 0, 0], [414, 0, 0], [415, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv3_2_bn_gamma\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv3_2_bn_beta\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv3_2_bn_moving_mean\", \n \"attrs\": {\n \"__init__\": \"[\\\"zero\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"rf_c1_det_context_conv3_2_bn_moving_var\", \n \"attrs\": {\n \"__init__\": \"[\\\"one\\\", {}]\", \n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"BatchNorm\", \n \"name\": \"rf_c1_det_context_conv3_2_bn\", \n \"attrs\": {\n \"eps\": \"2e-05\", \n \"fix_gamma\": \"False\", \n \"momentum\": \"0.9\"\n }, \n \"inputs\": [[416, 0, 0], [417, 0, 0], [418, 0, 0], [419, 0, 1], [420, 0, 1]]\n }, \n {\n \"op\": \"Concat\", \n \"name\": \"rf_c1_det_concat\", \n \"attrs\": {\n \"dim\": \"1\", \n \"num_args\": \"3\"\n }, \n \"inputs\": [[387, 0, 0], [404, 0, 0], [421, 0, 0]]\n }, \n {\n \"op\": \"Activation\", \n \"name\": \"rf_c1_det_concat_relu\", \n \"attrs\": {\"act_type\": \"relu\"}, \n \"inputs\": [[422, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_cls_score_stride8_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_cls_score_stride8_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"face_rpn_cls_score_stride8\", \n \"attrs\": {\n \"kernel\": \"(1, 1)\", \n \"num_filter\": \"4\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[423, 0, 0], [424, 0, 0], [425, 0, 0]]\n }, \n {\n \"op\": \"Reshape\", \n \"name\": \"face_rpn_cls_score_reshape_stride8\", \n \"attrs\": {\"shape\": \"(0, 2, -1, 0)\"}, \n \"inputs\": [[426, 0, 0]]\n }, \n {\n \"op\": \"SoftmaxActivation\", \n \"name\": \"face_rpn_cls_prob_stride8\", \n \"attrs\": {\"mode\": \"channel\"}, \n \"inputs\": [[427, 0, 0]]\n }, \n {\n \"op\": \"Reshape\", \n \"name\": \"face_rpn_cls_prob_reshape_stride8\", \n \"attrs\": {\"shape\": \"(0, 4, -1, 0)\"}, \n \"inputs\": [[428, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_bbox_pred_stride8_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_bbox_pred_stride8_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"face_rpn_bbox_pred_stride8\", \n \"attrs\": {\n \"kernel\": \"(1, 1)\", \n \"num_filter\": \"8\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[423, 0, 0], [430, 0, 0], [431, 0, 0]]\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_landmark_pred_stride8_weight\", \n \"attrs\": {\n \"__init__\": \"[\\\"normal\\\", {\\\"sigma\\\": 0.01}]\", \n \"__lr_mult__\": \"1.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"null\", \n \"name\": \"face_rpn_landmark_pred_stride8_bias\", \n \"attrs\": {\n \"__init__\": \"[\\\"constant\\\", {\\\"value\\\": 0.0}]\", \n \"__lr_mult__\": \"2.0\", \n \"__wd_mult__\": \"0.0\"\n }, \n \"inputs\": []\n }, \n {\n \"op\": \"Convolution\", \n \"name\": \"face_rpn_landmark_pred_stride8\", \n \"attrs\": {\n \"kernel\": \"(1, 1)\", \n \"num_filter\": \"20\", \n \"pad\": \"(0, 0)\", \n \"stride\": \"(1, 1)\"\n }, \n \"inputs\": [[423, 0, 0], [433, 0, 0], [434, 0, 0]]\n }\n ], \n \"arg_nodes\": [\n 0, \n 1, \n 3, \n 4, \n 5, \n 6, \n 9, \n 11, \n 12, \n 13, \n 14, \n 17, \n 19, \n 20, \n 21, \n 22, \n 25, \n 27, \n 28, \n 29, \n 30, \n 33, \n 35, \n 36, \n 37, \n 38, \n 41, \n 43, \n 44, \n 45, \n 46, \n 49, \n 51, \n 52, \n 53, \n 54, \n 57, \n 59, \n 60, \n 61, \n 62, \n 65, \n 67, \n 68, \n 69, \n 70, \n 73, \n 75, \n 76, \n 77, \n 78, \n 81, \n 83, \n 84, \n 85, \n 86, \n 89, \n 91, \n 92, \n 93, \n 94, \n 97, \n 99, \n 100, \n 101, \n 102, \n 105, \n 107, \n 108, \n 109, \n 110, \n 113, \n 115, \n 116, \n 117, \n 118, \n 121, \n 123, \n 124, \n 125, \n 126, \n 129, \n 131, \n 132, \n 133, \n 134, \n 137, \n 139, \n 140, \n 141, \n 142, \n 145, \n 147, \n 148, \n 149, \n 150, \n 153, \n 155, \n 156, \n 157, \n 158, \n 161, \n 163, \n 164, \n 165, \n 166, \n 169, \n 171, \n 172, \n 173, \n 174, \n 177, \n 179, \n 180, \n 181, \n 182, \n 185, \n 187, \n 188, \n 189, \n 190, \n 193, \n 195, \n 196, \n 197, \n 198, \n 201, \n 203, \n 204, \n 205, \n 206, \n 209, \n 211, \n 212, \n 213, \n 214, \n 217, \n 218, \n 220, \n 221, \n 222, \n 223, \n 226, \n 227, \n 229, \n 230, \n 231, \n 232, \n 234, \n 235, \n 237, \n 238, \n 239, \n 240, \n 243, \n 244, \n 246, \n 247, \n 248, \n 249, \n 251, \n 252, \n 254, \n 255, \n 256, \n 257, \n 260, \n 261, \n 263, \n 264, \n 265, \n 266, \n 270, \n 271, \n 276, \n 277, \n 279, \n 280, \n 282, \n 283, \n 285, \n 286, \n 287, \n 288, \n 294, \n 295, \n 297, \n 298, \n 299, \n 300, \n 303, \n 304, \n 306, \n 307, \n 308, \n 309, \n 311, \n 312, \n 314, \n 315, \n 316, \n 317, \n 320, \n 321, \n 323, \n 324, \n 325, \n 326, \n 328, \n 329, \n 331, \n 332, \n 333, \n 334, \n 337, \n 338, \n 340, \n 341, \n 342, \n 343, \n 347, \n 348, \n 353, \n 354, \n 356, \n 357, \n 359, \n 360, \n 362, \n 363, \n 364, \n 365, \n 371, \n 372, \n 374, \n 375, \n 376, \n 377, \n 380, \n 381, \n 383, \n 384, \n 385, \n 386, \n 388, \n 389, \n 391, \n 392, \n 393, \n 394, \n 397, \n 398, \n 400, \n 401, \n 402, \n 403, \n 405, \n 406, \n 408, \n 409, \n 410, \n 411, \n 414, \n 415, \n 417, \n 418, \n 419, \n 420, \n 424, \n 425, \n 430, \n 431, \n 433, \n 434\n ], \n \"node_row_ptr\": [\n 0, \n 1, \n 2, \n 3, \n 4, \n 5, \n 6, \n 7, \n 10, \n 11, \n 12, \n 13, \n 14, \n 15, \n 16, \n 17, \n 20, \n 21, \n 22, \n 23, \n 24, \n 25, \n 26, \n 27, \n 30, \n 31, \n 32, \n 33, \n 34, \n 35, \n 36, \n 37, \n 40, \n 41, \n 42, \n 43, \n 44, \n 45, \n 46, \n 47, \n 50, \n 51, \n 52, \n 53, \n 54, \n 55, \n 56, \n 57, \n 60, \n 61, \n 62, \n 63, \n 64, \n 65, \n 66, \n 67, \n 70, \n 71, \n 72, \n 73, \n 74, \n 75, \n 76, \n 77, \n 80, \n 81, \n 82, \n 83, \n 84, \n 85, \n 86, \n 87, \n 90, \n 91, \n 92, \n 93, \n 94, \n 95, \n 96, \n 97, \n 100, \n 101, \n 102, \n 103, \n 104, \n 105, \n 106, \n 107, \n 110, \n 111, \n 112, \n 113, \n 114, \n 115, \n 116, \n 117, \n 120, \n 121, \n 122, \n 123, \n 124, \n 125, \n 126, \n 127, \n 130, \n 131, \n 132, \n 133, \n 134, \n 135, \n 136, \n 137, \n 140, \n 141, \n 142, \n 143, \n 144, \n 145, \n 146, \n 147, \n 150, \n 151, \n 152, \n 153, \n 154, \n 155, \n 156, \n 157, \n 160, \n 161, \n 162, \n 163, \n 164, \n 165, \n 166, \n 167, \n 170, \n 171, \n 172, \n 173, \n 174, \n 175, \n 176, \n 177, \n 180, \n 181, \n 182, \n 183, \n 184, \n 185, \n 186, \n 187, \n 190, \n 191, \n 192, \n 193, \n 194, \n 195, \n 196, \n 197, \n 200, \n 201, \n 202, \n 203, \n 204, \n 205, \n 206, \n 207, \n 210, \n 211, \n 212, \n 213, \n 214, \n 215, \n 216, \n 217, \n 220, \n 221, \n 222, \n 223, \n 224, \n 225, \n 226, \n 227, \n 230, \n 231, \n 232, \n 233, \n 234, \n 235, \n 236, \n 237, \n 240, \n 241, \n 242, \n 243, \n 244, \n 245, \n 246, \n 247, \n 250, \n 251, \n 252, \n 253, \n 254, \n 255, \n 256, \n 257, \n 260, \n 261, \n 262, \n 263, \n 264, \n 265, \n 266, \n 267, \n 270, \n 271, \n 272, \n 273, \n 274, \n 275, \n 276, \n 277, \n 278, \n 281, \n 282, \n 283, \n 284, \n 285, \n 286, \n 287, \n 288, \n 289, \n 292, \n 293, \n 294, \n 295, \n 296, \n 297, \n 298, \n 299, \n 302, \n 303, \n 304, \n 305, \n 306, \n 307, \n 308, \n 309, \n 310, \n 313, \n 314, \n 315, \n 316, \n 317, \n 318, \n 319, \n 320, \n 323, \n 324, \n 325, \n 326, \n 327, \n 328, \n 329, \n 330, \n 331, \n 334, \n 335, \n 336, \n 337, \n 338, \n 339, \n 340, \n 341, \n 342, \n 343, \n 344, \n 345, \n 346, \n 347, \n 348, \n 349, \n 350, \n 351, \n 352, \n 353, \n 354, \n 355, \n 358, \n 359, \n 360, \n 361, \n 362, \n 363, \n 364, \n 365, \n 366, \n 367, \n 368, \n 369, \n 372, \n 373, \n 374, \n 375, \n 376, \n 377, \n 378, \n 379, \n 380, \n 383, \n 384, \n 385, \n 386, \n 387, \n 388, \n 389, \n 390, \n 393, \n 394, \n 395, \n 396, \n 397, \n 398, \n 399, \n 400, \n 401, \n 404, \n 405, \n 406, \n 407, \n 408, \n 409, \n 410, \n 411, \n 414, \n 415, \n 416, \n 417, \n 418, \n 419, \n 420, \n 421, \n 422, \n 425, \n 426, \n 427, \n 428, \n 429, \n 430, \n 431, \n 432, \n 433, \n 434, \n 435, \n 436, \n 437, \n 438, \n 439, \n 440, \n 441, \n 442, \n 443, \n 444, \n 445, \n 446, \n 449, \n 450, \n 451, \n 452, \n 453, \n 454, \n 455, \n 456, \n 457, \n 458, \n 459, \n 460, \n 463, \n 464, \n 465, \n 466, \n 467, \n 468, \n 469, \n 470, \n 471, \n 474, \n 475, \n 476, \n 477, \n 478, \n 479, \n 480, \n 481, \n 484, \n 485, \n 486, \n 487, \n 488, \n 489, \n 490, \n 491, \n 492, \n 495, \n 496, \n 497, \n 498, \n 499, \n 500, \n 501, \n 502, \n 505, \n 506, \n 507, \n 508, \n 509, \n 510, \n 511, \n 512, \n 513, \n 516, \n 517, \n 518, \n 519, \n 520, \n 521, \n 522, \n 523, \n 524, \n 525, \n 526, \n 527, \n 528, \n 529, \n 530\n ], \n \"heads\": [[272, 0, 0], [349, 0, 0], [426, 0, 0], [269,0,0], [346,0,0], [423,0,0]], \n \"attrs\": {\"mxnet_version\": [\"int\", 10300]}\n}", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# Mixture Density Networks with Edward, Keras and TensorFlow\n\nThis notebook explains how to implement Mixture Density Networks (MDN) with Edward, Keras and TensorFlow.\nKeep in mind that if you want to use Keras and TensorFlow, like we do in this notebook, you need to set the backend of Keras to TensorFlow, [here](http://keras.io/backend/) it is explained how to do that.\n\nIn you are not familiar with MDNs have a look at the [following blog post](http://cbonnett.github.io/MDN.html) or at orginal [paper](http://research.microsoft.com/en-us/um/people/cmbishop/downloads/Bishop-NCRG-94-004.pdf) by Bishop.\n\nEdward implements many probability distribution functions that are TensorFlow compatible, this makes it attractive to use Edward for MDNs. \n\nHere are all the distributions that are currently implemented in Edward, there are more to come:\n\n1. [Bernoulli](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L49)\n2. [Beta](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L58)\n3. [Binomial](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L68)\n4. [Chi Squared](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L79)\n5. [Dirichlet](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L89)\n6. [Exponential](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L109)\n7. [Gamma](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L118)\n8. [Geometric](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L129)\n9. [Inverse Gamma](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L138)\n10. [log Normal](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L155)\n11. [Multinomial](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L165)\n12. [Multivariate Normal](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L194)\n13. [Negative Binomial](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L283)\n14. [Normal](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L294)\n15. [Poisson](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L310)\n16. [Student-t](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L319)\n17. [Truncated Normal](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L333)\n18. [Uniform](https://github.com/blei-lab/edward/blob/master/edward/stats/distributions.py#L352)\n\nLet's start with the necessary imports.", "_____no_output_____" ] ], [ [ "# imports\n%matplotlib inline\nimport matplotlib.pyplot as plt\nimport seaborn as sns\n\nimport edward as ed\nimport numpy as np\nimport tensorflow as tf\n\nfrom edward.stats import norm # Normal distribution from Edward. \nfrom keras import backend as K\nfrom keras.layers import Dense\nfrom sklearn.cross_validation import train_test_split", "_____no_output_____" ] ], [ [ "We will need some functions to plot the results later on, these are defined in the next code block. ", "_____no_output_____" ] ], [ [ "from scipy.stats import norm as normal\ndef plot_normal_mix(pis, mus, sigmas, ax, label='', comp=True):\n \"\"\"\n Plots the mixture of Normal models to axis=ax\n comp=True plots all components of mixture model\n \"\"\"\n x = np.linspace(-10.5, 10.5, 250)\n final = np.zeros_like(x)\n for i, (weight_mix, mu_mix, sigma_mix) in enumerate(zip(pis, mus, sigmas)):\n temp = normal.pdf(x, mu_mix, sigma_mix) * weight_mix\n final = final + temp\n if comp:\n ax.plot(x, temp, label='Normal ' + str(i))\n ax.plot(x, final, label='Mixture of Normals ' + label)\n ax.legend(fontsize=13)\n \ndef sample_from_mixture(x, pred_weights, pred_means, pred_std, amount):\n \"\"\"\n Draws samples from mixture model. \n Returns 2 d array with input X and sample from prediction of Mixture Model\n \"\"\"\n samples = np.zeros((amount, 2))\n n_mix = len(pred_weights[0])\n to_choose_from = np.arange(n_mix)\n for j,(weights, means, std_devs) in enumerate(zip(pred_weights, pred_means, pred_std)):\n index = np.random.choice(to_choose_from, p=weights)\n samples[j,1]= normal.rvs(means[index], std_devs[index], size=1)\n samples[j,0]= x[j]\n if j == amount -1:\n break\n return samples", "_____no_output_____" ] ], [ [ "## Making some toy-data to play with.\n\nThis is the same toy-data problem set as used in the [blog post](http://blog.otoro.net/2015/11/24/mixture-density-networks-with-tensorflow/) by Otoro where he explains MDNs. This is an inverse problem as you can see, for every ```X``` there are multiple ```y``` solutions.", "_____no_output_____" ] ], [ [ "def build_toy_dataset(nsample=40000):\n y_data = np.float32(np.random.uniform(-10.5, 10.5, (1, nsample))).T\n r_data = np.float32(np.random.normal(size=(nsample, 1))) # random noise\n x_data = np.float32(np.sin(0.75 * y_data) * 7.0 + y_data * 0.5 + r_data * 1.0)\n return train_test_split(x_data, y_data, random_state=42, train_size=0.1)\n\nX_train, X_test, y_train, y_test = build_toy_dataset()\nprint(\"Size of features in training data: {:s}\".format(X_train.shape))\nprint(\"Size of output in training data: {:s}\".format(y_train.shape))\nprint(\"Size of features in test data: {:s}\".format(X_test.shape))\nprint(\"Size of output in test data: {:s}\".format(y_test.shape))\n\nsns.regplot(X_train, y_train, fit_reg=False)", "Size of features in training data: (4000, 1)\nSize of output in training data: (4000, 1)\nSize of features in test data: (36000, 1)\nSize of output in test data: (36000, 1)\n" ] ], [ [ "### Building a MDN using Edward, Keras and TF\n\nWe will define a class that can be used to construct MDNs. In this notebook we will be using a mixture of Normal Distributions. The advantage of defining a class is that we can easily reuse this to build other MDNs with different amount of mixture components. Furthermore, this makes it play nicely with Edward.", "_____no_output_____" ] ], [ [ "class MixtureDensityNetwork:\n \"\"\"\n Mixture density network for outputs y on inputs x.\n p((x,y), (z,theta))\n = sum_{k=1}^K pi_k(x; theta) Normal(y; mu_k(x; theta), sigma_k(x; theta))\n where pi, mu, sigma are the output of a neural network taking x\n as input and with parameters theta. There are no latent variables\n z, which are hidden variables we aim to be Bayesian about.\n \"\"\"\n def __init__(self, K):\n self.K = K # here K is the amount of Mixtures \n\n def mapping(self, X):\n \"\"\"pi, mu, sigma = NN(x; theta)\"\"\"\n hidden1 = Dense(15, activation='relu')(X) # fully-connected layer with 15 hidden units\n hidden2 = Dense(15, activation='relu')(hidden1) \n self.mus = Dense(self.K)(hidden2) # the means \n self.sigmas = Dense(self.K, activation=K.exp)(hidden2) # the variance\n self.pi = Dense(self.K, activation=K.softmax)(hidden2) # the mixture components\n\n def log_prob(self, xs, zs=None):\n \"\"\"log p((xs,ys), (z,theta)) = sum_{n=1}^N log p((xs[n,:],ys[n]), theta)\"\"\"\n # Note there are no parameters we're being Bayesian about. The\n # parameters are baked into how we specify the neural networks.\n X, y = xs\n self.mapping(X)\n result = tf.exp(norm.logpdf(y, self.mus, self.sigmas))\n result = tf.mul(result, self.pi)\n result = tf.reduce_sum(result, 1)\n result = tf.log(result)\n return tf.reduce_sum(result)", "_____no_output_____" ] ], [ [ "We can set a seed in Edward so we can reproduce all the random components. The following line:\n\n```ed.set_seed(42)```\n\nsets the seed in Numpy and TensorFlow under the [hood](https://github.com/blei-lab/edward/blob/master/edward/util.py#L191). We use the class we defined above to initiate the MDN with 20 mixtures, this now can be used as an Edward model.", "_____no_output_____" ] ], [ [ "ed.set_seed(42)\nmodel = MixtureDensityNetwork(20)", "_____no_output_____" ] ], [ [ "In the following code cell we define the TensorFlow placeholders that are then used to define the Edward data model.\nThe following line passes the ```model``` and ```data``` to ```MAP``` from Edward which is then used to initialise the TensorFlow variables. \n\n```inference = ed.MAP(model, data)``` \n\nMAP is a Bayesian concept and stands for Maximum A Posteriori, it tries to find the set of parameters which maximizes the posterior distribution. In the example here we don't have a prior, in a Bayesian context this means we have a flat prior. For a flat prior MAP is equivalent to Maximum Likelihood Estimation. Edward is designed to be Bayesian about its statistical inference. The cool thing about MDN's with Edward is that we could easily include priors! ", "_____no_output_____" ] ], [ [ "X = tf.placeholder(tf.float32, shape=(None, 1))\ny = tf.placeholder(tf.float32, shape=(None, 1))\ndata = ed.Data([X, y]) # Make Edward Data model\n\ninference = ed.MAP(model, data) # Make the inference model\nsess = tf.Session() # Start TF session \nK.set_session(sess) # Pass session info to Keras\ninference.initialize(sess=sess) # Initialize all TF variables using the Edward interface ", "_____no_output_____" ] ], [ [ "Having done that we can train the MDN in TensorFlow just like we normally would, and we can get out the predictions we are interested in from ```model```, in this case: \n\n* ```model.pi``` the mixture components, \n* ```model.mus``` the means,\n* ```model.sigmas``` the standard deviations. \n\nThis is done in the last line of the code cell :\n```\npred_weights, pred_means, pred_std = sess.run([model.pi, model.mus, model.sigmas], \n feed_dict={X: X_test})\n```\n\nThe default minimisation technique used is ADAM with a decaying scale factor.\nThis can be seen [here](https://github.com/blei-lab/edward/blob/master/edward/inferences.py#L94) in the code base of Edward. Having a decaying scale factor is not the standard way of using ADAM, this is inspired by the Automatic Differentiation Variational Inference [(ADVI)](http://arxiv.org/abs/1603.00788) work where it was used in the RMSPROP minimizer. \n\nThe loss that is minimised in the ```MAP``` model from Edward is the negative log-likelihood, this calculation uses the ```log_prob``` method in the ```MixtureDensityNetwork``` class we defined above. \nThe ```build_loss``` method in the ```MAP``` class can be found [here](https://github.com/blei-lab/edward/blob/master/edward/inferences.py#L396). \n\nHowever the method ```inference.loss``` used below, returns the log-likelihood, so we expect this quantity to be maximized.", "_____no_output_____" ] ], [ [ "NEPOCH = 1000\ntrain_loss = np.zeros(NEPOCH)\ntest_loss = np.zeros(NEPOCH)\nfor i in range(NEPOCH):\n _, train_loss[i] = sess.run([inference.train, inference.loss],\n feed_dict={X: X_train, y: y_train})\n test_loss[i] = sess.run(inference.loss, feed_dict={X: X_test, y: y_test})\n \npred_weights, pred_means, pred_std = sess.run([model.pi, model.mus, model.sigmas], \n feed_dict={X: X_test})", "_____no_output_____" ] ], [ [ "We can plot the log-likelihood of the training and test sample as function of training epoch.\nKeep in mind that ```inference.loss``` returns the total log-likelihood, so not the loss per data point, so in the plotting routine we divide by the size of the train and test data respectively. \nWe see that it converges after 400 training steps.", "_____no_output_____" ] ], [ [ "fig, axes = plt.subplots(nrows=1, ncols=1, figsize=(16, 3.5))\nplt.plot(np.arange(NEPOCH), test_loss/len(X_test), label='Test')\nplt.plot(np.arange(NEPOCH), train_loss/len(X_train), label='Train')\nplt.legend(fontsize=20)\nplt.xlabel('Epoch', fontsize=15)\nplt.ylabel('Log-likelihood', fontsize=15)", "_____no_output_____" ] ], [ [ "Next we can have a look at how some individual examples perform. Keep in mind this is an inverse problem\nso we can't get the answer correct, we can hope that the truth lies in area where the model has high probability.\nIn the next plot the truth is the vertical grey line while the blue line is the prediction of the mixture density network. As you can see, we didn't do too bad.", "_____no_output_____" ] ], [ [ "obj = [0, 4, 6]\nfig, axes = plt.subplots(nrows=3, ncols=1, figsize=(16, 6))\n\nplot_normal_mix(pred_weights[obj][0], pred_means[obj][0], pred_std[obj][0], axes[0], comp=False)\naxes[0].axvline(x=y_test[obj][0], color='black', alpha=0.5)\n\nplot_normal_mix(pred_weights[obj][2], pred_means[obj][2], pred_std[obj][2], axes[1], comp=False)\naxes[1].axvline(x=y_test[obj][2], color='black', alpha=0.5)\n\nplot_normal_mix(pred_weights[obj][1], pred_means[obj][1], pred_std[obj][1], axes[2], comp=False)\naxes[2].axvline(x=y_test[obj][1], color='black', alpha=0.5)", "_____no_output_____" ] ], [ [ "We can check the ensemble by drawing samples of the prediction and plotting the density of those. \nSeems the MDN learned what it needed too.", "_____no_output_____" ] ], [ [ "a = sample_from_mixture(X_test, pred_weights, pred_means, pred_std, amount=len(X_test))\nsns.jointplot(a[:,0], a[:,1], kind=\"hex\", color=\"#4CB391\", ylim=(-10,10), xlim=(-14,14))", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<H3>Importing Required Libraries", "_____no_output_____" ] ], [ [ "from pyspark.sql import SparkSession\nfrom pyspark.sql import functions as F\n", "_____no_output_____" ] ], [ [ "<H3>Getting Spark Session", "_____no_output_____" ] ], [ [ "spark = SparkSession.builder.getOrCreate()", "_____no_output_____" ] ], [ [ "<H3>Reading CSV", "_____no_output_____" ] ], [ [ "df = spark.read.csv(\"Big_Cities_Health_Data_Inventory.csv\", header=True)", "_____no_output_____" ], [ "df.show(10)", "+------------------+--------------------+----+------+---------------+-----+--------------------+--------------------------+--------------------+-------+-----+\n|Indicator Category| Indicator|Year|Gender|Race/ Ethnicity|Value| Place|BCHC Requested Methodology| Source|Methods|Notes|\n+------------------+--------------------+----+------+---------------+-----+--------------------+--------------------------+--------------------+-------+-----+\n| HIV/AIDS|AIDS Diagnoses Ra...|2013| Both| All| 30.4|Atlanta (Fulton C...| AIDS cases diagno...|Diagnoses numbers...| null| null|\n| HIV/AIDS|AIDS Diagnoses Ra...|2012| Both| All| 39.6|Atlanta (Fulton C...| AIDS cases diagno...|Diagnoses numbers...| null| null|\n| HIV/AIDS|AIDS Diagnoses Ra...|2011| Both| All| 41.7|Atlanta (Fulton C...| AIDS cases diagno...|Diagnoses numbers...| null| null|\n| Cancer|All Types of Canc...|2013| Male| All|195.8|Atlanta (Fulton C...| 2012, 2013, 2014;...|National Center f...| null| null|\n| Cancer|All Types of Canc...|2013|Female| All|135.5|Atlanta (Fulton C...| 2012, 2013, 2014;...|National Center f...| null| null|\n| Cancer|All Types of Canc...|2013| Both| All|159.3|Atlanta (Fulton C...| 2012, 2013, 2014;...|National Center f...| null| null|\n| Cancer|All Types of Canc...|2012| Male| All|199.2|Atlanta (Fulton C...| 2012, 2013, 2014;...|National Center f...| null| null|\n| Cancer|All Types of Canc...|2012|Female| All|137.6|Atlanta (Fulton C...| 2012, 2013, 2014;...|National Center f...| null| null|\n| Cancer|All Types of Canc...|2012| Both| All|160.3|Atlanta (Fulton C...| 2012, 2013, 2014;...|National Center f...| null| null|\n| Cancer|All Types of Canc...|2011| Male| All|196.2|Atlanta (Fulton C...| 2012, 2013, 2014;...|National Center f...| null| null|\n+------------------+--------------------+----+------+---------------+-----+--------------------+--------------------------+--------------------+-------+-----+\nonly showing top 10 rows\n\n" ] ], [ [ "<H3>Printing Schema", "_____no_output_____" ] ], [ [ "df.printSchema()", "root\n |-- Indicator Category: string (nullable = true)\n |-- Indicator: string (nullable = true)\n |-- Year: string (nullable = true)\n |-- Gender: string (nullable = true)\n |-- Race/ Ethnicity: string (nullable = true)\n |-- Value: string (nullable = true)\n |-- Place: string (nullable = true)\n |-- BCHC Requested Methodology: string (nullable = true)\n |-- Source: string (nullable = true)\n |-- Methods: string (nullable = true)\n |-- Notes: string (nullable = true)\n\n" ] ], [ [ "<H3>Dropping Unwanted Columns", "_____no_output_____" ] ], [ [ "df = df.drop(\"Notes\", \"Methods\", \"Source\", \"BCHC Requested Methodology\")", "_____no_output_____" ], [ "df.printSchema()", "root\n |-- Indicator Category: string (nullable = true)\n |-- Indicator: string (nullable = true)\n |-- Year: string (nullable = true)\n |-- Gender: string (nullable = true)\n |-- Race/ Ethnicity: string (nullable = true)\n |-- Value: string (nullable = true)\n |-- Place: string (nullable = true)\n\n" ] ], [ [ "<H3>Counting Null Values", "_____no_output_____" ] ], [ [ "df.select([F.count(F.when(F.isnan(c) | F.col(c).isNull(), c)).alias(c) for c in df.columns]).show()", "+------------------+---------+----+------+---------------+-----+-----+\n|Indicator Category|Indicator|Year|Gender|Race/ Ethnicity|Value|Place|\n+------------------+---------+----+------+---------------+-----+-----+\n| 0| 28| 28| 218| 212| 231| 218|\n+------------------+---------+----+------+---------------+-----+-----+\n\n" ] ], [ [ "Since there are several null values in the columns as shown in the table above, first steps would be to remove / replace null values in each column", "_____no_output_____" ], [ "<H3>Working with Null Values", "_____no_output_____" ] ], [ [ "df.filter(df[\"Indicator\"].isNull()).show(28)", "+--------------------+---------+----+------+---------------+-----+-----+\n| Indicator Category|Indicator|Year|Gender|Race/ Ethnicity|Value|Place|\n+--------------------+---------+----+------+---------------+-----+-----+\n| FOR THE POPULATI...| null|null| null| null| null| null|\n| 12 MONTHS (S1701)\"| null|null| null| null| null| null|\n| (S1701)\"| null|null| null| null| null| null|\n| (S1701)\"| null|null| null| null| null| null|\n|from the flu shot...| null|null| null| null| null| null|\n|from the flu shot...| null|null| null| null| null| null|\n|from the flu shot...| null|null| null| null| null| null|\n|from the flu shot...| null|null| null| null| null| null|\n|from the flu shot...| null|null| null| null| null| null|\n|from the flu shot...| null|null| null| null| null| null|\n|from the flu shot...| null|null| null| null| null| null|\n|(percent of respo...| null|null| null| null| null| null|\n|(percent of respo...| null|null| null| null| null| null|\n|(percent of respo...| null|null| null| null| null| null|\n| your nose?\"\" \"| null|null| null| null| null| null|\n| your nose?\"\" \"| null|null| null| null| null| null|\n| your nose?\"\" \"| null|null| null| null| null| null|\n| your nose?\"\" \"| null|null| null| null| null| null|\n| your nose?\"\" \"| null|null| null| null| null| null|\n| your nose?\"\" \"| null|null| null| null| null| null|\n| your nose?\"\" \"| null|null| null| null| null| null|\n|(percent of respo...| null|null| null| null| null| null|\n|(see note above a...| null|null| null| null| null| null|\n|(see note above a...| null|null| null| null| null| null|\n|(see note above a...| null|null| null| null| null| null|\n|(see note above a...| null|null| null| null| null| null|\n|(see note above a...| null|null| null| null| null| null|\n|(see note above a...| null|null| null| null| null| null|\n+--------------------+---------+----+------+---------------+-----+-----+\n\n" ] ], [ [ "Since all the rows that have null values in Indicator have null values for other columns like Year, Gender, Race and etc, it would be better to remove these observations", "_____no_output_____" ] ], [ [ "# Counting total number of rows in the dataset to compare with the rows after null value rows are removed.\nrows_count_pre = df.count()\nprint(\"Total number of rows before deleting: \",rows_count_pre)", "Total number of rows before deleting: 13730\n" ], [ "# deleting all the rows where there are null values in the columns mentioned below\ndf = df.na.drop(subset=[\"Indicator\", \"Year\", \"Gender\", \"Race/ Ethnicity\", \"Value\", \"Place\"])", "_____no_output_____" ], [ "rows_count_post = df.count()\nprint(\"Total number of rows after deleting: \",rows_count_post)", "Total number of rows after deleting: 13499\n" ], [ "total_rows_removed = rows_count_pre - rows_count_post\nprint(\"Total number of rows deleted: \", total_rows_removed)", "Total number of rows deleted: 231\n" ], [ "#Checking the null values again to see if the dataset is clean\ndf.select([F.count(F.when(F.isnan(c) | F.col(c).isNull(), c)).alias(c) for c in df.columns]).show()", "+------------------+---------+----+------+---------------+-----+-----+\n|Indicator Category|Indicator|Year|Gender|Race/ Ethnicity|Value|Place|\n+------------------+---------+----+------+---------------+-----+-----+\n| 0| 0| 0| 0| 0| 0| 0|\n+------------------+---------+----+------+---------------+-----+-----+\n\n" ] ], [ [ "The results above show that all the rows with null values have been deleted from the dataset. This completes the step of removing all the null values from the dataset", "_____no_output_____" ], [ "<H3>Splitting the Place Column into City and State Columns", "_____no_output_____" ] ], [ [ "split_col = F.split(df[\"Place\"], ',')\ndf = df.withColumn(\"City_County\", split_col.getItem(0))\ndf = df.withColumn(\"State\", split_col.getItem(1))\ndf.select(\"City_County\", \"State\").show(truncate=False)", "+-----------------------+-----+\n|City_County |State|\n+-----------------------+-----+\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n|Atlanta (Fulton County)| GA |\n+-----------------------+-----+\nonly showing top 20 rows\n\n" ], [ "Creating a User Defined Function to take care of the City_County column to extract the city. Same can be done using", "_____no_output_____" ], [ "import re\ndef extract_city(city_str):\n result = re.sub(r'\\([^)]*\\)', '', city_str)\n return result", "_____no_output_____" ], [ "from pyspark.sql.types import StringType\nudfExtract = F.udf(extract_city, StringType())\ndf = df.withColumn(\"City\", udfExtract(df[\"City_County\"]))\ndf.select(\"City\", \"State\").show(truncate=False)", "+--------+-----+\n|City |State|\n+--------+-----+\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n|Atlanta | GA |\n+--------+-----+\nonly showing top 20 rows\n\n" ] ], [ [ "This sums up the cleaning process of data using PySpark. Below is the final state of the dataset", "_____no_output_____" ] ], [ [ "df.show()", "+--------------------+--------------------+----+------+---------------+-----+--------------------+--------------------+-----+--------+\n| Indicator Category| Indicator|Year|Gender|Race/ Ethnicity|Value| Place| City_County|State| City|\n+--------------------+--------------------+----+------+---------------+-----+--------------------+--------------------+-----+--------+\n| HIV/AIDS|AIDS Diagnoses Ra...|2013| Both| All| 30.4|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| HIV/AIDS|AIDS Diagnoses Ra...|2012| Both| All| 39.6|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| HIV/AIDS|AIDS Diagnoses Ra...|2011| Both| All| 41.7|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| Cancer|All Types of Canc...|2013| Male| All|195.8|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| Cancer|All Types of Canc...|2013|Female| All|135.5|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| Cancer|All Types of Canc...|2013| Both| All|159.3|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| Cancer|All Types of Canc...|2012| Male| All|199.2|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| Cancer|All Types of Canc...|2012|Female| All|137.6|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| Cancer|All Types of Canc...|2012| Both| All|160.3|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| Cancer|All Types of Canc...|2011| Male| All|196.2|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| Cancer|All Types of Canc...|2011|Female| All|147.0|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| Cancer|All Types of Canc...|2011| Both| All|165.2|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| Cancer|All Types of Canc...|2013| Both| Black|208.3|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| Cancer|All Types of Canc...|2012| Both| Black|202.7|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n|Maternal and Chil...|Infant Mortality ...|2012| Both| White| 4.5|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| Cancer|All Types of Canc...|2011| Both| Black|216.0|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| Cancer|All Types of Canc...|2013| Both| White|128.8|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| Cancer|All Types of Canc...|2012| Both| White|133.7|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n| Cancer|All Types of Canc...|2011| Both| White|132.0|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n|Life Expectancy a...|All-Cause Mortali...|2012|Female| All|578.4|Atlanta (Fulton C...|Atlanta (Fulton C...| GA|Atlanta |\n+--------------------+--------------------+----+------+---------------+-----+--------------------+--------------------+-----+--------+\nonly showing top 20 rows\n\n" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown", "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Exploratory Data Analysis\n\n* Dataset taken from https://github.com/Tariq60/LIAR-PLUS", "_____no_output_____" ], [ "## 1. Import Libraries", "_____no_output_____" ] ], [ [ "import os\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\n\nTRAIN_PATH = \"../data/raw/dataset/tsv/train2.tsv\"\nVAL_PATH = \"../data/raw/dataset/tsv/val2.tsv\"\nTEST_PATH = \"../data/raw/dataset/tsv/test2.tsv\"\n\ncolumns = [\"id\", \"statement_json\", \"label\", \"statement\", \"subject\", \"speaker\", \"speaker_title\", \"state_info\",\n \"party_affiliation\", \"barely_true_count\", \"false_count\", \"half_true_count\", \"mostly_true_count\",\n \"pants_fire_count\", \"context\", \"justification\"]\n\n", "_____no_output_____" ] ], [ [ "## 2. Read the dataset", "_____no_output_____" ] ], [ [ "train_df = pd.read_csv(TRAIN_PATH, sep=\"\\t\", names=columns)\nval_df = pd.read_csv(VAL_PATH, sep=\"\\t\", names=columns)\ntest_df = pd.read_csv(TEST_PATH, sep=\"\\t\", names=columns)", "_____no_output_____" ], [ "print(f\"Length of train set: {len(train_df)}\")\nprint(f\"Length of validation set: {len(val_df)}\")\nprint(f\"Length of test set: {len(test_df)}\")", "Length of train set: 10242\nLength of validation set: 1284\nLength of test set: 1267\n" ], [ "train_df.head()", "_____no_output_____" ] ], [ [ "## 3. Data Cleaning", "_____no_output_____" ], [ "* Some of the most important coloumns are \"label\", \"statement\".\n* Now we should check if any of them have null values.", "_____no_output_____" ] ], [ [ "print(\"Do we have empty strings in `label`?\")\npd.isna(train_df[\"label\"]).value_counts()", "Do we have empty strings in `label`?\n" ] ], [ [ "* 2 entries without any label\n* What exactly are those 2 entries?", "_____no_output_____" ] ], [ [ "train_df.loc[pd.isna(train_df[\"label\"]), :].index", "_____no_output_____" ], [ "train_df.loc[[2143]]\n", "_____no_output_____" ], [ "train_df.loc[[9377]]", "_____no_output_____" ] ], [ [ "* All the coloumns of those 2 entries are blank\n* Drop those 2 entries", "_____no_output_____" ] ], [ [ "train_df.dropna(subset=[\"label\"], inplace=True)\nlen(train_df)", "_____no_output_____" ] ], [ [ "## 4. Some Feature Analysis", "_____no_output_____" ], [ "### 4.1 Party Affiliation", "_____no_output_____" ] ], [ [ "print(train_df[\"party_affiliation\"].value_counts())\n\nif not os.path.exists(\"./img\"):\n os.makedirs(\"./img\")\n\nfig = plt.figure(figsize=(10, 6))\nparty_affil_plot = train_df[\"party_affiliation\"].value_counts().plot.bar()\nplt.tight_layout(pad=1)\nplt.savefig(\"img/party_affil_plot.png\", dpi=200)", "republican 4497\ndemocrat 3336\nnone 1744\norganization 219\nindependent 147\nnewsmaker 56\nlibertarian 40\nactivist 39\njournalist 38\ncolumnist 35\ntalk-show-host 26\nstate-official 20\nlabor-leader 11\ntea-party-member 10\nbusiness-leader 9\ngreen 3\neducation-official 2\nliberal-party-canada 1\ngovernment-body 1\nModerate 1\ndemocratic-farmer-labor 1\nocean-state-tea-party-action 1\nconstitution-party 1\nName: party_affiliation, dtype: int64\n" ] ], [ [ "### 4.2 States Stats", "_____no_output_____" ] ], [ [ "print(train_df[\"state_info\"].value_counts())\n\nfig = plt.figure(figsize=(10, 6))\nstate_info_plot = train_df[\"state_info\"].value_counts().plot.bar()\nplt.tight_layout(pad=1)\nplt.savefig(\"img/state_info_plot.png\", dpi=200)", "Texas 1009\nFlorida 997\nWisconsin 713\nNew York 657\nIllinois 556\n ... \nQatar 1\nVirginia 1\nUnited Kingdom 1\nChina 1\nRhode Island 1\nName: state_info, Length: 84, dtype: int64\n" ] ], [ [ "* Apparently, we have a state_info entry with value as \"Virginia director, Coalition to Stop Gun Violence\".\nIt should be replaced with \"Virginia\" only", "_____no_output_____" ] ], [ [ "train_df[train_df[\"state_info\"]==\"Virginia director, Coalition to Stop Gun Violence\"]", "_____no_output_____" ], [ "indx = train_df[train_df[\"state_info\"]==\"Virginia director, Coalition to Stop Gun Violence\"].index[0]\ntrain_df.loc[indx, \"state_info\"] = \"Virginia\"\n\nfig = plt.figure(figsize=(10, 6))\nstate_info_plot = train_df[\"state_info\"].value_counts().plot.bar()\nplt.tight_layout(pad=1)\nplt.savefig(\"img/state_info_plot.png\", dpi=200)", "_____no_output_____" ] ], [ [ "### 4.3 Label Distribution", "_____no_output_____" ] ], [ [ "print(train_df[\"label\"].value_counts())\n\nfig = plt.figure(figsize=(10, 6))\nlabel_stats_plot = train_df[\"label\"].value_counts().plot.bar()\nplt.tight_layout(pad=1)\nplt.savefig(\"img/label_stats_plot.png\", dpi=100)", "half-true 2114\nfalse 1995\nmostly-true 1962\ntrue 1676\nbarely-true 1654\npants-fire 839\nName: label, dtype: int64\n" ] ], [ [ "### 4.4 Speaker Distribution", "_____no_output_____" ] ], [ [ "print(train_df.speaker.value_counts())\n\nfig = plt.figure(figsize=(10, 6))\nspeaker_stats_plot = train_df[\"speaker\"].value_counts()[:10].plot.bar()\nplt.tight_layout(pad=1)\nplt.title(\"Speakers\")\nplt.savefig(\"img/speaker_stats_plot.png\", dpi=100)", "barack-obama 488\ndonald-trump 273\nhillary-clinton 239\nmitt-romney 176\nscott-walker 149\n ... \nlorraine-fende 1\nnfederation-o-independent-business-virginia 1\njim-moore 1\nscott-surovell 1\nalan-powell 1\nName: speaker, Length: 2910, dtype: int64\n" ], [ "print(train_df.speaker_title.value_counts())\n\nfig = plt.figure(figsize=(10, 6))\nspeaker_title_stats_plot = train_df[\"speaker_title\"].value_counts()[:10].plot.bar()\nplt.tight_layout(pad=1)\nplt.title(\"Speaker Title\")\nplt.savefig(\"img/speaker_title_stats_plot.png\", dpi=100)", "President 492\nU.S. Senator 479\nGovernor 391\nPresident-Elect 273\nU.S. senator 263\n ... \nPundit and communications consultant 1\nHarrisonburg city councilman 1\nTheme park company 1\nExecutive director, NARAL Pro-Choice Virginia 1\nPresident, The Whitman Strategy Group 1\nName: speaker_title, Length: 1184, dtype: int64\n" ] ], [ [ "### 4.5 Democrats vs Republicans\n\n* Let's see how the 2 main parties compete with each other in terms of\ntruthfulness in the labels", "_____no_output_____" ] ], [ [ "fig = plt.figure(figsize=(8,4))\n\nplt.suptitle(\"Party-wise Label\")\nax1 = fig.add_subplot(121)\nparty_wise = train_df[train_df[\"party_affiliation\"]==\"democrat\"][\"label\"].value_counts().to_frame()\nax1.pie(party_wise[\"label\"], labels=party_wise.index, autopct='%1.1f%%',\n startangle=90)\nax1.set_title(\"Democrat\")\n\nplt.suptitle(\"Party-wise Label\")\nax2 = fig.add_subplot(122)\nparty_wise = train_df[train_df[\"party_affiliation\"]==\"republican\"][\"label\"].value_counts().to_frame()\nax2.pie(party_wise[\"label\"], labels=party_wise.index, autopct='%1.1f%%',\n startangle=90)\nax2.set_title(\"Republican\")\nplt.tight_layout()\nplt.savefig(\"img/dems_gop_label_plot.png\", dpi=200)", "_____no_output_____" ] ], [ [ "* We can combine some labels to get a more simplified plot\n", "_____no_output_____" ] ], [ [ "def get_binary_label(label):\n if label in [\"pants-fire\", \"barely-true\", \"false\"]:\n return False\n elif label in [\"true\", \"half-true\", \"mostly-true\"]:\n return True\n\ntrain_df[\"binary_label\"] = train_df.label.apply(get_binary_label)", "_____no_output_____" ], [ "fig = plt.figure(figsize=(8,4))\n\nplt.suptitle(\"Party-wise Label\")\nax1 = fig.add_subplot(121)\nparty_wise = train_df[train_df[\"party_affiliation\"]==\"democrat\"][\"binary_label\"].value_counts().to_frame()\nax1.pie(party_wise[\"binary_label\"], labels=party_wise.index, autopct='%1.1f%%',\n startangle=90)\nax1.set_title(\"Democrat\")\n\nplt.suptitle(\"Party-wise Label\")\nax2 = fig.add_subplot(122)\nparty_wise = train_df[train_df[\"party_affiliation\"]==\"republican\"][\"binary_label\"].value_counts().to_frame()\nax2.pie(party_wise[\"binary_label\"], labels=party_wise.index, autopct='%1.1f%%',\n startangle=90)\nax2.set_title(\"Republican\")\nplt.tight_layout()\nplt.savefig(\"img/dems_gop_binary_label_plot.png\", dpi=200)", "_____no_output_____" ] ], [ [ "## 5. Sentiment Analysis", "_____no_output_____" ] ], [ [ "from textblob import TextBlob\n\npol = lambda x: TextBlob(x).sentiment.polarity\nsub = lambda x: TextBlob(x).sentiment.subjectivity\n\ntrain_df['polarity_true'] = train_df[train_df[\"binary_label\"]==True]['statement'].apply(pol)\ntrain_df['subjectivity_true'] = train_df[train_df[\"binary_label\"]==True]['statement'].apply(sub)\n\nplt.rcParams['figure.figsize'] = [10, 8]\n\nx = train_df[\"polarity_true\"]\ny = train_df[\"subjectivity_true\"]\nplt.scatter(x, y, color='blue')\n\nplt.title('Sentiment Analysis', fontsize=20)\nplt.xlabel('<-- Negative ---------------- Positive -->', fontsize=10)\nplt.ylabel('<-- Facts ---------------- Opinions -->', fontsize=10)\nplt.savefig(\"img/sa_true.png\", format=\"png\", dpi=200)\nplt.show()", "_____no_output_____" ], [ "train_df['polarity_false'] = train_df[train_df[\"binary_label\"]==False]['statement'].apply(pol)\ntrain_df['subjectivity_false'] = train_df[train_df[\"binary_label\"]==False]['statement'].apply(sub)\n\nplt.rcParams['figure.figsize'] = [10, 8]\n\nx = train_df[\"polarity_false\"]\ny = train_df[\"subjectivity_false\"]\nplt.scatter(x, y, color='blue')\n\nplt.title('Sentiment Analysis', fontsize=20)\nplt.xlabel('<-- Negative ---------------- Positive -->', fontsize=10)\nplt.ylabel('<-- Facts ---------------- Opinions -->', fontsize=10)\nplt.savefig(\"img/sa_false.png\", format=\"png\", dpi=200)\nplt.show()", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "#hide\n#skip\n! [[ -e /content ]] && pip install -Uqq fastai # upgrade fastai on colab", "_____no_output_____" ], [ "#default_exp collab\n#default_class_lvl 3", "_____no_output_____" ], [ "#export\nfrom fastai.tabular.all import *", "_____no_output_____" ], [ "#hide\nfrom nbdev.showdoc import *", "_____no_output_____" ] ], [ [ "# Collaborative filtering\n\n> Tools to quickly get the data and train models suitable for collaborative filtering", "_____no_output_____" ], [ "This module contains all the high-level functions you need in a collaborative filtering application to assemble your data, get a model and train it with a `Learner`. We will go other those in order but you can also check the [collaborative filtering tutorial](http://docs.fast.ai/tutorial.collab).", "_____no_output_____" ], [ "## Gather the data", "_____no_output_____" ] ], [ [ "#export\nclass TabularCollab(TabularPandas):\n \"Instance of `TabularPandas` suitable for collaborative filtering (with no continuous variable)\"\n with_cont=False", "_____no_output_____" ] ], [ [ "This is just to use the internal of the tabular application, don't worry about it.", "_____no_output_____" ] ], [ [ "#export\nclass CollabDataLoaders(DataLoaders):\n \"Base `DataLoaders` for collaborative filtering.\"\n @delegates(DataLoaders.from_dblock)\n @classmethod\n def from_df(cls, ratings, valid_pct=0.2, user_name=None, item_name=None, rating_name=None, seed=None, path='.', **kwargs):\n \"Create a `DataLoaders` suitable for collaborative filtering from `ratings`.\"\n user_name = ifnone(user_name, ratings.columns[0])\n item_name = ifnone(item_name, ratings.columns[1])\n rating_name = ifnone(rating_name, ratings.columns[2])\n cat_names = [user_name,item_name]\n splits = RandomSplitter(valid_pct=valid_pct, seed=seed)(range_of(ratings))\n to = TabularCollab(ratings, [Categorify], cat_names, y_names=[rating_name], y_block=TransformBlock(), splits=splits)\n return to.dataloaders(path=path, **kwargs)\n\n @classmethod\n def from_csv(cls, csv, **kwargs):\n \"Create a `DataLoaders` suitable for collaborative filtering from `csv`.\"\n return cls.from_df(pd.read_csv(csv), **kwargs)\n\nCollabDataLoaders.from_csv = delegates(to=CollabDataLoaders.from_df)(CollabDataLoaders.from_csv)", "_____no_output_____" ] ], [ [ "This class should not be used directly, one of the factory methods should be preferred instead. All those factory methods accept as arguments:\n\n- `valid_pct`: the random percentage of the dataset to set aside for validation (with an optional `seed`)\n- `user_name`: the name of the column containing the user (defaults to the first column)\n- `item_name`: the name of the column containing the item (defaults to the second column)\n- `rating_name`: the name of the column containing the rating (defaults to the third column)\n- `path`: the folder where to work\n- `bs`: the batch size\n- `val_bs`: the batch size for the validation `DataLoader` (defaults to `bs`)\n- `shuffle_train`: if we shuffle the training `DataLoader` or not\n- `device`: the PyTorch device to use (defaults to `default_device()`)", "_____no_output_____" ] ], [ [ "show_doc(CollabDataLoaders.from_df)", "_____no_output_____" ] ], [ [ "Let's see how this works on an example:", "_____no_output_____" ] ], [ [ "path = untar_data(URLs.ML_SAMPLE)\nratings = pd.read_csv(path/'ratings.csv')\nratings.head()", "_____no_output_____" ], [ "dls = CollabDataLoaders.from_df(ratings, bs=64)\ndls.show_batch()", "_____no_output_____" ], [ "show_doc(CollabDataLoaders.from_csv)", "_____no_output_____" ], [ "dls = CollabDataLoaders.from_csv(path/'ratings.csv', bs=64)", "_____no_output_____" ] ], [ [ "## Models", "_____no_output_____" ], [ "fastai provides two kinds of models for collaborative filtering: a dot-product model and a neural net. ", "_____no_output_____" ] ], [ [ "#export\nclass EmbeddingDotBias(Module):\n \"Base dot model for collaborative filtering.\"\n def __init__(self, n_factors, n_users, n_items, y_range=None):\n self.y_range = y_range\n (self.u_weight, self.i_weight, self.u_bias, self.i_bias) = [Embedding(*o) for o in [\n (n_users, n_factors), (n_items, n_factors), (n_users,1), (n_items,1)\n ]]\n\n def forward(self, x):\n users,items = x[:,0],x[:,1]\n dot = self.u_weight(users)* self.i_weight(items)\n res = dot.sum(1) + self.u_bias(users).squeeze() + self.i_bias(items).squeeze()\n if self.y_range is None: return res\n return torch.sigmoid(res) * (self.y_range[1]-self.y_range[0]) + self.y_range[0]\n\n @classmethod\n def from_classes(cls, n_factors, classes, user=None, item=None, y_range=None):\n \"Build a model with `n_factors` by inferring `n_users` and `n_items` from `classes`\"\n if user is None: user = list(classes.keys())[0]\n if item is None: item = list(classes.keys())[1]\n res = cls(n_factors, len(classes[user]), len(classes[item]), y_range=y_range)\n res.classes,res.user,res.item = classes,user,item\n return res\n\n def _get_idx(self, arr, is_item=True):\n \"Fetch item or user (based on `is_item`) for all in `arr`\"\n assert hasattr(self, 'classes'), \"Build your model with `EmbeddingDotBias.from_classes` to use this functionality.\"\n classes = self.classes[self.item] if is_item else self.classes[self.user]\n c2i = {v:k for k,v in enumerate(classes)}\n try: return tensor([c2i[o] for o in arr])\n except Exception as e:\n print(f\"\"\"You're trying to access {'an item' if is_item else 'a user'} that isn't in the training data.\n If it was in your original data, it may have been split such that it's only in the validation set now.\"\"\")\n\n def bias(self, arr, is_item=True):\n \"Bias for item or user (based on `is_item`) for all in `arr`\"\n idx = self._get_idx(arr, is_item)\n layer = (self.i_bias if is_item else self.u_bias).eval().cpu()\n return to_detach(layer(idx).squeeze(),gather=False)\n\n def weight(self, arr, is_item=True):\n \"Weight for item or user (based on `is_item`) for all in `arr`\"\n idx = self._get_idx(arr, is_item)\n layer = (self.i_weight if is_item else self.u_weight).eval().cpu()\n return to_detach(layer(idx),gather=False)", "_____no_output_____" ] ], [ [ "The model is built with `n_factors` (the length of the internal vectors), `n_users` and `n_items`. For a given user and item, it grabs the corresponding weights and bias and returns\n``` python\ntorch.dot(user_w, item_w) + user_b + item_b\n```\nOptionally, if `y_range` is passed, it applies a `SigmoidRange` to that result.", "_____no_output_____" ] ], [ [ "x,y = dls.one_batch()\nmodel = EmbeddingDotBias(50, len(dls.classes['userId']), len(dls.classes['movieId']), y_range=(0,5)\n ).to(x.device)\nout = model(x)\nassert (0 <= out).all() and (out <= 5).all()", "_____no_output_____" ], [ "show_doc(EmbeddingDotBias.from_classes)", "_____no_output_____" ] ], [ [ "`y_range` is passed to the main init. `user` and `item` are the names of the keys for users and items in `classes` (default to the first and second key respectively). `classes` is expected to be a dictionary key to list of categories like the result of `dls.classes` in a `CollabDataLoaders`:", "_____no_output_____" ] ], [ [ "dls.classes", "_____no_output_____" ] ], [ [ "Let's see how it can be used in practice:", "_____no_output_____" ] ], [ [ "model = EmbeddingDotBias.from_classes(50, dls.classes, y_range=(0,5)\n ).to(x.device)\nout = model(x)\nassert (0 <= out).all() and (out <= 5).all()", "_____no_output_____" ] ], [ [ "Two convenience methods are added to easily access the weights and bias when a model is created with `EmbeddingDotBias.from_classes`:", "_____no_output_____" ] ], [ [ "show_doc(EmbeddingDotBias.weight)", "_____no_output_____" ] ], [ [ "The elements of `arr` are expected to be class names (which is why the model needs to be created with `EmbeddingDotBias.from_classes`)", "_____no_output_____" ] ], [ [ "mov = dls.classes['movieId'][42] \nw = model.weight([mov])\ntest_eq(w, model.i_weight(tensor([42])))", "_____no_output_____" ], [ "show_doc(EmbeddingDotBias.bias)", "_____no_output_____" ] ], [ [ "The elements of `arr` are expected to be class names (which is why the model needs to be created with `EmbeddingDotBias.from_classes`)", "_____no_output_____" ] ], [ [ "mov = dls.classes['movieId'][42] \nb = model.bias([mov])\ntest_eq(b, model.i_bias(tensor([42])))", "_____no_output_____" ], [ "#export \nclass EmbeddingNN(TabularModel):\n \"Subclass `TabularModel` to create a NN suitable for collaborative filtering.\"\n @delegates(TabularModel.__init__)\n def __init__(self, emb_szs, layers, **kwargs):\n super().__init__(emb_szs=emb_szs, n_cont=0, out_sz=1, layers=layers, **kwargs)", "_____no_output_____" ], [ "show_doc(EmbeddingNN)", "_____no_output_____" ] ], [ [ "`emb_szs` should be a list of two tuples, one for the users, one for the items, each tuple containing the number of users/items and the corresponding embedding size (the function `get_emb_sz` can give a good default). All the other arguments are passed to `TabularModel`.", "_____no_output_____" ] ], [ [ "emb_szs = get_emb_sz(dls.train_ds, {})\nmodel = EmbeddingNN(emb_szs, [50], y_range=(0,5)\n ).to(x.device)\nout = model(x)\nassert (0 <= out).all() and (out <= 5).all()", "_____no_output_____" ] ], [ [ "## Create a `Learner`", "_____no_output_____" ], [ "The following function lets us quickly create a `Learner` for collaborative filtering from the data.", "_____no_output_____" ] ], [ [ "# export\n@log_args(to_return=True, but_as=Learner.__init__)\n@delegates(Learner.__init__)\ndef collab_learner(dls, n_factors=50, use_nn=False, emb_szs=None, layers=None, config=None, y_range=None, loss_func=None, **kwargs):\n \"Create a Learner for collaborative filtering on `dls`.\"\n emb_szs = get_emb_sz(dls, ifnone(emb_szs, {}))\n if loss_func is None: loss_func = MSELossFlat()\n if config is None: config = tabular_config()\n if y_range is not None: config['y_range'] = y_range\n if layers is None: layers = [n_factors]\n if use_nn: model = EmbeddingNN(emb_szs=emb_szs, layers=layers, **config)\n else: model = EmbeddingDotBias.from_classes(n_factors, dls.classes, y_range=y_range)\n return Learner(dls, model, loss_func=loss_func, **kwargs)", "_____no_output_____" ] ], [ [ "If `use_nn=False`, the model used is an `EmbeddingDotBias` with `n_factors` and `y_range`. Otherwise, it's a `EmbeddingNN` for which you can pass `emb_szs` (will be inferred from the `dls` with `get_emb_sz` if you don't provide any), `layers` (defaults to `[n_factors]`) `y_range`, and a `config` that you can create with `tabular_config` to customize your model. \n\n`loss_func` will default to `MSELossFlat` and all the other arguments are passed to `Learner`.", "_____no_output_____" ] ], [ [ "learn = collab_learner(dls, y_range=(0,5))", "_____no_output_____" ], [ "learn.fit_one_cycle(1)", "_____no_output_____" ] ], [ [ "## Export -", "_____no_output_____" ] ], [ [ "#hide\nfrom nbdev.export import *\nnotebook2script()", "Converted 00_torch_core.ipynb.\nConverted 01_layers.ipynb.\nConverted 02_data.load.ipynb.\nConverted 03_data.core.ipynb.\nConverted 04_data.external.ipynb.\nConverted 05_data.transforms.ipynb.\nConverted 06_data.block.ipynb.\nConverted 07_vision.core.ipynb.\nConverted 08_vision.data.ipynb.\nConverted 09_vision.augment.ipynb.\nConverted 09b_vision.utils.ipynb.\nConverted 09c_vision.widgets.ipynb.\nConverted 10_tutorial.pets.ipynb.\nConverted 11_vision.models.xresnet.ipynb.\nConverted 12_optimizer.ipynb.\nConverted 13_callback.core.ipynb.\nConverted 13a_learner.ipynb.\nConverted 13b_metrics.ipynb.\nConverted 14_callback.schedule.ipynb.\nConverted 14a_callback.data.ipynb.\nConverted 15_callback.hook.ipynb.\nConverted 15a_vision.models.unet.ipynb.\nConverted 16_callback.progress.ipynb.\nConverted 17_callback.tracker.ipynb.\nConverted 18_callback.fp16.ipynb.\nConverted 18a_callback.training.ipynb.\nConverted 19_callback.mixup.ipynb.\nConverted 20_interpret.ipynb.\nConverted 20a_distributed.ipynb.\nConverted 21_vision.learner.ipynb.\nConverted 22_tutorial.imagenette.ipynb.\nConverted 23_tutorial.vision.ipynb.\nConverted 24_tutorial.siamese.ipynb.\nConverted 24_vision.gan.ipynb.\nConverted 30_text.core.ipynb.\nConverted 31_text.data.ipynb.\nConverted 32_text.models.awdlstm.ipynb.\nConverted 33_text.models.core.ipynb.\nConverted 34_callback.rnn.ipynb.\nConverted 35_tutorial.wikitext.ipynb.\nConverted 36_text.models.qrnn.ipynb.\nConverted 37_text.learner.ipynb.\nConverted 38_tutorial.text.ipynb.\nConverted 39_tutorial.transformers.ipynb.\nConverted 40_tabular.core.ipynb.\nConverted 41_tabular.data.ipynb.\nConverted 42_tabular.model.ipynb.\nConverted 43_tabular.learner.ipynb.\nConverted 44_tutorial.tabular.ipynb.\nConverted 45_collab.ipynb.\nConverted 46_tutorial.collab.ipynb.\nConverted 50_tutorial.datablock.ipynb.\nConverted 60_medical.imaging.ipynb.\nConverted 61_tutorial.medical_imaging.ipynb.\nConverted 65_medical.text.ipynb.\nConverted 70_callback.wandb.ipynb.\nConverted 71_callback.tensorboard.ipynb.\nConverted 72_callback.neptune.ipynb.\nConverted 73_callback.captum.ipynb.\nConverted 74_callback.cutmix.ipynb.\nConverted 97_test_utils.ipynb.\nConverted 99_pytorch_doc.ipynb.\nConverted index.ipynb.\nConverted tutorial.ipynb.\n" ] ] ]

[ "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "code", "code", "code", "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "from rsna_retro.imports import *\nfrom rsna_retro.metadata import *\nfrom rsna_retro.preprocess import *\nfrom rsna_retro.train import *\nfrom rsna_retro.train3d import *\nfrom rsna_retro.trainfull3d_labels import *", "Loading imports\n" ], [ "torch.cuda.set_device(4)", "_____no_output_____" ], [ "dls = get_3d_dls_aug(Meta.df_comb, sz=128, bs=32, grps=Meta.grps_stg1)", "_____no_output_____" ] ], [ [ "## Model", "_____no_output_____" ] ], [ [ "def get_3d_head(p=0.0):\n pool, feat = (nn.AdaptiveAvgPool3d(1), 64)\n m = nn.Sequential(Batchify(),\n ConvLayer(512,512,stride=2,ndim=3), # 8\n ConvLayer(512,1024,stride=2,ndim=3), # 4\n ConvLayer(1024,1024,stride=2,ndim=3), # 2\n nn.AdaptiveAvgPool3d((1, 1, 1)), Batchify(), Flat3d(), nn.Dropout(p),\n nn.Linear(1024, 6))\n init_cnn(m)\n return m", "_____no_output_____" ], [ "m = get_3d_head()\nconfig=dict(custom_head=m)\nlearn = get_learner(dls, xresnet18, get_loss(), config=config)", "_____no_output_____" ], [ "hook = ReshapeBodyHook(learn.model[0])\nlearn.add_cb(RowLoss())", "_____no_output_____" ], [ "# learn.load(f'runs/baseline_stg1_xresnet18-3', strict=False)", "_____no_output_____" ], [ "name = 'trainfull3d_labels_partial3d_new'", "_____no_output_____" ] ], [ [ "## Training", "_____no_output_____" ] ], [ [ "learn.lr_find()", "_____no_output_____" ], [ "do_fit(learn, 8, 1e-3)\nlearn.save(f'runs/{name}-1')", "_____no_output_____" ], [ "learn.load(f'runs/{name}-1')\nlearn.dls = get_3d_dls_aug(Meta.df_comb, sz=256, bs=12, grps=Meta.grps_stg1)\ndo_fit(learn, 4, 1e-4)\nlearn.save(f'runs/{name}-2')", "_____no_output_____" ], [ "learn.load(f'runs/{name}-2')\nlearn.dls = get_3d_dls_aug(Meta.df_comb, sz=384, bs=4, path=path_jpg, grps=Meta.grps_stg1)\ndo_fit(learn, 2, 1e-5)\nlearn.save(f'runs/{name}-3')", "_____no_output_____" ] ] ]

[ "code", "markdown", "code", "markdown", "code" ]

[ [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "#Import Modules", "_____no_output_____" ] ], [ [ "import cv2\nimport numpy as np\nfrom google.colab.patches import cv2_imshow", "_____no_output_____" ] ], [ [ "#Load Image ", "_____no_output_____" ] ], [ [ "#image is loaded using cv2.imread() method,here flag is 0 ,specifies to load image in GRAYSCALE mode.\n'''\nSyntax:\n cv2.imread(path,flag)\nParameters:\n path: string representing the path of the image to be read.\n flag: specifies the way in which image should be read.\n'''\nimg=cv2.imread(\"input.png\",0)\ncv2_imshow(img)", "_____no_output_____" ] ], [ [ "#Apply scaling Operation\n\n\n", "_____no_output_____" ] ], [ [ "# To perform scaling operation,cv2.resize() method is used.\n'''\nSyntax:\n cv2.resize(image,(width,height)=None,fx=1,fy=1,interpolation)\nParameters:\n image: input image.\n (width,height): determining the size of output image ; optional parameter.\n fx: scaling factor for x-axis,default=1.\n fy: scaling factor for y-axis,default=1.\n interpolation: interpolation method to be used.\n'''\nscaled_up_x=cv2.resize(img,None,fx=2,fy=1,interpolation=cv2.INTER_CUBIC)\nscaled_down_x=cv2.resize(img,None,fx=0.5,fy=1,interpolation=cv2.INTER_LINEAR)\nscaled_up_y=cv2.resize(img,None,fx=1,fy=2,interpolation=cv2.INTER_CUBIC)\nscaled_down_y=cv2.resize(img,None,fx=1,fy=0.5,interpolation=cv2.INTER_LINEAR)", "_____no_output_____" ] ], [ [ "#Display the scaled image", "_____no_output_____" ] ], [ [ "cv2_imshow(scaled_up_x)\ncv2_imshow(scaled_down_x)\ncv2_imshow(scaled_up_y)\ncv2_imshow(scaled_down_y)", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# ***Introduction to Radar Using Python and MATLAB***\n## Andy Harrison - Copyright (C) 2019 Artech House\n<br/>\n\n# Pulse Train Ambiguity Function\n***", "_____no_output_____" ], [ "Referring to Section 8.6.1, the amibguity function for a coherent pulse train is found by employing the generic waveform technique outlined in Section 8.6.3.\n***", "_____no_output_____" ], [ "Begin by getting the library path", "_____no_output_____" ] ], [ [ "import lib_path", "_____no_output_____" ] ], [ [ "Set the pulsewidth (s), the pulse repetition interval (s) and the number of pulses", "_____no_output_____" ] ], [ [ "pulsewidth = 0.4\n\npri = 1.0\n\nnumber_of_pulses = 6", "_____no_output_____" ] ], [ [ "Generate the time delay (s) using the `linspace` routine from `scipy`", "_____no_output_____" ] ], [ [ "from numpy import linspace\n\n\n# Set the time delay\n\ntime_delay = linspace(-number_of_pulses * pri, number_of_pulses * pri, 5000)", "_____no_output_____" ] ], [ [ "Calculate the ambiguity function for the pulse train", "_____no_output_____" ] ], [ [ "from Libs.ambiguity.ambiguity_function import pulse_train\n\nfrom numpy import finfo\n\n\nambiguity = pulse_train(time_delay, finfo(float).eps, pulsewidth, pri, number_of_pulses)", "_____no_output_____" ] ], [ [ "Plot the zero-Doppler cut using the `matplotlib` routines", "_____no_output_____" ] ], [ [ "from matplotlib import pyplot as plt\n\n\n# Set the figure size\n\nplt.rcParams[\"figure.figsize\"] = (15, 10)\n\n\n\n# Plot the ambiguity function\n\nplt.plot(time_delay, ambiguity, '')\n\n\n\n# Set the x and y axis labels\n\nplt.xlabel(\"Time (s)\", size=12)\n\nplt.ylabel(\"Relative Amplitude\", size=12)\n\n\n\n# Turn on the grid\n\nplt.grid(linestyle=':', linewidth=0.5)\n\n\n\n# Set the plot title and labels\n\nplt.title('Pulse Train Ambiguity Function', size=14)\n\n\n\n# Set the tick label size\n\nplt.tick_params(labelsize=12)", "_____no_output_____" ] ], [ [ "Set the Doppler mismatch frequencies using the `linspace` routine", "_____no_output_____" ] ], [ [ "doppler_frequency = linspace(-2.0 / pulsewidth, 2.0 / pulsewidth, 1000)", "_____no_output_____" ] ], [ [ "Calculate the ambiguity function for the pulse train", "_____no_output_____" ] ], [ [ "ambiguity = pulse_train(finfo(float).eps, doppler_frequency, pulsewidth, pri, number_of_pulses)", "_____no_output_____" ] ], [ [ "Display the zero-range cut for the pulse train", "_____no_output_____" ] ], [ [ "plt.plot(doppler_frequency, ambiguity, '')\n\n\n# Set the x and y axis labels\n\nplt.xlabel(\"Doppler (Hz)\", size=12)\n\nplt.ylabel(\"Relative Amplitude\", size=12)\n\n\n\n# Turn on the grid\n\nplt.grid(linestyle=':', linewidth=0.5)\n\n\n\n# Set the plot title and labels\n\nplt.title('Pulse Train Ambiguity Function', size=14)\n\n\n\n# Set the tick label size\n\nplt.tick_params(labelsize=12)", "_____no_output_____" ] ], [ [ "Set the time delay and Doppler mismatch frequency and create the two-dimensional grid using the `meshgrid` routine from `scipy`", "_____no_output_____" ] ], [ [ "from numpy import meshgrid\n\n\n# Set the time delay\n\ntime_delay = linspace(-number_of_pulses * pri, number_of_pulses * pri, 1000)\n\n\n\n# Set the Doppler mismatch\n\ndoppler_frequency = linspace(-2.0 / pulsewidth, 2.0 / pulsewidth, 1000)\n\n\n\n# Create the grid\n\nt, f = meshgrid(time_delay, doppler_frequency)", "_____no_output_____" ] ], [ [ "Calculate the ambiguity function for the pulse train", "_____no_output_____" ] ], [ [ "ambiguity = pulse_train(t, f, pulsewidth, pri, number_of_pulses)", "_____no_output_____" ] ], [ [ "Display the two-dimensional contour plot for the pulse train ambiguity function", "_____no_output_____" ] ], [ [ "# Plot the ambiguity function\nfrom numpy import finfo\n\nplt.contour(t, f, ambiguity + finfo('float').eps, 20, cmap='jet', vmin=-0.2, vmax=1.0)\n\n\n# Set the x and y axis labels\n\nplt.xlabel(\"Time (s)\", size=12)\n\nplt.ylabel(\"Doppler (Hz)\", size=12)\n\n\n\n# Turn on the grid\n\nplt.grid(linestyle=':', linewidth=0.5)\n\n\n\n# Set the plot title and labels\n\nplt.title('Pulse Pulse Ambiguity Function', size=14)\n\n\n\n# Set the tick label size\n\nplt.tick_params(labelsize=12)", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# 5.2 Fourier transform and Fourier series\n\nWe make use of the theory of tempered distributions (see\n[@strichartz2003guide] for an introduction) and we begin by collecting\nsome results of independent interest, which will also be important\nlater.\n\n## 5.2.1 Fourier transform\n\nBefore studying the Fourier transform, we first consider Schwartz space\nwhich is defined below.\n\n````{prf:definition}\nThe Schwartz space\n$\\mathcal{S}\\left(\\mathbb{R}^{n}\\right)$ is the topological vector space\nof functions $f: \\mathbb{R}^{n} \\rightarrow \\mathbb{C}$ such that\n$f \\in C^{\\infty}\\left(\\mathbb{R}^{n}\\right)$ and\n\n$$\n x^{\\alpha} \\partial^{\\beta} f(x) \\rightarrow 0 \\quad \\text { as }|x| \\rightarrow \\infty\n$$\n\nfor every pair of multi-indices $\\alpha, \\beta \\in \\mathbb{N}_{0}^{n} .$\nFor $\\alpha, \\beta \\in \\mathbb{N}_{0}^{n}$ and\n$f \\in \\mathcal{S}\\left(\\mathbb{R}^{n}\\right)$ let (5.10)\n\n$$\n \\|f\\|_{\\alpha, \\beta}=\\sup _{\\mathbb{R}^{n}}\\left|x^{\\alpha} \\partial^{\\beta} f\\right|\n$$\n\nA sequence of functions $\\left\\{f_{k}: k \\in \\mathbb{N}\\right\\}$\nconverges to a function $f$ in $\\mathcal{S}\\left(\\mathbb{R}^{n}\\right)$\nif\n\n$$\n \\left\\|f_{n}-f\\right\\|_{\\alpha, \\beta} \\rightarrow 0 \\quad \\text { as } k \\rightarrow \\infty\n$$\n\nfor every $\\alpha, \\beta \\in \\mathbb{N}_{0}^{n}$.\n````\n\nThe Schwartz space consists of smooth functions whose derivatives and\nthe function itself decay at infinity faster than any power. Schwartz\nfunctions are rapidly decreasing. When there is no ambiguity, we will\nwrite $\\mathcal{S}\\left(\\mathbb{R}^{n}\\right)$ as $\\mathcal{S}$. Roughly\nspeaking, tempered distributions grow no faster than a polynomial at\ninfinity.\n\n````{prf:definition}\nA tempered distribution $T$ on $\\mathbb{R}^{n}$ is a continuous linear\nfunctional\n$T: \\mathcal{S}\\left(\\mathbb{R}^{n}\\right) \\rightarrow \\mathbb{C} .$ The\ntopological vector space of tempered distributions is denoted by\n$\\mathcal{S}^{\\prime}\\left(\\mathbb{R}^{n}\\right)$ or\n$\\mathcal{S}^{\\prime} .$ If $\\langle T, f\\rangle$ denotes the value of\n$T \\in \\mathcal{S}^{\\prime}$ acting on $f \\in \\mathcal{S}$ then a\nsequence $\\left\\{T_{k}\\right\\}$ converges to $T$ in\n$\\mathcal{S}^{\\prime}$, written $T_{k} \\rightarrow T$, if\n\n$$\n \\left\\langle T_{k}, f\\right\\rangle \\rightarrow\\langle T, f\\rangle\n$$\n\nfor every $f \\in \\mathcal{S}$.\n````\n\nSince $\\mathcal{D} \\subset \\mathcal{S}$ is densely and continuously\nimbedded, we have $\\mathcal{S}^{\\prime} \\subset \\mathcal{D}^{\\prime} .$\nMoreover, a distribution $T \\in \\mathcal{D}^{\\prime}$ extends uniquely\nto a tempered distribution $T \\in \\mathcal{S}^{\\prime}$ if and only if\nit is continuous on $\\mathcal{D}$ with respect to the topology on\n$\\mathcal{S}$. Every function $f \\in L_{\\text {loc }}^{1}$ defines a\nregular distribution $T_{f} \\in \\mathcal{D}^{\\prime}$ by\n\n$$\n \\left\\langle T_{f}, \\phi\\right\\rangle=\\int f \\phi d x \\quad \\text { for all } \\phi \\in \\mathcal{D}.\n$$\n\nIf $|f| \\leq p$ is bounded by some polynomial $p,$ then $T_{f}$ extends\nto a tempered distribution $T_{f} \\in \\mathcal{S}^{\\prime}$, but this is\nnot the case for functions $f$ that grow too rapidly at infinity.\n\nThe Schwartz space is a natural one to use for the Fourier transform.\nDifferentiation and multiplication exchange roles under the Fourier\ntransform and therefore so do the properties of smoothness and rapid\ndecrease. As a result, the Fourier transform is an automorphism of the\nSchwartz space. By duality, the Fourier transform is also an\nautomorphism of the space of tempered distributions.\n\n````{prf:definition}\nThe Fourier transform of a function $f \\in \\mathcal{S}\\left(\\mathbb{R}^{n}\\right)$\nis the function $\\hat{f}: \\mathbb{R}^{n} \\rightarrow \\mathbb{C}$ defined\nby \n\n$$\n \\hat{f}(\\omega)= \\int f(x) e^{-2 \\pi i\\omega \\cdot x} d x.\n$$\n\nThe inverse Fourier transform of $f$ is the function\n$\\check{f}: \\mathbb{R}^{n} \\rightarrow \\mathbb{C}$ defined by\n\n$$\n \\check{f}(x)=\\int f(\\omega) e^{2 \\pi i\\omega \\cdot x} d k.\n$$\n\n````\n\n````{prf:definition}\nThe Fourier transform of a tempered distribution $f \\in \\mathcal{S}'$ is defined by\n\n$$\n \\langle \\hat{f}, \\phi\\rangle = \\langle f, \\hat \\phi\\rangle,\\quad \\forall \\phi\\in \\mathcal{S}.\n$$\n\n````\n\nThe support of a continuous function $f$ is the closure of the set\n$\\{x\\in \\mathbb{R}: f(x)\\neq 0\\}$.\n\n```{admonition} Properties\nThe Fourier transform has the following properties\n\n1. If $f\\in \\mathcal{S}'$ and the support of $\\hat f$ is $\\{0\\}$, then\n $f$ is a polynomial.\n\n2. If $f\\in \\mathcal{S}'$ and the support of $\\hat f$ is a single point\n $\\{a\\}$, then $f(x)=e^{2\\pi iax}P(x)$, where $P(x)$ is a polynomial.\n```\n\n## 5.2.2 Poisson summation formula\n\n``` {prf:theorem}\nLet $f \\in L^{1}(\\mathbb{R})$ and $f$ is continuous. Then we have for\nalmost all $(x, \\omega ) \\in \\mathbb{R} \\times \\hat{\\mathbb{R}}$ that\n\n$$\n T \\sum_{n \\in \\mathbb{Z}} f(x+n T) e^{-2 \\pi i \\omega (x+n T)}=\\sum_{n \\in \\mathbb{Z}} \\hat{f}\\left(\\omega +\\frac{n}{T}\\right) e^{2 \\pi i n x / T}\n$$\n\nwhere both sides converge absolutely.\n\nIn addition, let $\\Lambda$ be the lattice in $\\mathbb{R}^{d}$ consisting\nof points with integer coordinates. For a function $f$ in\n$L^{1}\\left(\\mathbb{R}^{d}\\right)$ and $f$ is continuous, we have\n\n$$\n \\sum_{\\omega \\in \\Lambda} f(x+\\omega )=\\sum_{\\nu \\in \\Lambda} \\hat{f}(\\omega ) e^{2 \\pi i x \\cdot \\omega }.\n$$\n\nwhere both series converge absolutely and uniformly on $\\Lambda$.\n```\n\n```{prf;proof}\n*Proof.* We just give a proof of a simple case that\n$f: \\mathbb{R} \\rightarrow \\mathbb{C}$ is a Schwarz function (see\nDefinition [\\[def:schwarz\\]](#def:schwarz){reference-type=\"ref\"\nreference=\"def:schwarz\"}). Let: $$F(x)=\\sum_{n \\in \\mathbb{Z}} f(x+n).$$\nThen $F(x)$ is 1-periodic (because of absolute convergence), and has\nFourier coefficients: $$\\begin{aligned}\n\\hat{F}_{\\omega } &=\\int_{0}^{1} \\sum_{n \\in \\mathbb{Z}} f(x+n) e^{-2 \\pi i \\omega x} \\mathrm{~d} x \\\\\n&=\\sum_{n \\in \\mathbb{Z}} \\int_{0}^{1} f(x+n) e^{-2 \\pi i \\omega x} \\mathrm{~d} x \\quad \\text { because } f \\text { is Schwarz, so convergence is uniform}\\\\\n&=\\sum_{n \\in \\mathbb{Z}} \\int_{n}^{n+1} f(x) e^{-2 \\pi i\\omega x} \\mathrm{~d} x \\\\\n&=\\int_{\\mathbb{R}} f(x) e^{-2 \\pi i \\omega x} \\mathrm{~d} x\\\\\n&=\\hat{f}(k)\\\\\n\\end{aligned}$$ where $\\hat{f}$ is the Fourier transform of $f$.\n\nTherefore by the definition of the Fourier series of $f:$\n\n$$\n F(x) =\\sum_{\\omega \\in \\mathbb{Z}} \\hat{f}(k) e^{2\\pi i \\omega x}.\n$$\n\nChoosing $x=0$ in this formula:\n$$\\sum_{n \\in \\mathbb{Z}} f(n)=\\sum_{\\omega \\in \\mathbb{Z}} \\hat{f}(\\omega )$$\nas required. ◻\n```\n\n## 5.2.3 A special cut-off function\n\nLet us first state the following simple result that can be obtained by\nfollowing a calculation given in Section 3 of [@johnson2015saddle].\n\n```{prf:lemma}\nGiven $\\alpha>1$, consider \n\n$$\n \\label{alpha-g}\n g(t) = \\begin{cases} \n e^{-(1-t^2)^{1 - \\alpha}} & t\\in (-1,1) \\\\\n 0 & \\text{otherwise}.\n \\end{cases}\n$$\n\nthen there is a constant $c_\\alpha$ such that\n \n$$\n |\\hat{g}(\\omega )|\\lesssim e^{-c_\\alpha|\\omega |^{1-\\alpha^{-1}}},\n$$\n\n```\n\n```{prf:proof}\n*Proof.* Consider the asymptotic behavior of the Fourier transform\n\n$$\n F(\\omega )=\\int_{-\\infty}^{\\infty} g(t) e^{2\\pi i \\omega t} dt=2 \\operatorname{Re} \\int_{0}^{1} e^{2\\pi i \\omega t- (1-t^{2})^{1-\\alpha}} dt\n$$\n\nfor $|\\operatorname{Re} \\omega | \\gg 1.$ (Without loss of generality, we\ncan restrict ourselves to real $\\omega \\geq 0$). With a change of\nvariable $x=1-t$,\n\n$$\n F(\\omega )=2 \\operatorname{Re} \\int_{0}^{1} e^{f(x)} dx\n$$\n\nwith $f(x)=2\\pi i \\omega - 2\\pi i \\omega x- (2x-x^2)^{1-\\alpha}\\approx \\tilde f(x)+O\\left(x^{2-\\alpha}\\right)$\nand\n\n$$\n \\tilde f(x) = 2\\pi i \\omega - 2\\pi i \\omega x - (2 x)^{1-\\alpha}.\n$$\n\nThe saddle point is the $x=x_0$ where $f'(x_0)=0$. Since\n$\\tilde f'(x)=-2\\pi i \\omega + (\\alpha-1)2^{1-\\alpha} x^{-\\alpha},$\n\n$$\n x_{0} \\approx \\tilde x_0=\\left (2^{-\\alpha} (\\alpha-1) / i \\omega \\pi \\right )^{1 / \\alpha} \\sim \\omega ^{-1 / \\alpha}.\n$$\n\nTherefore $\\tilde f(\\tilde x_{0}) \\sim \\omega ^{(\\alpha-1) / \\alpha}$\nasymptotically. The second derivative is\n\n$$\n \\tilde f'' (\\tilde x_{0} )=-2^{1-\\alpha} \\alpha(\\alpha-1) \\tilde x_{0}^{-\\alpha-1}=-i^{(\\alpha+1) / \\alpha} 2 A \\omega ^{(\\alpha+1)/\\alpha},\n$$\n\nwhere $$A=2\\alpha (\\alpha-1)^{-1/\\alpha}\\pi^{(\\alpha+1)/\\alpha}.$$ Now,\n\n$$\n \\begin{split}\n \\tilde f(x)\\approx &\\tilde f(\\tilde x_0) + {\\tilde f''(\\tilde x_0)\\over 2} (x-\\tilde x_0)^2\n \\\\\n =&2\\pi i \\omega - (\\alpha - 1)^{1\\over \\alpha}(i\\omega \\pi )^{\\alpha -1\\over \\alpha} - (\\alpha - 1)^{1-\\alpha\\over \\alpha} (i\\omega \\pi )^{\\alpha -1\\over \\alpha}\n \\\\\n &-i^{(\\alpha+1) / \\alpha} A \\omega ^{(\\alpha+1)/\\alpha}(x- 2^{-1}(\\alpha - 1)^{-{1\\over \\alpha}}(i\\omega \\pi )^{-{1\\over \\alpha}} )^2.\n \\end{split} \n$$ \n\nChoose a contour $x=i^{-1 / \\alpha}u$, in which case\n\n$$\n \\tilde f(x) \\approx \\tilde f(\\tilde x_{0}) -i^{(\\alpha-1) / \\alpha} A \\omega ^{(\\alpha+1) / \\alpha}\\left(u-u_{0}\\right)^{2},\n$$\n\nwhich is a path of descent so we can perform a Gaussian integral.\n\nRecall that the integral of \n\n$$\n \\label{gaussInt}\n \\int_{-\\infty}^{\\infty} e^{-a u^{2}} d u=\\sqrt{\\pi / a}\n$$\n\nas long as Re$a>0,$ which is true here. Note also that, in the limit as $\\omega$\nbecomes large, the integrand becomes zero except close to\n$u=\\sqrt{1 / 2 \\omega },$ so we can neglect the rest of the contour and\ntreat the integral over $u$ as going from $-\\infty$ to $\\infty$.\n(Thankfully, the width of the Gaussian $\\Delta u \\sim \\omega ^{-3 / 4}$\ngoes to zero faster than the location of the maximum\n$u_{0} \\sim \\omega ^{-1 / 2},$ so we don't have to worry about the $u=0$\norigin). Also note that the change of variables from $x$ to $u$ gives us\nthe Jacobian factor for $$dx=i^{-1 / \\alpha}d u.$$ Thus, when all is\nsaid and done, we obtain the exact asymptotic form of the Fourier\nintegral for $\\omega \\gg 1$: \n\n$$\n \\begin{split}\n F(\\omega ) \\approx &2 \\operatorname{Re}\\int_{0}^{1} e^{\\tilde f(\\tilde x_0) - i^{(\\alpha-1) / \\alpha} A \\omega ^{(\\alpha+1) / \\alpha}\\left(u-u_{0}\\right)^{2}} dx\n \\\\\n =&2 \\operatorname{Re} e^{\\tilde f(\\tilde x_0)} i^{-1 / \\alpha} \\int_{-\\infty}^{\\infty} e^{- i^{(\\alpha-1) \\over \\alpha} A \\omega ^{(\\alpha+1) / \\alpha}\\left(u-u_{0}\\right)^{2}} du\n \\\\\n =&2 \\operatorname{Re} e^{\\tilde f(\\tilde x_0)} \\pi^{1/2}i^{-1 / \\alpha} i^{(1-\\alpha) \\over 2\\alpha} A^{-1/2} \\omega ^{-(\\alpha+1) / 2\\alpha}\\qquad \\text{ by \\eqref{gaussInt}} \n \\\\\n =&2 \\operatorname{Re}\\left[\\sqrt{\\frac{\\pi}{(i \\omega )^{(\\alpha+1) / \\alpha} A}} e^{\\tilde f(\\tilde x_0)}\\right]\n \\\\\n \\approx &2 \\operatorname{Re}\\left[\\sqrt{\\frac{\\pi}{(i \\omega )^{(\\alpha+1) / \\alpha} A}} e^{ 2\\pi i \\omega - 2\\pi i \\omega \\tilde x_{0}- \\left[\\left(2-\\tilde x_{0}\\right) \\tilde x_{0}\\right]^{1-\\alpha}}\\right]\n \\end{split}\n$$\n\nwith $x_{0}$ and $A$ given above. Notice that\n$\\tilde x_0\\sim \\omega ^{-1 / \\alpha}$. Thus,\n\n$$\n |F(\\omega ) | \\approx e^{-c_\\alpha|\\omega |^{1-\\alpha^{-1}}}.\n$$ ◻\n```\n", "_____no_output_____" ], [ "## 5.2.4 Fourier transform of polynomials\n\nWe begin by noting that an activation function $\\sigma$, which satisfies\na polynomial growth condition $|\\sigma(x)| \\leq C(1 + |x|)^n$ for some\nconstants $C$ and $n$, is a tempered distribution. As a result, we make\nthis assumption on our activation functions in the following theorems.\nWe briefly note that this condition is sufficient, but not necessary\n(for instance an integrable function need not satisfy a pointwise\npolynomial growth bound) for $\\sigma$ to be represent a tempered\ndistribution.\n\nWe begin by studying the convolution of $\\sigma$ with a Gaussian\nmollifier. Let $\\eta$ be a Gaussian mollifier\n\n$$\n \\eta(x) = \\frac{1}{\\sqrt{\\pi}}e^{-x^2}.\n$$\n\nSet\n$\\eta_\\epsilon=\\frac{1}{\\epsilon}\\eta(\\frac{x}{\\epsilon})$. Then\nconsider \n\n$$\n \\sigma_{\\epsilon}(x):=\\sigma\\ast{\\eta_\\epsilon}(x)=\\int_{\\mathbb{R}}\\sigma(x-y){\\eta_\\epsilon}(y)dy\n$$ (sigma-epsilon)\n\nfor a given activation function $\\sigma$. It is clear that\n$\\sigma_{\\epsilon}\\in C^\\infty(\\mathbb{R})$. Moreover, by considering\nthe Fourier transform (as a tempered distribution) we see that\n\n$$\n \\hat{\\sigma}_{\\epsilon} = \\hat{\\sigma}\\hat{\\eta}_{\\epsilon} = \\hat{\\sigma}\\eta_{\\epsilon^{-1}}.\n$$ (eq_278)\n\nWe begin by stating a lemma which characterizes the set of polynomials\nin terms of their Fourier transform.\n\n```{prf:lemma}\n:label: polynomial_lemma\nGiven\na tempered distribution $\\sigma$, the following statements are\nequivalent:\n\n1. $\\sigma$ is a polynomial\n\n2. $\\sigma_\\epsilon$ given by {eq}`sigma-epsilon` is a polynomial for any $\\epsilon>0$.\n\n3. $\\text{supp}(\\hat{\\sigma})\\subset \\{0\\}$.\n```\n\n```{prf:proof}\n*Proof.* We begin by proving that (3) and (1) are equivalent. This\nfollows from a characterization of distributions supported at a single\npoint (see [@strichartz2003guide], section 6.3). In particular, a\ndistribution supported at $0$ must be a finite linear combination of\nDirac masses and their derivatives. In particular, if $\\hat{\\sigma}$ is\nsupported at $0$, then\n\n$$\n \\hat{\\sigma} = \\displaystyle\\sum_{i=1}^n a_i\\delta^{(i)}.\n$$\n\nTaking the inverse Fourier transform and noting that the inverse Fourier transform\nof $\\delta^{(i)}$ is $c_ix^i$, we see that $\\sigma$ is a polynomial.\nThis shows that (3) implies (1), for the converse we simply take the\nFourier transform of a polynomial and note that it is a finite linear\ncombination of Dirac masses and their derivatives.\n\nFinally, we prove the equivalence of (2) and (3). For this it suffices\nto show that $\\hat{\\sigma}$ is supported at $0$ iff\n$\\hat{\\sigma}_\\epsilon$ is supported at $0$. This follows from equation\n{eq}`eq_278` and the\nfact that $\\eta_{\\epsilon^{-1}}$ is nowhere vanishing. ◻\n```\n\nAs an application of Lemma {prf:ref}`polynomial_lemma`, we give a simple proof of the result in\nthe next section.", "_____no_output_____" ] ] ]

[ "markdown" ]

[ [ "markdown", "markdown" ] ]

[ "BSD-3-Clause" ]

[ "BSD-3-Clause" ]

[ "BSD-3-Clause" ]

[ [ [ "%matplotlib inline", "_____no_output_____" ] ], [ [ "\n============================================\n4D Neuroimaging/BTi phantom dataset tutorial\n============================================\n\nHere we read 4DBTi epochs data obtained with a spherical phantom\nusing four different dipole locations. For each condition we\ncompute evoked data and compute dipole fits.\n\nData are provided by Jean-Michel Badier from MEG center in Marseille, France.\n", "_____no_output_____" ] ], [ [ "# Authors: Alex Gramfort <[email protected]>\n#\n# License: BSD (3-clause)\n\nimport os.path as op\nimport numpy as np\nfrom mayavi import mlab\nfrom mne.datasets import phantom_4dbti\nimport mne", "_____no_output_____" ] ], [ [ "Read data and compute a dipole fit at the peak of the evoked response\n\n", "_____no_output_____" ] ], [ [ "data_path = phantom_4dbti.data_path()\nraw_fname = op.join(data_path, '%d/e,rfhp1.0Hz')\n\ndipoles = list()\nsphere = mne.make_sphere_model(r0=(0., 0., 0.), head_radius=0.080)\n\nt0 = 0.07 # peak of the response\n\npos = np.empty((4, 3))\n\nfor ii in range(4):\n raw = mne.io.read_raw_bti(raw_fname % (ii + 1,),\n rename_channels=False, preload=True)\n raw.info['bads'] = ['A173', 'A213', 'A232']\n events = mne.find_events(raw, 'TRIGGER', mask=4350, mask_type='not_and')\n epochs = mne.Epochs(raw, events=events, event_id=8192, tmin=-0.2, tmax=0.4,\n preload=True)\n evoked = epochs.average()\n evoked.plot(time_unit='s')\n cov = mne.compute_covariance(epochs, tmax=0.)\n dip = mne.fit_dipole(evoked.copy().crop(t0, t0), cov, sphere)[0]\n pos[ii] = dip.pos[0]", "_____no_output_____" ] ], [ [ "Compute localisation errors\n\n", "_____no_output_____" ] ], [ [ "actual_pos = 0.01 * np.array([[0.16, 1.61, 5.13],\n [0.17, 1.35, 4.15],\n [0.16, 1.05, 3.19],\n [0.13, 0.80, 2.26]])\nactual_pos = np.dot(actual_pos, [[0, 1, 0], [-1, 0, 0], [0, 0, 1]])\n\nerrors = 1e3 * np.linalg.norm(actual_pos - pos, axis=1)\nprint(\"errors (mm) : %s\" % errors)", "_____no_output_____" ] ], [ [ "Plot the dipoles in 3D\n\n", "_____no_output_____" ] ], [ [ "def plot_pos(pos, color=(0., 0., 0.)):\n mlab.points3d(pos[:, 0], pos[:, 1], pos[:, 2], scale_factor=0.005,\n color=color)\n\n\nmne.viz.plot_alignment(evoked.info, bem=sphere, surfaces=[])\n# Plot the position of the actual dipole\nplot_pos(actual_pos, color=(1., 0., 0.))\n# Plot the position of the estimated dipole\nplot_pos(pos, color=(1., 1., 0.))", "_____no_output_____" ] ] ]

[ "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "from sklearn.model_selection import train_test_split\nimport pandas as pd\nimport numpy as np\nfrom sklearn.preprocessing import StandardScaler\nfrom keras.utils import to_categorical\nfrom keras.models import Sequential\nfrom keras.layers import Dense", "Using TensorFlow backend.\nC:\\Users\\heine\\anaconda3\\lib\\site-packages\\tensorflow\\python\\framework\\dtypes.py:516: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\nC:\\Users\\heine\\anaconda3\\lib\\site-packages\\tensorflow\\python\\framework\\dtypes.py:517: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\nC:\\Users\\heine\\anaconda3\\lib\\site-packages\\tensorflow\\python\\framework\\dtypes.py:518: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\nC:\\Users\\heine\\anaconda3\\lib\\site-packages\\tensorflow\\python\\framework\\dtypes.py:519: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\nC:\\Users\\heine\\anaconda3\\lib\\site-packages\\tensorflow\\python\\framework\\dtypes.py:520: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\nC:\\Users\\heine\\anaconda3\\lib\\site-packages\\tensorflow\\python\\framework\\dtypes.py:525: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\nC:\\Users\\heine\\anaconda3\\lib\\site-packages\\tensorboard\\compat\\tensorflow_stub\\dtypes.py:541: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\nC:\\Users\\heine\\anaconda3\\lib\\site-packages\\tensorboard\\compat\\tensorflow_stub\\dtypes.py:542: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\nC:\\Users\\heine\\anaconda3\\lib\\site-packages\\tensorboard\\compat\\tensorflow_stub\\dtypes.py:543: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\nC:\\Users\\heine\\anaconda3\\lib\\site-packages\\tensorboard\\compat\\tensorflow_stub\\dtypes.py:544: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\nC:\\Users\\heine\\anaconda3\\lib\\site-packages\\tensorboard\\compat\\tensorflow_stub\\dtypes.py:545: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\nC:\\Users\\heine\\anaconda3\\lib\\site-packages\\tensorboard\\compat\\tensorflow_stub\\dtypes.py:550: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\n" ], [ "#df = pd.read_csv(\".\\\\Data_USD.csv\", header=None,skiprows=1)\ndf = pd.read_csv(\".\\\\Data_USD.csv\")\ndf.head().to_csv(\".\\\\test.csv\")", "_____no_output_____" ], [ "T=df.groupby(\"SEX\") ", "_____no_output_____" ], [ "T.describe()", "_____no_output_____" ], [ "df.tail()", "_____no_output_____" ], [ "# X = df.drop('Y_Value',axis =1).values\n# y = df['Y_Value'].values\nX = df.drop('DEFAULT_PAYMENT_NEXT_MO',axis =1).values\nX[2999,0]", "_____no_output_____" ], [ "X.shape", "_____no_output_____" ], [ "y = df['DEFAULT_PAYMENT_NEXT_MO'].values\n#y.reshape(-1,1)", "_____no_output_____" ], [ "#print(X.shape)\nX.shape", "_____no_output_____" ], [ "#print(y.shape)\ny.shape", "_____no_output_____" ], [ "X_train, X_test, y_train, y_test = train_test_split (X,y,test_size=0.2, random_state=42)", "_____no_output_____" ], [ "y_test.T", "_____no_output_____" ], [ "X_test.shape", "_____no_output_____" ], [ "from sklearn.preprocessing import StandardScaler\n\nX_scaler = StandardScaler().fit(X_train)", "_____no_output_____" ], [ "X_scaler", "_____no_output_____" ], [ "X_train_scaled = X_scaler.transform(X_train)\nX_test_scaled = X_scaler.transform(X_test)", "_____no_output_____" ], [ "X_train_scaled", "_____no_output_____" ], [ "y_train_categorical = to_categorical(y_train)\ny_test_categorical = to_categorical(y_test)", "_____no_output_____" ], [ "from keras.models import Sequential\n\n#instantiate\nmodel = Sequential()", "_____no_output_____" ], [ "from keras.layers import Dense\n\nnumber_inputs = 10\nnumber_hidden = 30\n\nmodel.add(Dense(units = number_hidden, activation ='relu', input_dim=number_inputs))\nmodel.add(Dense(units = 35, activation ='relu')) #second hidden layer\nmodel.add(Dense(units = 25, activation ='relu')) #second hidden layer\nmodel.add(Dense(units = 15, activation ='relu')) #second hidden layer\nmodel.add(Dense(units = 5, activation ='relu')) #third hidden layer", "_____no_output_____" ], [ "number_classes =2 ## yes or no\nmodel.add(Dense(units = number_classes, activation = 'softmax'))", "_____no_output_____" ], [ "model.summary()", "Model: \"sequential_1\"\n_________________________________________________________________\nLayer (type) Output Shape Param # \n=================================================================\ndense_1 (Dense) (None, 30) 330 \n_________________________________________________________________\ndense_2 (Dense) (None, 35) 1085 \n_________________________________________________________________\ndense_3 (Dense) (None, 25) 900 \n_________________________________________________________________\ndense_4 (Dense) (None, 15) 390 \n_________________________________________________________________\ndense_5 (Dense) (None, 5) 80 \n_________________________________________________________________\ndense_6 (Dense) (None, 2) 12 \n=================================================================\nTotal params: 2,797\nTrainable params: 2,797\nNon-trainable params: 0\n_________________________________________________________________\n" ], [ "#compile the model\nmodel.compile(optimizer = 'sgd' ,\n loss = 'categorical_crossentropy',\n metrics =['accuracy'])", "_____no_output_____" ], [ "#train the model\n\nmodel.fit(X_train_scaled, y_train_categorical, epochs=100,shuffle = True,verbose =2)", "WARNING:tensorflow:From C:\\Users\\heine\\anaconda3\\lib\\site-packages\\keras\\backend\\tensorflow_backend.py:422: The name tf.global_variables is deprecated. Please use tf.compat.v1.global_variables instead.\n\nEpoch 1/100\n - 2s - loss: 0.5336 - accuracy: 0.7782\nEpoch 2/100\n - 1s - loss: 0.4987 - accuracy: 0.7782\nEpoch 3/100\n - 1s - loss: 0.4804 - accuracy: 0.7782\nEpoch 4/100\n - 2s - loss: 0.4696 - accuracy: 0.7965\nEpoch 5/100\n - 2s - loss: 0.4625 - accuracy: 0.8047\nEpoch 6/100\n - 2s - loss: 0.4579 - accuracy: 0.8060\nEpoch 7/100\n - 1s - loss: 0.4553 - accuracy: 0.8067\nEpoch 8/100\n - 1s - loss: 0.4538 - accuracy: 0.8055\nEpoch 9/100\n - 1s - loss: 0.4527 - accuracy: 0.8055\nEpoch 10/100\n - 1s - loss: 0.4516 - accuracy: 0.8053\nEpoch 11/100\n - 1s - loss: 0.4506 - accuracy: 0.8062\nEpoch 12/100\n - 1s - loss: 0.4496 - accuracy: 0.8050\nEpoch 13/100\n - 1s - loss: 0.4488 - accuracy: 0.8055\nEpoch 14/100\n - 1s - loss: 0.4479 - accuracy: 0.8055\nEpoch 15/100\n - 2s - loss: 0.4474 - accuracy: 0.8060\nEpoch 16/100\n - 1s - loss: 0.4465 - accuracy: 0.8065\nEpoch 17/100\n - 1s - loss: 0.4461 - accuracy: 0.8067\nEpoch 18/100\n - 1s - loss: 0.4453 - accuracy: 0.8062\nEpoch 19/100\n - 1s - loss: 0.4449 - accuracy: 0.8062\nEpoch 20/100\n - 1s - loss: 0.4445 - accuracy: 0.8065\nEpoch 21/100\n - 1s - loss: 0.4442 - accuracy: 0.8060\nEpoch 22/100\n - 1s - loss: 0.4436 - accuracy: 0.8067\nEpoch 23/100\n - 1s - loss: 0.4430 - accuracy: 0.8070\nEpoch 24/100\n - 1s - loss: 0.4425 - accuracy: 0.8064\nEpoch 25/100\n - 1s - loss: 0.4423 - accuracy: 0.8076\nEpoch 26/100\n - 1s - loss: 0.4424 - accuracy: 0.8059\nEpoch 27/100\n - 2s - loss: 0.4420 - accuracy: 0.8066\nEpoch 28/100\n - 1s - loss: 0.4415 - accuracy: 0.8073\nEpoch 29/100\n - 1s - loss: 0.4415 - accuracy: 0.8068\nEpoch 30/100\n - 1s - loss: 0.4410 - accuracy: 0.8056\nEpoch 31/100\n - 1s - loss: 0.4408 - accuracy: 0.8070\nEpoch 32/100\n - 1s - loss: 0.4400 - accuracy: 0.8075\nEpoch 33/100\n - 1s - loss: 0.4402 - accuracy: 0.8071\nEpoch 34/100\n - 1s - loss: 0.4397 - accuracy: 0.8075\nEpoch 35/100\n - 1s - loss: 0.4398 - accuracy: 0.8080\nEpoch 36/100\n - 1s - loss: 0.4394 - accuracy: 0.8061\nEpoch 37/100\n - 1s - loss: 0.4391 - accuracy: 0.8078\nEpoch 38/100\n - 1s - loss: 0.4390 - accuracy: 0.8076\nEpoch 39/100\n - 1s - loss: 0.4387 - accuracy: 0.8082\nEpoch 40/100\n - 1s - loss: 0.4383 - accuracy: 0.8076\nEpoch 41/100\n - 1s - loss: 0.4378 - accuracy: 0.8074\nEpoch 42/100\n - 1s - loss: 0.4382 - accuracy: 0.8071\nEpoch 43/100\n - 1s - loss: 0.4378 - accuracy: 0.8080\nEpoch 44/100\n - 1s - loss: 0.4374 - accuracy: 0.8075\nEpoch 45/100\n - 1s - loss: 0.4371 - accuracy: 0.8070\nEpoch 46/100\n - 1s - loss: 0.4369 - accuracy: 0.8079\nEpoch 47/100\n - 1s - loss: 0.4369 - accuracy: 0.8075\nEpoch 48/100\n - 1s - loss: 0.4362 - accuracy: 0.8079\nEpoch 49/100\n - 1s - loss: 0.4366 - accuracy: 0.8080\nEpoch 50/100\n - 1s - loss: 0.4357 - accuracy: 0.8086\nEpoch 51/100\n - 2s - loss: 0.4355 - accuracy: 0.8087\nEpoch 52/100\n - 2s - loss: 0.4357 - accuracy: 0.8076\nEpoch 53/100\n - 2s - loss: 0.4352 - accuracy: 0.8073\nEpoch 54/100\n - 1s - loss: 0.4353 - accuracy: 0.8069\nEpoch 55/100\n - 1s - loss: 0.4353 - accuracy: 0.8085\nEpoch 56/100\n - 1s - loss: 0.4350 - accuracy: 0.8087\nEpoch 57/100\n - 1s - loss: 0.4348 - accuracy: 0.8080\nEpoch 58/100\n - 1s - loss: 0.4347 - accuracy: 0.8074\nEpoch 59/100\n - 1s - loss: 0.4346 - accuracy: 0.8085\nEpoch 60/100\n - 1s - loss: 0.4340 - accuracy: 0.8083\nEpoch 61/100\n - 1s - loss: 0.4337 - accuracy: 0.8091\nEpoch 62/100\n - 2s - loss: 0.4335 - accuracy: 0.8085\nEpoch 63/100\n - 1s - loss: 0.4334 - accuracy: 0.8093\nEpoch 64/100\n - 1s - loss: 0.4335 - accuracy: 0.8095\nEpoch 65/100\n - 1s - loss: 0.4332 - accuracy: 0.8087\nEpoch 66/100\n - 1s - loss: 0.4330 - accuracy: 0.8089\nEpoch 67/100\n - 1s - loss: 0.4331 - accuracy: 0.8093\nEpoch 68/100\n - 1s - loss: 0.4333 - accuracy: 0.8080\nEpoch 69/100\n - 1s - loss: 0.4331 - accuracy: 0.8088\nEpoch 70/100\n - 1s - loss: 0.4326 - accuracy: 0.8090\nEpoch 71/100\n - 1s - loss: 0.4324 - accuracy: 0.8087\nEpoch 72/100\n - 1s - loss: 0.4326 - accuracy: 0.8092\nEpoch 73/100\n - 1s - loss: 0.4318 - accuracy: 0.8092\nEpoch 74/100\n - 1s - loss: 0.4319 - accuracy: 0.8098\nEpoch 75/100\n - 1s - loss: 0.4314 - accuracy: 0.8085\nEpoch 76/100\n - 1s - loss: 0.4315 - accuracy: 0.8083\nEpoch 77/100\n - 1s - loss: 0.4314 - accuracy: 0.8101\nEpoch 78/100\n - 1s - loss: 0.4311 - accuracy: 0.8105\nEpoch 79/100\n - 1s - loss: 0.4315 - accuracy: 0.8091\nEpoch 80/100\n - 1s - loss: 0.4312 - accuracy: 0.8092\nEpoch 81/100\n - 1s - loss: 0.4312 - accuracy: 0.8101\nEpoch 82/100\n - 1s - loss: 0.4312 - accuracy: 0.8100\nEpoch 83/100\n - 1s - loss: 0.4302 - accuracy: 0.8102\nEpoch 84/100\n - 1s - loss: 0.4311 - accuracy: 0.8090\nEpoch 85/100\n - 1s - loss: 0.4302 - accuracy: 0.8095\nEpoch 86/100\n - 1s - loss: 0.4311 - accuracy: 0.8096\nEpoch 87/100\n - 2s - loss: 0.4308 - accuracy: 0.8105\nEpoch 88/100\n - 1s - loss: 0.4297 - accuracy: 0.8108\nEpoch 89/100\n - 1s - loss: 0.4301 - accuracy: 0.8095\nEpoch 90/100\n - 1s - loss: 0.4301 - accuracy: 0.8093\nEpoch 91/100\n - 1s - loss: 0.4303 - accuracy: 0.8096\nEpoch 92/100\n - 1s - loss: 0.4299 - accuracy: 0.8113\nEpoch 93/100\n - 1s - loss: 0.4291 - accuracy: 0.8096\nEpoch 94/100\n - 1s - loss: 0.4296 - accuracy: 0.8102\nEpoch 95/100\n - 1s - loss: 0.4291 - accuracy: 0.8098\nEpoch 96/100\n - 1s - loss: 0.4293 - accuracy: 0.8111\nEpoch 97/100\n - 1s - loss: 0.4291 - accuracy: 0.8103\nEpoch 98/100\n - 1s - loss: 0.4286 - accuracy: 0.8105\nEpoch 99/100\n - 2s - loss: 0.4287 - accuracy: 0.8110\nEpoch 100/100\n - 1s - loss: 0.4283 - accuracy: 0.8118\n" ], [ "model.save(\"ccneuralnetwork.h5\")", "_____no_output_____" ], [ "#quantify the model\nmodel_loss, model_accuracy = model.evaluate(X_test_scaled,y_test_categorical,verbose =2)\nprint( model_loss )\nprint (model_accuracy)", "0.45065889310836793\n0.8054999709129333\n" ] ], [ [ "F1, Precision Recall, and Confusion Matrix", "_____no_output_____" ] ], [ [ "from sklearn.metrics import precision_recall_fscore_support\nfrom sklearn.metrics import recall_score\nfrom sklearn.metrics import classification_report", "_____no_output_____" ], [ "y_prediction = model.predict_classes(X_test)", "_____no_output_____" ], [ "y_prediction.reshape(-1,1)", "_____no_output_____" ], [ "print(\"Recall score:\"+ str(recall_score(y_test, y_prediction)))", "Recall score:0.0\n" ], [ "print(classification_report(y_test, y_prediction,\n target_names=[\"default\", \"non_default\"]))", " precision recall f1-score support\n\n default 0.78 1.00 0.88 4687\n non_default 0.00 0.00 0.00 1313\n\n accuracy 0.78 6000\n macro avg 0.39 0.50 0.44 6000\nweighted avg 0.61 0.78 0.69 6000\n\n" ], [ "import itertools\nimport matplotlib.pyplot as plt\nfrom sklearn.metrics import confusion_matrix", "_____no_output_____" ], [ "def plot_confusion_matrix(cm, classes,\n normalize=False,\n title='Confusion matrix',\n cmap=plt.cm.Blues):\n \"\"\"\n This function prints and plots the confusion matrix.\n Normalization can be applied by setting `normalize=True`.\n \"\"\"\n if normalize:\n cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n print(\"Normalized confusion matrix\")\n else:\n print('Confusion matrix, without normalization')\n\n print(cm)\n\n plt.imshow(cm, interpolation='nearest', cmap=cmap)\n plt.title(title)\n plt.colorbar()\n tick_marks = np.arange(len(classes))\n plt.xticks(tick_marks, classes, rotation=45)\n plt.yticks(tick_marks, classes)\n\n fmt = '.2f' if normalize else 'd'\n thresh = cm.max() / 2.\n for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):\n plt.text(j, i, format(cm[i, j], fmt),\n horizontalalignment=\"center\",\n color=\"red\" if cm[i, j] > thresh else \"black\")\n\n plt.tight_layout()\n plt.ylabel('True label')\n plt.xlabel('Predicted label')\n\n# Compute confusion matrix\ncnf_matrix = confusion_matrix(y_test, y_prediction)\nnp.set_printoptions(precision=2)\n\n# Plot non-normalized confusion matrix\nplt.figure()\nplot_confusion_matrix(cnf_matrix, classes=['Defualt', 'Non_default'],\n title='Confusion matrix, without normalization')\n\n# Plot normalized confusion matrix\nplt.figure()\nplot_confusion_matrix(cnf_matrix, classes=['Defualt', 'Non_default'], normalize=True,\n title='Normalized confusion matrix')\n\nplt.show()", "Confusion matrix, without normalization\n[[4687 0]\n [1313 0]]\nNormalized confusion matrix\n[[1. 0.]\n [1. 0.]]\n" ] ] ]

[ "code", "markdown", "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Nonlinear recharge models\n*R.A. Collenteur, University of Graz*\n\nThis notebook explains the use of the `RechargeModel` stress model to simulate the combined effect of precipitation and potential evaporation on the groundwater levels. For the computation of the groundwater recharge, three recharge models are currently available:\n\n- `Linear` ([Berendrecht et al., 2003](#References); [von Asmuth et al., 2008](#References))\n- `Berendrecht` ([Berendrecht et al., 2006](#References))\n- `FlexModel` ([Collenteur et al., 2021](#References))\n\nThe first model is a simple linear function of precipitation and potential evaporation while the latter two are simulate a nonlinear response of recharge to precipitation using a soil-water balance concepts. Detailed descriptions of these models can be found in articles listed in the [References](#References) at the end of this notebook.\n\n<div class=\"alert alert-info\">\n \n<b>Tip</b> \n \nTo run this notebook and the related non-linear recharge models, it is strongly recommended to install Numba (http://numba.pydata.org). This Just-In-Time (JIT) compiler compiles the computationally intensive part of the recharge calculation, making the non-linear model as fast as the Linear recharge model.\n \n</div>\n", "_____no_output_____" ] ], [ [ "import pandas as pd\nimport pastas as ps\nimport matplotlib.pyplot as plt\n\nps.show_versions(numba=True)\nps.set_log_level(\"INFO\")", "Python version: 3.8.2 (default, Mar 25 2020, 11:22:43) \n[Clang 4.0.1 (tags/RELEASE_401/final)]\nNumpy version: 1.20.2\nScipy version: 1.6.2\nPandas version: 1.1.5\nPastas version: 0.18.0b\nMatplotlib version: 3.3.4\nnumba version: 0.51.2\n" ] ], [ [ "## Read Input data\nInput data handling is similar to other stressmodels. The only thing that is necessary to check is that the precipitation and evaporation are provided in mm/day. This is necessary because the parameters for the nonlinear recharge models are defined in mm for the length unit and days for the time unit. It is possible to use other units, but this would require manually setting the initial values and parameter boundaries for the recharge models.", "_____no_output_____" ] ], [ [ "head = pd.read_csv(\"../data/B32C0639001.csv\", parse_dates=['date'], \n index_col='date', squeeze=True) \n\n# Make this millimeters per day\nevap = ps.read_knmi(\"../data/etmgeg_260.txt\", variables=\"EV24\").series * 1e3\nrain = ps.read_knmi(\"../data/etmgeg_260.txt\", variables=\"RH\").series * 1e3\n\nfig, axes = plt.subplots(3,1, figsize=(10,6), sharex=True)\nhead.plot(ax=axes[0], x_compat=True, linestyle=\" \", marker=\".\")\nevap.plot(ax=axes[1], x_compat=True)\nrain.plot(ax=axes[2], x_compat=True)\naxes[0].set_ylabel(\"Head [m]\")\naxes[1].set_ylabel(\"Evap [mm/d]\")\naxes[2].set_ylabel(\"Rain [mm/d]\")\n\nplt.xlim(\"1985\", \"2005\");", "INFO: Inferred frequency for time series EV24 260: freq=D\nINFO: Inferred frequency for time series RH 260: freq=D\n" ] ], [ [ "## Make a basic model\nThe normal workflow may be used to create and calibrate the model.\n1. Create a Pastas `Model` instance\n2. Choose a recharge model. All recharge models can be accessed through the recharge subpackage (`ps.rch`).\n3. Create a `RechargeModel` object and add it to the model\n4. Solve and visualize the model\n\n", "_____no_output_____" ] ], [ [ "ml = ps.Model(head)\n\n# Select a recharge model\nrch = ps.rch.FlexModel()\n#rch = ps.rch.Berendrecht()\n#rch = ps.rch.Linear()\n\nrm = ps.RechargeModel(rain, evap, recharge=rch, rfunc=ps.Gamma, name=\"rch\")\nml.add_stressmodel(rm)\n\nml.solve(noise=True, tmin=\"1990\", report=\"basic\")\nml.plots.results(figsize=(10,6));", "INFO: Cannot determine frequency of series head: freq=None. The time series is irregular.\nINFO: Inferred frequency for time series RH 260: freq=D\nINFO: Inferred frequency for time series EV24 260: freq=D\n" ] ], [ [ "## Analyze the estimated recharge flux\nAfter the parameter estimation we can take a look at the recharge flux computed by the model. The flux is easy to obtain using the `get_stress` method of the model object, which automatically provides the optimal parameter values that were just estimated. After this, we can for example look at the yearly recharge flux estimated by the Pastas model.", "_____no_output_____" ] ], [ [ "recharge = ml.get_stress(\"rch\").resample(\"A\").sum()\nax = recharge.plot.bar(figsize=(10,3))\nax.set_xticklabels(recharge.index.year)\nplt.ylabel(\"Recharge [mm/year]\");", "_____no_output_____" ] ], [ [ "## A few things to keep in mind:\nBelow are a few things to keep in mind while using the (nonlinear) recharge models.\n\n- The use of an appropriate warmup period is necessary, so make sure the precipitation and evaporation are available some time (e.g., one year) before the calibration period.\n- Make sure that the units of the precipitation fluxes are in mm/day and that the DatetimeIndex matches exactly.\n- It may be possible to fix or vary certain parameters, dependent on the problem. Obtaining better initial parameters may be possible by solving without a noise model first (`ml.solve(noise=False)`) and then solve it again using a noise model.\n- For relatively shallow groundwater levels, it may be better to use the `Exponential` response function as the the non-linear models already cause a delayed response.\n\n## References\n- Berendrecht, W. L., Heemink, A. W., van Geer, F. C., and Gehrels, J. C. (2003) [Decoupling of modeling and measuring interval in groundwater time series analysis based on response characteristics](https://doi.org/10.1016/S0022-1694(03)00075-1), Journal of Hydrology, 278, 1–16.\n- Berendrecht, W. L., Heemink, A. W., van Geer, F. C., and Gehrels, J. C. (2006) [A non-linear state space approach to model groundwater fluctuations](https://www.sciencedirect.com/science/article/abs/pii/S0309170805002113), Advances in Water Resources, 29, 959–973.\n- Collenteur, R., Bakker, M., Klammler, G., and Birk, S. (2021) [Estimation of groundwater recharge from groundwater levels using nonlinear transfer function noise models and comparison to lysimeter data](https://doi.org/10.5194/hess-2020-392), Hydrol. Earth Syst. Sci., 25, 2931–2949.\n- Von Asmuth, J.R., Maas, K., Bakker, M. and Petersen, J. (2008) [Modeling Time Series of Ground Water Head Fluctuations Subjected to Multiple Stresses](https://doi.org/10.1111/j.1745-6584.2007.00382.x). Groundwater, 46: 30-40.\n\n## Data Sources\nIn this notebook we analysed a head time series near the town of De Bilt in the Netherlands. Data is obtained from the following resources:\n- The heads (`B32C0639001.csv`) are downloaded from https://www.dinoloket.nl/ \n- The precipitation and potential evaporation (`etmgeg_260.txt`) are downloaded from https://knmi.nl", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import pandas as pd\nimport numpy as np\n\ndf1 = pd.read_csv('../data/raw/drug_deaths.csv')\ndf1.info()", "<class 'pandas.core.frame.DataFrame'>\nRangeIndex: 5105 entries, 0 to 5104\nData columns (total 42 columns):\n # Column Non-Null Count Dtype \n--- ------ -------------- ----- \n 0 Unnamed: 0 5105 non-null int64 \n 1 ID 5105 non-null object \n 2 Date 5103 non-null object \n 3 DateType 5103 non-null float64\n 4 Age 5102 non-null float64\n 5 Sex 5099 non-null object \n 6 Race 5092 non-null object \n 7 ResidenceCity 4932 non-null object \n 8 ResidenceCounty 4308 non-null object \n 9 ResidenceState 3556 non-null object \n 10 DeathCity 5100 non-null object \n 11 DeathCounty 4005 non-null object \n 12 Location 5081 non-null object \n 13 LocationifOther 590 non-null object \n 14 DescriptionofInjury 4325 non-null object \n 15 InjuryPlace 5039 non-null object \n 16 InjuryCity 3349 non-null object \n 17 InjuryCounty 2364 non-null object \n 18 InjuryState 1424 non-null object \n 19 COD 5105 non-null object \n 20 OtherSignifican 169 non-null object \n 21 Heroin 5105 non-null int64 \n 22 Cocaine 5105 non-null int64 \n 23 Fentanyl 5105 non-null object \n 24 Fentanyl_Analogue 5105 non-null float64\n 25 Oxycodone 5105 non-null int64 \n 26 Oxymorphone 5105 non-null int64 \n 27 Ethanol 5105 non-null int64 \n 28 Hydrocodone 5105 non-null int64 \n 29 Benzodiazepine 5105 non-null int64 \n 30 Methadone 5105 non-null int64 \n 31 Amphet 5105 non-null int64 \n 32 Tramad 5105 non-null int64 \n 33 Morphine_NotHeroin 5105 non-null object \n 34 Hydromorphone 5105 non-null int64 \n 35 Other 435 non-null object \n 36 OpiateNOS 5105 non-null int64 \n 37 AnyOpioid 5105 non-null object \n 38 MannerofDeath 5095 non-null object \n 39 DeathCityGeo 5105 non-null object \n 40 ResidenceCityGeo 5012 non-null object \n 41 InjuryCityGeo 5027 non-null object \ndtypes: float64(3), int64(13), object(26)\nmemory usage: 1.6+ MB\n" ], [ "#Drop columns not needed in the analysis or with many null values\ndf1 = df1.drop(['Unnamed: 0','DateType','COD','ResidenceCity','ResidenceCounty', 'ResidenceState','DeathCity',\n 'DeathCounty','Location','LocationifOther','DescriptionofInjury','InjuryPlace','InjuryCity',\n 'InjuryCounty','InjuryState','OtherSignifican', 'Other','DeathCityGeo','ResidenceCityGeo',\n 'InjuryCityGeo'],axis = 1)\n\n#Rename with the full drug name\ndf1 = df1.rename(columns={\"Amphet\": \"Amphetamine\", \"Tramad\": \"Tramadol\"})\n\ndf1.info()", "<class 'pandas.core.frame.DataFrame'>\nRangeIndex: 5105 entries, 0 to 5104\nData columns (total 22 columns):\n # Column Non-Null Count Dtype \n--- ------ -------------- ----- \n 0 ID 5105 non-null object \n 1 Date 5103 non-null object \n 2 Age 5102 non-null float64\n 3 Sex 5099 non-null object \n 4 Race 5092 non-null object \n 5 Heroin 5105 non-null int64 \n 6 Cocaine 5105 non-null int64 \n 7 Fentanyl 5105 non-null object \n 8 Fentanyl_Analogue 5105 non-null float64\n 9 Oxycodone 5105 non-null int64 \n 10 Oxymorphone 5105 non-null int64 \n 11 Ethanol 5105 non-null int64 \n 12 Hydrocodone 5105 non-null int64 \n 13 Benzodiazepine 5105 non-null int64 \n 14 Methadone 5105 non-null int64 \n 15 Amphetamine 5105 non-null int64 \n 16 Tramadol 5105 non-null int64 \n 17 Morphine_NotHeroin 5105 non-null object \n 18 Hydromorphone 5105 non-null int64 \n 19 OpiateNOS 5105 non-null int64 \n 20 AnyOpioid 5105 non-null object \n 21 MannerofDeath 5095 non-null object \ndtypes: float64(2), int64(12), object(8)\nmemory usage: 877.5+ KB\n" ], [ "df1 = df1.assign(Date = pd.to_datetime(df1[\"Date\"]))\ndf1 = df1.dropna()\ndf1.info()", "<class 'pandas.core.frame.DataFrame'>\nInt64Index: 5080 entries, 1 to 5104\nData columns (total 22 columns):\n # Column Non-Null Count Dtype \n--- ------ -------------- ----- \n 0 ID 5080 non-null object \n 1 Date 5080 non-null datetime64[ns]\n 2 Age 5080 non-null float64 \n 3 Sex 5080 non-null object \n 4 Race 5080 non-null object \n 5 Heroin 5080 non-null int64 \n 6 Cocaine 5080 non-null int64 \n 7 Fentanyl 5080 non-null object \n 8 Fentanyl_Analogue 5080 non-null float64 \n 9 Oxycodone 5080 non-null int64 \n 10 Oxymorphone 5080 non-null int64 \n 11 Ethanol 5080 non-null int64 \n 12 Hydrocodone 5080 non-null int64 \n 13 Benzodiazepine 5080 non-null int64 \n 14 Methadone 5080 non-null int64 \n 15 Amphetamine 5080 non-null int64 \n 16 Tramadol 5080 non-null int64 \n 17 Morphine_NotHeroin 5080 non-null object \n 18 Hydromorphone 5080 non-null int64 \n 19 OpiateNOS 5080 non-null int64 \n 20 AnyOpioid 5080 non-null object \n 21 MannerofDeath 5080 non-null object \ndtypes: datetime64[ns](1), float64(2), int64(12), object(7)\nmemory usage: 912.8+ KB\n" ], [ " df1.drop_duplicates(['ID'])", "_____no_output_____" ], [ "df1 = df1.astype({'Age': 'int64', 'Fentanyl_Analogue': 'int64'})\ndf1.info()", "<class 'pandas.core.frame.DataFrame'>\nInt64Index: 5080 entries, 1 to 5104\nData columns (total 22 columns):\n # Column Non-Null Count Dtype \n--- ------ -------------- ----- \n 0 ID 5080 non-null object \n 1 Date 5080 non-null datetime64[ns]\n 2 Age 5080 non-null int64 \n 3 Sex 5080 non-null object \n 4 Race 5080 non-null object \n 5 Heroin 5080 non-null int64 \n 6 Cocaine 5080 non-null int64 \n 7 Fentanyl 5080 non-null object \n 8 Fentanyl_Analogue 5080 non-null int64 \n 9 Oxycodone 5080 non-null int64 \n 10 Oxymorphone 5080 non-null int64 \n 11 Ethanol 5080 non-null int64 \n 12 Hydrocodone 5080 non-null int64 \n 13 Benzodiazepine 5080 non-null int64 \n 14 Methadone 5080 non-null int64 \n 15 Amphetamine 5080 non-null int64 \n 16 Tramadol 5080 non-null int64 \n 17 Morphine_NotHeroin 5080 non-null object \n 18 Hydromorphone 5080 non-null int64 \n 19 OpiateNOS 5080 non-null int64 \n 20 AnyOpioid 5080 non-null object \n 21 MannerofDeath 5080 non-null object \ndtypes: datetime64[ns](1), int64(14), object(7)\nmemory usage: 912.8+ KB\n" ], [ "err1 = pd.isnull(pd.to_numeric(df1['Fentanyl'], errors='coerce'))\nerr2 = pd.isnull(pd.to_numeric(df1['Morphine_NotHeroin'], errors='coerce'))\nerr3 = pd.isnull(pd.to_numeric(df1['AnyOpioid'], errors='coerce'))\ndf1[err1]", "_____no_output_____" ], [ "df1[err2]", "_____no_output_____" ], [ "df1[err3]", "_____no_output_____" ], [ "df2 = df1.drop(index =[507,2808,3741,3801,62,583,4199,4384,263,456,3213,3499,4794])\ndf2", "_____no_output_____" ], [ "df2 = df2.astype({'Fentanyl': 'int64', 'Morphine_NotHeroin': 'int64', 'AnyOpioid': 'int64'})\ndf2.info()", "<class 'pandas.core.frame.DataFrame'>\nInt64Index: 5067 entries, 1 to 5104\nData columns (total 22 columns):\n # Column Non-Null Count Dtype \n--- ------ -------------- ----- \n 0 ID 5067 non-null object \n 1 Date 5067 non-null datetime64[ns]\n 2 Age 5067 non-null int64 \n 3 Sex 5067 non-null object \n 4 Race 5067 non-null object \n 5 Heroin 5067 non-null int64 \n 6 Cocaine 5067 non-null int64 \n 7 Fentanyl 5067 non-null int64 \n 8 Fentanyl_Analogue 5067 non-null int64 \n 9 Oxycodone 5067 non-null int64 \n 10 Oxymorphone 5067 non-null int64 \n 11 Ethanol 5067 non-null int64 \n 12 Hydrocodone 5067 non-null int64 \n 13 Benzodiazepine 5067 non-null int64 \n 14 Methadone 5067 non-null int64 \n 15 Amphetamine 5067 non-null int64 \n 16 Tramadol 5067 non-null int64 \n 17 Morphine_NotHeroin 5067 non-null int64 \n 18 Hydromorphone 5067 non-null int64 \n 19 OpiateNOS 5067 non-null int64 \n 20 AnyOpioid 5067 non-null int64 \n 21 MannerofDeath 5067 non-null object \ndtypes: datetime64[ns](1), int64(17), object(4)\nmemory usage: 910.5+ KB\n" ], [ "import matplotlib.pyplot as plt\nimport seaborn as sns\n\nyear = pd.to_datetime(df['Date']).dt.year.value_counts()\nplt.figure(figsize = (10, 5))\nwith plt.style.context('fivethirtyeight'):\n graph1 = sns.barplot(x = year.index.astype('int64'), y = year.values.astype('int64'), \n palette=sns.cubehelix_palette(8))\nplt.tight_layout()\nplt.ylabel('Deaths')\nplt.show()", "_____no_output_____" ], [ " ", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "%pylab inline\nfrom sklearn import datasets ## load the iris data. \niris = datasets.load_iris()\nX = iris.data \nC = iris.target\nX.shape\n", "Populating the interactive namespace from numpy and matplotlib\n" ], [ "## DEAL WITH DATA\n#total = np.matrix()\nimport os\nimport glob\npath = \"./data\"\n\nfor filename in glob.glob(os.path.join(path, '*.txt')):\n print (filename)\n with open(filename) as f:\n content = f.readlines()\n# you may also want to remove whitespace characters like `\\n` at the end of each line\n content = [x.strip() for x in content]\n mat1 = np.array([float(s) for s in content[0].split(' ')])\n mat2 = np.array([float(s) for s in content[1].split(' ')])\n mat3 = np.array([float(s) for s in content[2].split(' ')])\n mat4 = np.array([float(s) for s in content[3].split(' ')])\n# A1 = vstack((A1, mat1))\n# A2 = vstack((A2, mat2))\n# A3 = vstack((A3, mat3))\n# A4 = vstack((A4, mat4))\n \n ", "./data/filename-Cashew+chicken.txt\n1.5 0.0 0.0 0.0 1.0 0.0 2.0 0.0 0.0 2.0 0.0 0.0 0.0 0.0 0.0 0.0 1.5 0.0 1.0 0.0 0.0 0.0 0.0\n[1.5, 0.0, 0.0, 0.0, 1.0, 0.0, 2.0, 0.0, 0.0, 2.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.5, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]\n./data/filename-Pavlova.txt\n1.0 0.0 0.3333333333333333 0.0 3.1666666666666665 0.0 1.5 0.0 0.5 0.0 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.0 2.0 0.0 0.0 0.0 0.0\n[1.0, 0.0, 0.3333333333333333, 0.0, 3.1666666666666665, 0.0, 1.5, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 2.0, 0.0, 0.0, 0.0, 0.0]\n./data/filename-bolognese+sauce.txt\n0.0 0.0 0.0 0.0 0.5 0.0 1.0 0.0 0.0 5.0 0.0 0.0 0.0 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.0 0.0 0.0\n[0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 1.0, 0.0, 0.0, 5.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]\n./data/filename-ITALIAN+NACHOS.txt\n0.25 0.2 0.0 0.0 0.0 0.2 1.15 0.0 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.2 0.0 0.0 0.0\n[0.25, 0.2, 0.0, 0.0, 0.0, 0.2, 1.15, 0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.2, 0.0, 0.0, 0.0]\n./data/filename-Greek+Salad.txt\n1.75 1.45 0.5 0.0 1.5 0.45 3.4 0.0 0.0 0.0 0.0 0.25 0.0 0.3333333333333333 0.0 0.0 0.0 0.0 3.0 0.7 0.6666666666666666 1.0 0.0\n[1.75, 1.45, 0.5, 0.0, 1.5, 0.45, 3.4, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 0.3333333333333333, 0.0, 0.0, 0.0, 0.0, 3.0, 0.7, 0.6666666666666666, 1.0, 0.0]\n./data/filename-dal.txt\n3.5 0.2 0.0 0.0 2.5 0.2 1.4 0.0 0.5 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.5 1.0 0.2 0.0 0.0 0.0\n[3.5, 0.2, 0.0, 0.0, 2.5, 0.2, 1.4, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 1.0, 0.2, 0.0, 0.0, 0.0]\n./data/filename-Nem+cuon.txt\n4.25 0.5 0.125 0.125 1.5 0.0 7.0 0.0 0.5 2.5 0.0 1.6666666666666667 0.0 0.3333333333333333 0.0 0.0 0.0 0.0 1.0 0.5 0.0 0.0 0.0\n[4.25, 0.5, 0.125, 0.125, 1.5, 0.0, 7.0, 0.0, 0.5, 2.5, 0.0, 1.6666666666666667, 0.0, 0.3333333333333333, 0.0, 0.0, 0.0, 0.0, 1.0, 0.5, 0.0, 0.0, 0.0]\n./data/filename-Veg+Dum+Biryani.txt\n6.666666666666666 0.6166666666666667 0.962121212121212 0.09090909090909091 1.0 0.2 2.047727272727273 0.0 1.0 1.0909090909090908 0.0 0.9583333333333334 0.0 0.16666666666666666 0.0 0.0 0.0 0.0 1.0 0.2 0.0 1.0 0.0\n[6.666666666666666, 0.6166666666666667, 0.962121212121212, 0.09090909090909091, 1.0, 0.2, 2.047727272727273, 0.0, 1.0, 1.0909090909090908, 0.0, 0.9583333333333334, 0.0, 0.16666666666666666, 0.0, 0.0, 0.0, 0.0, 1.0, 0.2, 0.0, 1.0, 0.0]\n./data/filename-pho.txt\n9.833333333333334 0.5 1.5909090909090908 1.1818181818181819 2.5 1.3823529411764706 8.712121212121211 0.0 2.0 6.181818181818182 0.0 0.8333333333333334 0.0 3.1666666666666665 0.0 0.0 1.0588235294117647 1.0588235294117647 1.0 0.0 0.0 0.0 0.0\n[9.833333333333334, 0.5, 1.5909090909090908, 1.1818181818181819, 2.5, 1.3823529411764706, 8.712121212121211, 0.0, 2.0, 6.181818181818182, 0.0, 0.8333333333333334, 0.0, 3.1666666666666665, 0.0, 0.0, 1.0588235294117647, 1.0588235294117647, 1.0, 0.0, 0.0, 0.0, 0.0]\n./data/filename-Ginger+beef .txt\n4.5 2.25 1.5833333333333333 0.0 1.1666666666666665 0.8823529411764706 9.875 0.0 0.5 1.0 0.0 1.9583333333333335 0.0 2.1666666666666665 0.0 0.0 0.058823529411764705 1.0588235294117647 0.0 0.0 0.0 0.0 0.0\n[4.5, 2.25, 1.5833333333333333, 0.0, 1.1666666666666665, 0.8823529411764706, 9.875, 0.0, 0.5, 1.0, 0.0, 1.9583333333333335, 0.0, 2.1666666666666665, 0.0, 0.0, 0.058823529411764705, 1.0588235294117647, 0.0, 0.0, 0.0, 0.0, 0.0]\n./data/filename-Mapo+tofu.txt\n3.9166666666666665 0.2 0.0 0.0 0.5 1.0823529411764705 6.483333333333333 0.0 1.0 3.0 0.0 0.8333333333333334 0.0 3.1666666666666665 0.0 0.0 2.5588235294117645 0.058823529411764705 1.0 0.2 1.0 0.0 0.0\n[3.9166666666666665, 0.2, 0.0, 0.0, 0.5, 1.0823529411764705, 6.483333333333333, 0.0, 1.0, 3.0, 0.0, 0.8333333333333334, 0.0, 3.1666666666666665, 0.0, 0.0, 2.5588235294117645, 0.058823529411764705, 1.0, 0.2, 1.0, 0.0, 0.0]\n./data/filename-Pasta+e+fagioli.txt\n4.833333333333333 0.5 0.25 0.0 1.5 0.25 5.541666666666667 0.0 0.0 1.0 0.0 1.125 0.0 2.5 0.0 0.0 0.0 0.5 2.0 0.0 1.0 0.0 0.0\n[4.833333333333333, 0.5, 0.25, 0.0, 1.5, 0.25, 5.541666666666667, 0.0, 0.0, 1.0, 0.0, 1.125, 0.0, 2.5, 0.0, 0.0, 0.0, 0.5, 2.0, 0.0, 1.0, 0.0, 0.0]\n./data/filename-Caesar+Salad.txt\n1.0833333333333333 1.25 0.7954545454545454 0.09090909090909091 1.5 0.0 4.5643939393939394 0.0 1.0 1.0909090909090908 0.0 0.125 0.0 0.0 0.0 0.0 0.5 0.0 1.0 0.0 0.0 1.0 0.0\n[1.0833333333333333, 1.25, 0.7954545454545454, 0.09090909090909091, 1.5, 0.0, 4.5643939393939394, 0.0, 1.0, 1.0909090909090908, 0.0, 0.125, 0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0]\n./data/filename-Bruschetta+de+Flageolets .txt\n2.0833333333333335 1.0 1.0454545454545454 0.09090909090909091 2.5 0.0 4.6893939393939394 0.0 0.0 1.0909090909090908 0.0 1.5 0.0 0.0 0.0 0.0 0.0 0.0 2.0 0.0 0.0 0.0 0.0\n[2.0833333333333335, 1.0, 1.0454545454545454, 0.09090909090909091, 2.5, 0.0, 4.6893939393939394, 0.0, 0.0, 1.0909090909090908, 0.0, 1.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 2.0, 0.0, 0.0, 0.0, 0.0]\n./data/filename-Spag+bol.txt\n0.0 0.0 0.0 0.0 0.5 0.0 1.0 0.0 0.0 5.0 0.0 0.0 0.0 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.0 0.0 0.0\n[0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 1.0, 0.0, 0.0, 5.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]\n./data/filename-Chicken+parmigiana.txt\n5.583333333333333 0.0 0.5454545454545454 0.09090909090909091 2.0 0.0 14.18939393939394 0.0 0.0 2.090909090909091 0.0 1.0 0.0 0.6666666666666666 0.0 0.0 0.0 0.5 0.0 0.0 3.333333333333333 0.0 0.0\n[5.583333333333333, 0.0, 0.5454545454545454, 0.09090909090909091, 2.0, 0.0, 14.18939393939394, 0.0, 0.0, 2.090909090909091, 0.0, 1.0, 0.0, 0.6666666666666666, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 3.333333333333333, 0.0, 0.0]\n./data/filename-chicken+Salad.txt\n5.2666666666666675 0.2 3.2333333333333334 0.0 4.533333333333333 0.0 12.433333333333334 0.0 1.0 8.0 0.0 0.0 0.0 3.333333333333333 0.0 0.0 0.0 0.0 4.0 0.0 0.0 0.0 0.0\n[5.2666666666666675, 0.2, 3.2333333333333334, 0.0, 4.533333333333333, 0.0, 12.433333333333334, 0.0, 1.0, 8.0, 0.0, 0.0, 0.0, 3.333333333333333, 0.0, 0.0, 0.0, 0.0, 4.0, 0.0, 0.0, 0.0, 0.0]\n./data/filename-CARBONARA.txt\n0.5833333333333334 0.0 0.0 0.0 2.0 0.0 1.9166666666666665 0.0 0.0 0.0 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0\n[0.5833333333333334, 0.0, 0.0, 0.0, 2.0, 0.0, 1.9166666666666665, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]\n./data/filename-Lamingtons.txt\n0.0 0.0 1.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.0 0.0\n[0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.0]\n./data/filename-Hyd+Dum+Ka+Chicken+Biryani.txt\n16.5 0.3666666666666667 2.1666666666666665 1.6666666666666665 1.5666666666666667 0.2 3.9 0.3333333333333333 0.0 2.0 0.0 0.8333333333333334 0.0 0.16666666666666666 0.0 0.0 0.39999999999999997 1.5 4.0 0.2 2.2 0.0 0.0\n[16.5, 0.3666666666666667, 2.1666666666666665, 1.6666666666666665, 1.5666666666666667, 0.2, 3.9, 0.3333333333333333, 0.0, 2.0, 0.0, 0.8333333333333334, 0.0, 0.16666666666666666, 0.0, 0.0, 0.39999999999999997, 1.5, 4.0, 0.2, 2.2, 0.0, 0.0]\n./data/filename-Torta+de+Acelga.txt\n2.5 0.0 0.0 0.0 1.0 0.0 3.5 2.0 1.5 0.0 0.0 1.5 0.0 0.3333333333333333 0.0 0.0 0.0 0.0 0.0 0.0 0.6666666666666666 1.0 0.0\n[2.5, 0.0, 0.0, 0.0, 1.0, 0.0, 3.5, 2.0, 1.5, 0.0, 0.0, 1.5, 0.0, 0.3333333333333333, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.6666666666666666, 1.0, 0.0]\n./data/filename-French+Toast.txt\n4.916666666666667 0.0 1.6666666666666665 0.0 4.066666666666666 0.0 2.138888888888889 0.1111111111111111 0.0 1.0 1.0 3.0 0.0 0.5 0.0 0.0 0.06666666666666667 0.3333333333333333 3.0 0.0 1.2 0.0 0.0\n[4.916666666666667, 0.0, 1.6666666666666665, 0.0, 4.066666666666666, 0.0, 2.138888888888889, 0.1111111111111111, 0.0, 1.0, 1.0, 3.0, 0.0, 0.5, 0.0, 0.0, 0.06666666666666667, 0.3333333333333333, 3.0, 0.0, 1.2, 0.0, 0.0]\n./data/filename-AFFOGATO.txt\n1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 2.0 0.0 0.0 0.0 0.0\n[1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 2.0, 0.0, 0.0, 0.0, 0.0]\n./data/filename-General+Zuo+chicken.txt\n5.25 0.5 0.7083333333333333 0.125 10.666666666666666 0.25 6.375 0.0 2.5 10.0 0.0 3.625 0.0 6.0 0.0 1.0 6.5 0.5 11.0 0.0 1.0 0.0 0.0\n[5.25, 0.5, 0.7083333333333333, 0.125, 10.666666666666666, 0.25, 6.375, 0.0, 2.5, 10.0, 0.0, 3.625, 0.0, 6.0, 0.0, 1.0, 6.5, 0.5, 11.0, 0.0, 1.0, 0.0, 0.0]\n./data/filename-Chicken+curry.txt\n13.25 0.25 0.25 1.0 1.1428571428571428 0.0 6.125 0.3333333333333333 0.0 2.0 0.0 0.125 0.8571428571428571 1.0 0.0 0.0 0.3333333333333333 2.0 1.0 0.0 0.3333333333333333 0.0 0.0\n[13.25, 0.25, 0.25, 1.0, 1.1428571428571428, 0.0, 6.125, 0.3333333333333333, 0.0, 2.0, 0.0, 0.125, 0.8571428571428571, 1.0, 0.0, 0.0, 0.3333333333333333, 2.0, 1.0, 0.0, 0.3333333333333333, 0.0, 0.0]\n./data/filename-mango+lessi.txt\n0.6666666666666666 0.0 2.6666666666666665 0.0 0.0 0.0 0.4444444444444444 0.2222222222222222 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0\n[0.6666666666666666, 0.0, 2.6666666666666665, 0.0, 0.0, 0.0, 0.4444444444444444, 0.2222222222222222, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]\n./data/filename-Pap+en+vleis.txt\n4.0 0.0 0.5454545454545454 0.09090909090909091 2.0 0.0 0.2727272727272727 0.0 0.0 8.09090909090909 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0\n[4.0, 0.0, 0.5454545454545454, 0.09090909090909091, 2.0, 0.0, 0.2727272727272727, 0.0, 0.0, 8.09090909090909, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]\n./data/filename-Bruschetta.txt\n2.0833333333333335 1.0 1.0454545454545454 0.09090909090909091 2.5 0.0 4.6893939393939394 0.0 0.0 1.0909090909090908 0.0 1.5 0.0 0.0 0.0 0.0 0.0 0.0 2.0 0.0 0.0 0.0 0.0\n[2.0833333333333335, 1.0, 1.0454545454545454, 0.09090909090909091, 2.5, 0.0, 4.6893939393939394, 0.0, 0.0, 1.0909090909090908, 0.0, 1.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 2.0, 0.0, 0.0, 0.0, 0.0]\n./data/filename-masala.txt\n10.916666666666668 0.16666666666666666 0.16666666666666666 0.0 1.5 0.0 1.25 0.3333333333333333 1.5 0.0 0.0 0.0 0.0 2.5 0.0 0.0 0.3333333333333333 3.0 0.0 0.0 0.3333333333333333 0.0 0.0\n[10.916666666666668, 0.16666666666666666, 0.16666666666666666, 0.0, 1.5, 0.0, 1.25, 0.3333333333333333, 1.5, 0.0, 0.0, 0.0, 0.0, 2.5, 0.0, 0.0, 0.3333333333333333, 3.0, 0.0, 0.0, 0.3333333333333333, 0.0, 0.0]\n./data/filename-noodle+soup.txt\n1.0 0.0 0.3333333333333333 0.0 1.1666666666666665 0.0 0.5 0.0 0.5 1.0 0.0 0.0 0.0 0.0 0.0 0.0 1.5 0.0 0.0 1.0 0.0 1.0 0.0\n[1.0, 0.0, 0.3333333333333333, 0.0, 1.1666666666666665, 0.0, 0.5, 0.0, 0.5, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.5, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0]\n./data/filename-Kong+pao+chicken.txt\n5.25 1.9500000000000002 1.0833333333333333 0.0 5.166666666666667 0.4 7.375000000000001 0.0 2.0 2.0 0.0 2.041666666666667 0.0 6.333333333333333 0.0 0.0 2.5 0.5 1.0 0.4 1.0 0.0 0.0\n[5.25, 1.9500000000000002, 1.0833333333333333, 0.0, 5.166666666666667, 0.4, 7.375000000000001, 0.0, 2.0, 2.0, 0.0, 2.041666666666667, 0.0, 6.333333333333333, 0.0, 0.0, 2.5, 0.5, 1.0, 0.4, 1.0, 0.0, 0.0]\n" ], [ "\nfrom sklearn import decomposition\n## pca do singular value decomposition, accomplishes dimentionality reduction by only keeping the top 3 e-val.\n## it also makes the data less sparse by whitening.\npca = decomposition.PCA(n_components=3, whiten = True)\npca.fit(X) # different convention: row vs col !!!\nprint (pca.components_.T, pca.explained_variance_)\nE= pca.components_\nK = E.T.dot(E).dot(X.T)\n", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "//Create a SparkContext to initialize Spark\n// val conf = new SparkConf()\n// conf.setMaster(\"local\")\n// conf.setAppName(\"Word Count\")\n// val sc = new SparkContext(conf)\n\nval textFile = sc.textFile(\"data/shakespeare.txt\")\n\n//word count\nval counts = textFile.flatMap(line => line.split(\" \"))\n .map(word => (word, 1))\n .reduceByKey(_ + _)\n\ncounts.foreach(println)\n// System.out.println(\"Total words: \" + counts.count());\ncounts.saveAsTextFile(\"output/shakespeareWordCount\")", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Place Stock Trades into Senator Dataframe", "_____no_output_____" ], [ "## 1. Understand the Senator Trading Report (STR) Dataframe", "_____no_output_____" ] ], [ [ "import pandas as pd\n#https://docs.google.com/spreadsheets/d/1lH_LpTgRlfzKvpRnWYgoxlkWvJj0v1r3zN3CeWMAgqI/edit?usp=sharing\ntry:\n sen_df = pd.read_csv(\"Senator Stock Trades/Senate Stock Watcher 04_16_2020 All Transactions.csv\")\nexcept:\n sen_df = pd.read_csv(\"https://github.com/pkm29/big_data_final_project/raw/master/Senate%20Stock%20Trades/Senate%20Stock%20Watcher%2004_16_2020%20All%20Transactions.csv\")\nsen_df.head()", "_____no_output_____" ], [ "sen_df.type.unique()", "_____no_output_____" ] ], [ [ "There are 4 types of trades.\nExchanges: Exchange 1 stock for another\nSale (Full): Selling all of their stock\nPurchase: Buying a stock\nSale (Partial): Selling some of that particular stock", "_____no_output_____" ] ], [ [ "n_exchanges = len(sen_df.loc[sen_df['type'] == \"Exchange\"])\nn_trades = len(sen_df)\nprint(\"There are \" +str(n_exchanges) +\" exchange trades out of a total of \" +str(n_trades)+ \" trades.\")\nsen_df = sen_df.loc[sen_df['type'] != \"Exchange\"]\n", "There are 84 exchange trades out of a total of 8600 trades.\n" ] ], [ [ "At this point in time, I will exclude exchange trades because they are so few and wish to build the basic structure of the project. As you can see, this would require splitting up the exchange into two rows with each company and so on. I may include this step later if time permits. ", "_____no_output_____" ], [ "There should now be 8516 trades remaining in the dataframe. Let's make sure this is so.", "_____no_output_____" ] ], [ [ "n_trades = len(sen_df)\nprint(\"There are \" +str(n_trades)+ \" trades in the dataframe\")", "There are 8516 trades in the dataframe\n" ], [ "n_blank_ticker = len(sen_df.loc[sen_df['ticker'] == \"--\"])\nprint(\"There are \" +str(n_blank_ticker) +\" trades w/o a ticker out of a total of \" +str(n_trades)+ \" trades\")\nsen_df = sen_df.loc[sen_df['ticker'] != \"--\"]", "There are 1872 trades w/o a ticker out of a total of 8516 trades\n" ] ], [ [ "For the same reasons we excluded exchange trades, we will also exclude trades without a ticker (which all public stocks have - the ticker is their identifier on the stock exchange). Eliminating trades without a ticker takes out trades of other types of securities (corporate bonds, municipal securities, non-public stock).", "_____no_output_____" ], [ "There should now be 6644 trades remaining in the dataframe. Let's make sure this is so.", "_____no_output_____" ] ], [ [ "n_trades = len(sen_df)\nprint(\"There are \" +str(n_trades)+ \" trades in the dataframe\")", "There are 6644 trades in the dataframe\n" ] ], [ [ "## 2. Add Data to STR Dataframe ", "_____no_output_____" ], [ "### Import Data", "_____no_output_____" ], [ "In this step we will be using company information such as market cap and industry from online lists provided by the NYSE, NASDAQ, and ASXL exchange. Links can be found here:https://stackoverflow.com/questions/25338608/download-all-stock-symbol-list-of-a-market", "_____no_output_____" ] ], [ [ "ticker_list = list()\ntry:\n NYSE_df = pd.read_csv(\"NYSEcompanylist.csv\")\nexcept:\n NYSE_df = pd.read_csv(\"https://github.com/pkm29/big_data_final_project/raw/master/Stocks/NYSEcompanylist.csv\")\n \ntry:\n NASDAQ_df = pd.read_csv(\"NASDAQcompanylist.csv\")\nexcept:\n NASDAQ_df = pd.read_csv(\"https://github.com/pkm29/big_data_final_project/raw/master/Stocks/NASDAQcompanylist.csv\")\n \nticker_list.append(NYSE_df)\nticker_list.append(NASDAQ_df)\n\nNYSE_df.head()", "_____no_output_____" ], [ "NASDAQ_df.head()", "_____no_output_____" ], [ "\"\"\"\nAdd data for Berkshire Hathaway, Lions Gate Entertainment, and Royal Dutch Shell to the NYSE company list. While\n#these companies are in the company list, their fields are empty. Also, change the tickers of these companies to \n#match Senate Stock Data (since dashes are used instead of periods in that dataset, we make sure the same is true \nin the NYSE company list). What matters is consistent convention here.\n\"\"\"\n\nrow_count = 0\nreplacement_count = 0\n\nfor row_tuple in NYSE_df.itertuples():\n \n if replacement_count == 4:\n break\n \n if row_tuple.Symbol == \"BRK.B\":\n #row_tuple.Symbol = \"BRK-B\"\n NYSE_df.at[row_count, 'Symbol'] = \"BRK-B\"\n #Shares outstanding reported in Q1 2020 financial reports, stock price from May 6, when this data is dated\n #row_tuple.MarketCap = \"$420.02B\"\n NYSE_df.at[row_count, 'MarketCap'] = \"$420.02B\"\n #row_tuple.Sector = \"Miscellaneous\"\n NYSE_df.at[row_count, 'Sector'] = \"Miscellaneous\"\n #row_tuple.industry = \"Conglomerate\"\n NYSE_df.at[row_count, 'industry'] = \"Conglomerate\"\n replacement_count = replacement_count + 1\n \n if row_tuple.Symbol == \"LGF.B\":\n #row_tuple.Symbol = \"LGF-B\"\n #Shares outstanding reported in Q1 2020 financial reports, stock price from May 6, when this data is dated\n #row_tuple.MarketCap = \"$14.62B\"\n #row_tuple.Sector = \"Consumer Services\"\n #row_tuple.industry = \"Movies/Entertainment\"\n \n NYSE_df.at[row_count, 'Symbol'] = \"LGF-B\"\n NYSE_df.at[row_count, 'MarketCap'] = \"$14.62B\"\n NYSE_df.at[row_count, 'Sector'] = \"Consumer Services\"\n NYSE_df.at[row_count, 'industry'] = \"Movies/Entertainment\" \n replacement_count = replacement_count + 1\n\n if row_tuple.Symbol == \"RDS.A\":\n #row_tuple.Symbol = \"RDS-A\"\n #Shares outstanding reported in Q1 2020 financial reports, stock price from May 6, when this data is dated\n #row_tuple.MarketCap = \"$122.28B\"\n #row_tuple.Sector = \"Energy\"\n #row_tuple.industry = \"Oil & Gas Production\"\n \n NYSE_df.at[row_count, 'Symbol'] = \"RDS-A\"\n NYSE_df.at[row_count, 'MarketCap'] = \"$122.28B\"\n NYSE_df.at[row_count, 'Sector'] = \"Energy\"\n NYSE_df.at[row_count, 'industry'] = \"Oil & Gas Production\" \n replacement_count = replacement_count + 1\n\n if row_tuple.Symbol == \"RDS.B\":\n #row_tuple.Symbol = \"RDS-B\"\n #Shares outstanding reported in Q1 2020 financial reports, stock price from May 6, when this data is dated\n #row_tuple.MarketCap = \"$122.09B\"\n #row_tuple.Sector = \"Energy\"\n #row_tuple.industry = \"Oil & Gas Production\"\n \n NYSE_df.at[row_count, 'Symbol'] = \"RDS-B\"\n NYSE_df.at[row_count, 'MarketCap'] = \"$122.09B\"\n NYSE_df.at[row_count, 'Sector'] = \"Energy\"\n NYSE_df.at[row_count, 'industry'] = \"Oil & Gas Production\"\n replacement_count = replacement_count + 1\n \n row_count = row_count + 1\n \n#Confirm changes have been made successfully\nfor row_tuple in NYSE_df.itertuples():\n \n if row_tuple.Symbol == \"BRK-B\":\n print (row_tuple)\n \n if row_tuple.Symbol == \"LGF-B\":\n print (row_tuple)\n \n if row_tuple.Symbol == \"RDS-A\":\n print (row_tuple)\n \n if row_tuple.Symbol == \"RDS-B\":\n print (row_tuple)\n", "Pandas(Index=365, Symbol='BRK-B', Name='Berkshire Hathaway Inc.', LastSale=nan, MarketCap='$420.02B', IPOyear=nan, Sector='Miscellaneous', industry='Conglomerate', _8='https://old.nasdaq.com/symbol/brk.b', _9=nan)\nPandas(Index=1700, Symbol='LGF-B', Name='Lions Gate Entertainment Corporation', LastSale=nan, MarketCap='$14.62B', IPOyear=2016.0, Sector='Consumer Services', industry='Movies/Entertainment', _8='https://old.nasdaq.com/symbol/lgf.b', _9=nan)\nPandas(Index=2426, Symbol='RDS-A', Name='Royal Dutch Shell PLC', LastSale=nan, MarketCap='$122.28B', IPOyear=nan, Sector='Energy', industry='Oil & Gas Production', _8='https://old.nasdaq.com/symbol/rds.a', _9=nan)\nPandas(Index=2427, Symbol='RDS-B', Name='Royal Dutch Shell PLC', LastSale=nan, MarketCap='$122.09B', IPOyear=nan, Sector='Energy', industry='Oil & Gas Production', _8='https://old.nasdaq.com/symbol/rds.b', _9=nan)\n" ], [ "#There are also 2 instances where a wrong ticker for Berkshire Hathaway is found in the Senate Stock data\n#(BRKB is used as opposed to BRK-B). Thus, we correct for those instances here.\n\n#Find indices of these two trades\nfor row_tuple in sen_df.itertuples():\n if row_tuple.ticker == \"BRKB\":\n print (row_tuple)\n \n#We can see that the indices are 1207 and 4611, so we will manually modify the ticker field of these trades.\n\nsen_df.at[1207, 'ticker'] = \"BRK-B\"\nsen_df.at[4611, 'ticker'] = \"BRK-B\"\n\nlen(sen_df)", "Pandas(Index=1207, transaction_date='12/06/2018', owner='Self', ticker='BRKB', asset_description='Berkshire Hathaway Inc.', asset_type='Stock', type='Purchase', amount='$15,001 - $50,000', comment='--', senator='Jerry Moran, ', ptr_link='https://efdsearch.senate.gov/search/view/ptr/dafb67b1-d578-402d-9e26-c3c2a575b42c/')\nPandas(Index=4611, transaction_date='03/02/2017', owner='Spouse', ticker='BRKB', asset_description='Berkshire Hathaway Inc.', asset_type='Stock', type='Purchase', amount='$15,001 - $50,000', comment='--', senator='Susan M Collins', ptr_link='https://efdsearch.senate.gov/search/view/ptr/ea48ecad-1959-4f7a-ace0-778a71d06437/')\n" ], [ "#Get sector data for each stock trade\n\nsector_data = list()\nfor row_tuple in sen_df.itertuples():\n tic = row_tuple.ticker\n count = 0\n for row_tuple_tic in NYSE_df.itertuples():\n sym = row_tuple_tic.Symbol\n if tic == sym:\n count = count+1\n if row_tuple_tic.Sector == \"n/a\":\n sector_data.append(\"none\")\n else:\n sector_data.append(row_tuple_tic.Sector)\n break\n if count == 0:\n for row_tuple_tic in NASDAQ_df.itertuples():\n sym = row_tuple_tic.Symbol\n if tic == sym:\n count = count+1\n if row_tuple_tic.Sector == \"n/a\":\n sector_data.append(\"none\")\n else:\n sector_data.append(row_tuple_tic.Sector)\n break\n if count == 0:\n sector_data.append(\"none\")\n \nprint(sector_data[0:9])", "['Capital Goods', 'Capital Goods', 'none', 'none', 'Basic Industries', 'Consumer Services', 'Finance', 'Technology', 'Health Care']\n" ], [ "#make sure length matches number of rows in df\nprint(len(sector_data))\n\n#counter for how many times the stock traded by senator not found in exchange data set\nno_ticker_cnt = 0\n\nfor i in sector_data:\n if i == \"none\":\n no_ticker_cnt = no_ticker_cnt + 1\n \nprint(no_ticker_cnt)", "6644\n1141\n" ], [ "#Get industry data for each stock trade\n\nindustry_data = list()\nfor row_tuple in sen_df.itertuples():\n tic = row_tuple.ticker\n count = 0\n for row_tuple_tic in NYSE_df.itertuples():\n sym = row_tuple_tic.Symbol\n if tic == sym:\n count = count+1\n if row_tuple_tic.industry == \"n/a\":\n industry_data.append(\"none\")\n else:\n industry_data.append(row_tuple_tic.industry)\n break\n if count == 0:\n for row_tuple_tic in NASDAQ_df.itertuples():\n sym = row_tuple_tic.Symbol\n if tic == sym:\n count = count+1\n if row_tuple_tic.industry == \"n/a\":\n industry_data.append(\"none\")\n else:\n industry_data.append(row_tuple_tic.industry)\n break\n if count == 0:\n industry_data.append(\"none\")\n \nprint(industry_data[0:9])", "['Biotechnology: Laboratory Analytical Instruments', 'Industrial Machinery/Components', 'none', 'none', 'Paints/Coatings', 'Building operators', 'Major Banks', 'Semiconductors', 'Major Pharmaceuticals']\n" ], [ "#make sure length matches number of rows in df\nprint(len(industry_data))\n\n#counter for how many times the stock traded by senator not found in exchange data set\nno_ticker_cnt = 0\n\nfor i in industry_data:\n if i == \"none\":\n no_ticker_cnt = no_ticker_cnt + 1\n \nprint(no_ticker_cnt)", "6644\n1141\n" ], [ "#Get market cap data for each stock trade\n\nmktcap_data = list()\nfor row_tuple in sen_df.itertuples():\n tic = row_tuple.ticker\n count = 0\n for row_tuple_tic in NYSE_df.itertuples():\n sym = row_tuple_tic.Symbol\n if tic == sym:\n count = count+1\n if row_tuple_tic.MarketCap == \"n/a\":\n mktcap_data.append(\"none\")\n else:\n mktcap_data.append(row_tuple_tic.MarketCap)\n break\n if count == 0:\n for row_tuple_tic in NASDAQ_df.itertuples():\n sym = row_tuple_tic.Symbol\n if tic == sym:\n count = count+1\n if row_tuple_tic.MarketCap == \"n/a\":\n mktcap_data.append(\"none\")\n else:\n mktcap_data.append(row_tuple_tic.MarketCap)\n break\n if count == 0:\n mktcap_data.append(\"none\")\n \nprint(mktcap_data[0:9])", "['$47.15B', '$8.58B', 'none', 'none', '$8.74B', '$34.86B', '$92.93B', '$52.6B', '$190.9B']\n" ], [ "#make sure length matches number of rows in df\nprint(len(mktcap_data))\n\n#counter for how many times the stock traded by senator not found in exchange data set\nno_ticker_cnt = 0\n\nfor i in mktcap_data:\n if i == \"none\":\n no_ticker_cnt = no_ticker_cnt + 1\n \nprint(no_ticker_cnt)", "6644\n1141\n" ], [ "#add new columns to df\n\nsen_df['mkt_cap'] = mktcap_data\nsen_df['sector'] = sector_data\nsen_df['industry'] = industry_data\n\nsen_df = sen_df.fillna(\"none\")\nsen_df.head()", "_____no_output_____" ], [ "\"\"\"\nPrint out names of companies with missing data to find out why we have so many misses (~17% of our data).\nThere seem to be 3 reasons for this:\n1. Companies merging with another or being acquired (or even acquiring and taking the acquired company's name - very rare)\n2. Foreign companies (listed abroad)\n3. American companies listed abroad - this applies to a very small number of trades\n\"\"\"\n\nfrom collections import Counter\n\ncompany_missing_data = list()\nfor row_tuple in sen_df.itertuples():\n if row_tuple.mkt_cap == \"none\":\n company_missing_data.append(row_tuple.asset_description)\n\nprint(Counter(company_missing_data))\n", "Counter({'First Data Corporation': 76, 'Revere Bank': 36, 'Halyard Health, Inc.': 36, 'CBS Corporation (NYSE)': 35, 'Whole Foods Market, Inc.': 29, 'Williams Partners L.P. (NYSE)': 27, 'Versum Materials, Inc.': 25, 'Bollore': 24, 'SPDR S&amp;P 500 ETF': 22, 'Celgene Corporation': 20, 'Exelis Inc. Common Stock Ex-Dis (NYSE)': 18, 'USG Corporation': 16, 'CBS Corporation': 15, 'Weight Watchers International, Inc.': 14, 'Cablevision Systems Corporation (NYSE)': 13, 'Michael Kors Holdings Limited (NYSE)': 13, 'VCA Inc.': 12, 'ACE Limited (NYSE)': 12, 'Nestl\\\\u00e9 S.A.': 11, 'DowDuPont Inc.': 11, 'Celgene Corporation (NASDAQ)': 11, 'KLX Inc. (NASDAQ)': 11, 'Axiall Corporation': 11, 'United Technologies Corporation': 10, 'Energy Transfer Partners, L.P.': 9, 'Targa Resources Partners LP (NYSE)': 9, 'Energy Transfer Equity, L.P. (NYSE)': 9, 'Qlik Technologies, Inc. (NASDAQ)': 9, 'Deutsche Global Infrastructure A (NASDAQ)': 9, 'Cohen &amp; Steers Dividend Value A (NASDAQ)': 9, 'Raytheon Company': 8, 'Buckeye Partners, L.P.': 8, 'Avon Products, Inc.': 8, 'Cheniere Energy, Inc.': 7, 'Express Scripts Holding Company': 7, \"Cabela's Incorporated\": 7, 'B/E Aerospace Inc. (NASDAQ)': 7, 'Express Scripts Holding Company (NASDAQ)': 7, 'ViacomCBS Inc.': 6, 'SPDR S&amp;P Biotech ETF': 6, 'Roche Holding AG': 6, 'SPDR S&amp;P Semiconductor ETF': 6, 'Comcast Corporation (NASDAQ)': 6, 'VCA Inc. (NASDAQ)': 6, 'Time Warner Inc. (NYSE)': 6, 'BlackRock Mid-Cap Growth Equity Inv A (NASDAQ)': 6, 'Harris Corporation': 5, 'Fanuc Corporation': 5, 'Valero Energy Partners LP': 5, 'iShares Transportation Average': 5, 'CR Bard Inc. (NYSE)': 5, 'Red Hat, Inc. (NYSE)': 5, 'Red Hat, Inc.': 4, 'Liberty Media Corp. Reg.Sh.C FO': 4, 'Praxair, Inc.': 4, 'Cogentix Medical, Inc.': 4, 'Ensco plc (NYSE)': 4, 'Utilities Select Sector SPDR ETF (NYSEArca)': 4, 'Praxair Inc.': 4, 'Quintiles IMS Holdings, Inc.': 4, 'Axiall Corporation (NYSE)': 4, 'Valero Energy Partners LP (NYSE)': 4, 'Raytheon Company (NYSE)': 4, 'Alerian MLP ETF (NYSEArca)': 4, 'United Technologies Corp. (NYSE)': 4, 'Valeant Pharmaceuticals International, Inc. (NYSE)': 4, 'Ivy Asset Strategy C (NASDAQ)': 4, 'Lehigh Gas Partners LP (NYSE)': 4, 'LinkedIn Corporation (NYSE)': 4, 'Siemens Aktiengesellschaft': 3, 'Andeavor Logistics LP': 3, 'PHILLIPS EDISON GR': 3, 'SPDR Gold Shares': 3, 'Time Warner Inc.': 3, 'Energy Transfer Equity, L.P.': 3, 'Vanguard FTSE Emerging Markets ETF': 3, 'Reynolds American Inc.': 3, 'PureFunds ISE Cyber Security ETF': 3, 'C. R. Bard, Inc.': 3, 'Torchmark Corporation': 3, 'Linear Technology Corporation': 3, 'DCP Midstream Partners LP': 3, 'iShares MSCI Emerging Markets': 3, 'Williams Partners L.P.': 3, 'Avon Products Inc. (NYSE)': 3, 'Buckeye Partners, L.P. (NYSE)': 3, 'EMC Corporation': 3, 'SPY Feb 2016 put 180.000': 3, 'iShares Core S&amp;P Mid-Cap (NYSEArca)': 3, 'Guggenheim S&amp;P 500 Equal Weight ETF (NYSEArca)': 3, 'iShares International Select Dividend (NYSEArca)': 3, 'iShares Global Telecom (NYSEArca)': 3, 'SPDR S&amp;P Dividend ETF (NYSEArca)': 3, 'SPDR S&amp;P International Dividend ETF (NYSEArca)': 3, 'iShares Latin America 40 (NYSEArca)': 3, 'iShares MSCI Pacific ex Japan (NYSEArca)': 3, 'BioMed Realty Trust Inc. (NYSE)': 3, 'EMC Corporation (NYSE)': 3, 'Nestl (OTC Markets)': 3, 'Legg Mason Batterymarch Intl Eq C (NASDAQ)': 3, 'Triangle Capital Corporation (NYSE)': 3, 'Rockwood Holdings, Inc. (NYSE)': 3, 'Praxair Inc. (NYSE)': 3, 'PartnerRe Ltd. (NYSE)': 3, 'Sprott Physical Gold and Silver Trust': 2, 'Tencent Holdings Limited': 2, 'BB&amp;T Corporation <div class=\"text-muted\"><em>Rate/Coupon:</em> 5.25%<br> <em>Matures:</em> 11/01/2019</div>': 2, 'BB&amp;T Corporation <div class=\"text-muted\"><em>Rate/Coupon:</em> 5.25<br> <em>Matures:</em> 11/01/2019</div>': 2, 'VISA - CLASSE A': 2, 'SPDR Bloomberg Barclays High Yield Bond ETF': 2, 'FACEBOOK INC FACEBOOK CLASS A': 2, 'iShares Russell 1000 Growth ETF': 2, 'Vanguard FTSE Europe ETF': 2, 'iShares US Energy ETF': 2, 'SoftBank Group Corp.': 2, 'Reckitt Benckiser Group plc': 2, 'Pernod Ricard SA': 2, \"L'Or\\\\u00e9al S.A.\": 2, 'Murata Manufacturing Co., Ltd.': 2, 'Naspers Limited': 2, 'Keyence Corporation': 2, 'Carlsberg A/S': 2, 'Asahi Kasei Corporation': 2, 'Invesco S&amp;P SmallCap Low Volatility ETF': 2, 'Zayo Group Holdings, Inc.': 2, 'Mainstay Medical International plc': 2, 'ProShares UltraShort 20+ Year Treasury': 2, 'SPDR S&amp;P 500 ETF <div class=\"text-muted\">Option Type: Put <br><em>Strike price:</em> $210.00 <br> <em>Expires:</em> 06/30/2017 </div>': 2, 'RSP Permian, Inc.': 2, 'BNP Paribas SA': 2, 'Singapore Telecommunications Limited': 2, 'Shiseido Company, Limited': 2, 'Regal Entertainment Group': 2, 'SPDR S&amp;amp;P 500 ETF': 2, 'Linn Energy, LLC (NASDAQ)': 2, 'iShares MSCI Emerging Markets (NYSEArca)': 2, 'Health Care Select Sector SPDR ETF (NYSEArca)': 2, 'SPDR S&amp;P 500 ETF (NYSEArca)': 2, 'PowerShares Emerging Markets Sov Dbt ETF (NYSEArca)': 2, 'AMG Yacktman Service': 2, 'Dell Technologies Inc.': 2, 'XEROX CORP <div class=\"text-muted\"><em>Rate/Coupon:</em> 4.5%<br> <em>Matures:</em> 05/15/2021</div>': 2, 'SunTrust Banks, Inc.': 2, 'Aquesta Financial Holdings, Inc.': 2, 'SPDR EURO STOXX 50 ETF': 2, 'SPDR S&amp;P 500 ETF Trust (NYSEArca)': 2, 'Industrial Select Sector SPDR ETF': 2, 'Newfield Exploration Co.': 2, 'Permanent Portfolio (NASDAQ)': 2, 'ProShares Short 20+ Year Treasury': 2, 'United Technologies Corporation (NYSE)': 2, 'JP MORGAN CHASE (OTC Markets)': 2, 'Vanguard Consumer Staples ETF': 2, 'Linear Technology Corporation (NASDAQ)': 2, 'Harris Corporation (NYSE)': 2, 'Hubbell Inc. (NYSE)': 2, 'Pengrowth Energy Corporation (NYSE)': 2, 'Protective Life Corporation (NYSE)': 2, 'Oppenheimer SteelPath MLP Income C (NASDAQ)': 2, 'Crosstex Energy, Inc. (NASDAQ)': 2, 'Columbia Marsico Growth Z (NASDAQ)': 2, 'Atmel Corporation (NASDAQ)': 2, 'ProShares Short 20+ Year Treasury (NYSEArca)': 2, 'PowerShares DB Precious Metals': 2, 'ProShares VIX Short-Term Futures ETF': 2, 'ALPS Alerian MLP ETF': 2, 'ProShares VIX Short-Term Futures (NYSEArca)': 2, 'Halyard Health, Inc. (NYSE)': 2, 'Northern Oil and Gas, Inc. (AMEX)': 2, 'BlackRock Equity Dividend Inv A (NASDAQ)': 2, 'Roche Holding AG (OTC Markets)': 2, 'Zimmer Holdings, Inc. (NYSE)': 2, 'WCM Focused International Growth Fund Institutiona': 1, 'SPDR S&amp;P 500 ETF Trust': 1, 'Vanguard FTSE Emerging Markets Index Fund ETF Shar': 1, 'Vanguard Intermediate-Term Bond Index Fund ETF Sha': 1, 'Vanguard GNMA Fund Admiral Shares': 1, 'Templeton Global Bond Fund Advisor Class': 1, 'iShares Edge MSCI USA Momentum Factor ETF': 1, 'BlackRock Global Allocation Fund, Inc. Institution': 1, 'Glenmede Small Cap Equity Portfolio Class Advisor': 1, 'AMG TimesSquare Small Cap Growth Fund Class Z': 1, 'Vanguard PRIMECAP Fund Admiral Shares': 1, 'Royal Dutch Shell plc': 1, 'Anadarko Petroleum Corporation': 1, 'Safran SA': 1, 'EssilorLuxottica Soci\\\\u00e9t\\\\u00e9 anonyme': 1, 'Airbus SE': 1, 'AIA Group Limited': 1, 'T. Rowe Price New Horizons': 1, 'JPMorgan Mid Cap Value L': 1, 'Fidelity Contrafund': 1, 'The Toronto-Dominion Bank <div class=\"text-muted\"><em>Rate/Coupon:</em> 3.25<br> <em>Matures:</em> 06/11/2021</div>': 1, 'American Funds Growth Fund of Amer F1': 1, 'BlackRock Global Allocation Instl': 1, 'Shin-Etsu Chemical Co., Ltd.': 1, 'International Business Machines Corporation': 1, 'Guggenheim S&amp;P 500 Eq Wt Technology ETF': 1, 'iShares Transportation Average ETF': 1, 'iShares US Pharmaceuticals ETF': 1, 'US Global Jets ETF': 1, 'Entellus Medical, Inc.': 1, 'ETFMG Prime Cyber Security ETF': 1, 'The Priceline Group Inc.': 1, 'South32 Limited': 1, 'LVMH Mo\\\\u00ebt Hennessy Louis Vuitton S.E.': 1, 'Compass Group PLC': 1, 'BASF SE': 1, 'AXA SA': 1, 'Vtech Holdings Limited': 1, 'SMC Corporation': 1, 'SCANA Corporation': 1, 'Pacer Funds Trust - Pacer Trendpilot 450 ETF': 1, 'Dr Pepper Snapple Group, Inc.': 1, 'Zynga Inc.': 1, 'Newfield Exploration Company': 1, 'Potash Corporation of Saskatchewan Inc.': 1, 'Bioverativ Inc.': 1, 'Victory Sycamore Established Value I': 1, 'Agrium Inc.': 1, 'Tesoro Logistics LP': 1, 'FPL Group, Inc. 5 7/8% Preferre': 1, 'ONEOK Partners, L.P.': 1, 'iShares US Broker-Dealers &amp; Secs Exchs': 1, 'DB 3x German Bund Futures ETN': 1, 'NSRGY- Nestle (OTC Markets)': 1, 'American Funds Tax-Exempt Bond A': 1, 'iShares Russell 1000 Value': 1, 'Under Armour, Inc. Class C Comm': 1, 'Shire plc': 1, 'Aqua America Inc.': 1, 'Columbia Pipeline Group, Inc.': 1, 'iShares Edge MSCI USA Momentum Factor': 1, 'SYSCO': 1, 'Starwood Hotels &amp; Resorts Worldwide Inc.': 1, 'ADT CORPORATION': 1, 'ITC Holdings Corp.': 1, 'RSP Permian, Inc. (NYSE)': 1, 'AGL Resources Inc. (NYSE)': 1, 'iShares US Pharmaceuticals (NYSEArca)': 1, 'STARBUCKS (Swiss)': 1, 'Noble Energy, Inc. (NYSE)': 1, 'JP MORGAN CHASE (OTC Markets) <div class=\"text-muted\"><em>Rate/Coupon:</em> 6.75<br> <em>Matures:</em> 02/01/24</div>': 1, 'Morgan Stanley Inst Active Intl Allc I (NASDAQ)': 1, 'Invesco Equally-Wtd S&amp;P 500 Y (NASDAQ)': 1, 'Wells Fargo Advantage S/T Muni Bd Adm (NASDAQ)': 1, 'Invesco Municipal Income Y (NASDAQ)': 1, 'Invesco SmallCapValue Y (NASDAQ)': 1, 'Invesco Comstock Y (NASDAQ)': 1, 'Morgan Stanley Inst Emerging Mkts I (NASDAQ)': 1, 'The Priceline Group Inc. (NASDAQ)': 1, 'ITC Holdings Corp. (NYSE)': 1, 'PowerShares DB 3x German Bund F (NYSE)': 1, 'Convergys Corporation (NYSE)': 1, 'SPDR EURO STOXX Small Cap ETF (NYSEArca)': 1, 'SPDR Series Trust - SPDR S&amp;P Retail ETF (NYSEArca)': 1, 'SPDR S&amp;P 600 Small Cap ETF (NYSEArca) <div class=\"text-muted\"><em>Description:</em>&nbsp;ETF</div>': 1, 'Kraft Foods Group, Inc. (NASDAQ)': 1, 'iShares Transportation Average (NYSEArca)': 1, 'WageWorks, Inc. (NYSE)': 1, 'Vitamin Shoppe, Inc. (NYSE)': 1, 'Monotype Imaging Holdings Inc. (NASDAQ)': 1, 'Tumi Holdings, Inc. (NYSE)': 1, 'Tangoe, Inc. (NASDAQ)': 1, 'SciQuest, Inc. (NASDAQ)': 1, 'Stone Energy Corp. (NYSE)': 1, 'Roadrunner Transportation Systems, Inc. (NYSE)': 1, 'Popeyes Louisiana Kitchen, Inc. (NASDAQ)': 1, 'NeuStar, Inc. (NYSE)': 1, 'Mattress Firm Holding Corp. (NASDAQ)': 1, 'LifeLock, Inc. (NYSE)': 1, 'IPC Healthcare, Inc. (NASDAQ)': 1, 'HFF, Inc. (NYSE)': 1, 'Financial Engines, Inc. (NASDAQ)': 1, \"Del Frisco's Restaurant Group, Inc. (NASDAQ)\": 1, 'Cyberonics Inc. (NASDAQ)': 1, 'Control4 Corporation (NASDAQ)': 1, 'Constant Contact, Inc. (NASDAQ)': 1, 'Cantel Medical Corp. (NYSE)': 1, 'CLARCOR Inc. (NYSE)': 1, 'AmSurg Corp. (NASDAQ)': 1, 'AmTrust Financial Services, Inc. (NASDAQ)': 1, 'The Advisory Board Company (NASDAQ)': 1, 'Coach, Inc. (NYSE)': 1, 'Boulder Total Return Fund, Inc. (NYSE)': 1, 'MainStay Marketfield C (NASDAQ)': 1, 'Precision Castparts Corp. (NYSE)': 1, 'Capital One Financial Corp Pfd (NYSE)': 1, 'Isis Pharmaceuticals, Inc. (NASDAQ)': 1, 'Vanguard Emerging Markets Stock Idx ETF': 1, 'SPDR S&amp;P Retail ETF': 1, 'Northern Tier Energy LP (NYSE)': 1, 'Raytheon Co. (NYSE)': 1, 'Yahoo! Inc. (NASDAQ)': 1, 'Vectren Corporation (NYSE)': 1, 'Validus Holdings, Ltd. (NYSE)': 1, 'Starwood Hotels &amp; Resorts Worldwide Inc. (NYSE)': 1, 'Penn West Petroleum Ltd. (NYSE)': 1, 'Landauer Inc. (NYSE)': 1, 'Hatteras Financial Corp (NYSE)': 1, 'FirstMerit Corporation (NASDAQ)': 1, 'CYS Investments, Inc. (NYSE)': 1, 'Baker Hughes Incorporated (NYSE)': 1, 'Windstream Holdings, Inc. (NASDAQ)': 1, 'Symantec Corporation (NASDAQ)': 1})\n" ], [ "#Get a view of how many industries are found in our senate stock data.\n\nindustry_dict = Counter(industry_data)\n\nindustry_list = list()\nfor x in industry_dict:\n industry_list.append(x)\n \nprint(industry_list[0:9])\nn_industries = len(industry_list)\n#since 'none' is included in our list\nn_industries = n_industries - 1\nprint(\"There are \" + str(n_industries) + \" industries covered by the trades of senators.\")", "Counter({'none': 1141, 'Major Banks': 350, 'Major Pharmaceuticals': 317, 'Natural Gas Distribution': 203, 'Industrial Machinery/Components': 187, 'Computer Manufacturing': 180, 'Oil & Gas Production': 166, 'Consumer Electronics/Appliances': 165, 'Telecommunications Equipment': 157, 'Computer Software: Prepackaged Software': 138, 'Television Services': 137, 'Clothing/Shoe/Accessory Stores': 135, 'Semiconductors': 132, 'Integrated oil Companies': 131, 'Major Chemicals': 112, 'Computer Software: Programming, Data Processing': 112, 'Department/Specialty Retail Stores': 106, 'Beverages (Production/Distribution)': 102, 'Services-Misc. Amusement & Recreation': 97, 'Biotechnology: Biological Products (No Diagnostic Substances)': 96, 'Package Goods/Cosmetics': 91, 'Auto Manufacturing': 90, 'Medical/Dental Instruments': 83, 'Computer peripheral equipment': 81, 'Business Services': 78, 'Diversified Commercial Services': 77, 'Real Estate Investment Trusts': 76, 'Hotels/Resorts': 73, 'Consumer Electronics/Video Chains': 66, 'Plastic Products': 65, 'Investment Bankers/Brokers/Service': 63, 'Restaurants': 59, 'Broadcasting': 57, 'Air Freight/Delivery Services': 56, nan: 56, 'Containers/Packaging': 55, 'Apparel': 49, 'Medical/Nursing Services': 49, 'Packaged Foods': 46, 'Oilfield Services/Equipment': 46, 'Catalog/Specialty Distribution': 44, 'Aerospace': 43, 'Agricultural Chemicals': 43, 'EDP Services': 40, 'Paints/Coatings': 39, 'Electric Utilities: Central': 38, 'Biotechnology: Laboratory Analytical Instruments': 35, 'Movies/Entertainment': 35, 'Savings Institutions': 34, 'Radio And Television Broadcasting And Communications Equipment': 34, 'Computer Communications Equipment': 33, 'Property-Casualty Insurers': 33, 'Power Generation': 33, 'Railroads': 33, 'Finance: Consumer Services': 30, 'Food Chains': 26, 'Life Insurance': 25, 'Biotechnology: Electromedical & Electrotherapeutic Apparatus': 24, 'RETAIL: Building Materials': 24, 'Food Distributors': 23, 'Investment Managers': 23, 'Specialty Chemicals': 22, 'Conglomerate': 20, 'Other Consumer Services': 20, 'Industrial Specialties': 20, 'Oil Refining/Marketing': 19, 'Metal Fabrications': 18, 'Environmental Services': 18, 'Trucking Freight/Courier Services': 18, 'Construction/Ag Equipment/Trucks': 17, 'Auto Parts:O.E.M.': 17, 'Automotive Aftermarket': 17, 'Medical Specialities': 16, 'Specialty Insurers': 15, 'Recreational Products/Toys': 14, 'Shoe Manufacturing': 14, 'Marine Transportation': 14, 'Farming/Seeds/Milling': 11, 'Newspapers/Magazines': 10, 'Transportation Services': 10, 'Military/Government/Technical': 10, 'Biotechnology: Commercial Physical & Biological Resarch': 10, 'Other Pharmaceuticals': 10, 'Wholesale Distributors': 9, 'Accident &Health Insurance': 9, 'Office Equipment/Supplies/Services': 9, 'Home Furnishings': 8, 'Other Specialty Stores': 8, 'Electronic Components': 8, 'Rental/Leasing Companies': 7, 'Paper': 7, 'Building operators': 6, 'Specialty Foods': 6, 'Precious Metals': 5, 'Medical Electronics': 5, 'Commercial Banks': 5, 'Mining & Quarrying of Nonmetallic Minerals (No Fuels)': 4, 'Coal Mining': 4, 'Electrical Products': 3, 'Building Products': 3, 'Electronics Distribution': 3, 'Water Supply': 3, 'Oil/Gas Transmission': 3, 'Professional Services': 2, 'Misc Health and Biotechnology Services': 2, 'Biotechnology: In Vitro & In Vivo Diagnostic Substances': 2, 'Meat/Poultry/Fish': 2, 'Textiles': 1, 'Ordnance And Accessories': 1, 'Homebuilding': 1, 'Engineering & Construction': 1, 'Steel/Iron Ore': 1, 'Tools/Hardware': 1, 'Advertising': 1, 'Hospital/Nursing Management': 1, 'Fluid Controls': 1})\n['Biotechnology: Laboratory Analytical Instruments', 'Industrial Machinery/Components', 'none', 'Paints/Coatings', 'Building operators', 'Major Banks', 'Semiconductors', 'Major Pharmaceuticals', 'Newspapers/Magazines']\nThere are 115 industries covered by the trades of senators.\n" ], [ "import string\n\nindustry_size_data = list()\n\nfor row_tuple in sen_df.itertuples():\n industry_size = row_tuple.industry\n \n if industry_size == 'none':\n industry_size_data.append(\"none\")\n continue\n \n size = row_tuple.mkt_cap\n factor = 0\n x = size.find(\"M\")\n if x != -1:\n factor = 1000000\n else:\n factor = 1000000000\n \n size = size.lstrip(\"$\")\n size = size.rstrip(\"MB\")\n size = float(size)\n size = size*factor\n \n if size < 500000000:\n industry_size = industry_size + \"1\"\n industry_size_data.append(industry_size)\n continue\n elif size < 1000000000:\n industry_size = industry_size + \"2\"\n industry_size_data.append(industry_size)\n continue\n elif size < 10000000000:\n industry_size = industry_size + \"3\"\n industry_size_data.append(industry_size)\n continue\n elif size < 50000000000:\n industry_size = industry_size + \"4\"\n industry_size_data.append(industry_size)\n continue\n elif size < 100000000000:\n industry_size = industry_size + \"5\"\n industry_size_data.append(industry_size)\n continue\n elif size < 500000000000:\n industry_size = industry_size + \"6\"\n industry_size_data.append(industry_size)\n continue\n else:\n industry_size = industry_size + \"7\"\n industry_size_data.append(industry_size)\n continue\n \nprint(industry_size_data[0:9])\nprint(len(industry_size_data))", "['Biotechnology: Laboratory Analytical Instruments4', 'Industrial Machinery/Components3', 'none', 'none', 'Paints/Coatings3', 'Building operators4', 'Major Banks5', 'Semiconductors5', 'Major Pharmaceuticals6']\n6644\n" ], [ "#add the new column to df\n\nsen_df['classification'] = industry_size_data\nsen_df.head()", "_____no_output_____" ], [ "#create a list of all the classifications per industry across whole dataframe, to get a view of the breakdown in\n#classifications across each industry\n\nclassification_industry_breakdown = list()\n\nfor x in industry_list:\n y = list()\n for row_tuple in sen_df.itertuples():\n if row_tuple.industry == x:\n y.append(row_tuple.classification)\n classification_industry_breakdown.append(y)\n \nprint(classification_industry_breakdown[0:9])", "[['Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments3', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments3', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments3', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4', 'Biotechnology: Laboratory Analytical Instruments4'], ['Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6', 'Industrial Machinery/Components5', 'Industrial Machinery/Components5', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components5', 'Industrial Machinery/Components5', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components4', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components5', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components4', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components4', 'Industrial Machinery/Components5', 'Industrial Machinery/Components5', 'Industrial Machinery/Components5', 'Industrial Machinery/Components4', 'Industrial Machinery/Components5', 'Industrial Machinery/Components6', 'Industrial Machinery/Components5', 'Industrial Machinery/Components6', 'Industrial Machinery/Components4', 'Industrial Machinery/Components3', 'Industrial Machinery/Components6', 'Industrial Machinery/Components5', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6', 'Industrial Machinery/Components5', 'Industrial Machinery/Components5', 'Industrial Machinery/Components4', 'Industrial Machinery/Components5', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6', 'Industrial Machinery/Components5', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components4', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components5', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components5', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components2', 'Industrial Machinery/Components4', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components4', 'Industrial Machinery/Components3', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components3', 'Industrial Machinery/Components4', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6', 'Industrial Machinery/Components6', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components4', 'Industrial Machinery/Components2', 'Industrial Machinery/Components4', 'Industrial Machinery/Components3', 'Industrial Machinery/Components3', 'Industrial Machinery/Components4', 'Industrial Machinery/Components3', 'Industrial Machinery/Components2', 'Industrial Machinery/Components4', 'Industrial Machinery/Components6'], ['none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none', 'none'], ['Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings4', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings3', 'Paints/Coatings4', 'Paints/Coatings4', 'Paints/Coatings4', 'Paints/Coatings3'], ['Building operators4', 'Building operators4', 'Building operators4', 'Building operators4', 'Building operators4', 'Building operators3'], ['Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks3', 'Major Banks3', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks3', 'Major Banks6', 'Major Banks3', 'Major Banks4', 'Major Banks5', 'Major Banks4', 'Major Banks3', 'Major Banks3', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks4', 'Major Banks6', 'Major Banks6', 'Major Banks4', 'Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks6', 'Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks6', 'Major Banks3', 'Major Banks5', 'Major Banks6', 'Major Banks3', 'Major Banks5', 'Major Banks5', 'Major Banks6', 'Major Banks4', 'Major Banks6', 'Major Banks4', 'Major Banks6', 'Major Banks5', 'Major Banks4', 'Major Banks3', 'Major Banks4', 'Major Banks3', 'Major Banks3', 'Major Banks6', 'Major Banks6', 'Major Banks3', 'Major Banks6', 'Major Banks3', 'Major Banks6', 'Major Banks6', 'Major Banks3', 'Major Banks3', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks3', 'Major Banks3', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks4', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks6', 'Major Banks3', 'Major Banks5', 'Major Banks6', 'Major Banks3', 'Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks6', 'Major Banks4', 'Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks5', 'Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks3', 'Major Banks6', 'Major Banks3', 'Major Banks3', 'Major Banks6', 'Major Banks3', 'Major Banks3', 'Major Banks4', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks6', 'Major Banks4', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks4', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks4', 'Major Banks3', 'Major Banks5', 'Major Banks4', 'Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks6', 'Major Banks5', 'Major Banks4', 'Major Banks4', 'Major Banks4', 'Major Banks6', 'Major Banks3', 'Major Banks6', 'Major Banks5', 'Major Banks6', 'Major Banks5', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks4', 'Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks4', 'Major Banks4', 'Major Banks4', 'Major Banks6', 'Major Banks3', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks3', 'Major Banks4', 'Major Banks4', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks4', 'Major Banks5', 'Major Banks3', 'Major Banks4', 'Major Banks4', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks4', 'Major Banks4', 'Major Banks4', 'Major Banks3', 'Major Banks4', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks6', 'Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks4', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks4', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks4', 'Major Banks4', 'Major Banks4', 'Major Banks3', 'Major Banks4', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks4', 'Major Banks6', 'Major Banks4', 'Major Banks2', 'Major Banks5', 'Major Banks5', 'Major Banks5', 'Major Banks6', 'Major Banks3', 'Major Banks4', 'Major Banks4', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks4', 'Major Banks6', 'Major Banks6', 'Major Banks3', 'Major Banks3', 'Major Banks5', 'Major Banks5', 'Major Banks5', 'Major Banks4', 'Major Banks6', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks3', 'Major Banks6', 'Major Banks4', 'Major Banks6', 'Major Banks2', 'Major Banks5', 'Major Banks6', 'Major Banks3', 'Major Banks3', 'Major Banks2', 'Major Banks3', 'Major Banks6', 'Major Banks4', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks5', 'Major Banks6', 'Major Banks6', 'Major Banks6', 'Major Banks5', 'Major Banks5'], ['Semiconductors5', 'Semiconductors5', 'Semiconductors5', 'Semiconductors5', 'Semiconductors5', 'Semiconductors4', 'Semiconductors6', 'Semiconductors6', 'Semiconductors3', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors4', 'Semiconductors4', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors4', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors4', 'Semiconductors6', 'Semiconductors4', 'Semiconductors4', 'Semiconductors4', 'Semiconductors5', 'Semiconductors4', 'Semiconductors6', 'Semiconductors6', 'Semiconductors5', 'Semiconductors6', 'Semiconductors5', 'Semiconductors3', 'Semiconductors5', 'Semiconductors6', 'Semiconductors6', 'Semiconductors4', 'Semiconductors4', 'Semiconductors3', 'Semiconductors6', 'Semiconductors4', 'Semiconductors4', 'Semiconductors4', 'Semiconductors6', 'Semiconductors4', 'Semiconductors6', 'Semiconductors4', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors3', 'Semiconductors3', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors3', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors4', 'Semiconductors4', 'Semiconductors4', 'Semiconductors4', 'Semiconductors4', 'Semiconductors4', 'Semiconductors4', 'Semiconductors4', 'Semiconductors4', 'Semiconductors4', 'Semiconductors4', 'Semiconductors6', 'Semiconductors4', 'Semiconductors4', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors6', 'Semiconductors3', 'Semiconductors4', 'Semiconductors3', 'Semiconductors3', 'Semiconductors4', 'Semiconductors3', 'Semiconductors3', 'Semiconductors3', 'Semiconductors4', 'Semiconductors5', 'Semiconductors3', 'Semiconductors3', 'Semiconductors3', 'Semiconductors6', 'Semiconductors6', 'Semiconductors3', 'Semiconductors3', 'Semiconductors3', 'Semiconductors3', 'Semiconductors4', 'Semiconductors4', 'Semiconductors4', 'Semiconductors4', 'Semiconductors4', 'Semiconductors4', 'Semiconductors3', 'Semiconductors4'], ['Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals3', 'Major Pharmaceuticals3', 'Major Pharmaceuticals6', 'Major Pharmaceuticals3', 'Major Pharmaceuticals3', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals1', 'Major Pharmaceuticals6', 'Major Pharmaceuticals3', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals5', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals5', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals3', 'Major Pharmaceuticals3', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals5', 'Major Pharmaceuticals1', 'Major Pharmaceuticals6', 'Major Pharmaceuticals5', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals4', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals3', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals2', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals3', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals5', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals3', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals4', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals4', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals5', 'Major Pharmaceuticals5', 'Major Pharmaceuticals6', 'Major Pharmaceuticals5', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals3', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals4', 'Major Pharmaceuticals5', 'Major Pharmaceuticals4', 'Major Pharmaceuticals5', 'Major Pharmaceuticals3', 'Major Pharmaceuticals6', 'Major Pharmaceuticals4', 'Major Pharmaceuticals6', 'Major Pharmaceuticals5', 'Major Pharmaceuticals5', 'Major Pharmaceuticals5', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals5', 'Major Pharmaceuticals5', 'Major Pharmaceuticals5', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals5', 'Major Pharmaceuticals4', 'Major Pharmaceuticals3', 'Major Pharmaceuticals3', 'Major Pharmaceuticals3', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals3', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals3', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals5', 'Major Pharmaceuticals5', 'Major Pharmaceuticals5', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals5', 'Major Pharmaceuticals3', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals3', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals3', 'Major Pharmaceuticals3', 'Major Pharmaceuticals1', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals3', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals6', 'Major Pharmaceuticals4'], ['Newspapers/Magazines3', 'Newspapers/Magazines3', 'Newspapers/Magazines3', 'Newspapers/Magazines3', 'Newspapers/Magazines3', 'Newspapers/Magazines3', 'Newspapers/Magazines3', 'Newspapers/Magazines3', 'Newspapers/Magazines2', 'Newspapers/Magazines3']]\n" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "!pip install datadotworld\n!pip install datadotworld[pandas]", "_____no_output_____" ], [ "# !dw configure", "_____no_output_____" ], [ "from google.colab import drive\n\nimport pandas as pd\nimport matplotlib.pyplot as plt\nimport numpy as np\n\n# drive.mount('/content/drive')", "_____no_output_____" ], [ "HOME = '/content/drive/My Drive/Colab Notebooks/matrix/dataworkshop_matrix'\n%cd {HOME}", "/content/drive/My Drive/Colab Notebooks/matrix/dataworkshop_matrix\n" ], [ "import datadotworld as dw\n\ndata = dw.load_dataset('datafiniti/mens-shoe-prices')\ndf = data.dataframes['7004_1']", "/usr/local/lib/python3.6/dist-packages/datadotworld/models/dataset.py:209: UserWarning: Unable to set data frame dtypes automatically using 7004_1 schema. Data types may need to be adjusted manually. Error: Integer column has NA values in column 10\n 'Error: {}'.format(resource_name, e))\n/usr/local/lib/python3.6/dist-packages/datadotworld/util.py:121: DtypeWarning: Columns (39,45) have mixed types. Specify dtype option on import or set low_memory=False.\n return self._loader_func()\n" ], [ "df_usd = df[ df.prices_currency == 'USD'].copy()\ndf_usd['prices_amountmin'] = df_usd.prices_amountmin.astype(np.float)\nfilter_max = np.percentile(df_usd['prices_amountmin'], 99)\ndf_usd_filter = df_usd[ df_usd['prices_amountmin'] < filter_max]\ndf_usd_filter['prices_amountmin'].hist(bins=100)", "_____no_output_____" ], [ "df_usd_filter.to_csv('data/men_shoes.csv', index=False)", "_____no_output_____" ], [ "from sklearn.tree import DecisionTreeRegressor\nfrom sklearn.metrics import mean_absolute_error\nfrom sklearn.model_selection import cross_val_score", "_____no_output_____" ], [ "df = pd.read_csv('data/men_shoes.csv', low_memory=False)", "_____no_output_____" ], [ "df.shape", "_____no_output_____" ], [ "df.columns", "_____no_output_____" ], [ "mean_amountmin = np.mean(df['prices_amountmin'])\nmean_amountmin", "_____no_output_____" ], [ "y_true = df['prices_amountmin']\nm = y_true.shape[0]\nm", "_____no_output_____" ], [ "y_pred = [mean_amountmin] * m", "_____no_output_____" ], [ "mae = mean_absolute_error(y_true, y_pred)\nmae", "_____no_output_____" ], [ "np.log1p(df['prices_amountmin']).hist(bins=100)", "_____no_output_____" ], [ "y_pred = [np.median(df['prices_amountmin'])] * m\nmae = mean_absolute_error(y_true, y_pred)\nmae", "_____no_output_____" ], [ "price_log_mean = np.mean(np.log1p(df['prices_amountmin']))\ny_pred = [np.expm1(price_log_mean)] * m\nmae = mean_absolute_error(y_true, y_pred)\nmae", "_____no_output_____" ], [ "df['brand_cat'] = df['brand'].factorize()[0]", "_____no_output_____" ], [ "def run_model(feats):\n X = df[ feats ].values\n y = df['prices_amountmin'].values\n model = DecisionTreeRegressor(max_depth=5)\n score = cross_val_score(model, X, y, scoring='neg_mean_absolute_error')\n return np.mean(score), np.std(score)", "_____no_output_____" ], [ "run_model(feats=['brand_cat'])", "_____no_output_____" ], [ "df['manufacturer_cat'] = df['manufacturer'].factorize()[0]", "_____no_output_____" ], [ "run_model(feats=['manufacturer_cat'])", "_____no_output_____" ], [ "run_model(feats=['manufacturer_cat', 'brand_cat'])", "_____no_output_____" ], [ "!git add day4.ipynb\n!git config --global user.email \"[email protected]\"\n!git config --global user.name \"Marek Zalecki\"\n!git commit -m \"Decision Tree Model\"\n!git push -u origin master", "[master 2ddf606] Decision Tree Model\n 1 file changed, 1 insertion(+)\n create mode 100644 day4.ipynb\nTo https://github.com/mzignis/dataworkshop_matrix.git\n ! [rejected] master -> master (fetch first)\nerror: failed to push some refs to 'https://[email protected]/mzignis/dataworkshop_matrix.git'\nhint: Updates were rejected because the remote contains work that you do\nhint: not have locally. This is usually caused by another repository pushing\nhint: to the same ref. You may want to first integrate the remote changes\nhint: (e.g., 'git pull ...') before pushing again.\nhint: See the 'Note about fast-forwards' in 'git push --help' for details.\n" ], [ "!git pull", "remote: Enumerating objects: 3, done.\u001b[K\nremote: Counting objects: 33% (1/3)\u001b[K\rremote: Counting objects: 66% (2/3)\u001b[K\rremote: Counting objects: 100% (3/3)\u001b[K\rremote: Counting objects: 100% (3/3), done.\u001b[K\nremote: Compressing objects: 50% (1/2)\u001b[K\rremote: Compressing objects: 100% (2/2)\u001b[K\rremote: Compressing objects: 100% (2/2), done.\u001b[K\nremote: Total 2 (delta 1), reused 0 (delta 0), pack-reused 0\u001b[K\nUnpacking objects: 100% (2/2), done.\nFrom https://github.com/mzignis/dataworkshop_matrix\n 2e4714e..052c777 master -> origin/master\nRemoving day_3.ipynb\nhint: Waiting for your editor to close the file... error: unable to start editor 'editor'\nNot committing merge; use 'git commit' to complete the merge.\n" ], [ "", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import trajectories\nimport numpy as np\nimport matplotlib.pyplot as plt\nfrom ctypes import cdll, c_float\n%matplotlib inline", "_____no_output_____" ], [ "LIB_NAME = \"../../../build/libekf_python.dylib\"", "_____no_output_____" ], [ "lib = cdll.LoadLibrary(LIB_NAME)\n\nPoint = c_float * 2\n\n\nestimator_predict = lib.estimator_predict\nestimator_predict.restype = None\n\nestimator_process_distance_measurement = lib.estimator_process_distance_measurement\nestimator_process_distance_measurement.restype = None\nestimator_process_distance_measurement.argtypes = (Point, c_float)\n\nestimator_get_x = lib.estimator_get_x\nestimator_get_x.restype = c_float\n\nestimator_get_y = lib.estimator_get_y\nestimator_get_y.restype = c_float\n\n", "_____no_output_____" ], [ "# Simulation start\nf = 200\nf_uwb = 10\nvmax = 0.4\nN = 10000\nvariance = (vmax / (2 * f_uwb))**2\n\nBEACON_POS = [\n (-1.5, 0),\n (1.5, 1),\n (1.5, -1),\n]\n\n\nx, xhat = [], []\ny, yhat = [], []\nts = []\n\n", "_____no_output_____" ], [ "for i, p in zip(range(N), trajectories.generate_circular_traj(1, np.deg2rad(10), 1/f)):\n # feeds the input into Kalman\n estimator_predict()\n\n if i % (f / f_uwb) == 0:\n for beacon in BEACON_POS:\n z = np.sqrt((beacon[0] - p.pos[0])**2 + (beacon[1] - p.pos[1])**2)\n z += np.random.normal(0, 0.03)\n estimator_process_distance_measurement(Point(*beacon), z)\n\n # Saves the data\n ts.append(p.timestamp)\n x.append(p.pos[0])\n xhat.append(estimator_get_x())\n y.append(p.pos[1])\n yhat.append(estimator_get_y())\n\n", "_____no_output_____" ], [ "plt.plot(x, y)\nplt.plot(xhat, yhat)\nplt.plot([x for x, y in BEACON_POS],[y for x, y in BEACON_POS], 'x')\nplt.legend(('Simulated trajectory', 'EKF output', 'anchors'))\nplt.title('trajectory')\nplt.xlabel('x [m]')\nplt.ylabel('y [m]')\nplt.gcf().savefig('uwb_only_trajectory.pdf')\nplt.show()\n\nerror = [np.sqrt((x-xh)**2+(y-yh)**2) for x,xh,y,yh in zip(x, xhat,y,yhat)]\nplt.plot(ts, error)\nplt.xlabel('timestamp')\nplt.ylabel('error [m]')\nplt.ylim(0, 0.1)\nplt.title('Position error (RMS = {:.3f} m)'.format(np.mean(error)))\nplt.gcf().savefig('uwb_only_error.pdf')\nplt.show()", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\nimport seaborn as sns \nimport matplotlib.pyplot as plt\nfrom sklearn.preprocessing import LabelEncoder", "_____no_output_____" ], [ "match1 = pd.read_csv(\"match1.csv\", index_col = 0, parse_dates = True)", "_____no_output_____" ], [ "match1[\"date\"] = pd.to_datetime(match1[\"date\"])", "_____no_output_____" ], [ "match1[\"year\"] = pd.DatetimeIndex(match1['date']).year\nmatch1[\"month\"] = pd.DatetimeIndex(match1['date']).month\nmatch1[\"day\"] = pd.DatetimeIndex(match1['date']).day\nmatch1[\"week_num\"] = match1['date'].dt.week", "_____no_output_____" ], [ "match1.to_csv(\"match_with_date_cols.csv\")", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "import numpy as np\nimport os\nimport matplotlib.pyplot as plt\n%matplotlib inline", "_____no_output_____" ], [ "results_dir = \"prime_number\"\ninput = '1000'\nrequests = '2000'\nfiles = os.listdir(results_dir)\n\ncpuMap = {}\nramMap = {}\n\nfor file in files:\n tokens = file.split(\"_\")\n cpuList = []\n ramList = []\n if tokens[1] == input and tokens[2] == requests:\n lines = [line.rstrip('\\n') for line in open(\"{0}/{1}\".format(results_dir, file))]\n for line in lines:\n params = line.split(\" \")\n cpuList.append(int(params[0]))\n ramList.append(int(params[1]))\n cpuMap[tokens[0]] = cpuList\n ramMap[tokens[0]] = ramList\n \n \n \n ", "_____no_output_____" ], [ "plt.figure(figsize=[10, 5])\n#plt.xticks(labels)\ninvokerDataToPrint = ['1', '2', '3', '4', '5', '6', '7']\nfor e in invokerDataToPrint:\n plt.plot(cpuMap[e], label='Invoker #{0}: CPU'.format(e))\nplt.ylabel('CPU Load, %')\nplt.xlabel('Time, seconds')\nplt.legend()\nplt.show()", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code" ] ]

[ "BSD-Source-Code" ]

[ "BSD-Source-Code" ]

[ "BSD-Source-Code" ]

[ [ [ "empty" ] ] ]

[ "empty" ]

[ [ "empty" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# Collaborative filtering on Google Analytics data\n\nThis notebook demonstrates how to implement a WALS matrix refactorization approach to do collaborative filtering.", "_____no_output_____" ] ], [ [ "import os\nPROJECT = \"qwiklabs-gcp-00-34ffb0f0dc65\" # REPLACE WITH YOUR PROJECT ID\nBUCKET = \"cloud-training-demos-ml\" # REPLACE WITH YOUR BUCKET NAME\nREGION = \"us-central1\" # REPLACE WITH YOUR BUCKET REGION e.g. us-central1\n\n# Do not change these\nos.environ[\"PROJECT\"] = PROJECT\nos.environ[\"BUCKET\"] = BUCKET\nos.environ[\"REGION\"] = REGION\nos.environ[\"TFVERSION\"] = \"1.13\"", "_____no_output_____" ], [ "%%bash\ngcloud config set project $PROJECT\ngcloud config set compute/region $REGION", "Updated property [core/project].\nUpdated property [compute/region].\n" ], [ "import tensorflow as tf\nprint(tf.__version__)", "1.13.1\n" ] ], [ [ "## Create raw dataset\n<p>\nFor collaborative filtering, we don't need to know anything about either the users or the content. Essentially, all we need to know is userId, itemId, and rating that the particular user gave the particular item.\n<p>\nIn this case, we are working with newspaper articles. The company doesn't ask their users to rate the articles. However, we can use the time-spent on the page as a proxy for rating.\n<p>\nNormally, we would also add a time filter to this (\"latest 7 days\"), but our dataset is itself limited to a few days.", "_____no_output_____" ] ], [ [ "from google.cloud import bigquery\nbq = bigquery.Client(project = PROJECT)\n\nsql = \"\"\"\n#standardSQL\nWITH CTE_visitor_page_content AS (\n SELECT\n fullVisitorID,\n (SELECT MAX(IF(index=10, value, NULL)) FROM UNNEST(hits.customDimensions)) AS latestContentId, \n (LEAD(hits.time, 1) OVER (PARTITION BY fullVisitorId ORDER BY hits.time ASC) - hits.time) AS session_duration \n FROM\n `cloud-training-demos.GA360_test.ga_sessions_sample`, \n UNNEST(hits) AS hits\n WHERE \n # only include hits on pages\n hits.type = \"PAGE\"\n\n GROUP BY \n fullVisitorId,\n latestContentId,\n hits.time )\n\n-- Aggregate web stats\nSELECT \n fullVisitorID as visitorId,\n latestContentId as contentId,\n SUM(session_duration) AS session_duration\nFROM\n CTE_visitor_page_content\nWHERE\n latestContentId IS NOT NULL \nGROUP BY\n fullVisitorID, \n latestContentId\nHAVING \n session_duration > 0\nORDER BY \n latestContentId \n\"\"\"\n\ndf = bq.query(sql).to_dataframe()\ndf.head()", "_____no_output_____" ], [ "stats = df.describe()\nstats", "_____no_output_____" ], [ "df[[\"session_duration\"]].plot(kind=\"hist\", logy=True, bins=100, figsize=[8,5])", "_____no_output_____" ], [ "# The rating is the session_duration scaled to be in the range 0-1. This will help with training.\nmedian = stats.loc[\"50%\", \"session_duration\"]\ndf[\"rating\"] = 0.3 * df[\"session_duration\"] / median\ndf.loc[df[\"rating\"] > 1, \"rating\"] = 1\ndf[[\"rating\"]].plot(kind=\"hist\", logy=True, bins=100, figsize=[8,5])", "_____no_output_____" ], [ "del df[\"session_duration\"]", "_____no_output_____" ], [ "%%bash\nrm -rf data\nmkdir data", "_____no_output_____" ], [ "df.to_csv(path_or_buf = \"data/collab_raw.csv\", index = False, header = False)", "_____no_output_____" ], [ "!head data/collab_raw.csv", "7337153711992174438,100074831,0.2321051400452234\n5190801220865459604,100170790,1.0\n2293633612703952721,100510126,0.2481776360816793\n5874973374932455844,100510126,0.16690549004998828\n1173698801255170595,100676857,0.05464232805149575\n883397426232997550,10083328,0.9487035095774818\n1808867070685560283,100906145,1.0\n7615995624631762562,100906145,0.48418654214351925\n5519169380728479914,100915139,0.20026163722525925\n3427736932800080345,100950628,0.558924688331153\n" ] ], [ [ "## Create dataset for WALS\n<p>\nThe raw dataset (above) won't work for WALS:\n<ol>\n<li> The userId and itemId have to be 0,1,2 ... so we need to create a mapping from visitorId (in the raw data) to userId and contentId (in the raw data) to itemId.\n<li> We will need to save the above mapping to a file because at prediction time, we'll need to know how to map the contentId in the table above to the itemId.\n<li> We'll need two files: a \"rows\" dataset where all the items for a particular user are listed; and a \"columns\" dataset where all the users for a particular item are listed.\n</ol>\n\n<p>\n\n### Mapping", "_____no_output_____" ] ], [ [ "import pandas as pd\nimport numpy as np\ndef create_mapping(values, filename):\n with open(filename, 'w') as ofp:\n value_to_id = {value:idx for idx, value in enumerate(values.unique())}\n for value, idx in value_to_id.items():\n ofp.write(\"{},{}\\n\".format(value, idx))\n return value_to_id\n\ndf = pd.read_csv(filepath_or_buffer = \"data/collab_raw.csv\",\n header = None,\n names = [\"visitorId\", \"contentId\", \"rating\"],\n dtype = {\"visitorId\": str, \"contentId\": str, \"rating\": np.float})\ndf.to_csv(path_or_buf = \"data/collab_raw.csv\", index = False, header = False)\nuser_mapping = create_mapping(df[\"visitorId\"], \"data/users.csv\")\nitem_mapping = create_mapping(df[\"contentId\"], \"data/items.csv\")", "_____no_output_____" ], [ "!head -3 data/*.csv", "==> data/collab_raw.csv <==\n7337153711992174438,100074831,0.2321051400452234\n5190801220865459604,100170790,1.0\n2293633612703952721,100510126,0.2481776360816793\n\n==> data/items.csv <==\n727741,5272\n179038175,626\n299458287,4513\n\n==> data/users.csv <==\n6319375062712956077,33748\n7933447845885715412,47057\n5774017011910110015,76528\n" ], [ "df[\"userId\"] = df[\"visitorId\"].map(user_mapping.get)\ndf[\"itemId\"] = df[\"contentId\"].map(item_mapping.get)", "_____no_output_____" ], [ "mapped_df = df[[\"userId\", \"itemId\", \"rating\"]]\nmapped_df.to_csv(path_or_buf = \"data/collab_mapped.csv\", index = False, header = False)\nmapped_df.head()", "_____no_output_____" ] ], [ [ "### Creating rows and columns datasets", "_____no_output_____" ] ], [ [ "import pandas as pd\nimport numpy as np\nmapped_df = pd.read_csv(filepath_or_buffer = \"data/collab_mapped.csv\", header = None, names = [\"userId\", \"itemId\", \"rating\"])\nmapped_df.head()", "_____no_output_____" ], [ "NITEMS = np.max(mapped_df[\"itemId\"]) + 1\nNUSERS = np.max(mapped_df[\"userId\"]) + 1\nmapped_df[\"rating\"] = np.round(mapped_df[\"rating\"].values, 2)\nprint(\"{} items, {} users, {} interactions\".format( NITEMS, NUSERS, len(mapped_df) ))", "5721 items, 82902 users, 279594 interactions\n" ], [ "grouped_by_items = mapped_df.groupby(\"itemId\")\niter = 0\nfor item, grouped in grouped_by_items:\n print(item, grouped[\"userId\"].values, grouped[\"rating\"].values)\n iter = iter + 1\n if iter > 5:\n break", "0 [0] [0.23]\n1 [1] [1.]\n2 [2 3] [0.25 0.17]\n3 [4] [0.05]\n4 [5] [0.95]\n5 [6 7] [1. 0.48]\n" ], [ "import tensorflow as tf\ngrouped_by_items = mapped_df.groupby(\"itemId\")\nwith tf.python_io.TFRecordWriter(\"data/users_for_item\") as ofp:\n for item, grouped in grouped_by_items:\n example = tf.train.Example(features = tf.train.Features(feature = {\n \"key\": tf.train.Feature(int64_list = tf.train.Int64List(value = [item])),\n \"indices\": tf.train.Feature(int64_list = tf.train.Int64List(value = grouped[\"userId\"].values)),\n \"values\": tf.train.Feature(float_list = tf.train.FloatList(value = grouped[\"rating\"].values))\n }))\n ofp.write(example.SerializeToString())", "_____no_output_____" ], [ "grouped_by_users = mapped_df.groupby(\"userId\")\nwith tf.python_io.TFRecordWriter(\"data/items_for_user\") as ofp:\n for user, grouped in grouped_by_users:\n example = tf.train.Example(features = tf.train.Features(feature = {\n \"key\": tf.train.Feature(int64_list = tf.train.Int64List(value = [user])),\n \"indices\": tf.train.Feature(int64_list = tf.train.Int64List(value = grouped[\"itemId\"].values)),\n \"values\": tf.train.Feature(float_list = tf.train.FloatList(value = grouped[\"rating\"].values))\n }))\n ofp.write(example.SerializeToString())", "_____no_output_____" ], [ "!ls -lrt data", "total 31908\n-rw-r--r-- 1 jupyter jupyter 13152765 Jul 31 20:41 collab_raw.csv\n-rw-r--r-- 1 jupyter jupyter 2134511 Jul 31 20:41 users.csv\n-rw-r--r-- 1 jupyter jupyter 82947 Jul 31 20:41 items.csv\n-rw-r--r-- 1 jupyter jupyter 7812739 Jul 31 20:41 collab_mapped.csv\n-rw-r--r-- 1 jupyter jupyter 2252828 Jul 31 20:41 users_for_item\n-rw-r--r-- 1 jupyter jupyter 7217822 Jul 31 20:41 items_for_user\n" ] ], [ [ "To summarize, we created the following data files from collab_raw.csv:\n<ol>\n<li> ```collab_mapped.csv``` is essentially the same data as in ```collab_raw.csv``` except that ```visitorId``` and ```contentId``` which are business-specific have been mapped to ```userId``` and ```itemId``` which are enumerated in 0,1,2,.... The mappings themselves are stored in ```items.csv``` and ```users.csv``` so that they can be used during inference.\n<li> ```users_for_item``` contains all the users/ratings for each item in TFExample format\n<li> ```items_for_user``` contains all the items/ratings for each user in TFExample format\n</ol>", "_____no_output_____" ], [ "## Train with WALS\n\nOnce you have the dataset, do matrix factorization with WALS using the [WALSMatrixFactorization](https://www.tensorflow.org/versions/master/api_docs/python/tf/contrib/factorization/WALSMatrixFactorization) in the contrib directory.\nThis is an estimator model, so it should be relatively familiar.\n<p>\nAs usual, we write an input_fn to provide the data to the model, and then create the Estimator to do train_and_evaluate.\nBecause it is in contrib and hasn't moved over to tf.estimator yet, we use tf.contrib.learn.Experiment to handle the training loop.", "_____no_output_____" ] ], [ [ "import os\nimport tensorflow as tf\nfrom tensorflow.python.lib.io import file_io\nfrom tensorflow.contrib.factorization import WALSMatrixFactorization\n \ndef read_dataset(mode, args):\n def decode_example(protos, vocab_size):\n features = {\n \"key\": tf.FixedLenFeature(shape = [1], dtype = tf.int64),\n \"indices\": tf.VarLenFeature(dtype = tf.int64),\n \"values\": tf.VarLenFeature(dtype = tf.float32)}\n parsed_features = tf.parse_single_example(serialized = protos, features = features)\n values = tf.sparse_merge(sp_ids = parsed_features[\"indices\"], sp_values = parsed_features[\"values\"], vocab_size = vocab_size)\n # Save key to remap after batching\n # This is a temporary workaround to assign correct row numbers in each batch.\n # You can ignore details of this part and remap_keys().\n key = parsed_features[\"key\"]\n decoded_sparse_tensor = tf.SparseTensor(indices = tf.concat(values = [values.indices, [key]], axis = 0), \n values = tf.concat(values = [values.values, [0.0]], axis = 0), \n dense_shape = values.dense_shape)\n return decoded_sparse_tensor\n \n \n def remap_keys(sparse_tensor):\n # Current indices of our SparseTensor that we need to fix\n bad_indices = sparse_tensor.indices # shape = (current_batch_size * (number_of_items/users[i] + 1), 2)\n # Current values of our SparseTensor that we need to fix\n bad_values = sparse_tensor.values # shape = (current_batch_size * (number_of_items/users[i] + 1),)\n\n # Since batch is ordered, the last value for a batch index is the user\n # Find where the batch index chages to extract the user rows\n # 1 where user, else 0\n user_mask = tf.concat(values = [bad_indices[1:,0] - bad_indices[:-1,0], tf.constant(value = [1], dtype = tf.int64)], axis = 0) # shape = (current_batch_size * (number_of_items/users[i] + 1), 2)\n\n # Mask out the user rows from the values\n good_values = tf.boolean_mask(tensor = bad_values, mask = tf.equal(x = user_mask, y = 0)) # shape = (current_batch_size * number_of_items/users[i],)\n item_indices = tf.boolean_mask(tensor = bad_indices, mask = tf.equal(x = user_mask, y = 0)) # shape = (current_batch_size * number_of_items/users[i],)\n user_indices = tf.boolean_mask(tensor = bad_indices, mask = tf.equal(x = user_mask, y = 1))[:, 1] # shape = (current_batch_size,)\n\n good_user_indices = tf.gather(params = user_indices, indices = item_indices[:,0]) # shape = (current_batch_size * number_of_items/users[i],)\n\n # User and item indices are rank 1, need to make rank 1 to concat\n good_user_indices_expanded = tf.expand_dims(input = good_user_indices, axis = -1) # shape = (current_batch_size * number_of_items/users[i], 1)\n good_item_indices_expanded = tf.expand_dims(input = item_indices[:, 1], axis = -1) # shape = (current_batch_size * number_of_items/users[i], 1)\n good_indices = tf.concat(values = [good_user_indices_expanded, good_item_indices_expanded], axis = 1) # shape = (current_batch_size * number_of_items/users[i], 2)\n\n remapped_sparse_tensor = tf.SparseTensor(indices = good_indices, values = good_values, dense_shape = sparse_tensor.dense_shape)\n return remapped_sparse_tensor\n\n \n def parse_tfrecords(filename, vocab_size):\n if mode == tf.estimator.ModeKeys.TRAIN:\n num_epochs = None # indefinitely\n else:\n num_epochs = 1 # end-of-input after this\n\n files = tf.gfile.Glob(filename = os.path.join(args[\"input_path\"], filename))\n\n # Create dataset from file list\n dataset = tf.data.TFRecordDataset(files)\n dataset = dataset.map(map_func = lambda x: decode_example(x, vocab_size))\n dataset = dataset.repeat(count = num_epochs)\n dataset = dataset.batch(batch_size = args[\"batch_size\"])\n dataset = dataset.map(map_func = lambda x: remap_keys(x))\n return dataset.make_one_shot_iterator().get_next()\n \n def _input_fn():\n features = {\n WALSMatrixFactorization.INPUT_ROWS: parse_tfrecords(\"items_for_user\", args[\"nitems\"]),\n WALSMatrixFactorization.INPUT_COLS: parse_tfrecords(\"users_for_item\", args[\"nusers\"]),\n WALSMatrixFactorization.PROJECT_ROW: tf.constant(True)\n }\n return features, None\n\n return _input_fn", "_____no_output_____" ] ], [ [ "This code is helpful in developing the input function. You don't need it in production.", "_____no_output_____" ] ], [ [ "def try_out():\n with tf.Session() as sess:\n fn = read_dataset(\n mode = tf.estimator.ModeKeys.EVAL, \n args = {\"input_path\": \"data\", \"batch_size\": 4, \"nitems\": NITEMS, \"nusers\": NUSERS})\n feats, _ = fn()\n \n print(feats[\"input_rows\"].eval())\n print(feats[\"input_rows\"].eval())\n\ntry_out()", "SparseTensorValue(indices=array([[ 0, 0],\n [ 0, 3522],\n [ 0, 3583],\n [ 1, 1],\n [ 1, 2359],\n [ 1, 3133],\n [ 1, 4864],\n [ 1, 4901],\n [ 1, 4906],\n [ 1, 5667],\n [ 2, 2],\n [ 3, 2],\n [ 3, 1467]]), values=array([0.23, 0.05, 0.18, 1. , 0.11, 0.55, 0.3 , 0.72, 0.46, 0.3 , 0.25,\n 0.17, 0.13], dtype=float32), dense_shape=array([ 4, 5721]))\nSparseTensorValue(indices=array([[ 4, 3],\n [ 5, 4],\n [ 5, 5042],\n [ 5, 5525],\n [ 5, 5553],\n [ 6, 5],\n [ 7, 5]]), values=array([0.05, 0.95, 0.63, 1. , 0.16, 1. , 0.48], dtype=float32), dense_shape=array([ 4, 5721]))\n" ], [ "def find_top_k(user, item_factors, k):\n all_items = tf.matmul(a = tf.expand_dims(input = user, axis = 0), b = tf.transpose(a = item_factors))\n topk = tf.nn.top_k(input = all_items, k = k)\n return tf.cast(x = topk.indices, dtype = tf.int64)\n \ndef batch_predict(args):\n import numpy as np\n with tf.Session() as sess:\n estimator = tf.contrib.factorization.WALSMatrixFactorization(\n num_rows = args[\"nusers\"], \n num_cols = args[\"nitems\"],\n embedding_dimension = args[\"n_embeds\"],\n model_dir = args[\"output_dir\"])\n \n # This is how you would get the row factors for out-of-vocab user data\n # row_factors = list(estimator.get_projections(input_fn=read_dataset(tf.estimator.ModeKeys.EVAL, args)))\n # user_factors = tf.convert_to_tensor(np.array(row_factors))\n\n # But for in-vocab data, the row factors are already in the checkpoint\n user_factors = tf.convert_to_tensor(value = estimator.get_row_factors()[0]) # (nusers, nembeds)\n # In either case, we have to assume catalog doesn\"t change, so col_factors are read in\n item_factors = tf.convert_to_tensor(value = estimator.get_col_factors()[0])# (nitems, nembeds)\n\n # For each user, find the top K items\n topk = tf.squeeze(input = tf.map_fn(fn = lambda user: find_top_k(user, item_factors, args[\"topk\"]), elems = user_factors, dtype = tf.int64))\n with file_io.FileIO(os.path.join(args[\"output_dir\"], \"batch_pred.txt\"), mode = 'w') as f:\n for best_items_for_user in topk.eval():\n f.write(\",\".join(str(x) for x in best_items_for_user) + '\\n')\n\ndef train_and_evaluate(args):\n train_steps = int(0.5 + (1.0 * args[\"num_epochs\"] * args[\"nusers\"]) / args[\"batch_size\"])\n steps_in_epoch = int(0.5 + args[\"nusers\"] / args[\"batch_size\"])\n print(\"Will train for {} steps, evaluating once every {} steps\".format(train_steps, steps_in_epoch))\n def experiment_fn(output_dir):\n return tf.contrib.learn.Experiment(\n tf.contrib.factorization.WALSMatrixFactorization(\n num_rows = args[\"nusers\"], \n num_cols = args[\"nitems\"],\n embedding_dimension = args[\"n_embeds\"],\n model_dir = args[\"output_dir\"]),\n train_input_fn = read_dataset(tf.estimator.ModeKeys.TRAIN, args),\n eval_input_fn = read_dataset(tf.estimator.ModeKeys.EVAL, args),\n train_steps = train_steps,\n eval_steps = 1,\n min_eval_frequency = steps_in_epoch\n )\n\n from tensorflow.contrib.learn.python.learn import learn_runner\n learn_runner.run(experiment_fn = experiment_fn, output_dir = args[\"output_dir\"])\n \n batch_predict(args)", "_____no_output_____" ], [ "import shutil\nshutil.rmtree(path = \"wals_trained\", ignore_errors=True)\ntrain_and_evaluate({\n \"output_dir\": \"wals_trained\",\n \"input_path\": \"data/\",\n \"num_epochs\": 0.05,\n \"nitems\": NITEMS,\n \"nusers\": NUSERS,\n\n \"batch_size\": 512,\n \"n_embeds\": 10,\n \"topk\": 3\n })", "Will train for 8 steps, evaluating once every 162 steps\nWARNING:tensorflow:From <ipython-input-25-4ad1e7c785ce>:49: run (from tensorflow.contrib.learn.python.learn.learn_runner) is deprecated and will be removed in a future version.\nInstructions for updating:\nUse tf.estimator.train_and_evaluate.\nWARNING:tensorflow:From /home/jupyter/.local/lib/python3.5/site-packages/tensorflow/contrib/learn/python/learn/estimators/estimator.py:1179: BaseEstimator.__init__ (from tensorflow.contrib.learn.python.learn.estimators.estimator) is deprecated and will be removed in a future version.\nInstructions for updating:\nPlease replace uses of any Estimator from tf.contrib.learn with an Estimator from tf.estimator.*\nWARNING:tensorflow:From /home/jupyter/.local/lib/python3.5/site-packages/tensorflow/contrib/learn/python/learn/estimators/estimator.py:427: RunConfig.__init__ (from tensorflow.contrib.learn.python.learn.estimators.run_config) is deprecated and will be removed in a future version.\nInstructions for updating:\nWhen switching to tf.estimator.Estimator, use tf.estimator.RunConfig instead.\nINFO:tensorflow:Using default config.\nINFO:tensorflow:Using config: {'_tf_config': gpu_options {\n per_process_gpu_memory_fraction: 1.0\n}\n, '_task_id': 0, '_is_chief': True, '_cluster_spec': <tensorflow.python.training.server_lib.ClusterSpec object at 0x7fb220516da0>, '_save_checkpoints_steps': None, '_task_type': None, '_tf_random_seed': None, '_save_summary_steps': 100, '_num_ps_replicas': 0, '_keep_checkpoint_max': 5, '_log_step_count_steps': 100, '_keep_checkpoint_every_n_hours': 10000, '_environment': 'local', '_eval_distribute': None, '_session_config': None, '_train_distribute': None, '_evaluation_master': '', '_num_worker_replicas': 0, '_device_fn': None, '_master': '', '_protocol': None, '_save_checkpoints_secs': 600, '_model_dir': 'wals_trained'}\nWARNING:tensorflow:From <ipython-input-25-4ad1e7c785ce>:45: Experiment.__init__ (from tensorflow.contrib.learn.python.learn.experiment) is deprecated and will be removed in a future version.\nInstructions for updating:\nPlease switch to tf.estimator.train_and_evaluate. You will also have to convert to a tf.estimator.Estimator.\nWARNING:tensorflow:From /home/jupyter/.local/lib/python3.5/site-packages/tensorflow/contrib/learn/python/learn/monitors.py:279: BaseMonitor.__init__ (from tensorflow.contrib.learn.python.learn.monitors) is deprecated and will be removed after 2016-12-05.\nInstructions for updating:\nMonitors are deprecated. Please use tf.train.SessionRunHook.\nWARNING:tensorflow:From /home/jupyter/.local/lib/python3.5/site-packages/tensorflow/python/framework/op_def_library.py:263: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.\nInstructions for updating:\nColocations handled automatically by placer.\nWARNING:tensorflow:From /home/jupyter/.local/lib/python3.5/site-packages/tensorflow/contrib/factorization/python/ops/wals.py:315: ModelFnOps.__new__ (from tensorflow.contrib.learn.python.learn.estimators.model_fn) is deprecated and will be removed in a future version.\nInstructions for updating:\nWhen switching to tf.estimator.Estimator, use tf.estimator.EstimatorSpec. You can use the `estimator_spec` method to create an equivalent one.\nINFO:tensorflow:Create CheckpointSaverHook.\nINFO:tensorflow:Graph was finalized.\nINFO:tensorflow:Running local_init_op.\nINFO:tensorflow:Done running local_init_op.\nINFO:tensorflow:Saving checkpoints for 0 into wals_trained/model.ckpt.\nINFO:tensorflow:SweepHook running init op.\nINFO:tensorflow:SweepHook running prep ops for the row sweep.\nINFO:tensorflow:Next fit step starting.\nINFO:tensorflow:loss = 96509.96, step = 1\nINFO:tensorflow:Next fit step starting.\nINFO:tensorflow:Next fit step starting.\nINFO:tensorflow:Next fit step starting.\nINFO:tensorflow:Next fit step starting.\nINFO:tensorflow:Next fit step starting.\nINFO:tensorflow:Next fit step starting.\nINFO:tensorflow:Next fit step starting.\nINFO:tensorflow:Saving checkpoints for 8 into wals_trained/model.ckpt.\nINFO:tensorflow:Loss for final step: 110142.75.\nWARNING:tensorflow:From /home/jupyter/.local/lib/python3.5/site-packages/tensorflow/python/ops/metrics_impl.py:363: to_float (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\nInstructions for updating:\nUse tf.cast instead.\nINFO:tensorflow:Starting evaluation at 2019-07-31T20:43:12Z\nINFO:tensorflow:Graph was finalized.\nWARNING:tensorflow:From /home/jupyter/.local/lib/python3.5/site-packages/tensorflow/python/training/saver.py:1266: checkpoint_exists (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version.\nInstructions for updating:\nUse standard file APIs to check for files with this prefix.\nINFO:tensorflow:Restoring parameters from wals_trained/model.ckpt-8\nINFO:tensorflow:Running local_init_op.\nINFO:tensorflow:Done running local_init_op.\nINFO:tensorflow:Evaluation [1/1]\nINFO:tensorflow:Finished evaluation at 2019-07-31-20:43:12\nINFO:tensorflow:Saving dict for global step 8: global_step = 8, loss = 96509.96\nINFO:tensorflow:Using default config.\nINFO:tensorflow:Using config: {'_tf_config': gpu_options {\n per_process_gpu_memory_fraction: 1.0\n}\n, '_task_id': 0, '_is_chief': True, '_cluster_spec': <tensorflow.python.training.server_lib.ClusterSpec object at 0x7fb2207064e0>, '_save_checkpoints_steps': None, '_task_type': None, '_tf_random_seed': None, '_save_summary_steps': 100, '_num_ps_replicas': 0, '_keep_checkpoint_max': 5, '_log_step_count_steps': 100, '_keep_checkpoint_every_n_hours': 10000, '_environment': 'local', '_eval_distribute': None, '_session_config': None, '_train_distribute': None, '_evaluation_master': '', '_num_worker_replicas': 0, '_device_fn': None, '_master': '', '_protocol': None, '_save_checkpoints_secs': 600, '_model_dir': 'wals_trained'}\n" ], [ "!ls wals_trained", "batch_pred.txt\t\t\t model.ckpt-0.index\ncheckpoint\t\t\t model.ckpt-0.meta\neval\t\t\t\t model.ckpt-8.data-00000-of-00001\nevents.out.tfevents.1564605788.r model.ckpt-8.index\ngraph.pbtxt\t\t\t model.ckpt-8.meta\nmodel.ckpt-0.data-00000-of-00001\n" ], [ "!head wals_trained/batch_pred.txt", "284,5609,36\n284,2754,42\n284,3168,534\n2621,5528,2694\n4409,5295,343\n5161,3267,3369\n5479,1335,55\n5479,1335,55\n4414,284,5572\n284,241,2359\n" ] ], [ [ "## Run as a Python module\n\nLet's run it as Python module for just a few steps.", "_____no_output_____" ] ], [ [ "os.environ[\"NITEMS\"] = str(NITEMS)\nos.environ[\"NUSERS\"] = str(NUSERS)", "_____no_output_____" ], [ "%%bash\nrm -rf wals.tar.gz wals_trained\ngcloud ml-engine local train \\\n --module-name=walsmodel.task \\\n --package-path=${PWD}/walsmodel \\\n -- \\\n --output_dir=${PWD}/wals_trained \\\n --input_path=${PWD}/data \\\n --num_epochs=0.01 --nitems=${NITEMS} --nusers=${NUSERS} \\\n --job-dir=./tmp", "Will train for 2 steps, evaluating once every 162 steps\n" ] ], [ [ "## Run on Cloud", "_____no_output_____" ] ], [ [ "%%bash\ngsutil -m cp data/* gs://${BUCKET}/wals/data", "_____no_output_____" ], [ "%%bash\nOUTDIR=gs://${BUCKET}/wals/model_trained\nJOBNAME=wals_$(date -u +%y%m%d_%H%M%S)\necho $OUTDIR $REGION $JOBNAME\ngsutil -m rm -rf $OUTDIR\ngcloud ml-engine jobs submit training $JOBNAME \\\n --region=$REGION \\\n --module-name=walsmodel.task \\\n --package-path=${PWD}/walsmodel \\\n --job-dir=$OUTDIR \\\n --staging-bucket=gs://$BUCKET \\\n --scale-tier=BASIC_GPU \\\n --runtime-version=$TFVERSION \\\n -- \\\n --output_dir=$OUTDIR \\\n --input_path=gs://${BUCKET}/wals/data \\\n --num_epochs=10 --nitems=${NITEMS} --nusers=${NUSERS} ", "_____no_output_____" ] ], [ [ "This took <b>10 minutes</b> for me.", "_____no_output_____" ], [ "## Get row and column factors\n\nOnce you have a trained WALS model, you can get row and column factors (user and item embeddings) from the checkpoint file. We'll look at how to use these in the section on building a recommendation system using deep neural networks.", "_____no_output_____" ] ], [ [ "def get_factors(args):\n with tf.Session() as sess:\n estimator = tf.contrib.factorization.WALSMatrixFactorization(\n num_rows = args[\"nusers\"], \n num_cols = args[\"nitems\"],\n embedding_dimension = args[\"n_embeds\"],\n model_dir = args[\"output_dir\"])\n \n row_factors = estimator.get_row_factors()[0]\n col_factors = estimator.get_col_factors()[0]\n return row_factors, col_factors", "_____no_output_____" ], [ "args = {\n \"output_dir\": \"gs://{}/wals/model_trained\".format(BUCKET),\n \"nitems\": NITEMS,\n \"nusers\": NUSERS,\n \"n_embeds\": 10\n }\n\nuser_embeddings, item_embeddings = get_factors(args)\nprint(user_embeddings[:3])\nprint(item_embeddings[:3])", "INFO:tensorflow:Using default config.\nINFO:tensorflow:Using config: {'_environment': 'local', '_is_chief': True, '_keep_checkpoint_every_n_hours': 10000, '_num_worker_replicas': 0, '_session_config': None, '_task_type': None, '_eval_distribute': None, '_tf_config': gpu_options {\n per_process_gpu_memory_fraction: 1.0\n}\n, '_master': '', '_log_step_count_steps': 100, '_model_dir': 'gs://qwiklabs-gcp-cbc8684b07fc2dbd-bucket/wals/model_trained', '_cluster_spec': <tensorflow.python.training.server_lib.ClusterSpec object at 0x7f4bd8302f28>, '_device_fn': None, '_keep_checkpoint_max': 5, '_task_id': 0, '_evaluation_master': '', '_save_checkpoints_steps': None, '_protocol': None, '_train_distribute': None, '_save_checkpoints_secs': 600, '_save_summary_steps': 100, '_tf_random_seed': None, '_num_ps_replicas': 0}\n[[ 3.3451824e-06 -1.1986867e-05 4.8447573e-06 -1.5209486e-05\n -1.7004859e-07 1.1976428e-05 9.8887876e-06 7.2386983e-06\n -7.0237149e-07 -7.9796819e-06]\n [-2.5300323e-03 1.4055537e-03 -9.8291773e-04 -4.2533795e-03\n -1.4166030e-03 -1.9530674e-03 8.5932651e-04 -1.5276540e-03\n 2.1342330e-03 1.2041229e-03]\n [ 9.5228699e-21 5.5453966e-21 2.2947056e-21 -5.8859543e-21\n 7.7516509e-21 -2.7640896e-20 2.3587296e-20 -3.9876822e-21\n 1.7312470e-20 2.5409211e-20]]\n[[-1.2125404e-06 -8.6304914e-05 4.4657736e-05 -6.8423047e-05\n 5.8551927e-06 9.7241784e-05 6.6776753e-05 1.6673854e-05\n -1.2708440e-05 -5.1148414e-05]\n [-1.1353870e-01 5.9097271e-02 -4.6105500e-02 -1.5460028e-01\n -1.9166643e-02 -7.3236257e-02 3.5582058e-02 -5.6805085e-02\n 7.5831160e-02 7.5306065e-02]\n [ 7.1989548e-20 4.4574543e-20 6.5149121e-21 -4.6291777e-20\n 8.8196718e-20 -2.3245078e-19 1.9459292e-19 4.0191465e-20\n 1.6273659e-19 2.2836562e-19]]\n" ] ], [ [ "You can visualize the embedding vectors using dimensional reduction techniques such as PCA.", "_____no_output_____" ] ], [ [ "import matplotlib.pyplot as plt\nfrom mpl_toolkits.mplot3d import Axes3D\nfrom sklearn.decomposition import PCA\n\npca = PCA(n_components = 3)\npca.fit(user_embeddings)\nuser_embeddings_pca = pca.transform(user_embeddings)\n\nfig = plt.figure(figsize = (8,8))\nax = fig.add_subplot(111, projection = \"3d\")\nxs, ys, zs = user_embeddings_pca[::150].T\nax.scatter(xs, ys, zs)", "_____no_output_____" ] ], [ [ "<pre>\n# Copyright 2018 Google Inc. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n</pre>", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ] ]

[ "BSD-2-Clause" ]

[ "BSD-2-Clause" ]

[ "BSD-2-Clause" ]

[ [ [ "<script async src=\"https://www.googletagmanager.com/gtag/js?id=UA-59152712-8\"></script>\n<script>\n window.dataLayer = window.dataLayer || [];\n function gtag(){dataLayer.push(arguments);}\n gtag('js', new Date());\n\n gtag('config', 'UA-59152712-8');\n</script>\n\n# Computing the 4-Velocity Time-Component $u^0$, the Magnetic Field Measured by a Comoving Observer $b^{\\mu}$, and the Poynting Vector $S^i$\n\n## Authors: Zach Etienne & Patrick Nelson\n\n[comment]: <> (Abstract: TODO)\n\n**Notebook Status:** <font color='green'><b> Validated </b></font>\n\n**Validation Notes:** This module has been validated against a trusted code (the hand-written smallbPoynET in WVUThorns_diagnostics, which itself is based on expressions in IllinoisGRMHD... which was validated against the original GRMHD code of the Illinois NR group)\n\n### NRPy+ Source Code for this module: [u0_smallb_Poynting__Cartesian.py](../edit/u0_smallb_Poynting__Cartesian/u0_smallb_Poynting__Cartesian.py)\n\n[comment]: <> (Introduction: TODO)", "_____no_output_____" ], [ "<a id='toc'></a>\n\n# Table of Contents\n$$\\label{toc}$$\n\nThis notebook is organized as follows\n\n1. [Step 1](#u0bu): Computing $u^0$ and $b^{\\mu}$\n 1. [Step 1.a](#4metric): Compute the 4-metric $g_{\\mu\\nu}$ and its inverse $g^{\\mu\\nu}$ from the ADM 3+1 variables, using the [`BSSN.ADMBSSN_tofrom_4metric`](../edit/BSSN/ADMBSSN_tofrom_4metric.py) ([**tutorial**](Tutorial-ADMBSSN_tofrom_4metric.ipynb)) NRPy+ module\n 1. [Step 1.b](#u0): Compute $u^0$ from the Valencia 3-velocity\n 1. [Step 1.c](#uj): Compute $u_j$ from $u^0$, the Valencia 3-velocity, and $g_{\\mu\\nu}$\n 1. [Step 1.d](#gamma): Compute $\\gamma=$ `gammaDET` from the ADM 3+1 variables\n 1. [Step 1.e](#beta): Compute $b^\\mu$\n1. [Step 2](#poynting_flux): Defining the Poynting Flux Vector $S^{i}$\n 1. [Step 2.a](#g): Computing $g^{i\\nu}$\n 1. [Step 2.b](#s): Computing $S^{i}$\n1. [Step 3](#code_validation): Code Validation against `u0_smallb_Poynting__Cartesian` NRPy+ module\n1. [Step 4](#appendix): Appendix: Proving Eqs. 53 and 56 in [Duez *et al* (2005)](https://arxiv.org/pdf/astro-ph/0503420.pdf)\n1. [Step 5](#latex_pdf_output): Output this notebook to $\\LaTeX$-formatted PDF file", "_____no_output_____" ], [ "<a id='u0bu'></a>\n\n# Step 1: Computing $u^0$ and $b^{\\mu}$ \\[Back to [top](#toc)\\]\n$$\\label{u0bu}$$\n\nFirst some definitions. The spatial components of $b^{\\mu}$ are simply the magnetic field as measured by an observer comoving with the plasma $B^{\\mu}_{\\rm (u)}$, divided by $\\sqrt{4\\pi}$. In addition, in the ideal MHD limit, $B^{\\mu}_{\\rm (u)}$ is orthogonal to the plasma 4-velocity $u^\\mu$, which sets the $\\mu=0$ component. \n\nNote also that $B^{\\mu}_{\\rm (u)}$ is related to the magnetic field as measured by a *normal* observer $B^i$ via a simple projection (Eq 21 in [Duez *et al* (2005)](https://arxiv.org/pdf/astro-ph/0503420.pdf)), which results in the expressions (Eqs 23 and 24 in [Duez *et al* (2005)](https://arxiv.org/pdf/astro-ph/0503420.pdf)):\n\n\\begin{align}\n\\sqrt{4\\pi} b^0 = B^0_{\\rm (u)} &= \\frac{u_j B^j}{\\alpha} \\\\\n\\sqrt{4\\pi} b^i = B^i_{\\rm (u)} &= \\frac{B^i + (u_j B^j) u^i}{\\alpha u^0}\\\\\n\\end{align}\n\n$B^i$ is related to the actual magnetic field evaluated in IllinoisGRMHD, $\\tilde{B}^i$ via\n\n$$B^i = \\frac{\\tilde{B}^i}{\\gamma},$$\n\nwhere $\\gamma$ is the determinant of the spatial 3-metric.\n\nThe above expressions will require that we compute\n1. the 4-metric $g_{\\mu\\nu}$ from the ADM 3+1 variables\n1. $u^0$ from the Valencia 3-velocity\n1. $u_j$ from $u^0$, the Valencia 3-velocity, and $g_{\\mu\\nu}$\n1. $\\gamma$ from the ADM 3+1 variables", "_____no_output_____" ], [ "<a id='4metric'></a>\n\n## Step 1.a: Compute the 4-metric $g_{\\mu\\nu}$ and its inverse $g^{\\mu\\nu}$ from the ADM 3+1 variables, using the [`BSSN.ADMBSSN_tofrom_4metric`](../edit/BSSN/ADMBSSN_tofrom_4metric.py) ([**tutorial**](Tutorial-ADMBSSN_tofrom_4metric.ipynb)) NRPy+ module \\[Back to [top](#toc)\\]\n$$\\label{4metric}$$\n\nWe are given $\\gamma_{ij}$, $\\alpha$, and $\\beta^i$ from ADMBase, so let's first compute \n\n$$\ng_{\\mu\\nu} = \\begin{pmatrix} \n-\\alpha^2 + \\beta^k \\beta_k & \\beta_i \\\\\n\\beta_j & \\gamma_{ij}\n\\end{pmatrix}.\n$$", "_____no_output_____" ] ], [ [ "# Step 1: Initialize needed Python/NRPy+ modules\nimport sympy as sp # SymPy: The Python computer algebra package upon which NRPy+ depends\nimport NRPy_param_funcs as par # NRPy+: Parameter interface\nimport indexedexp as ixp # NRPy+: Symbolic indexed expression (e.g., tensors, vectors, etc.) support\nimport reference_metric as rfm # NRPy+: Reference metric support\nfrom outputC import * # NRPy+: Basic C code output functionality\nimport BSSN.ADMBSSN_tofrom_4metric as AB4m # NRPy+: ADM/BSSN <-> 4-metric conversions\n\n# Set spatial dimension = 3\nDIM=3\n\nthismodule = \"smallbPoynET\"\n\n# Step 1.a: Compute the 4-metric $g_{\\mu\\nu}$ and its inverse \n# $g^{\\mu\\nu}$ from the ADM 3+1 variables, using the\n# BSSN.ADMBSSN_tofrom_4metric NRPy+ module\nimport BSSN.ADMBSSN_tofrom_4metric as AB4m\ngammaDD,betaU,alpha = AB4m.setup_ADM_quantities(\"ADM\")\nAB4m.g4DD_ito_BSSN_or_ADM(\"ADM\",gammaDD,betaU,alpha)\ng4DD = AB4m.g4DD\nAB4m.g4UU_ito_BSSN_or_ADM(\"ADM\",gammaDD,betaU,alpha)\ng4UU = AB4m.g4UU", "_____no_output_____" ] ], [ [ "<a id='u0'></a>\n\n## Step 1.b: Compute $u^0$ from the Valencia 3-velocity \\[Back to [top](#toc)\\]\n$$\\label{u0}$$\n\nAccording to Eqs. 9-11 of [the IllinoisGRMHD paper](https://arxiv.org/pdf/1501.07276.pdf), the Valencia 3-velocity $v^i_{(n)}$ is related to the 4-velocity $u^\\mu$ via\n\n\\begin{align}\n\\alpha v^i_{(n)} &= \\frac{u^i}{u^0} + \\beta^i \\\\\n\\implies u^i &= u^0 \\left(\\alpha v^i_{(n)} - \\beta^i\\right)\n\\end{align}\n\nDefining $v^i = \\frac{u^i}{u^0}$, we get\n\n$$v^i = \\alpha v^i_{(n)} - \\beta^i,$$\n\nand in terms of this variable we get\n\n\\begin{align}\ng_{00} \\left(u^0\\right)^2 + 2 g_{0i} u^0 u^i + g_{ij} u^i u^j &= \\left(u^0\\right)^2 \\left(g_{00} + 2 g_{0i} v^i + g_{ij} v^i v^j\\right)\\\\\n\\implies u^0 &= \\pm \\sqrt{\\frac{-1}{g_{00} + 2 g_{0i} v^i + g_{ij} v^i v^j}} \\\\\n&= \\pm \\sqrt{\\frac{-1}{(-\\alpha^2 + \\beta^2) + 2 \\beta_i v^i + \\gamma_{ij} v^i v^j}} \\\\\n&= \\pm \\sqrt{\\frac{1}{\\alpha^2 - \\gamma_{ij}\\left(\\beta^i + v^i\\right)\\left(\\beta^j + v^j\\right)}}\\\\\n&= \\pm \\sqrt{\\frac{1}{\\alpha^2 - \\alpha^2 \\gamma_{ij}v^i_{(n)}v^j_{(n)}}}\\\\\n&= \\pm \\frac{1}{\\alpha}\\sqrt{\\frac{1}{1 - \\gamma_{ij}v^i_{(n)}v^j_{(n)}}}\n\\end{align}\n\nGenerally speaking, numerical errors will occasionally drive expressions under the radical to either negative values or potentially enormous values (corresponding to enormous Lorentz factors). Thus a reliable approach for computing $u^0$ requires that we first rewrite the above expression in terms of the Lorentz factor squared: $\\Gamma^2=\\left(\\alpha u^0\\right)^2$:\n\\begin{align}\nu^0 &= \\pm \\frac{1}{\\alpha}\\sqrt{\\frac{1}{1 - \\gamma_{ij}v^i_{(n)}v^j_{(n)}}}\\\\\n\\implies \\left(\\alpha u^0\\right)^2 &= \\frac{1}{1 - \\gamma_{ij}v^i_{(n)}v^j_{(n)}} \\\\\n\\implies \\gamma_{ij}v^i_{(n)}v^j_{(n)} &= 1 - \\frac{1}{\\left(\\alpha u^0\\right)^2} \\\\\n&= 1 - \\frac{1}{\\Gamma^2}\n\\end{align}\n\nIn order for the bottom expression to hold true, the left-hand side must be between 0 and 1. Again, this is not guaranteed due to the appearance of numerical errors. In fact, a robust algorithm will not allow $\\Gamma^2$ to become too large (which might contribute greatly to the stress-energy of a given gridpoint), so let's define $\\Gamma_{\\rm max}$, the largest allowed Lorentz factor. \n\nThen our algorithm for computing $u^0$ is as follows:\n\nIf\n$$R=\\gamma_{ij}v^i_{(n)}v^j_{(n)}>1 - \\frac{1}{\\Gamma_{\\rm max}^2},$$ \nthen adjust the 3-velocity $v^i$ as follows:\n\n$$v^i_{(n)} = \\sqrt{\\frac{1 - \\frac{1}{\\Gamma_{\\rm max}^2}}{R}}v^i_{(n)}.$$\n\nAfter this rescaling, we are then guaranteed that if $R$ is recomputed, it will be set to its ceiling value $R=R_{\\rm max} = 1 - \\frac{1}{\\Gamma_{\\rm max}^2}$.\n\nThen, regardless of whether the ceiling on $R$ was applied, $u^0$ can be safely computed via\n\n$$\nu^0 = \\frac{1}{\\alpha \\sqrt{1-R}}.\n$$", "_____no_output_____" ] ], [ [ "ValenciavU = ixp.register_gridfunctions_for_single_rank1(\"AUX\",\"ValenciavU\",DIM=3)\n\n# Step 1: Compute R = 1 - 1/max(Gamma)\nR = sp.sympify(0)\nfor i in range(DIM):\n for j in range(DIM):\n R += gammaDD[i][j]*ValenciavU[i]*ValenciavU[j]\n\nGAMMA_SPEED_LIMIT = par.Cparameters(\"REAL\",thismodule,\"GAMMA_SPEED_LIMIT\",10.0) # Default value based on\n # IllinoisGRMHD.\n # GiRaFFE default = 2000.0\nRmax = 1 - 1/(GAMMA_SPEED_LIMIT*GAMMA_SPEED_LIMIT)\n\nrescaledValenciavU = ixp.zerorank1()\nfor i in range(DIM):\n rescaledValenciavU[i] = ValenciavU[i]*sp.sqrt(Rmax/R)\n\nrescaledu0 = 1/(alpha*sp.sqrt(1-Rmax))\nregularu0 = 1/(alpha*sp.sqrt(1-R))\n\ncomputeu0_Cfunction = \"\"\"\n/* Function for computing u^0 from Valencia 3-velocity. */\n/* Inputs: ValenciavU[], alpha, gammaDD[][], GAMMA_SPEED_LIMIT (C parameter) */\n/* Output: u0=u^0 and velocity-limited ValenciavU[] */\\n\\n\"\"\"\n\ncomputeu0_Cfunction += outputC([R,Rmax],[\"const double R\",\"const double Rmax\"],\"returnstring\",\n params=\"includebraces=False,CSE_varprefix=tmpR,outCverbose=False\")\n\ncomputeu0_Cfunction += \"if(R <= Rmax) \"\ncomputeu0_Cfunction += outputC(regularu0,\"u0\",\"returnstring\",\n params=\"includebraces=True,CSE_varprefix=tmpnorescale,outCverbose=False\")\ncomputeu0_Cfunction += \" else \"\ncomputeu0_Cfunction += outputC([rescaledValenciavU[0],rescaledValenciavU[1],rescaledValenciavU[2],rescaledu0],\n [\"ValenciavU0\",\"ValenciavU1\",\"ValenciavU2\",\"u0\"],\"returnstring\",\n params=\"includebraces=True,CSE_varprefix=tmprescale,outCverbose=False\")\n\nprint(computeu0_Cfunction)", "\n/* Function for computing u^0 from Valencia 3-velocity. */\n/* Inputs: ValenciavU[], alpha, gammaDD[][], GAMMA_SPEED_LIMIT (C parameter) */\n/* Output: u0=u^0 and velocity-limited ValenciavU[] */\n\nconst double tmpR0 = 2*ValenciavU0;\nconst double R = ((ValenciavU0)*(ValenciavU0))*gammaDD00 + ((ValenciavU1)*(ValenciavU1))*gammaDD11 + 2*ValenciavU1*ValenciavU2*gammaDD12 + ValenciavU1*gammaDD01*tmpR0 + ((ValenciavU2)*(ValenciavU2))*gammaDD22 + ValenciavU2*gammaDD02*tmpR0;\nconst double Rmax = 1 - 1/((GAMMA_SPEED_LIMIT)*(GAMMA_SPEED_LIMIT));\nif(R <= Rmax) {\n const double tmpnorescale0 = 2*ValenciavU0;\n u0 = 1/(alpha*sqrt(-((ValenciavU0)*(ValenciavU0))*gammaDD00 - ((ValenciavU1)*(ValenciavU1))*gammaDD11 - 2*ValenciavU1*ValenciavU2*gammaDD12 - ValenciavU1*gammaDD01*tmpnorescale0 - ((ValenciavU2)*(ValenciavU2))*gammaDD22 - ValenciavU2*gammaDD02*tmpnorescale0 + 1));\n}\n else {\n const double tmprescale0 = 2*ValenciavU0;\n const double tmprescale1 = sqrt((1 - 1/((GAMMA_SPEED_LIMIT)*(GAMMA_SPEED_LIMIT)))/(((ValenciavU0)*(ValenciavU0))*gammaDD00 + ((ValenciavU1)*(ValenciavU1))*gammaDD11 + 2*ValenciavU1*ValenciavU2*gammaDD12 + ValenciavU1*gammaDD01*tmprescale0 + ((ValenciavU2)*(ValenciavU2))*gammaDD22 + ValenciavU2*gammaDD02*tmprescale0));\n ValenciavU0 = ValenciavU0*tmprescale1;\n ValenciavU1 = ValenciavU1*tmprescale1;\n ValenciavU2 = ValenciavU2*tmprescale1;\n u0 = fabs(GAMMA_SPEED_LIMIT)/alpha;\n}\n\n" ] ], [ [ "<a id='uj'></a>\n\n## Step 1.c: Compute $u_j$ from $u^0$, the Valencia 3-velocity, and $g_{\\mu\\nu}$ \\[Back to [top](#toc)\\]\n$$\\label{uj}$$\n\nThe basic equation is\n\n\\begin{align}\nu_j &= g_{\\mu j} u^{\\mu} \\\\\n&= g_{0j} u^0 + g_{ij} u^i \\\\\n&= \\beta_j u^0 + \\gamma_{ij} u^i \\\\\n&= \\beta_j u^0 + \\gamma_{ij} u^0 \\left(\\alpha v^i_{(n)} - \\beta^i\\right) \\\\\n&= u^0 \\left(\\beta_j + \\gamma_{ij} \\left(\\alpha v^i_{(n)} - \\beta^i\\right) \\right)\\\\\n&= \\alpha u^0 \\gamma_{ij} v^i_{(n)} \\\\\n\\end{align}", "_____no_output_____" ] ], [ [ "u0 = par.Cparameters(\"REAL\",thismodule,\"u0\",1e300) # Will be overwritten in C code. Set to crazy value to ensure this.\n\nuD = ixp.zerorank1()\nfor i in range(DIM):\n for j in range(DIM):\n uD[j] += alpha*u0*gammaDD[i][j]*ValenciavU[i]", "_____no_output_____" ] ], [ [ "<a id='beta'></a>\n\n## Step 1.d: Compute $b^\\mu$ \\[Back to [top](#toc)\\]\n$$\\label{beta}$$\n\nWe compute $b^\\mu$ from the above expressions:\n\n\\begin{align}\n\\sqrt{4\\pi} b^0 = B^0_{\\rm (u)} &= \\frac{u_j B^j}{\\alpha} \\\\\n\\sqrt{4\\pi} b^i = B^i_{\\rm (u)} &= \\frac{B^i + (u_j B^j) u^i}{\\alpha u^0}\\\\\n\\end{align}\n\n$B^i$ is exactly equal to the $B^i$ evaluated in IllinoisGRMHD/GiRaFFE.\n\nPulling this together, we currently have available as input:\n+ $\\tilde{B}^i$\n+ $u_j$\n+ $u^0$,\n\nwith the goal of outputting now $b^\\mu$ and $b^2$:", "_____no_output_____" ] ], [ [ "M_PI = par.Cparameters(\"#define\",thismodule,\"M_PI\",\"\")\nBU = ixp.register_gridfunctions_for_single_rank1(\"AUX\",\"BU\",DIM=3)\n\n# uBcontraction = u_i B^i\nuBcontraction = sp.sympify(0)\nfor i in range(DIM):\n uBcontraction += uD[i]*BU[i]\n\n# uU = 3-vector representing u^i = u^0 \\left(\\alpha v^i_{(n)} - \\beta^i\\right)\nuU = ixp.zerorank1()\nfor i in range(DIM):\n uU[i] = u0*(alpha*ValenciavU[i] - betaU[i])\n\nsmallb4U = ixp.zerorank1(DIM=4)\nsmallb4U[0] = uBcontraction/(alpha*sp.sqrt(4*M_PI))\nfor i in range(DIM):\n smallb4U[1+i] = (BU[i] + uBcontraction*uU[i])/(alpha*u0*sp.sqrt(4*M_PI))", "_____no_output_____" ] ], [ [ "<a id='poynting_flux'></a>\n\n# Step 2: Defining the Poynting Flux Vector $S^{i}$ \\[Back to [top](#toc)\\]\n$$\\label{poynting_flux}$$\n\nThe Poynting flux is defined in Eq. 11 of [Kelly *et al.*](https://arxiv.org/pdf/1710.02132.pdf) (note that we choose the minus sign convention so that the Poynting luminosity across a spherical shell is $L_{\\rm EM} = \\int (-\\alpha T^i_{\\rm EM\\ 0}) \\sqrt{\\gamma} d\\Omega = \\int S^r \\sqrt{\\gamma} d\\Omega$, as in [Farris *et al.*](https://arxiv.org/pdf/1207.3354.pdf):\n\n$$\nS^i = -\\alpha T^i_{\\rm EM\\ 0} = -\\alpha\\left(b^2 u^i u_0 + \\frac{1}{2} b^2 g^i{}_0 - b^i b_0\\right)\n$$\n\n", "_____no_output_____" ], [ "<a id='s'></a>\n\n## Step 2.a: Computing $S^{i}$ \\[Back to [top](#toc)\\]\n$$\\label{s}$$\n\nGiven $g^{\\mu\\nu}$ computed above, we focus first on the $g^i{}_{0}$ term by computing \n$$\ng^\\mu{}_\\delta = g^{\\mu\\nu} g_{\\nu \\delta},\n$$\nand then the rest of the Poynting flux vector can be immediately computed from quantities defined above:\n$$\nS^i = -\\alpha T^i_{\\rm EM\\ 0} = -\\alpha\\left(b^2 u^i u_0 + \\frac{1}{2} b^2 g^i{}_0 - b^i b_0\\right)\n$$", "_____no_output_____" ] ], [ [ "# Step 2.a.i: compute g^\\mu_\\delta:\ng4UD = ixp.zerorank2(DIM=4)\nfor mu in range(4):\n for delta in range(4):\n for nu in range(4):\n g4UD[mu][delta] += g4UU[mu][nu]*g4DD[nu][delta]\n\n# Step 2.a.ii: compute b_{\\mu}\nsmallb4D = ixp.zerorank1(DIM=4)\nfor mu in range(4):\n for nu in range(4):\n smallb4D[mu] += g4DD[mu][nu]*smallb4U[nu]\n\n# Step 2.a.iii: compute u_0 = g_{mu 0} u^{mu} = g4DD[0][0]*u0 + g4DD[i][0]*uU[i]\nu_0 = g4DD[0][0]*u0\nfor i in range(DIM):\n u_0 += g4DD[i+1][0]*uU[i]\n \n# Step 2.a.iv: compute b^2, setting b^2 = smallb2etk, as gridfunctions with base names ending in a digit\n# are forbidden in NRPy+.\nsmallb2etk = sp.sympify(0)\nfor mu in range(4):\n smallb2etk += smallb4U[mu]*smallb4D[mu]\n\n# Step 2.a.v: compute S^i\nPoynSU = ixp.zerorank1()\nfor i in range(DIM):\n PoynSU[i] = -alpha * (smallb2etk*uU[i]*u_0 + sp.Rational(1,2)*smallb2etk*g4UD[i+1][0] - smallb4U[i+1]*smallb4D[0])", "_____no_output_____" ] ], [ [ "<a id='code_validation'></a>\n\n# Step 3: Code Validation against `u0_smallb_Poynting__Cartesian` NRPy+ module \\[Back to [top](#toc)\\]\n$$\\label{code_validation}$$\n\nHere, as a code validation check, we verify agreement in the SymPy expressions for u0, smallbU, smallb2etk, and PoynSU between\n\n1. this tutorial and \n2. the NRPy+ [u0_smallb_Poynting__Cartesian module](../edit/u0_smallb_Poynting__Cartesian/u0_smallb_Poynting__Cartesian.py).", "_____no_output_____" ] ], [ [ "import sys\nimport u0_smallb_Poynting__Cartesian.u0_smallb_Poynting__Cartesian as u0etc\nu0etc.compute_u0_smallb_Poynting__Cartesian(gammaDD,betaU,alpha,ValenciavU,BU)\n\nif u0etc.computeu0_Cfunction != computeu0_Cfunction:\n print(\"FAILURE: u0 C code has changed!\")\n sys.exit(1)\nelse:\n print(\"PASSED: u0 C code matches!\")\n\nfor i in range(4):\n print(\"u0etc.smallb4U[\"+str(i)+\"] - smallb4U[\"+str(i)+\"] = \" \n + str(u0etc.smallb4U[i]-smallb4U[i]))\n\nprint(\"u0etc.smallb2etk - smallb2etk = \" + str(u0etc.smallb2etk-smallb2etk))\n\nfor i in range(DIM):\n print(\"u0etc.PoynSU[\"+str(i)+\"] - PoynSU[\"+str(i)+\"] = \" \n + str(u0etc.PoynSU[i]-PoynSU[i]))", "PASSED: u0 C code matches!\nu0etc.smallb4U[0] - smallb4U[0] = 0\nu0etc.smallb4U[1] - smallb4U[1] = 0\nu0etc.smallb4U[2] - smallb4U[2] = 0\nu0etc.smallb4U[3] - smallb4U[3] = 0\nu0etc.smallb2etk - smallb2etk = 0\nu0etc.PoynSU[0] - PoynSU[0] = 0\nu0etc.PoynSU[1] - PoynSU[1] = 0\nu0etc.PoynSU[2] - PoynSU[2] = 0\n" ] ], [ [ "<a id='appendix'></a>\n\n# Step 4: Appendix: Proving Eqs. 53 and 56 in [Duez *et al* (2005)](https://arxiv.org/pdf/astro-ph/0503420.pdf)\n$$\\label{appendix}$$\n\n$u^\\mu u_\\mu = -1$ implies\n\n\\begin{align}\ng^{\\mu\\nu} u_\\mu u_\\nu &= g^{00} \\left(u_0\\right)^2 + 2 g^{0i} u_0 u_i + g^{ij} u_i u_j = -1 \\\\\n\\implies &g^{00} \\left(u_0\\right)^2 + 2 g^{0i} u_0 u_i + g^{ij} u_i u_j + 1 = 0\\\\\n& a x^2 + b x + c = 0\n\\end{align}\n\nThus we have a quadratic equation for $u_0$, with solution given by\n\n\\begin{align}\nu_0 &= \\frac{-b \\pm \\sqrt{b^2 - 4 a c}}{2 a} \\\\\n&= \\frac{-2 g^{0i}u_i \\pm \\sqrt{\\left(2 g^{0i} u_i\\right)^2 - 4 g^{00} (g^{ij} u_i u_j + 1)}}{2 g^{00}}\\\\\n&= \\frac{-g^{0i}u_i \\pm \\sqrt{\\left(g^{0i} u_i\\right)^2 - g^{00} (g^{ij} u_i u_j + 1)}}{g^{00}}\\\\\n\\end{align}\n\nNotice that (Eq. 4.49 in [Gourgoulhon](https://arxiv.org/pdf/gr-qc/0703035.pdf))\n$$\ng^{\\mu\\nu} = \\begin{pmatrix} \n-\\frac{1}{\\alpha^2} & \\frac{\\beta^i}{\\alpha^2} \\\\\n\\frac{\\beta^i}{\\alpha^2} & \\gamma^{ij} - \\frac{\\beta^i\\beta^j}{\\alpha^2}\n\\end{pmatrix},\n$$\nso we have\n\n\\begin{align}\nu_0 &= \\frac{-\\beta^i u_i/\\alpha^2 \\pm \\sqrt{\\left(\\beta^i u_i/\\alpha^2\\right)^2 + 1/\\alpha^2 (g^{ij} u_i u_j + 1)}}{1/\\alpha^2}\\\\\n&= -\\beta^i u_i \\pm \\sqrt{\\left(\\beta^i u_i\\right)^2 + \\alpha^2 (g^{ij} u_i u_j + 1)}\\\\\n&= -\\beta^i u_i \\pm \\sqrt{\\left(\\beta^i u_i\\right)^2 + \\alpha^2 \\left(\\left[\\gamma^{ij} - \\frac{\\beta^i\\beta^j}{\\alpha^2}\\right] u_i u_j + 1\\right)}\\\\\n&= -\\beta^i u_i \\pm \\sqrt{\\left(\\beta^i u_i\\right)^2 + \\alpha^2 \\left(\\gamma^{ij}u_i u_j + 1\\right) - \\beta^i\\beta^j u_i u_j}\\\\\n&= -\\beta^i u_i \\pm \\sqrt{\\alpha^2 \\left(\\gamma^{ij}u_i u_j + 1\\right)}\\\\\n\\end{align}\n\nNow, since \n\n$$\nu^0 = g^{\\alpha 0} u_\\alpha = -\\frac{1}{\\alpha^2} u_0 + \\frac{\\beta^i u_i}{\\alpha^2},\n$$\n\nwe get\n\n\\begin{align}\nu^0 &= \\frac{1}{\\alpha^2} \\left(u_0 + \\beta^i u_i\\right) \\\\\n&= \\pm \\frac{1}{\\alpha^2} \\sqrt{\\alpha^2 \\left(\\gamma^{ij}u_i u_j + 1\\right)}\\\\\n&= \\pm \\frac{1}{\\alpha} \\sqrt{\\gamma^{ij}u_i u_j + 1}\\\\\n\\end{align}\n\nBy convention, the relativistic Gamma factor is positive and given by $\\alpha u^0$, so we choose the positive root. Thus we have derived Eq. 53 in [Duez *et al* (2005)](https://arxiv.org/pdf/astro-ph/0503420.pdf):\n\n$$\nu^0 = \\frac{1}{\\alpha} \\sqrt{\\gamma^{ij}u_i u_j + 1}.\n$$\n\nNext we evaluate \n\n\\begin{align}\nu^i &= u_\\mu g^{\\mu i} \\\\\n&= u_0 g^{0 i} + u_j g^{i j}\\\\\n&= u_0 \\frac{\\beta^i}{\\alpha^2} + u_j \\left(\\gamma^{ij} - \\frac{\\beta^i\\beta^j}{\\alpha^2}\\right)\\\\\n&= \\gamma^{ij} u_j + u_0 \\frac{\\beta^i}{\\alpha^2} - u_j \\frac{\\beta^i\\beta^j}{\\alpha^2}\\\\\n&= \\gamma^{ij} u_j + \\frac{\\beta^i}{\\alpha^2} \\left(u_0 - u_j \\beta^j\\right)\\\\\n&= \\gamma^{ij} u_j - \\beta^i u^0,\\\\\n\\implies v^i &= \\frac{\\gamma^{ij} u_j}{u^0} - \\beta^i\n\\end{align}\n\nwhich is equivalent to Eq. 56 in [Duez *et al* (2005)](https://arxiv.org/pdf/astro-ph/0503420.pdf). Notice in the last step, we used the above definition of $u^0$.", "_____no_output_____" ], [ "<a id='latex_pdf_output'></a>\n\n# Step 5: Output this notebook to $\\LaTeX$-formatted PDF file \\[Back to [top](#toc)\\]\n$$\\label{latex_pdf_output}$$\n\nThe following code cell converts this Jupyter notebook into a proper, clickable $\\LaTeX$-formatted PDF file. After the cell is successfully run, the generated PDF may be found in the root NRPy+ tutorial directory, with filename\n[Tutorial-u0_smallb_Poynting-Cartesian.pdf](Tutorial-u0_smallb_Poynting-Cartesian.pdf) (Note that clicking on this link may not work; you may need to open the PDF file through another means.)", "_____no_output_____" ] ], [ [ "!jupyter nbconvert --to latex --template latex_nrpy_style.tplx --log-level='WARN' Tutorial-u0_smallb_Poynting-Cartesian.ipynb\n!pdflatex -interaction=batchmode Tutorial-u0_smallb_Poynting-Cartesian.tex\n!pdflatex -interaction=batchmode Tutorial-u0_smallb_Poynting-Cartesian.tex\n!pdflatex -interaction=batchmode Tutorial-u0_smallb_Poynting-Cartesian.tex\n!rm -f Tut*.out Tut*.aux Tut*.log", "[pandoc warning] Duplicate link reference `[comment]' \"source\" (line 22, column 1)\r\nThis is pdfTeX, Version 3.14159265-2.6-1.40.18 (TeX Live 2017/Debian) (preloaded format=pdflatex)\r\n restricted \\write18 enabled.\r\nentering extended mode\r\nThis is pdfTeX, Version 3.14159265-2.6-1.40.18 (TeX Live 2017/Debian) (preloaded format=pdflatex)\r\n restricted \\write18 enabled.\r\nentering extended mode\r\nThis is pdfTeX, Version 3.14159265-2.6-1.40.18 (TeX Live 2017/Debian) (preloaded format=pdflatex)\r\n restricted \\write18 enabled.\r\nentering extended mode\r\n" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# import re\n# import tensorflow as tf\n# from tensorflow.keras.preprocessing.text import text_to_word_sequence\n\n# tokens=text_to_word_sequence(\"manta.com/c/mmcdqky/lily-co\")\n\n# print(tokens)\n\n\n# #to map the features to a dictioanary and then convert it to a csv file.\n# # Feauture extraction \n# class feature_extractor(object):\n# def __init__(self,url):\n# self.url=url\n# self.length=len(url)\n# #self.domain=url.split('//')[-1].split('/')[0]\n# #def entropy(self):\n# #.com,.org,.net,.edu\n# #has www.\n# #.extension-- .htm,.html,.php,.js\n# # Pattern regex = Pattern.compile(\".com[,/.]\")\n# def domain(self):\n# if re.search(\".com[ .,/]\",self.url):\n# return 1\n# elif re.search(\".org[.,/]\",self.url):\n# return 2\n# elif re.search(\".net[.,/]\",self.url):\n# return 3\n# elif re.search(\".edu[.,/]\",self.url):\n# return 4\n# else:\n# return 0\n# #def extension(self):\n\n# def num_digits(self):\n# return sum(n.isdigit() for n in self.url)\n \n# def num_char(self):\n# return sum(n.alpha() for n in self.url)\n \n# def has_http(self):\n# if \"http\" in self.url:\n# return 1\n# else:\n# return 0\n \n# def has_https(self):\n# if \"https\" in self.url:\n# return 1\n# else:\n# return 0\n \n# #def num_special_char(self):\n# #\n \n# #def num\n\n\n\n# def clean(input):\n# tokensBySlash = str(input.encode('utf-8')).split('/')\n# allTokens=[]\n# for i in tokensBySlash:\n# tokens = str(i).split('-')\n# tokensByDot = []\n# for j in range(0,len(tokens)):\n# tempTokens = str(tokens[j]).split('.')\n# tokentsByDot = tokensByDot + tempTokens\n# allTokens = allTokens + tokens + tokensByDot\n# allTokens = list(set(allTokens))\n# if 'com' in allTokens:\n# allTokens.remove('com')\n# return allTokens", "['manta', 'com', 'c', 'mmcdqky', 'lily', 'co']\n" ], [ "from urllib.parse import urlparse\nurl=\"http://www.pn-wuppertal.de/links/2-linkseite/5-httpwwwkrebshilfede\"", "_____no_output_____" ], [ "def getTokens(input):\n tokensBySlash = str(input.encode('utf-8')).split('/')\n allTokens=[]\n for i in tokensBySlash:\n tokens = str(i).split('-')\n tokensByDot = []\n for j in range(0,len(tokens)):\n tempTokens = str(tokens[j]).split('.')\n tokentsByDot = tokensByDot + tempTokens\n allTokens = allTokens + tokens + tokensByDot\n allTokens = list(set(allTokens))\n if 'com' in allTokens:\n allTokens.remove('com')\n return allTokens", "_____no_output_____" ], [ "url=\"http://www.pn-wuppertal.de/links/2-linkseite/5-httpwwwkrebshilfede\"\nx=(lambda s: sum(not((i.isalpha()) and not(i.isnumeric())) for i in s))\nprint((url))", "13\n" ], [ "from urllib.parse import urlparse\nurl=\"http://www.pn-wuppertal.de/links/2-linkseite/5-httpwwwkrebshilfede\"\ndef fd_length(url):\n urlpath= urlparse(url).path\n try:\n return len(urlpath.split('/')[1])\n except:\n return 0\nprint(urlparse(url))\nprint(fd_length(urlparse(url)))", "ParseResult(scheme='http', netloc='www.pn-wuppertal.de', path='/links/2-linkseite/5-httpwwwkrebshilfede', params='', query='', fragment='')\n" ], [ "urlparse(url).scheme", "_____no_output_____" ], [ "s='https://www.yandex.ru'\nprint(urlparse(s))", "ParseResult(scheme='https', netloc='www.yandex.ru', path='', params='', query='', fragment='')\n" ], [ "s='yourbittorrent.com/?q=anthony-hamilton-soulife'\nprint(urlparse(s))\nprint(tldextract.extract(s))", "ParseResult(scheme='', netloc='', path='yourbittorrent.com/', params='', query='q=anthony-hamilton-soulife', fragment='')\nExtractResult(subdomain='', domain='yourbittorrent', suffix='com')\n" ], [ "from urllib.parse import urlparse\nimport tldextract\ns='movies.yahoo.com/shop?d=hv&cf=info&id=1800340831'\nprint(urlparse(s))\nprint(tldextract.extract(s).subdomain)", "ParseResult(scheme='', netloc='', path='movies.yahoo.com/shop', params='', query='d=hv&cf=info&id=1800340831', fragment='')\nmovies\n" ], [ "len(urlparse(s).query)", "_____no_output_____" ], [ "def tld_length(tld):\n try:\n return len(tld)\n except:\n return -1", "_____no_output_____" ], [ "import tldextract\n", "_____no_output_____" ], [ "from urllib.parse import urlparse\nimport tldextract\ns='http://peluqueriadeautor.com/index.php?option=com_virtuemart&page=shop.browse&category_id=31&Itemid=70'\ndef extension(s):\n domains={'com':1,'edu':2,'org':3,'net':4,'onion':5}\n if s in domains.keys():\n return domains[s]\n else:\n return 0\n#s=tldextract.extract(s).suffix\n#print(extension(s))\nprint(tldextract.extract(s))\nprint(urlparse(s))", "ExtractResult(subdomain='', domain='peluqueriadeautor', suffix='com')\nParseResult(scheme='http', netloc='peluqueriadeautor.com', path='/index.php', params='', query='option=com_virtuemart&page=shop.browse&category_id=31&Itemid=70', fragment='')\n" ], [ "from urllib.parse import urlparse\nimport tldextract\nprint(tldextract.extract(\"http://motthegioi.com/the-gioi-cuoi/clip-dai-gia-mac-ca-voi-co-ban-banh-my-185682.html\"))\nprint(urlparse(\"http://motthegioi.vn/the-gioi-cuoi/clip-dai-gia-mac-ca-voi-co-ban-banh-my-185682.html\"))", "ExtractResult(subdomain='', domain='motthegioi', suffix='com')\nParseResult(scheme='http', netloc='motthegioi.vn', path='/the-gioi-cuoi/clip-dai-gia-mac-ca-voi-co-ban-banh-my-185682.html', params='', query='', fragment='')\n" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "from sklearn.preprocessing import LabelBinarizer\n\nfrom keras.datasets import cifar10\nfrom keras.preprocessing.image import ImageDataGenerator\nfrom keras.models import Sequential, model_from_json\nfrom keras.callbacks import ModelCheckpoint, EarlyStopping, ReduceLROnPlateau\nfrom keras.constraints import maxnorm\nfrom keras import regularizers\nfrom keras.layers.normalization import BatchNormalization\nfrom keras.layers import Dense, Dropout, Activation, Flatten\nfrom keras.layers import Conv2D, MaxPooling2D\n\nfrom keras.applications import imagenet_utils\nfrom keras.preprocessing.image import img_to_array\n\nimport numpy as np \nimport json\nimport os\nimport cv2\nimport h5py\n\nimport matplotlib.pyplot as plt\n%matplotlib inline", "Using TensorFlow backend.\n" ], [ "from helpers import TrainingMonitor\nfrom helpers import Utils", "_____no_output_____" ], [ "output_path = \"../output/\"", "_____no_output_____" ], [ "import tensorflow as tf\nconfig = tf.ConfigProto()\nconfig.gpu_options.allow_growth = True\nsession = tf.Session(config=config)", "_____no_output_____" ], [ "import os\nos.environ[\"PATH\"] += os.pathsep + 'C:/Program Files (x86)/Graphviz2.38/bin/'", "_____no_output_____" ], [ "(x_train, y_train), (x_test, y_test) = cifar10.load_data()\n\nx_train = x_train.astype('float32')\nx_test = x_test.astype('float32')\n\nmean = np.mean(x_train, axis=0)\nx_train -= mean\nx_test -= mean\n\nlb = LabelBinarizer()\ny_train = lb.fit_transform(y_train) \ny_test = lb.fit_transform(y_test)", "_____no_output_____" ], [ "db_train = h5py.File(\"../input/datasets/cifar_rgbmean_train.hdf5\")\ndb_test = h5py.File(\"../input/datasets/cifar_rgbmean_test.hdf5\")\n\nx_train_rgbmean = db_train[\"images\"][:].astype('float32')\nx_test_rgbmean = db_test[\"images\"][:].astype('float32')\n\nmean = np.mean(x_train_rgbmean, axis=0)\nx_train_rgbmean -= mean\nx_test_rgbmean -= mean\n\ny_train_rgbmean = db_train[\"labels\"][:]\ny_test_rgbmean = db_test[\"labels\"][:]", "_____no_output_____" ], [ "json_file = open(output_path + 'saved/vgg_base_model_86.03.json', 'r')\nmodel_json = json_file.read()\njson_file.close()\nmodel = model_from_json(model_json)\nmodel.load_weights(output_path + \"saved/vgg_base_weight_86.03.hdf5\")", "_____no_output_____" ], [ "model.summary()", "_________________________________________________________________\nLayer (type) Output Shape Param # \n=================================================================\nconv2d_1 (Conv2D) (None, 32, 32, 32) 896 \n_________________________________________________________________\nbatch_normalization_1 (Batch (None, 32, 32, 32) 128 \n_________________________________________________________________\nconv2d_2 (Conv2D) (None, 32, 32, 32) 9248 \n_________________________________________________________________\nbatch_normalization_2 (Batch (None, 32, 32, 32) 128 \n_________________________________________________________________\nmax_pooling2d_1 (MaxPooling2 (None, 16, 16, 32) 0 \n_________________________________________________________________\ndropout_1 (Dropout) (None, 16, 16, 32) 0 \n_________________________________________________________________\nconv2d_3 (Conv2D) (None, 16, 16, 64) 18496 \n_________________________________________________________________\nbatch_normalization_3 (Batch (None, 16, 16, 64) 256 \n_________________________________________________________________\nconv2d_4 (Conv2D) (None, 16, 16, 64) 36928 \n_________________________________________________________________\nbatch_normalization_4 (Batch (None, 16, 16, 64) 256 \n_________________________________________________________________\nmax_pooling2d_2 (MaxPooling2 (None, 8, 8, 64) 0 \n_________________________________________________________________\ndropout_2 (Dropout) (None, 8, 8, 64) 0 \n_________________________________________________________________\nconv2d_5 (Conv2D) (None, 8, 8, 64) 36928 \n_________________________________________________________________\nbatch_normalization_5 (Batch (None, 8, 8, 64) 256 \n_________________________________________________________________\nconv2d_6 (Conv2D) (None, 8, 8, 64) 36928 \n_________________________________________________________________\nbatch_normalization_6 (Batch (None, 8, 8, 64) 256 \n_________________________________________________________________\nmax_pooling2d_3 (MaxPooling2 (None, 4, 4, 64) 0 \n_________________________________________________________________\ndropout_3 (Dropout) (None, 4, 4, 64) 0 \n_________________________________________________________________\nflatten_1 (Flatten) (None, 1024) 0 \n_________________________________________________________________\ndense_1 (Dense) (None, 512) 524800 \n_________________________________________________________________\nbatch_normalization_7 (Batch (None, 512) 2048 \n_________________________________________________________________\ndense_2 (Dense) (None, 512) 262656 \n_________________________________________________________________\nbatch_normalization_8 (Batch (None, 512) 2048 \n_________________________________________________________________\ndense_3 (Dense) (None, 10) 5130 \n=================================================================\nTotal params: 937,386\nTrainable params: 934,698\nNon-trainable params: 2,688\n_________________________________________________________________\n" ], [ "for (i, layer) in enumerate(model.layers): \n print(\"{}\\t{}\".format(i, layer.__class__.__name__))", "0\tConv2D\n1\tBatchNormalization\n2\tConv2D\n3\tBatchNormalization\n4\tMaxPooling2D\n5\tDropout\n6\tConv2D\n7\tBatchNormalization\n8\tConv2D\n9\tBatchNormalization\n10\tMaxPooling2D\n11\tDropout\n12\tConv2D\n13\tBatchNormalization\n14\tConv2D\n15\tBatchNormalization\n16\tMaxPooling2D\n17\tDropout\n18\tFlatten\n19\tDense\n20\tBatchNormalization\n21\tDense\n22\tBatchNormalization\n23\tDense\n" ], [ "from keras.utils import plot_model\nplot_model(model, to_file='models/baseline-vgg.png', show_shapes=True, show_layer_names=True)", "_____no_output_____" ], [ "model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])", "_____no_output_____" ], [ "filepath=output_path + \"progress/weights-{val_acc:.4f}.hdf5\"\nMC = ModelCheckpoint(filepath, monitor='val_acc', verbose=0, save_best_only=True, mode='max')\n\nfigPath = os.path.sep.join([output_path, \"monitor/{}.png\".format(os.getpid())])\njsonPath = os.path.sep.join([output_path, \"monitor/{}.json\".format(os.getpid())])\nTM = TrainingMonitor(figPath, jsonPath=jsonPath, startAt=0)\n\nRLR = ReduceLROnPlateau(factor=np.sqrt(0.1), cooldown=0, patience=5, min_lr=0.5e-6)\n\ncallbacks = [MC, TM, RLR]", "_____no_output_____" ], [ "history = model.fit(x_train_rgbmean, y_train_rgbmean,\n batch_size=64,\n epochs=1,\n validation_split=0.33,\n shuffle=\"batch\",\n callbacks=callbacks)", "Train on 33500 samples, validate on 16500 samples\nEpoch 1/1\n33500/33500 [==============================] - 15s 438us/step - loss: 0.1268 - acc: 0.9597 - val_loss: 0.1164 - val_acc: 0.9643\n" ], [ "scores = model.evaluate(x_test_rgbmean, y_test_rgbmean, verbose=0)\nprint(\"Train: %.2f%%; Val: %.2f%%; Test: %.2f%%\" % \n (np.max(history.history['acc'])*100, np.max(history.history['val_acc'])*100, scores[1]*100)\n )", "Train: 95.85%; Val: 96.32%; Test: 96.30%\n" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "hash('python')", "_____no_output_____" ], [ "hash([12, 5, 7])", "_____no_output_____" ], [ "hash((12, 5, 7))", "_____no_output_____" ], [ "hash({12, 5, 7})", "_____no_output_____" ], [ "{(12, 4, 6)}", "_____no_output_____" ], [ "{[12, 4, 6]}", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/jeffheaton/t81_558_deep_learning/blob/master/t81_558_class_04_1_feature_encode.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ], [ "# T81-558: Applications of Deep Neural Networks\n**Module 4: Training for Tabular Data**\n* Instructor: [Jeff Heaton](https://sites.wustl.edu/jeffheaton/), McKelvey School of Engineering, [Washington University in St. Louis](https://engineering.wustl.edu/Programs/Pages/default.aspx)\n* For more information visit the [class website](https://sites.wustl.edu/jeffheaton/t81-558/).", "_____no_output_____" ], [ "# Module 4 Material\n\n* **Part 4.1: Encoding a Feature Vector for Keras Deep Learning** [[Video]](https://www.youtube.com/watch?v=Vxz-gfs9nMQ&list=PLjy4p-07OYzulelvJ5KVaT2pDlxivl_BN) [[Notebook]](https://github.com/jeffheaton/t81_558_deep_learning/blob/master/t81_558_class_04_1_feature_encode.ipynb)\n* Part 4.2: Keras Multiclass Classification for Deep Neural Networks with ROC and AUC [[Video]](https://www.youtube.com/watch?v=-f3bg9dLMks&list=PLjy4p-07OYzulelvJ5KVaT2pDlxivl_BN) [[Notebook]](https://github.com/jeffheaton/t81_558_deep_learning/blob/master/t81_558_class_04_2_multi_class.ipynb)\n* Part 4.3: Keras Regression for Deep Neural Networks with RMSE [[Video]](https://www.youtube.com/watch?v=wNhBUC6X5-E&list=PLjy4p-07OYzulelvJ5KVaT2pDlxivl_BN) [[Notebook]](https://github.com/jeffheaton/t81_558_deep_learning/blob/master/t81_558_class_04_3_regression.ipynb)\n* Part 4.4: Backpropagation, Nesterov Momentum, and ADAM Neural Network Training [[Video]](https://www.youtube.com/watch?v=VbDg8aBgpck&list=PLjy4p-07OYzulelvJ5KVaT2pDlxivl_BN) [[Notebook]](https://github.com/jeffheaton/t81_558_deep_learning/blob/master/t81_558_class_04_4_backprop.ipynb)\n* Part 4.5: Neural Network RMSE and Log Loss Error Calculation from Scratch [[Video]](https://www.youtube.com/watch?v=wmQX1t2PHJc&list=PLjy4p-07OYzulelvJ5KVaT2pDlxivl_BN) [[Notebook]](https://github.com/jeffheaton/t81_558_deep_learning/blob/master/t81_558_class_04_5_rmse_logloss.ipynb)", "_____no_output_____" ], [ "# Google CoLab Instructions\n\nThe following code ensures that Google CoLab is running the correct version of TensorFlow.", "_____no_output_____" ] ], [ [ "try:\n %tensorflow_version 2.x\n COLAB = True\n print(\"Note: using Google CoLab\")\nexcept:\n print(\"Note: not using Google CoLab\")\n COLAB = False", "Note: not using Google CoLab\n" ] ], [ [ "# Part 4.1: Encoding a Feature Vector for Keras Deep Learning\n\nNeural networks can accept many types of data. We will begin with tabular data, where there are well defined rows and columns. This is the sort of data you would typically see in Microsoft Excel. An example of tabular data is shown below.\n\nNeural networks require numeric input. This numeric form is called a feature vector. Each row of training data typically becomes one vector. The individual input neurons each receive one feature (or column) from this vector. In this section, we will see how to encode the following tabular data into a feature vector.", "_____no_output_____" ] ], [ [ "import pandas as pd\n\npd.set_option('display.max_columns', 7) \npd.set_option('display.max_rows', 5)\n\ndf = pd.read_csv(\n \"https://data.heatonresearch.com/data/t81-558/jh-simple-dataset.csv\",\n na_values=['NA','?'])\n\npd.set_option('display.max_columns', 9)\npd.set_option('display.max_rows', 5)\n\ndisplay(df)", "_____no_output_____" ] ], [ [ "The following observations can be made from the above data:\n* The target column is the column that you seek to predict. There are several candidates here. However, we will initially use product. This field specifies what product someone bought.\n* There is an ID column. This column should not be fed into the neural network as it contains no information useful for prediction.\n* Many of these fields are numeric and might not require any further processing.\n* The income column does have some missing values.\n* There are categorical values: job, area, and product.\n\nTo begin with, we will convert the job code into dummy variables.", "_____no_output_____" ] ], [ [ "pd.set_option('display.max_columns', 7) \npd.set_option('display.max_rows', 5)\n\ndummies = pd.get_dummies(df['job'],prefix=\"job\")\nprint(dummies.shape)\n\npd.set_option('display.max_columns', 9)\npd.set_option('display.max_rows', 10)\n\ndisplay(dummies)", "(2000, 33)\n" ] ], [ [ "Because there are 33 different job codes, there are 33 dummy variables. We also specified a prefix, because the job codes (such as \"ax\") are not that meaningful by themselves. Something such as \"job_ax\" also tells us the origin of this field.\n\nNext, we must merge these dummies back into the main data frame. We also drop the original \"job\" field, as it is now represented by the dummies. ", "_____no_output_____" ] ], [ [ "pd.set_option('display.max_columns', 7) \npd.set_option('display.max_rows', 5)\n\ndf = pd.concat([df,dummies],axis=1)\ndf.drop('job', axis=1, inplace=True)\n\npd.set_option('display.max_columns', 9)\npd.set_option('display.max_rows', 10)\n\ndisplay(df)", "_____no_output_____" ] ], [ [ "We also introduce dummy variables for the area column.", "_____no_output_____" ] ], [ [ "pd.set_option('display.max_columns', 7) \npd.set_option('display.max_rows', 5)\n\ndf = pd.concat([df,pd.get_dummies(df['area'],prefix=\"area\")],axis=1)\ndf.drop('area', axis=1, inplace=True)\n\npd.set_option('display.max_columns', 9)\npd.set_option('display.max_rows', 10)\ndisplay(df)", "_____no_output_____" ] ], [ [ "The last remaining transformation is to fill in missing income values. ", "_____no_output_____" ] ], [ [ "med = df['income'].median()\ndf['income'] = df['income'].fillna(med)", "_____no_output_____" ] ], [ [ "There are more advanced ways of filling in missing values, but they require more analysis. The idea would be to see if another field might give a hint as to what the income were. For example, it might be beneficial to calculate a median income for each of the areas or job categories. This is something to keep in mind for the class Kaggle competition.\n\nAt this point, the Pandas dataframe is ready to be converted to Numpy for neural network training. We need to know a list of the columns that will make up *x* (the predictors or inputs) and *y* (the target). \n\nThe complete list of columns is:", "_____no_output_____" ] ], [ [ "print(list(df.columns))", "['id', 'income', 'aspect', 'subscriptions', 'dist_healthy', 'save_rate', 'dist_unhealthy', 'age', 'pop_dense', 'retail_dense', 'crime', 'product', 'job_11', 'job_al', 'job_am', 'job_ax', 'job_bf', 'job_by', 'job_cv', 'job_de', 'job_dz', 'job_e2', 'job_f8', 'job_gj', 'job_gv', 'job_kd', 'job_ke', 'job_kl', 'job_kp', 'job_ks', 'job_kw', 'job_mm', 'job_nb', 'job_nn', 'job_ob', 'job_pe', 'job_po', 'job_pq', 'job_pz', 'job_qp', 'job_qw', 'job_rn', 'job_sa', 'job_vv', 'job_zz', 'area_a', 'area_b', 'area_c', 'area_d']\n" ] ], [ [ "This includes both the target and predictors. We need a list with the target removed. We also remove **id** because it is not useful for prediction.", "_____no_output_____" ] ], [ [ "x_columns = df.columns.drop('product').drop('id')\nprint(list(x_columns))", "['income', 'aspect', 'subscriptions', 'dist_healthy', 'save_rate', 'dist_unhealthy', 'age', 'pop_dense', 'retail_dense', 'crime', 'job_11', 'job_al', 'job_am', 'job_ax', 'job_bf', 'job_by', 'job_cv', 'job_de', 'job_dz', 'job_e2', 'job_f8', 'job_gj', 'job_gv', 'job_kd', 'job_ke', 'job_kl', 'job_kp', 'job_ks', 'job_kw', 'job_mm', 'job_nb', 'job_nn', 'job_ob', 'job_pe', 'job_po', 'job_pq', 'job_pz', 'job_qp', 'job_qw', 'job_rn', 'job_sa', 'job_vv', 'job_zz', 'area_a', 'area_b', 'area_c', 'area_d']\n" ] ], [ [ "### Generate X and Y for a Classification Neural Network", "_____no_output_____" ], [ "We can now generate *x* and *y*. Note, this is how we generate y for a classification problem. Regression would not use dummies and would simply encode the numeric value of the target.", "_____no_output_____" ] ], [ [ "# Convert to numpy - Classification\nx_columns = df.columns.drop('product').drop('id')\nx = df[x_columns].values\ndummies = pd.get_dummies(df['product']) # Classification\nproducts = dummies.columns\ny = dummies.values", "_____no_output_____" ] ], [ [ "We can display the *x* and *y* matrices.", "_____no_output_____" ] ], [ [ "print(x)\nprint(y)", "[[5.08760000e+04 1.31000000e+01 1.00000000e+00 ... 0.00000000e+00\n 1.00000000e+00 0.00000000e+00]\n [6.03690000e+04 1.86250000e+01 2.00000000e+00 ... 0.00000000e+00\n 1.00000000e+00 0.00000000e+00]\n [5.51260000e+04 3.47666667e+01 1.00000000e+00 ... 0.00000000e+00\n 1.00000000e+00 0.00000000e+00]\n ...\n [2.85950000e+04 3.94250000e+01 3.00000000e+00 ... 0.00000000e+00\n 0.00000000e+00 1.00000000e+00]\n [6.79490000e+04 5.73333333e+00 0.00000000e+00 ... 0.00000000e+00\n 1.00000000e+00 0.00000000e+00]\n [6.14670000e+04 1.68916667e+01 0.00000000e+00 ... 0.00000000e+00\n 1.00000000e+00 0.00000000e+00]]\n[[0 1 0 ... 0 0 0]\n [0 0 1 ... 0 0 0]\n [0 1 0 ... 0 0 0]\n ...\n [0 0 0 ... 0 1 0]\n [0 0 1 ... 0 0 0]\n [0 0 1 ... 0 0 0]]\n" ] ], [ [ "The x and y values are now ready for a neural network. Make sure that you construct the neural network for a classification problem. Specifically,\n\n* Classification neural networks have an output neuron count equal to the number of classes.\n* Classification neural networks should use **categorical_crossentropy** and a **softmax** activation function on the output layer.", "_____no_output_____" ], [ "### Generate X and Y for a Regression Neural Network\n\nFor a regression neural network, the *x* values are generated the same. However, *y* does not use dummies. Make sure to replace **income** with your actual target.", "_____no_output_____" ] ], [ [ "y = df['income'].values", "_____no_output_____" ] ], [ [ "# Module 4 Assignment\n\nYou can find the first assignment here: [assignment 4](https://github.com/jeffheaton/t81_558_deep_learning/blob/master/assignments/assignment_yourname_class1.ipynb)", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown", "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Question 2\n\nColin has just joined hogwarts and brought a muggle camera with him, He is very\nexcited about all the magical things around him and wants to take many photos. But\nthe problem, it has limited storage of 500 MB. You, as his friend, have to help him by\ndoing the following task. Calculate the maximum dimensions of one image if colin wants\nto take 200 images back with him when he goes home for summer vacation. Assume all\nimage shapes to be squares.\nColin is very curious and expects explanations for all the steps you take for the above\ntask, so explain each step with theoretical details.", "_____no_output_____" ], [ "# Answer:", "_____no_output_____" ], [ "Given,<br>\n Storage limit = 500mb <br>\n Total Images = 200<br>\n No. of channels of RGB = 3<br>\n So,<br>\n <center>500x1024x1024 Bytes = 200.$x^{2}$.3</center>\n <center>2.5x1024x1024/3 = $x^{2}$</center>\n <center>x = 934.779831475</center>\n <center>x = 934 approx to lower to accomodate all channels</center>", "_____no_output_____" ], [ "# Question 3\nCCD Sensor dim = 10x10mm<br>\nPixels = 1024 x 1024<br>\nFocal length = 43.5cm<br>\nTarget height = 390pixels <br>\nMinimum distance = 240m<br>\nFind height of creature.", "_____no_output_____" ], [ "# Answer\n\n1 pixel height = 10/1024mm<br>\nSo, <br>\n <center>Image Height(i) = 390*(10/1024)mm<center>\n <center>Image forms at focal length -> v = 43.5cm = 430mm<center>\n <center>Height of creature = x mm <center>\n <center>u = Object distance = 240m = 240*(1000)mm<center>\n <center>object height = h<center>\n <center>v/u = i/h<center>\n <center>430/(240*1000) = 390*(10/1024)/h<center>\n <center>h = 2125.726744186mm -> <b>h = 2.1257267441860002m<b><center>\n \n \n", "_____no_output_____" ] ] ]

[ "markdown" ]

[ [ "markdown", "markdown", "markdown", "markdown", "markdown" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "**This notebook is an exercise in the [Intermediate Machine Learning](https://www.kaggle.com/learn/intermediate-machine-learning) course. You can reference the tutorial at [this link](https://www.kaggle.com/alexisbcook/categorical-variables).**\n\n---\n", "_____no_output_____" ], [ "By encoding **categorical variables**, you'll obtain your best results thus far!\n\n# Setup\n\nThe questions below will give you feedback on your work. Run the following cell to set up the feedback system.", "_____no_output_____" ] ], [ [ "# Set up code checking\nimport os\nif not os.path.exists(\"../input/train.csv\"):\n os.symlink(\"../input/home-data-for-ml-course/train.csv\", \"../input/train.csv\") \n os.symlink(\"../input/home-data-for-ml-course/test.csv\", \"../input/test.csv\") \nfrom learntools.core import binder\nbinder.bind(globals())\nfrom learntools.ml_intermediate.ex3 import *\nprint(\"Setup Complete\")", "Setup Complete\n" ] ], [ [ "In this exercise, you will work with data from the [Housing Prices Competition for Kaggle Learn Users](https://www.kaggle.com/c/home-data-for-ml-course). \n\n\n\nRun the next code cell without changes to load the training and validation sets in `X_train`, `X_valid`, `y_train`, and `y_valid`. The test set is loaded in `X_test`.", "_____no_output_____" ] ], [ [ "import pandas as pd\nfrom sklearn.model_selection import train_test_split\n\n# Read the data\nX = pd.read_csv('../input/train.csv', index_col='Id') \nX_test = pd.read_csv('../input/test.csv', index_col='Id')\n\n# Remove rows with missing target, separate target from predictors\nX.dropna(axis=0, subset=['SalePrice'], inplace=True)\ny = X.SalePrice\nX.drop(['SalePrice'], axis=1, inplace=True)\n\n# To keep things simple, we'll drop columns with missing values\ncols_with_missing = [col for col in X.columns if X[col].isnull().any()] \nX.drop(cols_with_missing, axis=1, inplace=True)\nX_test.drop(cols_with_missing, axis=1, inplace=True)\n\n# Break off validation set from training data\nX_train, X_valid, y_train, y_valid = train_test_split(X, y,\n train_size=0.8, test_size=0.2,\n random_state=0)", "_____no_output_____" ] ], [ [ "Use the next code cell to print the first five rows of the data.", "_____no_output_____" ] ], [ [ "X_train.head()", "_____no_output_____" ] ], [ [ "Notice that the dataset contains both numerical and categorical variables. You'll need to encode the categorical data before training a model.\n\nTo compare different models, you'll use the same `score_dataset()` function from the tutorial. This function reports the [mean absolute error](https://en.wikipedia.org/wiki/Mean_absolute_error) (MAE) from a random forest model.", "_____no_output_____" ] ], [ [ "from sklearn.ensemble import RandomForestRegressor\nfrom sklearn.metrics import mean_absolute_error\n\n# function for comparing different approaches\ndef score_dataset(X_train, X_valid, y_train, y_valid):\n model = RandomForestRegressor(n_estimators=100, random_state=0)\n model.fit(X_train, y_train)\n preds = model.predict(X_valid)\n return mean_absolute_error(y_valid, preds)", "_____no_output_____" ] ], [ [ "# Step 1: Drop columns with categorical data\n\nYou'll get started with the most straightforward approach. Use the code cell below to preprocess the data in `X_train` and `X_valid` to remove columns with categorical data. Set the preprocessed DataFrames to `drop_X_train` and `drop_X_valid`, respectively. ", "_____no_output_____" ] ], [ [ "# Fill in the lines below: drop columns in training and validation data\ndrop_X_train = X_train.select_dtypes(exclude=['object'])\ndrop_X_valid = X_valid.select_dtypes(exclude=['object'])\n\n# Check your answers\nstep_1.check()", "_____no_output_____" ], [ "# Lines below will give you a hint or solution code\n#step_1.hint()\n#step_1.solution()", "_____no_output_____" ] ], [ [ "Run the next code cell to get the MAE for this approach.", "_____no_output_____" ] ], [ [ "print(\"MAE from Approach 1 (Drop categorical variables):\")\nprint(score_dataset(drop_X_train, drop_X_valid, y_train, y_valid))", "MAE from Approach 1 (Drop categorical variables):\n17837.82570776256\n" ] ], [ [ "Before jumping into label encoding, we'll investigate the dataset. Specifically, we'll look at the `'Condition2'` column. The code cell below prints the unique entries in both the training and validation sets.", "_____no_output_____" ] ], [ [ "print(\"Unique values in 'Condition2' column in training data:\", X_train['Condition2'].unique())\nprint(\"\\nUnique values in 'Condition2' column in validation data:\", X_valid['Condition2'].unique())", "Unique values in 'Condition2' column in training data: ['Norm' 'PosA' 'Feedr' 'PosN' 'Artery' 'RRAe']\n\nUnique values in 'Condition2' column in validation data: ['Norm' 'RRAn' 'RRNn' 'Artery' 'Feedr' 'PosN']\n" ] ], [ [ "# Step 2: Label encoding\n\n### Part A\n\nIf you now write code to: \n- fit a label encoder to the training data, and then \n- use it to transform both the training and validation data, \n\nyou'll get an error. Can you see why this is the case? (_You'll need to use the above output to answer this question._)", "_____no_output_____" ] ], [ [ "# Check your answer (Run this code cell to receive credit!)\nstep_2.a.check()", "_____no_output_____" ], [ "#step_2.a.hint()", "_____no_output_____" ] ], [ [ "This is a common problem that you'll encounter with real-world data, and there are many approaches to fixing this issue. For instance, you can write a custom label encoder to deal with new categories. The simplest approach, however, is to drop the problematic categorical columns. \n\nRun the code cell below to save the problematic columns to a Python list `bad_label_cols`. Likewise, columns that can be safely label encoded are stored in `good_label_cols`.", "_____no_output_____" ] ], [ [ "# All categorical columns\nobject_cols = [col for col in X_train.columns if X_train[col].dtype == \"object\"]\n\n# Columns that can be safely label encoded\ngood_label_cols = [col for col in object_cols if \n set(X_train[col]) == set(X_valid[col])]\n \n# Problematic columns that will be dropped from the dataset\nbad_label_cols = list(set(object_cols)-set(good_label_cols))\n \nprint('Categorical columns that will be label encoded:', good_label_cols)\nprint('\\nCategorical columns that will be dropped from the dataset:', bad_label_cols)", "Categorical columns that will be label encoded: ['MSZoning', 'Street', 'LotShape', 'LandContour', 'LotConfig', 'BldgType', 'HouseStyle', 'ExterQual', 'CentralAir', 'KitchenQual', 'PavedDrive', 'SaleCondition']\n\nCategorical columns that will be dropped from the dataset: ['Neighborhood', 'LandSlope', 'Condition1', 'Heating', 'Foundation', 'RoofMatl', 'Condition2', 'RoofStyle', 'ExterCond', 'Exterior1st', 'Utilities', 'Functional', 'HeatingQC', 'SaleType', 'Exterior2nd']\n" ] ], [ [ "### Part B\n\nUse the next code cell to label encode the data in `X_train` and `X_valid`. Set the preprocessed DataFrames to `label_X_train` and `label_X_valid`, respectively. \n- We have provided code below to drop the categorical columns in `bad_label_cols` from the dataset. \n- You should label encode the categorical columns in `good_label_cols`. ", "_____no_output_____" ] ], [ [ "from sklearn.preprocessing import LabelEncoder\n\n# Drop categorical columns that will not be encoded\nlabel_X_train = X_train.drop(bad_label_cols, axis=1)\nlabel_X_valid = X_valid.drop(bad_label_cols, axis=1)\n\n# Apply label encoder \nlabel_encoder = LabelEncoder()\nfor col in good_label_cols:\n label_X_train[col] = label_encoder.fit_transform(label_X_train[col])\n label_X_valid[col] = label_encoder.transform(label_X_valid[col])\n \n# Check your answer\nstep_2.b.check()", "_____no_output_____" ], [ "# Lines below will give you a hint or solution code\n#step_2.b.hint()\n#step_2.b.solution()", "_____no_output_____" ] ], [ [ "Run the next code cell to get the MAE for this approach.", "_____no_output_____" ] ], [ [ "print(\"MAE from Approach 2 (Label Encoding):\") \nprint(score_dataset(label_X_train, label_X_valid, y_train, y_valid))", "MAE from Approach 2 (Label Encoding):\n17575.291883561644\n" ] ], [ [ "So far, you've tried two different approaches to dealing with categorical variables. And, you've seen that encoding categorical data yields better results than removing columns from the dataset.\n\nSoon, you'll try one-hot encoding. Before then, there's one additional topic we need to cover. Begin by running the next code cell without changes. ", "_____no_output_____" ] ], [ [ "# Get number of unique entries in each column with categorical data\nobject_nunique = list(map(lambda col: X_train[col].nunique(), object_cols))\nd = dict(zip(object_cols, object_nunique))\n\n# Print number of unique entries by column, in ascending order\nsorted(d.items(), key=lambda x: x[1])", "_____no_output_____" ] ], [ [ "# Step 3: Investigating cardinality\n\n### Part A\n\nThe output above shows, for each column with categorical data, the number of unique values in the column. For instance, the `'Street'` column in the training data has two unique values: `'Grvl'` and `'Pave'`, corresponding to a gravel road and a paved road, respectively.\n\nWe refer to the number of unique entries of a categorical variable as the **cardinality** of that categorical variable. For instance, the `'Street'` variable has cardinality 2.\n\nUse the output above to answer the questions below.", "_____no_output_____" ] ], [ [ "# Fill in the line below: How many categorical variables in the training data\n# have cardinality greater than 10?\nhigh_cardinality_numcols = 3\n\n# Fill in the line below: How many columns are needed to one-hot encode the \n# 'Neighborhood' variable in the training data?\nnum_cols_neighborhood = 25\n\n# Check your answers\nstep_3.a.check()", "_____no_output_____" ], [ "# Lines below will give you a hint or solution code\n#step_3.a.hint()\n#step_3.a.solution()", "_____no_output_____" ] ], [ [ "### Part B\n\nFor large datasets with many rows, one-hot encoding can greatly expand the size of the dataset. For this reason, we typically will only one-hot encode columns with relatively low cardinality. Then, high cardinality columns can either be dropped from the dataset, or we can use label encoding.\n\nAs an example, consider a dataset with 10,000 rows, and containing one categorical column with 100 unique entries. \n- If this column is replaced with the corresponding one-hot encoding, how many entries are added to the dataset? \n- If we instead replace the column with the label encoding, how many entries are added? \n\nUse your answers to fill in the lines below.", "_____no_output_____" ] ], [ [ "# Fill in the line below: How many entries are added to the dataset by \n# replacing the column with a one-hot encoding?\nOH_entries_added = 1e4*100 - 1e4\n\n# Fill in the line below: How many entries are added to the dataset by\n# replacing the column with a label encoding?\nlabel_entries_added = 0\n\n# Check your answers\nstep_3.b.check()", "_____no_output_____" ], [ "# Lines below will give you a hint or solution code\n#step_3.b.hint()\n#step_3.b.solution()", "_____no_output_____" ] ], [ [ "Next, you'll experiment with one-hot encoding. But, instead of encoding all of the categorical variables in the dataset, you'll only create a one-hot encoding for columns with cardinality less than 10.\n\nRun the code cell below without changes to set `low_cardinality_cols` to a Python list containing the columns that will be one-hot encoded. Likewise, `high_cardinality_cols` contains a list of categorical columns that will be dropped from the dataset.", "_____no_output_____" ] ], [ [ "# Columns that will be one-hot encoded\nlow_cardinality_cols = [col for col in object_cols if X_train[col].nunique() < 10]\n\n# Columns that will be dropped from the dataset\nhigh_cardinality_cols = list(set(object_cols)-set(low_cardinality_cols))\n\nprint('Categorical columns that will be one-hot encoded:', low_cardinality_cols)\nprint('\\nCategorical columns that will be dropped from the dataset:', high_cardinality_cols)", "Categorical columns that will be one-hot encoded: ['MSZoning', 'Street', 'LotShape', 'LandContour', 'Utilities', 'LotConfig', 'LandSlope', 'Condition1', 'Condition2', 'BldgType', 'HouseStyle', 'RoofStyle', 'RoofMatl', 'ExterQual', 'ExterCond', 'Foundation', 'Heating', 'HeatingQC', 'CentralAir', 'KitchenQual', 'Functional', 'PavedDrive', 'SaleType', 'SaleCondition']\n\nCategorical columns that will be dropped from the dataset: ['Neighborhood', 'Exterior2nd', 'Exterior1st']\n" ] ], [ [ "# Step 4: One-hot encoding\n\nUse the next code cell to one-hot encode the data in `X_train` and `X_valid`. Set the preprocessed DataFrames to `OH_X_train` and `OH_X_valid`, respectively. \n- The full list of categorical columns in the dataset can be found in the Python list `object_cols`.\n- You should only one-hot encode the categorical columns in `low_cardinality_cols`. All other categorical columns should be dropped from the dataset. ", "_____no_output_____" ] ], [ [ "from sklearn.preprocessing import OneHotEncoder\n\n# Use as many lines of code as you need!\nOH_encoder = OneHotEncoder(handle_unknown='ignore', sparse=False)\nOH_cols_train = pd.DataFrame(OH_encoder.fit_transform(X_train[low_cardinality_cols]))\nOH_cols_valid = pd.DataFrame(OH_encoder.transform(X_valid[low_cardinality_cols]))\n\n# One-hot encoding removed index; put it back\nOH_cols_train.index = X_train.index\nOH_cols_valid.index = X_valid.index\n\n# Remove categorical columns (will replace with one-hot encoding)\nnum_X_train = X_train.drop(object_cols, axis=1)\nnum_X_valid = X_valid.drop(object_cols, axis=1)\n\n# Add one-hot encoded columns to numerical features\nOH_X_train = pd.concat([num_X_train, OH_cols_train], axis=1)\nOH_X_valid = pd.concat([num_X_valid, OH_cols_valid], axis=1)\n\n# Check your answer\nstep_4.check()", "_____no_output_____" ], [ "# Lines below will give you a hint or solution code\n#step_4.hint()\n#step_4.solution()", "_____no_output_____" ] ], [ [ "Run the next code cell to get the MAE for this approach.", "_____no_output_____" ] ], [ [ "print(\"MAE from Approach 3 (One-Hot Encoding):\") \nprint(score_dataset(OH_X_train, OH_X_valid, y_train, y_valid))", "_____no_output_____" ] ], [ [ "# Generate test predictions and submit your results\n\nAfter you complete Step 4, if you'd like to use what you've learned to submit your results to the leaderboard, you'll need to preprocess the test data before generating predictions.\n\n**This step is completely optional, and you do not need to submit results to the leaderboard to successfully complete the exercise.**\n\nCheck out the previous exercise if you need help with remembering how to [join the competition](https://www.kaggle.com/c/home-data-for-ml-course) or save your results to CSV. Once you have generated a file with your results, follow the instructions below:\n1. Begin by clicking on the blue **Save Version** button in the top right corner of the window. This will generate a pop-up window. \n2. Ensure that the **Save and Run All** option is selected, and then click on the blue **Save** button.\n3. This generates a window in the bottom left corner of the notebook. After it has finished running, click on the number to the right of the **Save Version** button. This pulls up a list of versions on the right of the screen. Click on the ellipsis **(...)** to the right of the most recent version, and select **Open in Viewer**. This brings you into view mode of the same page. You will need to scroll down to get back to these instructions.\n4. Click on the **Output** tab on the right of the screen. Then, click on the file you would like to submit, and click on the blue **Submit** button to submit your results to the leaderboard.\n\nYou have now successfully submitted to the competition!\n\nIf you want to keep working to improve your performance, select the blue **Edit** button in the top right of the screen. Then you can change your code and repeat the process. There's a lot of room to improve, and you will climb up the leaderboard as you work.\n", "_____no_output_____" ] ], [ [ "# (Optional) Your code here", "_____no_output_____" ] ], [ [ "# Keep going\n\nWith missing value handling and categorical encoding, your modeling process is getting complex. This complexity gets worse when you want to save your model to use in the future. The key to managing this complexity is something called **pipelines**. \n\n**[Learn to use pipelines](https://www.kaggle.com/alexisbcook/pipelines)** to preprocess datasets with categorical variables, missing values and any other messiness your data throws at you.", "_____no_output_____" ], [ "---\n\n\n\n\n*Have questions or comments? Visit the [Learn Discussion forum](https://www.kaggle.com/learn-forum/161289) to chat with other Learners.*", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "##### Copyright 2018 The TensorFlow Authors.\n\nLicensed under the Apache License, Version 2.0 (the \"License\");", "_____no_output_____" ] ], [ [ "#@title Licensed under the Apache License, Version 2.0 (the \"License\"); { display-mode: \"form\" }\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# https://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.", "_____no_output_____" ] ], [ [ "<table class=\"tfo-notebook-buttons\" align=\"left\">\n <td>\n <a target=\"_blank\" href=\"https://www.tensorflow.org/beta/{PATH}\">\n <img src=\"https://www.tensorflow.org/images/tf_logo_32px.png\" />\n View on TensorFlow.org</a>\n </td>\n <td>\n <a target=\"_blank\" href=\"https://colab.research.google.com/github/tensorflow/docs/blob/master/site/{PATH}.ipynb\">\n <img src=\"https://www.tensorflow.org/images/colab_logo_32px.png\" />\n Run in Google Colab</a>\n </td>\n <td>\n <a target=\"_blank\" href=\"https://github.com/tensorflow/docs/blob/master/site/{PATH}.ipynb\">\n <img src=\"https://www.tensorflow.org/images/GitHub-Mark-32px.png\" />\n View source on GitHub</a>\n </td>\n</table>", "_____no_output_____" ], [ "# Using TPUs\n\nTensor Processing Units (TPUs) are Google's specialized ASICs designed to dramatically accelerate machine learning workloads. They are available on Google Colab, the TensorFlow Research Cloud and Google Compute Engine. \n\nIn this notebook, you can try training a convolutional neural network against the Fashion MNIST dataset on Cloud TPUs using tf.keras and Distribution Strategy.\n", "_____no_output_____" ], [ "## Learning Objectives\n\nIn this Colab, you will learn how to:\n\n* Write a standard 4-layer conv-net with drop-out and batch normalization in Keras.\n* Use TPUs and Distribution Strategy to train the model.\n* Run a prediction to see how well the model can predict fashion categories and output the result.", "_____no_output_____" ], [ "## Instructions\n\nTo use TPUs in Colab:\n\n1. On the main menu, click Runtime and select **Change runtime type**. Set \"TPU\" as the hardware accelerator.\n1. Click Runtime again and select **Runtime > Run All**. You can also run the cells manually with Shift-ENTER. ", "_____no_output_____" ], [ "## Data, Model, and Training\n\n### Download the Data\n\nBegin by downloading the fashion MNIST dataset using `tf.keras.datasets`, as shown below. We will also need to convert the data to `float32` format, as the data types supported by TPUs are limited right now.\n\nTPUs currently do not support Eager Execution, so we disable that with `disable_eager_execution()`.", "_____no_output_____" ] ], [ [ "from __future__ import absolute_import, division, print_function, unicode_literals\nimport numpy as np", "_____no_output_____" ], [ "from __future__ import absolute_import, division, print_function\n\n!pip install tensorflow-gpu==2.0.0-beta1\nimport tensorflow as tf\ntf.compat.v1.disable_eager_execution()\n\nimport numpy as np\n\n(x_train, y_train), (x_test, y_test) = tf.keras.datasets.fashion_mnist.load_data()\n\n# add empty color dimension\nx_train = np.expand_dims(x_train, -1)\nx_test = np.expand_dims(x_test, -1)\n\n# convert types to float32\nx_train = x_train.astype(np.float32)\nx_test = x_test.astype(np.float32)\ny_train = y_train.astype(np.float32)\ny_test = y_test.astype(np.float32)", "_____no_output_____" ] ], [ [ "### Initialize TPUStrategy\n\nWe first initialize the TPUStrategy object before creating the model, so that Keras knows that we are creating a model for TPUs. \n\nTo do this, we are first creating a TPUClusterResolver using the IP address of the TPU, and then creating a TPUStrategy object from the Cluster Resolver.", "_____no_output_____" ] ], [ [ "import os\n\nresolver = tf.distribute.cluster_resolver.TPUClusterResolver()\ntf.tpu.experimental.initialize_tpu_system(resolver)\nstrategy = tf.distribute.experimental.TPUStrategy(resolver)", "_____no_output_____" ] ], [ [ "### Define the Model\n\nThe following example uses a standard conv-net that has 4 layers with drop-out and batch normalization between each layer. Note that we are creating the model within a `strategy.scope`.\n", "_____no_output_____" ] ], [ [ "with strategy.scope():\n model = tf.keras.models.Sequential()\n model.add(tf.keras.layers.BatchNormalization(input_shape=x_train.shape[1:]))\n model.add(tf.keras.layers.Conv2D(64, (5, 5), padding='same', activation='elu'))\n model.add(tf.keras.layers.MaxPooling2D(pool_size=(2, 2), strides=(2,2)))\n model.add(tf.keras.layers.Dropout(0.25))\n\n model.add(tf.keras.layers.BatchNormalization())\n model.add(tf.keras.layers.Conv2D(128, (5, 5), padding='same', activation='elu'))\n model.add(tf.keras.layers.MaxPooling2D(pool_size=(2, 2)))\n model.add(tf.keras.layers.Dropout(0.25))\n\n model.add(tf.keras.layers.BatchNormalization())\n model.add(tf.keras.layers.Conv2D(256, (5, 5), padding='same', activation='elu'))\n model.add(tf.keras.layers.MaxPooling2D(pool_size=(2, 2), strides=(2,2)))\n model.add(tf.keras.layers.Dropout(0.25))\n\n model.add(tf.keras.layers.BatchNormalization())\n model.add(tf.keras.layers.Conv2D(512, (5, 5), padding='same', activation='elu'))\n model.add(tf.keras.layers.MaxPooling2D(pool_size=(2, 2), strides=(2,2)))\n model.add(tf.keras.layers.Dropout(0.25))\n\n model.add(tf.keras.layers.Flatten())\n model.add(tf.keras.layers.Dense(256))\n model.add(tf.keras.layers.Activation('elu'))\n model.add(tf.keras.layers.Dropout(0.5))\n model.add(tf.keras.layers.Dense(10))\n model.add(tf.keras.layers.Activation('softmax'))\n model.summary()", "_____no_output_____" ] ], [ [ "### Train on the TPU\n\nTo train on the TPU, we can simply call `model.compile` under the strategy scope, and then call `model.fit` to start training. In this case, we are training for 5 epochs with 60 steps per epoch, and running evaluation at the end of 5 epochs.\n\nIt may take a while for the training to start, as the data and model has to be transferred to the TPU and compiled before training can start.", "_____no_output_____" ] ], [ [ "with strategy.scope():\n model.compile(\n optimizer=tf.train.AdamOptimizer(learning_rate=1e-3),\n loss=tf.keras.losses.sparse_categorical_crossentropy,\n metrics=['sparse_categorical_accuracy']\n )\n\nmodel.fit(\n (x_train, y_train),\n epochs=5,\n steps_per_epoch=60,\n validation_data=(x_test, y_test),\n validation_freq=5,\n)", "_____no_output_____" ] ], [ [ "### Check our results with Inference\n\nNow that we are done training, we can see how well the model can predict fashion categories:", "_____no_output_____" ] ], [ [ "LABEL_NAMES = ['t_shirt', 'trouser', 'pullover', 'dress', 'coat', 'sandal', 'shirt', 'sneaker', 'bag', 'ankle_boots']\n\nfrom matplotlib import pyplot\n%matplotlib inline\n\ndef plot_predictions(images, predictions):\n n = images.shape[0]\n nc = int(np.ceil(n / 4))\n f, axes = pyplot.subplots(nc, 4)\n for i in range(nc * 4):\n y = i // 4\n x = i % 4\n axes[x, y].axis('off')\n \n label = LABEL_NAMES[np.argmax(predictions[i])]\n confidence = np.max(predictions[i])\n if i > n:\n continue\n axes[x, y].imshow(images[i])\n axes[x, y].text(0.5, -1.5, label + ': %.3f' % confidence, fontsize=12)\n\n pyplot.gcf().set_size_inches(8, 8) \n\nplot_predictions(np.squeeze(x_test[:16]), \n model.predict(x_test[:16]))", "_____no_output_____" ] ], [ [ "### What's next\n\n* Learn about [Cloud TPUs](https://cloud.google.com/tpu/docs) that Google designed and optimized specifically to speed up and scale up ML workloads for training and inference and to enable ML engineers and researchers to iterate more quickly.\n* Explore the range of [Cloud TPU tutorials and Colabs](https://cloud.google.com/tpu/docs/tutorials) to find other examples that can be used when implementing your ML project.\n\nOn Google Cloud Platform, in addition to GPUs and TPUs available on pre-configured [deep learning VMs](https://cloud.google.com/deep-learning-vm/), you will find [AutoML](https://cloud.google.com/automl/)*(beta)* for training custom models without writing code and [Cloud ML Engine](https://cloud.google.com/ml-engine/docs/) which will allows you to run parallel trainings and hyperparameter tuning of your custom models on powerful distributed hardware.", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown", "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/cseveriano/spatio-temporal-forecasting/blob/master/notebooks/thesis_experiments/20200924_eMVFTS_Wind_Energy_Raw.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ], [ "## Forecasting experiments for GEFCOM 2012 Wind Dataset", "_____no_output_____" ], [ "\n## Install Libs\n", "_____no_output_____" ] ], [ [ "!pip3 install -U git+https://github.com/PYFTS/pyFTS\n!pip3 install -U git+https://github.com/cseveriano/spatio-temporal-forecasting\n!pip3 install -U git+https://github.com/cseveriano/evolving_clustering\n!pip3 install -U git+https://github.com/cseveriano/fts2image\n!pip3 install -U hyperopt\n!pip3 install -U pyts", "Collecting git+https://github.com/PYFTS/pyFTS\n Cloning https://github.com/PYFTS/pyFTS to /tmp/pip-req-build-q6mtqzlz\n Running command git clone -q https://github.com/PYFTS/pyFTS /tmp/pip-req-build-q6mtqzlz\nBuilding wheels for collected packages: pyFTS\n Building wheel for pyFTS (setup.py) ... \u001b[?25l\u001b[?25hdone\n Created wheel for pyFTS: filename=pyFTS-1.6-cp36-none-any.whl size=207416 sha256=9c1df08abea59c8449f05671299dfcd6a888ea4ec4926f4cb375e86a08251613\n Stored in directory: /tmp/pip-ephem-wheel-cache-2te7wbk9/wheels/e7/32/a9/230470113df5a73242a5a6d05671cb646db97abf14bbce2644\nSuccessfully built pyFTS\nInstalling collected packages: pyFTS\nSuccessfully installed pyFTS-1.6\nCollecting git+https://github.com/cseveriano/spatio-temporal-forecasting\n Cloning https://github.com/cseveriano/spatio-temporal-forecasting to /tmp/pip-req-build-7u637wx0\n Running command git clone -q https://github.com/cseveriano/spatio-temporal-forecasting /tmp/pip-req-build-7u637wx0\nBuilding wheels for collected packages: spatio-temporal-forecasting\n Building wheel for spatio-temporal-forecasting (setup.py) ... \u001b[?25l\u001b[?25hdone\n Created wheel for spatio-temporal-forecasting: filename=spatio_temporal_forecasting-1.0-cp36-none-any.whl size=55633 sha256=4e40efc1c38541efa9b7168b24d990baca71acbc1c5d5858425fb0354dd58403\n Stored in directory: /tmp/pip-ephem-wheel-cache-mzfi7veh/wheels/d2/1f/6f/439795864246039ef36c6a3c88edf7935c803c2cf97133066a\nSuccessfully built spatio-temporal-forecasting\nInstalling collected packages: spatio-temporal-forecasting\nSuccessfully installed spatio-temporal-forecasting-1.0\nCollecting git+https://github.com/cseveriano/evolving_clustering\n Cloning https://github.com/cseveriano/evolving_clustering to /tmp/pip-req-build-_fyfrguj\n Running command git clone -q https://github.com/cseveriano/evolving_clustering /tmp/pip-req-build-_fyfrguj\nBuilding wheels for collected packages: evolclustering\n Building wheel for evolclustering (setup.py) ... \u001b[?25l\u001b[?25hdone\n Created wheel for evolclustering: filename=evolclustering-0.1-cp36-none-any.whl size=25744 sha256=f5bb287535544a3476690f61225544c738fe73987cecd7732f3220970544224e\n Stored in directory: /tmp/pip-ephem-wheel-cache-w3m1u0lx/wheels/aa/b7/f1/4d93077e1b97361934f5992c77be80f0769eba6e0e9da6e22d\nSuccessfully built evolclustering\nInstalling collected packages: evolclustering\nSuccessfully installed evolclustering-0.1\nCollecting git+https://github.com/cseveriano/fts2image\n Cloning https://github.com/cseveriano/fts2image to /tmp/pip-req-build-tzt5k5x3\n Running command git clone -q https://github.com/cseveriano/fts2image /tmp/pip-req-build-tzt5k5x3\nBuilding wheels for collected packages: fts2image\n Building wheel for fts2image (setup.py) ... \u001b[?25l\u001b[?25hdone\n Created wheel for fts2image: filename=fts2image-0.1.0-py2.py3-none-any.whl size=7975 sha256=037dd37f1af0118db94a88ea46340bcf2840c6cf7c0fae8dfce67eb554a703c0\n Stored in directory: /tmp/pip-ephem-wheel-cache-7ygvs3r0/wheels/22/a1/62/57410665915134ffe4bc11ede1f9a47b1f4b29d8aad9582d31\nSuccessfully built fts2image\nInstalling collected packages: fts2image\nSuccessfully installed fts2image-0.1.0\nCollecting hyperopt\n\u001b[?25l Downloading https://files.pythonhosted.org/packages/90/d5/c7e276f4f7bc65ac26391c435245e5ef8911b4393e3df5a74906c48afeaf/hyperopt-0.2.4-py2.py3-none-any.whl (964kB)\n\u001b[K |████████████████████████████████| 972kB 3.4MB/s \n\u001b[?25hRequirement already satisfied, skipping upgrade: future in /usr/local/lib/python3.6/dist-packages (from hyperopt) (0.16.0)\nRequirement already satisfied, skipping upgrade: networkx>=2.2 in /usr/local/lib/python3.6/dist-packages (from hyperopt) (2.5)\nRequirement already satisfied, skipping upgrade: cloudpickle in /usr/local/lib/python3.6/dist-packages (from hyperopt) (1.3.0)\nRequirement already satisfied, skipping upgrade: six in /usr/local/lib/python3.6/dist-packages (from hyperopt) (1.15.0)\nRequirement already satisfied, skipping upgrade: scipy in /usr/local/lib/python3.6/dist-packages (from hyperopt) (1.4.1)\nRequirement already satisfied, skipping upgrade: numpy in /usr/local/lib/python3.6/dist-packages (from hyperopt) (1.18.5)\nRequirement already satisfied, skipping upgrade: tqdm in /usr/local/lib/python3.6/dist-packages (from hyperopt) (4.41.1)\nRequirement already satisfied, skipping upgrade: decorator>=4.3.0 in /usr/local/lib/python3.6/dist-packages (from networkx>=2.2->hyperopt) (4.4.2)\nInstalling collected packages: hyperopt\n Found existing installation: hyperopt 0.1.2\n Uninstalling hyperopt-0.1.2:\n Successfully uninstalled hyperopt-0.1.2\nSuccessfully installed hyperopt-0.2.4\nCollecting pyts\n\u001b[?25l Downloading https://files.pythonhosted.org/packages/b6/2b/1a62c0d32b40ee85daa8f6a6160828537b3d846c9fe93253b38846c6ec1f/pyts-0.11.0-py3-none-any.whl (2.5MB)\n\u001b[K |████████████████████████████████| 2.5MB 3.3MB/s \n\u001b[?25hRequirement already satisfied, skipping upgrade: numpy>=1.17.5 in /usr/local/lib/python3.6/dist-packages (from pyts) (1.18.5)\nRequirement already satisfied, skipping upgrade: scikit-learn>=0.22.1 in /usr/local/lib/python3.6/dist-packages (from pyts) (0.22.2.post1)\nRequirement already satisfied, skipping upgrade: joblib>=0.12 in /usr/local/lib/python3.6/dist-packages (from pyts) (0.16.0)\nRequirement already satisfied, skipping upgrade: numba>=0.48.0 in /usr/local/lib/python3.6/dist-packages (from pyts) (0.48.0)\nRequirement already satisfied, skipping upgrade: scipy>=1.3.0 in /usr/local/lib/python3.6/dist-packages (from pyts) (1.4.1)\nRequirement already satisfied, skipping upgrade: llvmlite<0.32.0,>=0.31.0dev0 in /usr/local/lib/python3.6/dist-packages (from numba>=0.48.0->pyts) (0.31.0)\nRequirement already satisfied, skipping upgrade: setuptools in /usr/local/lib/python3.6/dist-packages (from numba>=0.48.0->pyts) (50.3.0)\nInstalling collected packages: pyts\nSuccessfully installed pyts-0.11.0\n" ], [ "import pandas as pd\nimport numpy as np\nfrom hyperopt import hp\nfrom spatiotemporal.util import parameter_tuning, sampling\nfrom spatiotemporal.util import experiments as ex\nfrom sklearn.metrics import mean_squared_error\nfrom google.colab import files\nimport matplotlib.pyplot as plt\nimport pickle\nimport math\nfrom pyFTS.benchmarks import Measures\nfrom pyts.decomposition import SingularSpectrumAnalysis\nfrom google.colab import files\nimport warnings \nwarnings.filterwarnings(\"ignore\", category=DeprecationWarning)\nimport datetime", "_____no_output_____" ] ], [ [ "## Aux Functions", "_____no_output_____" ] ], [ [ "def normalize(df):\n mindf = df.min()\n maxdf = df.max()\n return (df-mindf)/(maxdf-mindf)\n\ndef denormalize(norm, _min, _max):\n return [(n * (_max-_min)) + _min for n in norm]\n\n\ndef getRollingWindow(index):\n pivot = index\n train_start = pivot.strftime('%Y-%m-%d')\n pivot = pivot + datetime.timedelta(days=20)\n train_end = pivot.strftime('%Y-%m-%d')\n\n pivot = pivot + datetime.timedelta(days=1)\n test_start = pivot.strftime('%Y-%m-%d')\n pivot = pivot + datetime.timedelta(days=6)\n test_end = pivot.strftime('%Y-%m-%d')\n \n return train_start, train_end, test_start, test_end\n\ndef calculate_rolling_error(cv_name, df, forecasts, order_list):\n cv_results = pd.DataFrame(columns=['Split', 'RMSE', 'SMAPE'])\n\n limit = df.index[-1].strftime('%Y-%m-%d')\n\n test_end = \"\"\n index = df.index[0]\n\n for i in np.arange(len(forecasts)):\n\n train_start, train_end, test_start, test_end = getRollingWindow(index)\n test = df[test_start : test_end]\n\n yhat = forecasts[i]\n\n order = order_list[i]\n rmse = Measures.rmse(test.iloc[order:], yhat[:-1])\n \n smape = Measures.smape(test.iloc[order:], yhat[:-1])\n \n res = {'Split' : index.strftime('%Y-%m-%d') ,'RMSE' : rmse, 'SMAPE' : smape}\n cv_results = cv_results.append(res, ignore_index=True)\n cv_results.to_csv(cv_name+\".csv\") \n\n index = index + datetime.timedelta(days=7)\n \n return cv_results\n\ndef get_final_forecast(norm_forecasts):\n \n forecasts_final = []\n \n for i in np.arange(len(norm_forecasts)):\n f_raw = denormalize(norm_forecasts[i], min_raw, max_raw)\n\n forecasts_final.append(f_raw)\n \n return forecasts_final", "_____no_output_____" ], [ "from spatiotemporal.test import methods_space_oahu as ms\nfrom spatiotemporal.util import parameter_tuning, sampling\nfrom spatiotemporal.util import experiments as ex\nfrom sklearn.metrics import mean_squared_error\nimport numpy as np\nfrom hyperopt import fmin, tpe, hp, STATUS_OK, Trials\nfrom hyperopt import space_eval\nimport traceback\nfrom . import sampling\nimport pickle\n\ndef calculate_error(loss_function, test_df, forecast, offset):\n error = loss_function(test_df.iloc[(offset):], forecast)\n print(\"Error : \"+str(error))\n return error\n\ndef method_optimize(experiment, forecast_method, train_df, test_df, space, loss_function, max_evals):\n def objective(params):\n print(params)\n try:\n _output = list(params['output'])\n forecast = forecast_method(train_df, test_df, params)\n _step = params.get('step', 1)\n offset = params['order'] + _step - 1\n error = calculate_error(loss_function, test_df[_output], forecast, offset)\n except Exception:\n traceback.print_exc()\n error = 1000\n return {'loss': error, 'status': STATUS_OK}\n\n print(\"Running experiment: \" + experiment)\n trials = Trials()\n best = fmin(objective, space, algo=tpe.suggest, max_evals=max_evals, trials=trials)\n print('best parameters: ')\n print(space_eval(space, best))\n\n pickle.dump(best, open(\"best_\" + experiment + \".pkl\", \"wb\"))\n pickle.dump(trials, open(\"trials_\" + experiment + \".pkl\", \"wb\"))\n\n\ndef run_search(methods, data, train, loss_function, max_evals=100, resample=None):\n\n if resample:\n data = sampling.resample_data(data, resample)\n\n train_df, test_df = sampling.train_test_split(data, train)\n\n for experiment, method, space in methods:\n method_optimize(experiment, method, train_df, test_df, space, loss_function, max_evals)", "_____no_output_____" ] ], [ [ "## Load Dataset", "_____no_output_____" ] ], [ [ "import pandas as pd\nimport matplotlib.pyplot as plt\nimport numpy as np\nimport math\nfrom sklearn.metrics import mean_squared_error", "_____no_output_____" ], [ "#columns names\nwind_farms = ['wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7']\n\n# read raw dataset\nimport pandas as pd\ndf = pd.read_csv('https://query.data.world/s/3zx2jusk4z6zvlg2dafqgshqp3oao6', parse_dates=['date'], index_col=0)\ndf.index = pd.to_datetime(df.index, format=\"%Y%m%d%H\")\n\ninterval = ((df.index >= '2009-07') & (df.index <= '2010-08'))\ndf = df.loc[interval]\n\n\n#Normalize Data\n\n# Save Min-Max for Denorm\nmin_raw = df.min()\nmax_raw = df.max()\n\n# Perform Normalization\nnorm_df = normalize(df)\n\n# Tuning split\ntuning_df = norm_df[\"2009-07-01\":\"2009-07-31\"]\n\nnorm_df = norm_df[\"2009-08-01\":\"2010-08-30\"]\ndf = df[\"2009-08-01\":\"2010-08-30\"]", "_____no_output_____" ] ], [ [ "## Forecasting Methods", "_____no_output_____" ], [ "### Persistence", "_____no_output_____" ] ], [ [ "def persistence_forecast(train, test, step):\n predictions = []\n \n for t in np.arange(0,len(test), step):\n yhat = [test.iloc[t]] * step\n predictions.extend(yhat)\n \n return predictions\n\ndef rolling_cv_persistence(df, step):\n\n forecasts = []\n lags_list = []\n\n limit = df.index[-1].strftime('%Y-%m-%d')\n\n test_end = \"\"\n index = df.index[0]\n\n while test_end < limit :\n print(\"Index: \", index.strftime('%Y-%m-%d')) \n\n train_start, train_end, test_start, test_end = getRollingWindow(index)\n index = index + datetime.timedelta(days=7)\n \n train = df[train_start : train_end]\n test = df[test_start : test_end]\n \n yhat = persistence_forecast(train, test, step) \n \n lags_list.append(1)\n forecasts.append(yhat)\n\n return forecasts, lags_list", "_____no_output_____" ], [ "forecasts_raw, order_list = rolling_cv_persistence(norm_df, 1)\nforecasts_final = get_final_forecast(forecasts_raw)\n\ncalculate_rolling_error(\"rolling_cv_wind_raw_persistence\", norm_df, forecasts_final, order_list)", "Index: 2009-08-01\nIndex: 2009-08-08\nIndex: 2009-08-15\nIndex: 2009-08-22\nIndex: 2009-08-29\nIndex: 2009-09-05\nIndex: 2009-09-12\nIndex: 2009-09-19\nIndex: 2009-09-26\nIndex: 2009-10-03\nIndex: 2009-10-10\nIndex: 2009-10-17\nIndex: 2009-10-24\nIndex: 2009-10-31\nIndex: 2009-11-07\nIndex: 2009-11-14\nIndex: 2009-11-21\nIndex: 2009-11-28\nIndex: 2009-12-05\nIndex: 2009-12-12\nIndex: 2009-12-19\nIndex: 2009-12-26\nIndex: 2010-01-02\nIndex: 2010-01-09\nIndex: 2010-01-16\nIndex: 2010-01-23\nIndex: 2010-01-30\nIndex: 2010-02-06\nIndex: 2010-02-13\nIndex: 2010-02-20\nIndex: 2010-02-27\nIndex: 2010-03-06\nIndex: 2010-03-13\nIndex: 2010-03-20\nIndex: 2010-03-27\nIndex: 2010-04-03\nIndex: 2010-04-10\nIndex: 2010-04-17\nIndex: 2010-04-24\nIndex: 2010-05-01\nIndex: 2010-05-08\nIndex: 2010-05-15\nIndex: 2010-05-22\nIndex: 2010-05-29\nIndex: 2010-06-05\nIndex: 2010-06-12\nIndex: 2010-06-19\nIndex: 2010-06-26\nIndex: 2010-07-03\nIndex: 2010-07-10\n" ], [ "files.download('rolling_cv_wind_raw_persistence.csv')", "_____no_output_____" ] ], [ [ "### VAR", "_____no_output_____" ] ], [ [ "from statsmodels.tsa.api import VAR, DynamicVAR", "_____no_output_____" ], [ "def evaluate_VAR_models(test_name, train, validation,target, maxlags_list):\n var_results = pd.DataFrame(columns=['Order','RMSE'])\n best_score, best_cfg, best_model = float(\"inf\"), None, None\n \n for lgs in maxlags_list:\n model = VAR(train)\n results = model.fit(maxlags=lgs, ic='aic')\n \n order = results.k_ar\n forecast = []\n\n for i in range(len(validation)-order) :\n forecast.extend(results.forecast(validation.values[i:i+order],1))\n\n forecast_df = pd.DataFrame(columns=validation.columns, data=forecast)\n rmse = Measures.rmse(validation[target].iloc[order:], forecast_df[target].values)\n\n if rmse < best_score:\n best_score, best_cfg, best_model = rmse, order, results\n\n res = {'Order' : str(order) ,'RMSE' : rmse}\n print('VAR (%s) RMSE=%.3f' % (str(order),rmse))\n var_results = var_results.append(res, ignore_index=True)\n var_results.to_csv(test_name+\".csv\")\n \n print('Best VAR(%s) RMSE=%.3f' % (best_cfg, best_score))\n return best_model", "_____no_output_____" ], [ "def var_forecast(train, test, params):\n order = params['order']\n step = params['step']\n\n model = VAR(train.values)\n results = model.fit(maxlags=order)\n lag_order = results.k_ar\n print(\"Lag order:\" + str(lag_order))\n forecast = []\n\n for i in np.arange(0,len(test)-lag_order+1,step) :\n forecast.extend(results.forecast(test.values[i:i+lag_order],step))\n\n forecast_df = pd.DataFrame(columns=test.columns, data=forecast)\n return forecast_df.values, lag_order", "_____no_output_____" ], [ "def rolling_cv_var(df, params):\n forecasts = []\n order_list = []\n\n limit = df.index[-1].strftime('%Y-%m-%d')\n\n test_end = \"\"\n index = df.index[0]\n\n while test_end < limit :\n print(\"Index: \", index.strftime('%Y-%m-%d')) \n\n train_start, train_end, test_start, test_end = getRollingWindow(index)\n index = index + datetime.timedelta(days=7)\n \n train = df[train_start : train_end]\n test = df[test_start : test_end]\n \n # Concat train & validation for test\n yhat, lag_order = var_forecast(train, test, params)\n \n forecasts.append(yhat)\n order_list.append(lag_order)\n\n return forecasts, order_list", "_____no_output_____" ], [ "params_raw = {'order': 4, 'step': 1}\n\nforecasts_raw, order_list = rolling_cv_var(norm_df, params_raw)\n\nforecasts_final = get_final_forecast(forecasts_raw)\ncalculate_rolling_error(\"rolling_cv_wind_raw_var\", df, forecasts_final, order_list)", "Index: 2009-08-01\nLag order:4\nIndex: 2009-08-08\nLag order:4\nIndex: 2009-08-15\nLag order:4\nIndex: 2009-08-22\nLag order:4\nIndex: 2009-08-29\nLag order:4\nIndex: 2009-09-05\nLag order:4\nIndex: 2009-09-12\nLag order:4\nIndex: 2009-09-19\nLag order:4\nIndex: 2009-09-26\nLag order:4\nIndex: 2009-10-03\nLag order:4\nIndex: 2009-10-10\nLag order:4\nIndex: 2009-10-17\nLag order:4\nIndex: 2009-10-24\nLag order:4\nIndex: 2009-10-31\nLag order:4\nIndex: 2009-11-07\nLag order:4\nIndex: 2009-11-14\nLag order:4\nIndex: 2009-11-21\nLag order:4\nIndex: 2009-11-28\nLag order:4\nIndex: 2009-12-05\nLag order:4\nIndex: 2009-12-12\nLag order:4\nIndex: 2009-12-19\nLag order:4\nIndex: 2009-12-26\nLag order:4\nIndex: 2010-01-02\nLag order:4\nIndex: 2010-01-09\nLag order:4\nIndex: 2010-01-16\nLag order:4\nIndex: 2010-01-23\nLag order:4\nIndex: 2010-01-30\nLag order:4\nIndex: 2010-02-06\nLag order:4\nIndex: 2010-02-13\nLag order:4\nIndex: 2010-02-20\nLag order:4\nIndex: 2010-02-27\nLag order:4\nIndex: 2010-03-06\nLag order:4\nIndex: 2010-03-13\nLag order:4\nIndex: 2010-03-20\nLag order:4\nIndex: 2010-03-27\nLag order:4\nIndex: 2010-04-03\nLag order:4\nIndex: 2010-04-10\nLag order:4\nIndex: 2010-04-17\nLag order:4\nIndex: 2010-04-24\nLag order:4\nIndex: 2010-05-01\nLag order:4\nIndex: 2010-05-08\nLag order:4\nIndex: 2010-05-15\nLag order:4\nIndex: 2010-05-22\nLag order:4\nIndex: 2010-05-29\nLag order:4\nIndex: 2010-06-05\nLag order:4\nIndex: 2010-06-12\nLag order:4\nIndex: 2010-06-19\nLag order:4\nIndex: 2010-06-26\nLag order:4\nIndex: 2010-07-03\nLag order:4\nIndex: 2010-07-10\nLag order:4\n" ], [ "files.download('rolling_cv_wind_raw_var.csv')", "_____no_output_____" ] ], [ [ "### e-MVFTS", "_____no_output_____" ] ], [ [ "from spatiotemporal.models.clusteredmvfts.fts import evolvingclusterfts", "_____no_output_____" ], [ "def evolvingfts_forecast(train_df, test_df, params, train_model=True):\n\n _variance_limit = params['variance_limit']\n _defuzzy = params['defuzzy']\n _t_norm = params['t_norm']\n _membership_threshold = params['membership_threshold']\n _order = params['order']\n _step = params['step']\n\n\n model = evolvingclusterfts.EvolvingClusterFTS(variance_limit=_variance_limit, defuzzy=_defuzzy, t_norm=_t_norm,\n membership_threshold=_membership_threshold)\n\n model.fit(train_df.values, order=_order, verbose=False)\n\n forecast = model.predict(test_df.values, steps_ahead=_step)\n\n forecast_df = pd.DataFrame(data=forecast, columns=test_df.columns)\n return forecast_df.values", "_____no_output_____" ], [ "def rolling_cv_evolving(df, params):\n forecasts = []\n order_list = []\n\n limit = df.index[-1].strftime('%Y-%m-%d')\n\n test_end = \"\"\n index = df.index[0]\n\n first_time = True\n\n while test_end < limit :\n print(\"Index: \", index.strftime('%Y-%m-%d')) \n\n train_start, train_end, test_start, test_end = getRollingWindow(index)\n index = index + datetime.timedelta(days=7)\n \n train = df[train_start : train_end]\n test = df[test_start : test_end]\n \n # Concat train & validation for test\n yhat = list(evolvingfts_forecast(train, test, params, train_model=first_time))\n #yhat.append(yhat[-1]) #para manter o formato do vetor de metricas\n forecasts.append(yhat)\n order_list.append(params['order'])\n\n first_time = False\n\n return forecasts, order_list", "_____no_output_____" ], [ "params_raw = {'variance_limit': 0.001, 'order': 2, 'defuzzy': 'weighted', 't_norm': 'threshold', 'membership_threshold': 0.6, 'step':1}\n\nforecasts_raw, order_list = rolling_cv_evolving(norm_df, params_raw)\n\nforecasts_final = get_final_forecast(forecasts_raw)\ncalculate_rolling_error(\"rolling_cv_wind_raw_emvfts\", df, forecasts_final, order_list)", "Index: 2009-08-01\nIndex: 2009-08-08\nIndex: 2009-08-15\nIndex: 2009-08-22\nIndex: 2009-08-29\nIndex: 2009-09-05\nIndex: 2009-09-12\nIndex: 2009-09-19\nIndex: 2009-09-26\nIndex: 2009-10-03\nIndex: 2009-10-10\nIndex: 2009-10-17\nIndex: 2009-10-24\nIndex: 2009-10-31\nIndex: 2009-11-07\nIndex: 2009-11-14\nIndex: 2009-11-21\nIndex: 2009-11-28\nIndex: 2009-12-05\nIndex: 2009-12-12\nIndex: 2009-12-19\nIndex: 2009-12-26\nIndex: 2010-01-02\nIndex: 2010-01-09\nIndex: 2010-01-16\nIndex: 2010-01-23\nIndex: 2010-01-30\nIndex: 2010-02-06\nIndex: 2010-02-13\nIndex: 2010-02-20\nIndex: 2010-02-27\nIndex: 2010-03-06\nIndex: 2010-03-13\nIndex: 2010-03-20\nIndex: 2010-03-27\nIndex: 2010-04-03\nIndex: 2010-04-10\nIndex: 2010-04-17\nIndex: 2010-04-24\nIndex: 2010-05-01\nIndex: 2010-05-08\nIndex: 2010-05-15\nIndex: 2010-05-22\nIndex: 2010-05-29\nIndex: 2010-06-05\nIndex: 2010-06-12\nIndex: 2010-06-19\nIndex: 2010-06-26\nIndex: 2010-07-03\nIndex: 2010-07-10\n" ], [ "files.download('rolling_cv_wind_raw_emvfts.csv')", "_____no_output_____" ] ], [ [ "### MLP", "_____no_output_____" ] ], [ [ "from keras.models import Sequential\nfrom keras.layers import Dense\nfrom keras.layers import LSTM\nfrom keras.layers import Dropout\nfrom keras.constraints import maxnorm\nfrom keras.models import Sequential\nfrom keras.layers.core import Dense, Dropout, Activation\nfrom keras.layers.normalization import BatchNormalization", "_____no_output_____" ], [ "# convert series to supervised learning\ndef series_to_supervised(data, n_in=1, n_out=1, dropnan=True):\n n_vars = 1 if type(data) is list else data.shape[1]\n df = pd.DataFrame(data)\n cols, names = list(), list()\n # input sequence (t-n, ... t-1)\n for i in range(n_in, 0, -1):\n cols.append(df.shift(i))\n names += [('var%d(t-%d)' % (j+1, i)) for j in range(n_vars)]\n # forecast sequence (t, t+1, ... t+n)\n for i in range(0, n_out):\n cols.append(df.shift(-i))\n if i == 0:\n names += [('var%d(t)' % (j+1)) for j in range(n_vars)]\n else:\n names += [('var%d(t+%d)' % (j+1, i)) for j in range(n_vars)]\n # put it all together\n agg = pd.concat(cols, axis=1)\n agg.columns = names\n # drop rows with NaN values\n if dropnan:\n agg.dropna(inplace=True)\n return agg", "_____no_output_____" ] ], [ [ "#### MLP Parameter Tuning", "_____no_output_____" ] ], [ [ "from spatiotemporal.util import parameter_tuning, sampling\nfrom spatiotemporal.util import experiments as ex\nfrom sklearn.metrics import mean_squared_error\nfrom hyperopt import hp\nimport numpy as np", "_____no_output_____" ], [ "mlp_space = {'choice':\n\n hp.choice('num_layers',\n [\n {'layers': 'two',\n },\n\n {'layers': 'three',\n\n 'units3': hp.choice('units3', [8, 16, 64, 128, 256, 512]),\n 'dropout3': hp.choice('dropout3', [0, 0.25, 0.5, 0.75])\n }\n\n ]),\n 'units1': hp.choice('units1', [8, 16, 64, 128, 256, 512]),\n 'units2': hp.choice('units2', [8, 16, 64, 128, 256, 512]),\n\n 'dropout1': hp.choice('dropout1', [0, 0.25, 0.5, 0.75]),\n 'dropout2': hp.choice('dropout2', [0, 0.25, 0.5, 0.75]),\n\n 'batch_size': hp.choice('batch_size', [28, 64, 128, 256, 512]),\n 'order': hp.choice('order', [1, 2, 3]),\n 'input': hp.choice('input', [wind_farms]),\n 'output': hp.choice('output', [wind_farms]),\n 'epochs': hp.choice('epochs', [100, 200, 300])}\n", "_____no_output_____" ], [ "def mlp_tuning(train_df, test_df, params):\n _input = list(params['input'])\n _nlags = params['order']\n _epochs = params['epochs']\n _batch_size = params['batch_size']\n nfeat = len(train_df.columns)\n nsteps = params.get('step',1)\n nobs = _nlags * nfeat\n\n output_index = -nfeat*nsteps\n\n train_reshaped_df = series_to_supervised(train_df[_input], n_in=_nlags, n_out=nsteps)\n train_X, train_Y = train_reshaped_df.iloc[:, :nobs].values, train_reshaped_df.iloc[:, output_index:].values\n\n test_reshaped_df = series_to_supervised(test_df[_input], n_in=_nlags, n_out=nsteps)\n test_X, test_Y = test_reshaped_df.iloc[:, :nobs].values, test_reshaped_df.iloc[:, output_index:].values\n\n # design network\n model = Sequential()\n model.add(Dense(params['units1'], input_dim=train_X.shape[1], activation='relu'))\n model.add(Dropout(params['dropout1']))\n model.add(BatchNormalization())\n\n model.add(Dense(params['units2'], activation='relu'))\n model.add(Dropout(params['dropout2']))\n model.add(BatchNormalization())\n\n if params['choice']['layers'] == 'three':\n model.add(Dense(params['choice']['units3'], activation='relu'))\n model.add(Dropout(params['choice']['dropout3']))\n model.add(BatchNormalization())\n\n model.add(Dense(train_Y.shape[1], activation='sigmoid'))\n model.compile(loss='mse', optimizer='adam')\n\n # includes the call back object\n model.fit(train_X, train_Y, epochs=_epochs, batch_size=_batch_size, verbose=False, shuffle=False)\n\n # predict the test set\n forecast = model.predict(test_X, verbose=False)\n\n return forecast\n", "_____no_output_____" ], [ "methods = []\nmethods.append((\"EXP_OAHU_MLP\", mlp_tuning, mlp_space))\ntrain_split = 0.6\nrun_search(methods, tuning_df, train_split, Measures.rmse, max_evals=30, resample=None)", "Running experiment: EXP_OAHU_MLP\n{'batch_size': 256, 'choice': {'layers': 'two'}, 'dropout1': 0, 'dropout2': 0.25, 'epochs': 300, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 3, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 16, 'units2': 512}\nError : 0.11210207774258987\n{'batch_size': 64, 'choice': {'dropout3': 0.75, 'layers': 'three', 'units3': 8}, 'dropout1': 0.75, 'dropout2': 0.75, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 1, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 64, 'units2': 8}\nError : 0.16887562719906232\n{'batch_size': 512, 'choice': {'dropout3': 0.5, 'layers': 'three', 'units3': 128}, 'dropout1': 0.5, 'dropout2': 0.5, 'epochs': 300, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 1, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 64, 'units2': 8}\nError : 0.16832074683739862\n{'batch_size': 28, 'choice': {'dropout3': 0, 'layers': 'three', 'units3': 256}, 'dropout1': 0.25, 'dropout2': 0, 'epochs': 300, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 64, 'units2': 64}\nError : 0.12007328735895494\n{'batch_size': 28, 'choice': {'dropout3': 0.5, 'layers': 'three', 'units3': 256}, 'dropout1': 0.75, 'dropout2': 0.25, 'epochs': 300, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 3, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 512, 'units2': 16}\nError : 0.11256583928262713\n{'batch_size': 256, 'choice': {'dropout3': 0.25, 'layers': 'three', 'units3': 64}, 'dropout1': 0.5, 'dropout2': 0.5, 'epochs': 300, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 3, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 512, 'units2': 16}\nError : 0.14391026899955472\n{'batch_size': 256, 'choice': {'dropout3': 0.75, 'layers': 'three', 'units3': 64}, 'dropout1': 0, 'dropout2': 0, 'epochs': 300, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 1, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 16, 'units2': 64}\nError : 0.11037676055120181\n{'batch_size': 512, 'choice': {'dropout3': 0.75, 'layers': 'three', 'units3': 128}, 'dropout1': 0.25, 'dropout2': 0.25, 'epochs': 300, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 1, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 256, 'units2': 512}\nError : 0.15784381475268033\n{'batch_size': 512, 'choice': {'dropout3': 0.75, 'layers': 'three', 'units3': 256}, 'dropout1': 0.75, 'dropout2': 0, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 1, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 128, 'units2': 8}\nError : 0.16657000728035204\n{'batch_size': 512, 'choice': {'layers': 'two'}, 'dropout1': 0.75, 'dropout2': 0.25, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 3, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 512, 'units2': 8}\nError : 0.26202963425973014\n{'batch_size': 64, 'choice': {'layers': 'two'}, 'dropout1': 0.25, 'dropout2': 0.5, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 512, 'units2': 64}\nError : 0.08758667541932756\n{'batch_size': 28, 'choice': {'dropout3': 0, 'layers': 'three', 'units3': 256}, 'dropout1': 0.5, 'dropout2': 0.75, 'epochs': 100, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 1, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 64, 'units2': 16}\nError : 0.139826483409004\n{'batch_size': 128, 'choice': {'layers': 'two'}, 'dropout1': 0.5, 'dropout2': 0.75, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 16, 'units2': 256}\nError : 0.12880869981278525\n{'batch_size': 128, 'choice': {'dropout3': 0.25, 'layers': 'three', 'units3': 8}, 'dropout1': 0, 'dropout2': 0.75, 'epochs': 100, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 3, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 512, 'units2': 64}\nError : 0.16604021900218402\n{'batch_size': 128, 'choice': {'layers': 'two'}, 'dropout1': 0, 'dropout2': 0.5, 'epochs': 300, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 1, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 128, 'units2': 256}\nError : 0.09555621269300194\n{'batch_size': 256, 'choice': {'dropout3': 0.75, 'layers': 'three', 'units3': 64}, 'dropout1': 0.75, 'dropout2': 0.25, 'epochs': 300, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 8, 'units2': 8}\nError : 0.1711557976639845\n{'batch_size': 28, 'choice': {'layers': 'two'}, 'dropout1': 0.75, 'dropout2': 0, 'epochs': 100, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 3, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 8, 'units2': 64}\nError : 0.1638326118189065\n{'batch_size': 256, 'choice': {'layers': 'two'}, 'dropout1': 0.25, 'dropout2': 0, 'epochs': 100, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 1, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 8, 'units2': 512}\nError : 0.15831764665590864\n{'batch_size': 256, 'choice': {'layers': 'two'}, 'dropout1': 0.5, 'dropout2': 0.75, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 256, 'units2': 256}\nError : 0.14529388682505784\n{'batch_size': 64, 'choice': {'dropout3': 0.25, 'layers': 'three', 'units3': 512}, 'dropout1': 0.25, 'dropout2': 0.75, 'epochs': 300, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 1, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 8, 'units2': 8}\nError : 0.1414119809552915\n{'batch_size': 64, 'choice': {'layers': 'two'}, 'dropout1': 0, 'dropout2': 0.5, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 128, 'units2': 128}\nError : 0.09542121366565244\n{'batch_size': 64, 'choice': {'layers': 'two'}, 'dropout1': 0.25, 'dropout2': 0.5, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 128, 'units2': 128}\nError : 0.08515883577119714\n{'batch_size': 64, 'choice': {'layers': 'two'}, 'dropout1': 0.25, 'dropout2': 0.5, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 128, 'units2': 128}\nError : 0.084967455912928\n{'batch_size': 64, 'choice': {'layers': 'two'}, 'dropout1': 0.25, 'dropout2': 0.5, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 128, 'units2': 128}\nError : 0.08816597673392379\n{'batch_size': 64, 'choice': {'layers': 'two'}, 'dropout1': 0.25, 'dropout2': 0.5, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 128, 'units2': 128}\nError : 0.08461966850490099\n{'batch_size': 64, 'choice': {'layers': 'two'}, 'dropout1': 0.25, 'dropout2': 0.5, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 128, 'units2': 128}\nError : 0.08416671260635603\n{'batch_size': 64, 'choice': {'layers': 'two'}, 'dropout1': 0.25, 'dropout2': 0.5, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 128, 'units2': 128}\nError : 0.08203448953925911\n{'batch_size': 64, 'choice': {'layers': 'two'}, 'dropout1': 0.25, 'dropout2': 0.5, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 128, 'units2': 128}\nError : 0.09141701084487909\n{'batch_size': 64, 'choice': {'layers': 'two'}, 'dropout1': 0.25, 'dropout2': 0.5, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 128, 'units2': 128}\nError : 0.08625258845773652\n{'batch_size': 64, 'choice': {'layers': 'two'}, 'dropout1': 0.25, 'dropout2': 0.5, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 256, 'units2': 128}\nError : 0.0846710829000828\n100%|██████████| 30/30 [02:15<00:00, 4.52s/trial, best loss: 0.08203448953925911]\nbest parameters: \n{'batch_size': 64, 'choice': {'layers': 'two'}, 'dropout1': 0.25, 'dropout2': 0.5, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 128, 'units2': 128}\n" ] ], [ [ "#### MLP Forecasting", "_____no_output_____" ] ], [ [ "def mlp_multi_forecast(train_df, test_df, params):\n\n \n nfeat = len(train_df.columns)\n nlags = params['order']\n nsteps = params.get('step',1)\n nobs = nlags * nfeat\n\n output_index = -nfeat*nsteps\n\n train_reshaped_df = series_to_supervised(train_df, n_in=nlags, n_out=nsteps)\n train_X, train_Y = train_reshaped_df.iloc[:, :nobs].values, train_reshaped_df.iloc[:, output_index:].values\n\n test_reshaped_df = series_to_supervised(test_df, n_in=nlags, n_out=nsteps)\n test_X, test_Y = test_reshaped_df.iloc[:, :nobs].values, test_reshaped_df.iloc[:, output_index:].values\n \n # design network\n model = designMLPNetwork(train_X.shape[1], train_Y.shape[1], params)\n \n # fit network\n model.fit(train_X, train_Y, epochs=500, batch_size=1000, verbose=False, shuffle=False)\n \n forecast = model.predict(test_X)\n \n# fcst = [f[0] for f in forecast]\n fcst = forecast\n return fcst", "_____no_output_____" ], [ "def designMLPNetwork(input_shape, output_shape, params):\n model = Sequential()\n model.add(Dense(params['units1'], input_dim=input_shape, activation='relu'))\n model.add(Dropout(params['dropout1']))\n model.add(BatchNormalization())\n\n model.add(Dense(params['units2'], activation='relu'))\n model.add(Dropout(params['dropout2']))\n model.add(BatchNormalization())\n\n if params['choice']['layers'] == 'three':\n model.add(Dense(params['choice']['units3'], activation='relu'))\n model.add(Dropout(params['choice']['dropout3']))\n model.add(BatchNormalization())\n\n model.add(Dense(output_shape, activation='sigmoid'))\n model.compile(loss='mse', optimizer='adam')\n\n return model", "_____no_output_____" ], [ "def rolling_cv_mlp(df, params):\n \n forecasts = []\n order_list = []\n\n limit = df.index[-1].strftime('%Y-%m-%d')\n\n test_end = \"\"\n index = df.index[0]\n\n while test_end < limit :\n print(\"Index: \", index.strftime('%Y-%m-%d')) \n\n train_start, train_end, test_start, test_end = getRollingWindow(index)\n index = index + datetime.timedelta(days=7)\n \n train = df[train_start : train_end]\n test = df[test_start : test_end]\n\n # Perform forecast\n yhat = list(mlp_multi_forecast(train, test, params))\n \n yhat.append(yhat[-1]) #para manter o formato do vetor de metricas\n \n forecasts.append(yhat)\n order_list.append(params['order'])\n\n return forecasts, order_list", "_____no_output_____" ], [ "# Enter best params\nparams_raw = {'batch_size': 64, 'choice': {'layers': 'two'}, 'dropout1': 0.25, 'dropout2': 0.5, 'epochs': 200, 'input': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'order': 2, 'output': ('wp1', 'wp2', 'wp3', 'wp4', 'wp5', 'wp6', 'wp7'), 'units1': 128, 'units2': 128}\n\nforecasts_raw, order_list = rolling_cv_mlp(norm_df, params_raw)\n\nforecasts_final = get_final_forecast(forecasts_raw)\ncalculate_rolling_error(\"rolling_cv_wind_raw_mlp_multi\", df, forecasts_final, order_list)", "Index: 2009-08-01\nIndex: 2009-08-08\nIndex: 2009-08-15\nIndex: 2009-08-22\nIndex: 2009-08-29\nIndex: 2009-09-05\nIndex: 2009-09-12\nIndex: 2009-09-19\nIndex: 2009-09-26\nIndex: 2009-10-03\nIndex: 2009-10-10\nIndex: 2009-10-17\nIndex: 2009-10-24\nIndex: 2009-10-31\nIndex: 2009-11-07\nIndex: 2009-11-14\nIndex: 2009-11-21\nIndex: 2009-11-28\nIndex: 2009-12-05\nIndex: 2009-12-12\nIndex: 2009-12-19\nIndex: 2009-12-26\nIndex: 2010-01-02\nIndex: 2010-01-09\nIndex: 2010-01-16\nIndex: 2010-01-23\nIndex: 2010-01-30\nIndex: 2010-02-06\nIndex: 2010-02-13\nIndex: 2010-02-20\nIndex: 2010-02-27\nIndex: 2010-03-06\nIndex: 2010-03-13\nIndex: 2010-03-20\nIndex: 2010-03-27\nIndex: 2010-04-03\nIndex: 2010-04-10\nIndex: 2010-04-17\nIndex: 2010-04-24\nIndex: 2010-05-01\nIndex: 2010-05-08\nIndex: 2010-05-15\nIndex: 2010-05-22\nIndex: 2010-05-29\nIndex: 2010-06-05\nIndex: 2010-06-12\nIndex: 2010-06-19\nIndex: 2010-06-26\nIndex: 2010-07-03\nIndex: 2010-07-10\n" ], [ "files.download('rolling_cv_wind_raw_mlp_multi.csv')", "_____no_output_____" ] ], [ [ "### Granular FTS", "_____no_output_____" ] ], [ [ "from pyFTS.models.multivariate import granular\nfrom pyFTS.partitioners import Grid, Entropy\nfrom pyFTS.models.multivariate import variable\nfrom pyFTS.common import Membership\nfrom pyFTS.partitioners import Grid, Entropy", "_____no_output_____" ] ], [ [ "#### Granular Parameter Tuning", "_____no_output_____" ] ], [ [ "granular_space = {\n 'npartitions': hp.choice('npartitions', [100, 150, 200]),\n 'order': hp.choice('order', [1, 2]),\n 'knn': hp.choice('knn', [1, 2, 3, 4, 5]),\n 'alpha_cut': hp.choice('alpha_cut', [0, 0.1, 0.2, 0.3]),\n 'input': hp.choice('input', [['wp1', 'wp2', 'wp3']]),\n 'output': hp.choice('output', [['wp1', 'wp2', 'wp3']])}", "_____no_output_____" ], [ "def granular_tuning(train_df, test_df, params):\n _input = list(params['input'])\n _output = list(params['output'])\n _npartitions = params['npartitions']\n _order = params['order']\n _knn = params['knn']\n _alpha_cut = params['alpha_cut']\n _step = params.get('step',1)\n\n ## create explanatory variables\n exp_variables = []\n for vc in _input:\n exp_variables.append(variable.Variable(vc, data_label=vc, alias=vc,\n npart=_npartitions, func=Membership.trimf,\n data=train_df, alpha_cut=_alpha_cut))\n model = granular.GranularWMVFTS(explanatory_variables=exp_variables, target_variable=exp_variables[0], order=_order,\n knn=_knn)\n model.fit(train_df[_input], num_batches=1)\n\n if _step > 1:\n forecast = pd.DataFrame(columns=test_df.columns)\n length = len(test_df.index)\n\n for k in range(0,(length -(_order + _step - 1))):\n fcst = model.predict(test_df[_input], type='multivariate', start_at=k, steps_ahead=_step)\n forecast = forecast.append(fcst.tail(1))\n else:\n forecast = model.predict(test_df[_input], type='multivariate')\n\n return forecast[_output].values\n", "_____no_output_____" ], [ "methods = []\nmethods.append((\"EXP_WIND_GRANULAR\", granular_tuning, granular_space))", "_____no_output_____" ], [ "train_split = 0.6\nrun_search(methods, tuning_df, train_split, Measures.rmse, max_evals=10, resample=None)", "Running experiment: EXP_WIND_GRANULAR\n{'alpha_cut': 0.1, 'input': ('wp1', 'wp2', 'wp3'), 'knn': 1, 'npartitions': 100, 'order': 1, 'output': ('wp1', 'wp2', 'wp3')}\nError : 0.11669905532137337\n{'alpha_cut': 0.2, 'input': ('wp1', 'wp2', 'wp3'), 'knn': 1, 'npartitions': 200, 'order': 2, 'output': ('wp1', 'wp2', 'wp3')}\nError : 0.08229067276531199\n{'alpha_cut': 0.2, 'input': ('wp1', 'wp2', 'wp3'), 'knn': 2, 'npartitions': 200, 'order': 2, 'output': ('wp1', 'wp2', 'wp3')}\nError : 0.08140150942675548\n{'alpha_cut': 0.1, 'input': ('wp1', 'wp2', 'wp3'), 'knn': 1, 'npartitions': 200, 'order': 1, 'output': ('wp1', 'wp2', 'wp3')}\nError : 0.11527883387924612\n{'alpha_cut': 0.2, 'input': ('wp1', 'wp2', 'wp3'), 'knn': 1, 'npartitions': 150, 'order': 1, 'output': ('wp1', 'wp2', 'wp3')}\nError : 0.11642857063129212\n{'alpha_cut': 0.2, 'input': ('wp1', 'wp2', 'wp3'), 'knn': 3, 'npartitions': 100, 'order': 1, 'output': ('wp1', 'wp2', 'wp3')}\nError : 0.10363929653907107\n{'alpha_cut': 0.3, 'input': ('wp1', 'wp2', 'wp3'), 'knn': 5, 'npartitions': 200, 'order': 2, 'output': ('wp1', 'wp2', 'wp3')}\nError : 0.07916522355127716\n{'alpha_cut': 0.3, 'input': ('wp1', 'wp2', 'wp3'), 'knn': 3, 'npartitions': 200, 'order': 2, 'output': ('wp1', 'wp2', 'wp3')}\nError : 0.07938399286248478\n{'alpha_cut': 0.1, 'input': ('wp1', 'wp2', 'wp3'), 'knn': 2, 'npartitions': 150, 'order': 2, 'output': ('wp1', 'wp2', 'wp3')}\nError : 0.08056469602939852\n{'alpha_cut': 0.2, 'input': ('wp1', 'wp2', 'wp3'), 'knn': 5, 'npartitions': 100, 'order': 1, 'output': ('wp1', 'wp2', 'wp3')}\nError : 0.09920669569870488\n100%|██████████| 10/10 [00:09<00:00, 1.05trial/s, best loss: 0.07916522355127716]\nbest parameters: \n{'alpha_cut': 0.3, 'input': ('wp1', 'wp2', 'wp3'), 'knn': 5, 'npartitions': 200, 'order': 2, 'output': ('wp1', 'wp2', 'wp3')}\n" ] ], [ [ "#### Granular Forecasting", "_____no_output_____" ] ], [ [ "def granular_forecast(train_df, test_df, params):\n\n _input = list(params['input'])\n _output = list(params['output'])\n _npartitions = params['npartitions']\n _knn = params['knn']\n _alpha_cut = params['alpha_cut']\n _order = params['order']\n _step = params.get('step',1)\n\n ## create explanatory variables\n exp_variables = []\n for vc in _input:\n exp_variables.append(variable.Variable(vc, data_label=vc, alias=vc,\n npart=_npartitions, func=Membership.trimf,\n data=train_df, alpha_cut=_alpha_cut))\n model = granular.GranularWMVFTS(explanatory_variables=exp_variables, target_variable=exp_variables[0], order=_order,\n knn=_knn)\n model.fit(train_df[_input], num_batches=1)\n\n if _step > 1:\n forecast = pd.DataFrame(columns=test_df.columns)\n length = len(test_df.index)\n\n for k in range(0,(length -(_order + _step - 1))):\n fcst = model.predict(test_df[_input], type='multivariate', start_at=k, steps_ahead=_step)\n forecast = forecast.append(fcst.tail(1))\n else:\n forecast = model.predict(test_df[_input], type='multivariate')\n\n return forecast[_output].values\n", "_____no_output_____" ], [ "def rolling_cv_granular(df, params):\n \n forecasts = []\n order_list = []\n\n limit = df.index[-1].strftime('%Y-%m-%d')\n\n test_end = \"\"\n index = df.index[0]\n\n while test_end < limit :\n print(\"Index: \", index.strftime('%Y-%m-%d')) \n\n train_start, train_end, test_start, test_end = getRollingWindow(index)\n index = index + datetime.timedelta(days=7)\n \n train = df[train_start : train_end]\n test = df[test_start : test_end]\n\n\n # Perform forecast\n yhat = list(granular_forecast(train, test, params))\n \n yhat.append(yhat[-1]) #para manter o formato do vetor de metricas\n \n forecasts.append(yhat)\n order_list.append(params['order'])\n\n return forecasts, order_list", "_____no_output_____" ], [ "def granular_get_final_forecast(forecasts_raw, input):\n \n forecasts_final = []\n l_min = df[input].min()\n l_max = df[input].max()\n\n\n for i in np.arange(len(forecasts_raw)):\n f_raw = denormalize(forecasts_raw[i], l_min, l_max)\n\n forecasts_final.append(f_raw)\n \n return forecasts_final", "_____no_output_____" ], [ "# Enter best params\nparams_raw = {'alpha_cut': 0.3, 'input': ('wp1', 'wp2', 'wp3'), 'knn': 5, 'npartitions': 200, 'order': 2, 'output': ('wp1', 'wp2', 'wp3')}\n\nforecasts_raw, order_list = rolling_cv_granular(norm_df, params_raw)\n\nforecasts_final = granular_get_final_forecast(forecasts_raw, list(params_raw['input']))\ncalculate_rolling_error(\"rolling_cv_wind_raw_granular\", df[list(params_raw['input'])], forecasts_final, order_list)", "Index: 2009-08-01\nIndex: 2009-08-08\nIndex: 2009-08-15\nIndex: 2009-08-22\nIndex: 2009-08-29\nIndex: 2009-09-05\nIndex: 2009-09-12\nIndex: 2009-09-19\nIndex: 2009-09-26\nIndex: 2009-10-03\nIndex: 2009-10-10\nIndex: 2009-10-17\nIndex: 2009-10-24\nIndex: 2009-10-31\nIndex: 2009-11-07\nIndex: 2009-11-14\nIndex: 2009-11-21\nIndex: 2009-11-28\nIndex: 2009-12-05\nIndex: 2009-12-12\nIndex: 2009-12-19\nIndex: 2009-12-26\nIndex: 2010-01-02\nIndex: 2010-01-09\nIndex: 2010-01-16\nIndex: 2010-01-23\nIndex: 2010-01-30\nIndex: 2010-02-06\nIndex: 2010-02-13\nIndex: 2010-02-20\nIndex: 2010-02-27\nIndex: 2010-03-06\nIndex: 2010-03-13\nIndex: 2010-03-20\nIndex: 2010-03-27\nIndex: 2010-04-03\nIndex: 2010-04-10\nIndex: 2010-04-17\nIndex: 2010-04-24\nIndex: 2010-05-01\nIndex: 2010-05-08\nIndex: 2010-05-15\nIndex: 2010-05-22\nIndex: 2010-05-29\nIndex: 2010-06-05\nIndex: 2010-06-12\nIndex: 2010-06-19\nIndex: 2010-06-26\nIndex: 2010-07-03\nIndex: 2010-07-10\n" ], [ "files.download('rolling_cv_wind_raw_granular.csv')", "_____no_output_____" ] ], [ [ "## Result Analysis", "_____no_output_____" ] ], [ [ "import pandas as pd\nfrom google.colab import files", "_____no_output_____" ], [ "files.upload()", "_____no_output_____" ], [ "def createBoxplot(filename, data, xticklabels, ylabel):\n # Create a figure instance\n fig = plt.figure(1, figsize=(9, 6))\n\n # Create an axes instance\n ax = fig.add_subplot(111)\n\n # Create the boxplot\n bp = ax.boxplot(data, patch_artist=True)\n \n ## change outline color, fill color and linewidth of the boxes\n for box in bp['boxes']:\n # change outline color\n box.set( color='#7570b3', linewidth=2)\n # change fill color\n box.set( facecolor = '#AACCFF' )\n\n ## change color and linewidth of the whiskers\n for whisker in bp['whiskers']:\n whisker.set(color='#7570b3', linewidth=2)\n\n ## change color and linewidth of the caps\n for cap in bp['caps']:\n cap.set(color='#7570b3', linewidth=2)\n\n ## change color and linewidth of the medians\n for median in bp['medians']:\n median.set(color='#FFE680', linewidth=2)\n\n ## change the style of fliers and their fill\n for flier in bp['fliers']:\n flier.set(marker='o', color='#e7298a', alpha=0.5)\n \n ## Custom x-axis labels\n ax.set_xticklabels(xticklabels)\n ax.set_ylabel(ylabel)\n plt.show()\n fig.savefig(filename, bbox_inches='tight')", "_____no_output_____" ], [ "var_results = pd.read_csv(\"rolling_cv_wind_raw_var.csv\")\nevolving_results = pd.read_csv(\"rolling_cv_wind_raw_emvfts.csv\")\nmlp_results = pd.read_csv(\"rolling_cv_wind_raw_mlp_multi.csv\")\ngranular_results = pd.read_csv(\"rolling_cv_wind_raw_granular.csv\")", "_____no_output_____" ], [ "metric = 'RMSE'\nresults_data = [evolving_results[metric],var_results[metric], mlp_results[metric], granular_results[metric]]\nxticks = ['e-MVFTS','VAR','MLP','FIG-FTS']\n\nylab = 'RMSE'\ncreateBoxplot(\"e-mvfts_boxplot_rmse_solar\", results_data, xticks, ylab)", "_____no_output_____" ], [ "pd.options.display.float_format = '{:.2f}'.format", "_____no_output_____" ], [ "metric = 'RMSE'\nrmse_df = pd.DataFrame(columns=['e-MVFTS','VAR','MLP','FIG-FTS'])\n\nrmse_df[\"e-MVFTS\"] = evolving_results[metric]\nrmse_df[\"VAR\"] = var_results[metric]\nrmse_df[\"MLP\"] = mlp_results[metric]\nrmse_df[\"FIG-FTS\"] = granular_results[metric]", "_____no_output_____" ], [ "rmse_df.std()", "_____no_output_____" ], [ "metric = 'SMAPE'\nresults_data = [evolving_results[metric],var_results[metric], mlp_results[metric], granular_results[metric]]\nxticks = ['e-MVFTS','VAR','MLP','FIG-FTS']\n\nylab = 'SMAPE'\ncreateBoxplot(\"e-mvfts_boxplot_smape_solar\", results_data, xticks, ylab)", "_____no_output_____" ], [ "metric = 'SMAPE'\nsmape_df = pd.DataFrame(columns=['e-MVFTS','VAR','MLP','FIG-FTS'])\n\nsmape_df[\"e-MVFTS\"] = evolving_results[metric]\nsmape_df[\"VAR\"] = var_results[metric]\nsmape_df[\"MLP\"] = mlp_results[metric]\nsmape_df[\"FIG-FTS\"] = granular_results[metric]", "_____no_output_____" ], [ "smape_df.std()", "_____no_output_____" ], [ "metric = \"RMSE\"\n\ndata = pd.DataFrame(columns=[\"VAR\", \"Evolving\", \"MLP\", \"Granular\"])\n\ndata[\"VAR\"] = var_results[metric]\ndata[\"Evolving\"] = evolving_results[metric]\ndata[\"MLP\"] = mlp_results[metric]\ndata[\"Granular\"] = granular_results[metric]\n\nax = data.plot(figsize=(18,6))\nax.set(xlabel='Window', ylabel=metric)\nfig = ax.get_figure()\n#fig.savefig(path_images + exp_id + \"_prequential.png\")\n \nx = np.arange(len(data.columns.values))\nnames = data.columns.values\nvalues = data.mean().values\nplt.figure(figsize=(5,6))\nplt.bar(x, values, align='center', alpha=0.5, width=0.9)\nplt.xticks(x, names)\n#plt.yticks(np.arange(0, 1.1, 0.1))\nplt.ylabel(metric)\n#plt.savefig(path_images + exp_id + \"_bars.png\")", "_____no_output_____" ], [ "metric = \"SMAPE\"\n\ndata = pd.DataFrame(columns=[\"VAR\", \"Evolving\", \"MLP\", \"Granular\"])\n\ndata[\"VAR\"] = var_results[metric]\ndata[\"Evolving\"] = evolving_results[metric]\ndata[\"MLP\"] = mlp_results[metric]\ndata[\"Granular\"] = granular_results[metric]\n\nax = data.plot(figsize=(18,6))\nax.set(xlabel='Window', ylabel=metric)\nfig = ax.get_figure()\n#fig.savefig(path_images + exp_id + \"_prequential.png\")\n \nx = np.arange(len(data.columns.values))\nnames = data.columns.values\nvalues = data.mean().values\nplt.figure(figsize=(5,6))\nplt.bar(x, values, align='center', alpha=0.5, width=0.9)\nplt.xticks(x, names)\n#plt.yticks(np.arange(0, 1.1, 0.1))\nplt.ylabel(metric)\n#plt.savefig(path_images + exp_id + \"_bars.png\")", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# Use `Lale` `AIF360` scorers to calculate and mitigate bias for credit risk AutoAI model", "_____no_output_____" ], [ "This notebook contains the steps and code to demonstrate support of AutoAI experiments in Watson Machine Learning service. It introduces commands for bias detecting and mitigation performed with `lale.lib.aif360` module.\n\nSome familiarity with Python is helpful. This notebook uses Python 3.8.", "_____no_output_____" ], [ "## Contents\n\nThis notebook contains the following parts:\n\n1.\t[Setup](#setup)\n2.\t[Optimizer definition](#definition)\n3.\t[Experiment Run](#run)\n4.\t[Pipeline bias detection and mitigation](#bias)\n5. [Deployment and score](#scoring)\n6. [Clean up](#cleanup)\n7.\t[Summary and next steps](#summary)", "_____no_output_____" ], [ "<a id=\"setup\"></a>\n## 1. Set up the environment\n\nIf you are not familiar with <a href=\"https://console.ng.bluemix.net/catalog/services/ibm-watson-machine-learning/\" target=\"_blank\" rel=\"noopener no referrer\">Watson Machine Learning (WML) Service</a> and AutoAI experiments please read more about it in the sample notebook: <a href=\"https://github.com/IBM/watson-machine-learning-samples/blob/master/cloud/notebooks/python_sdk/experiments/autoai/Use%20AutoAI%20and%20Lale%20to%20predict%20credit%20risk.ipynb\" target=\"_blank\" rel=\"noopener no referrer\">\"Use AutoAI and Lale to predict credit risk with `ibm-watson-machine-learning`\"</a>", "_____no_output_____" ], [ "### Install and import the `ibm-watson-machine-learning`, `lale` ,`aif360` and dependencies.\n**Note:** `ibm-watson-machine-learning` documentation can be found <a href=\"http://ibm-wml-api-pyclient.mybluemix.net/\" target=\"_blank\" rel=\"noopener no referrer\">here</a>.", "_____no_output_____" ] ], [ [ "!pip install -U ibm-watson-machine-learning | tail -n 1\n!pip install -U scikit-learn==0.23.2 | tail -n 1\n!pip install -U autoai-libs | tail -n 1\n!pip install -U lale | tail -n 1\n!pip install -U aif360 | tail -n 1\n!pip install -U liac-arff | tail -n 1\n!pip install -U cvxpy | tail -n 1\n!pip install -U fairlearn | tail -n 1", "_____no_output_____" ] ], [ [ "### Connection to WML\n\nAuthenticate the Watson Machine Learning service on IBM Cloud. You need to provide Cloud `API key` and `location`.\n\n**Tip**: Your `Cloud API key` can be generated by going to the [**Users** section of the Cloud console](https://cloud.ibm.com/iam#/users). From that page, click your name, scroll down to the **API Keys** section, and click **Create an IBM Cloud API key**. Give your key a name and click **Create**, then copy the created key and paste it below. You can also get a service specific url by going to the [**Endpoint URLs** section of the Watson Machine Learning docs](https://cloud.ibm.com/apidocs/machine-learning). You can check your instance location in your <a href=\"https://console.ng.bluemix.net/catalog/services/ibm-watson-machine-learning/\" target=\"_blank\" rel=\"noopener no referrer\">Watson Machine Learning (WML) Service</a> instance details.\n\n\nYou can use [IBM Cloud CLI](https://cloud.ibm.com/docs/cli/index.html) to retrieve the instance `location`.\n\n```\nibmcloud login --apikey API_KEY -a https://cloud.ibm.com\nibmcloud resource service-instance WML_INSTANCE_NAME\n```\n\n\n**NOTE:** You can also get a service specific apikey by going to the [**Service IDs** section of the Cloud Console](https://cloud.ibm.com/iam/serviceids). From that page, click **Create**, and then copy the created key and paste it in the following cell. \n\n\n**Action**: Enter your `api_key` and `location` in the following cell.", "_____no_output_____" ] ], [ [ "api_key = 'PUT_YOUR_KEY_HERE'\nlocation = 'us-south'", "_____no_output_____" ], [ "wml_credentials = {\n \"apikey\": api_key,\n \"url\": 'https://' + location + '.ml.cloud.ibm.com'\n}", "_____no_output_____" ], [ "from ibm_watson_machine_learning import APIClient\n\nclient = APIClient(wml_credentials)", "_____no_output_____" ] ], [ [ "### Working with spaces\n\nYou need to create a space that will be used for your work. If you do not have a space, you can use [Deployment Spaces Dashboard](https://dataplatform.cloud.ibm.com/ml-runtime/spaces?context=cpdaas) to create one.\n\n- Click **New Deployment Space**\n- Create an empty space\n- Select Cloud Object Storage\n- Select Watson Machine Learning instance and press **Create**\n- Copy `space_id` and paste it below\n\n**Tip**: You can also use SDK to prepare the space for your work. More information can be found [here](https://github.com/IBM/watson-machine-learning-samples/blob/master/cloud/notebooks/python_sdk/instance-management/Space%20management.ipynb).\n\n**Action**: assign space ID below\n", "_____no_output_____" ] ], [ [ "space_id = 'PASTE YOUR SPACE ID HERE'", "_____no_output_____" ], [ "client.spaces.list(limit=10)", "_____no_output_____" ], [ "client.set.default_space(space_id)", "_____no_output_____" ] ], [ [ "### Connections to COS\n\nIn next cell we read the COS credentials from the space.", "_____no_output_____" ] ], [ [ "cos_credentials = client.spaces.get_details(space_id=space_id)['entity']['storage']['properties']", "_____no_output_____" ] ], [ [ "<a id=\"definition\"></a>\n## 2. Optimizer definition", "_____no_output_____" ], [ "### Training data connection\n\nDefine connection information to COS bucket and training data CSV file. This example uses the [German Credit Risk dataset](https://raw.githubusercontent.com/IBM/watson-machine-learning-samples/master/cloud/data/credit_risk/credit_risk_training_light.csv).\n\nThe code in next cell uploads training data to the bucket.", "_____no_output_____" ] ], [ [ "filename = 'german_credit_data_biased_training.csv'\ndatasource_name = 'bluemixcloudobjectstorage'\nbucketname = cos_credentials['bucket_name']", "_____no_output_____" ] ], [ [ "Download training data from git repository and split for training and test set.", "_____no_output_____" ] ], [ [ "import os, wget\nimport pandas as pd\nimport numpy as np\nfrom sklearn.model_selection import train_test_split\n\nurl = 'https://raw.githubusercontent.com/IBM/watson-machine-learning-samples/master/cloud/data/credit_risk/german_credit_data_biased_training.csv'\nif not os.path.isfile(filename): wget.download(url)\n\ncredit_risk_df = pd.read_csv(filename)\n\nX = credit_risk_df.drop(['Risk'], axis=1)\ny = credit_risk_df['Risk']\n\nX_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1)\n\ncredit_risk_df.head()", "_____no_output_____" ] ], [ [ "#### Create connection", "_____no_output_____" ] ], [ [ "conn_meta_props= {\n client.connections.ConfigurationMetaNames.NAME: f\"Connection to Database - {datasource_name} \",\n client.connections.ConfigurationMetaNames.DATASOURCE_TYPE: client.connections.get_datasource_type_uid_by_name(datasource_name),\n client.connections.ConfigurationMetaNames.DESCRIPTION: \"Connection to external Database\",\n client.connections.ConfigurationMetaNames.PROPERTIES: {\n 'bucket': bucketname,\n 'access_key': cos_credentials['credentials']['editor']['access_key_id'],\n 'secret_key': cos_credentials['credentials']['editor']['secret_access_key'],\n 'iam_url': 'https://iam.cloud.ibm.com/identity/token',\n 'url': cos_credentials['endpoint_url']\n }\n}\n\nconn_details = client.connections.create(meta_props=conn_meta_props)", "_____no_output_____" ] ], [ [ "**Note**: The above connection can be initialized alternatively with `api_key` and `resource_instance_id`. \nThe above cell can be replaced with:\n\n\n```\nconn_meta_props= {\n client.connections.ConfigurationMetaNames.NAME: f\"Connection to Database - {db_name} \",\n client.connections.ConfigurationMetaNames.DATASOURCE_TYPE: client.connections.get_datasource_type_uid_by_name(db_name),\n client.connections.ConfigurationMetaNames.DESCRIPTION: \"Connection to external Database\",\n client.connections.ConfigurationMetaNames.PROPERTIES: {\n 'bucket': bucket_name,\n 'api_key': cos_credentials['apikey'],\n 'resource_instance_id': cos_credentials['resource_instance_id'],\n 'iam_url': 'https://iam.cloud.ibm.com/identity/token',\n 'url': 'https://s3.us.cloud-object-storage.appdomain.cloud'\n }\n}\n\nconn_details = client.connections.create(meta_props=conn_meta_props)\n\n```", "_____no_output_____" ] ], [ [ "connection_id = client.connections.get_uid(conn_details)", "_____no_output_____" ] ], [ [ "Define connection information to training data and upload train dataset to COS bucket.\n", "_____no_output_____" ] ], [ [ "from ibm_watson_machine_learning.helpers import DataConnection, S3Location\n\n\ncredit_risk_conn = DataConnection(\n connection_asset_id=connection_id,\n location=S3Location(bucket=bucketname,\n path=filename))\n\ncredit_risk_conn._wml_client = client\ntraining_data_reference=[credit_risk_conn]\n\n\ncredit_risk_conn.write(data=X_train.join(y_train), remote_name=filename)", "_____no_output_____" ] ], [ [ "### Optimizer configuration\n\nProvide the input information for AutoAI optimizer:\n- `name` - experiment name\n- `prediction_type` - type of the problem\n- `prediction_column` - target column name\n- `scoring` - optimization metric\n- `daub_include_only_estimators` - estimators which will be included during AutoAI training. More available estimators can be found in `experiment.ClassificationAlgorithms` enum", "_____no_output_____" ] ], [ [ "from ibm_watson_machine_learning.experiment import AutoAI\n\nexperiment = AutoAI(wml_credentials, space_id=space_id)\n\npipeline_optimizer = experiment.optimizer(\n name='Credit Risk Bias detection in AutoAI',\n prediction_type=AutoAI.PredictionType.BINARY,\n prediction_column='Risk',\n scoring=AutoAI.Metrics.ROC_AUC_SCORE,\n include_only_estimators=[experiment.ClassificationAlgorithms.XGB] \n)", "_____no_output_____" ] ], [ [ "<a id=\"run\"></a>\n## 3. Experiment run\n\nCall the `fit()` method to trigger the AutoAI experiment. You can either use interactive mode (synchronous job) or background mode (asychronous job) by specifying `background_model=True`.", "_____no_output_____" ] ], [ [ "run_details = pipeline_optimizer.fit(\n training_data_reference=training_data_reference,\n background_mode=False)", "Training job c646c56a-b7b7-4090-bfe6-d57327a76da6 completed: 100%|████████| [03:21<00:00, 2.01s/it]\n" ], [ "pipeline_optimizer.get_run_status()", "_____no_output_____" ], [ "summary = pipeline_optimizer.summary()\nsummary", "_____no_output_____" ] ], [ [ "### Get selected pipeline model\n\nDownload pipeline model object from the AutoAI training job.", "_____no_output_____" ] ], [ [ "best_pipeline = pipeline_optimizer.get_pipeline()", "_____no_output_____" ] ], [ [ "<a id=\"bias\"></a>\n## 4. Bias detection and mitigation\n\nThe `fairness_info` dictionary contains some fairness-related metadata. The favorable and unfavorable label are values of the target class column that indicate whether the loan was granted or denied. A protected attribute is a feature that partitions the population into groups whose outcome should have parity. The credit-risk dataset has two protected attribute columns, sex and age. Each prottected attributes has privileged and unprivileged group.\n\nNote that to use fairness metrics from lale with numpy arrays `protected_attributes.feature` need to be passed as index of the column in dataset, not as name.", "_____no_output_____" ] ], [ [ "fairness_info = {'favorable_labels': ['No Risk'],\n 'protected_attributes': [\n {'feature': X.columns.get_loc('Sex'),'reference_group': ['male']},\n {'feature': X.columns.get_loc('Age'), 'reference_group': [[26, 40]]}]}\nfairness_info", "_____no_output_____" ] ], [ [ "### Calculate fairness metrics", "_____no_output_____" ], [ "We will calculate some model metrics. Accuracy describes how accurate is the model according to dataset. \nDisparate impact is defined by comparing outcomes between a privileged group and an unprivileged group, \nso it needs to check the protected attribute to determine group membership for the sample record at hand.\nThe third calculated metric takes the disparate impact into account along with accuracy. The best value of the score is 1.0.", "_____no_output_____" ] ], [ [ "import sklearn.metrics\nfrom lale.lib.aif360 import disparate_impact, accuracy_and_disparate_impact\n\naccuracy_scorer = sklearn.metrics.make_scorer(sklearn.metrics.accuracy_score)\nprint(f'accuracy {accuracy_scorer(best_pipeline, X_test.values, y_test.values):.1%}')\ndisparate_impact_scorer = disparate_impact(**fairness_info)\nprint(f'disparate impact {disparate_impact_scorer(best_pipeline, X_test.values, y_test.values):.2f}')\ncombined_scorer = accuracy_and_disparate_impact(**fairness_info)\nprint(f'accuracy and disparate impact metric {combined_scorer(best_pipeline, X_test.values, y_test.values):.2f}')", "accuracy 82.4%\ndisparate impact 0.68\naccuracy and disparate impact metric 0.26\n" ] ], [ [ "### Mitigation\n\n`Hyperopt` minimizes (best_score - score_returned_by_the_scorer), where best_score is an argument to Hyperopt and score_returned_by_the_scorer is the value returned by the scorer for each evaluation point. We will use the `Hyperopt` to tune hyperparametres of the AutoAI pipeline to get new and more fair model. \n", "_____no_output_____" ] ], [ [ "from sklearn.linear_model import LogisticRegression as LR\nfrom sklearn.tree import DecisionTreeClassifier as Tree\nfrom sklearn.neighbors import KNeighborsClassifier as KNN\nfrom lale.lib.lale import Hyperopt\nfrom lale.lib.aif360 import FairStratifiedKFold\nfrom lale import wrap_imported_operators\n\nwrap_imported_operators()", "_____no_output_____" ], [ "prefix = best_pipeline.remove_last().freeze_trainable()\nprefix.visualize()", "_____no_output_____" ], [ "new_pipeline = prefix >> (LR | Tree | KNN)\nnew_pipeline.visualize()", "_____no_output_____" ], [ "fair_cv = FairStratifiedKFold(**fairness_info, n_splits=3)\n\npipeline_fairer = new_pipeline.auto_configure(\n X_train.values, y_train.values, optimizer=Hyperopt, cv=fair_cv,\n max_evals=10, scoring=combined_scorer, best_score=1.0)", "100%|██████████| 10/10 [01:13<00:00, 7.35s/trial, best loss: 0.27222222222222214]\n" ] ], [ [ "As with any trained model, we can evaluate and visualize the result.", "_____no_output_____" ] ], [ [ "print(f'accuracy {accuracy_scorer(pipeline_fairer, X_test.values, y_test.values):.1%}')\nprint(f'disparate impact {disparate_impact_scorer(pipeline_fairer, X_test.values, y_test.values):.2f}')\nprint(f'accuracy and disparate impact metric {combined_scorer(pipeline_fairer, X_test.values, y_test.values):.2f}')\npipeline_fairer.visualize()", "accuracy 75.8%\ndisparate impact 0.86\naccuracy and disparate impact metric 0.63\n" ] ], [ [ "As the result demonstrates, the best model found by AI Automation\nhas lower accuracy and much better disparate impact as the one we saw\nbefore. Also, it has tuned the repair level and\nhas picked and tuned a classifier. These results may vary by dataset and search space.", "_____no_output_____" ], [ "You can get source code of the created pipeline. You just need to change the below cell type `Raw NBCovert` to `code`.", "_____no_output_____" ] ], [ [ "pipeline_fairer.pretty_print(ipython_display=True, show_imports=False)", "_____no_output_____" ] ], [ [ "<a id=\"scoring\"></a>\n## 5. Deploy and Score\nIn this section you will learn how to deploy and score Lale pipeline model using WML instance.", "_____no_output_____" ], [ "#### Custom software_specification", "_____no_output_____" ], [ "Created model is AutoAI model refined with Lale. We will create new software specification based on default Python 3.7 \nenvironment extended by `autoai-libs` package.", "_____no_output_____" ] ], [ [ "base_sw_spec_uid = client.software_specifications.get_uid_by_name(\"default_py3.7\")\nprint(\"Id of default Python 3.7 software specification is: \", base_sw_spec_uid)", "Id of default Python 3.7 software specification is: e4429883-c883-42b6-87a8-f419d64088cd\n" ], [ "url = 'https://raw.githubusercontent.com/IBM/watson-machine-learning-samples/master/cloud/configs/config.yaml'\nif not os.path.isfile('config.yaml'): wget.download(url)", "_____no_output_____" ], [ "!cat config.yaml", "name: python37\nchannels:\n - defaults\ndependencies:\n - pip:\n - autoai-libs\n\nprefix: /opt/anaconda3/envs/python37" ] ], [ [ "`config.yaml` file describes details of package extention. Now you need to store new package extention with `APIClient`.", "_____no_output_____" ] ], [ [ "meta_prop_pkg_extn = {\n client.package_extensions.ConfigurationMetaNames.NAME: \"Scikt with autoai-libs\",\n client.package_extensions.ConfigurationMetaNames.DESCRIPTION: \"Pkg extension for autoai-libs\",\n client.package_extensions.ConfigurationMetaNames.TYPE: \"conda_yml\"\n}\n\npkg_extn_details = client.package_extensions.store(meta_props=meta_prop_pkg_extn, file_path=\"config.yaml\")\npkg_extn_uid = client.package_extensions.get_uid(pkg_extn_details)\npkg_extn_url = client.package_extensions.get_href(pkg_extn_details)", "Creating package extensions\nSUCCESS\n" ] ], [ [ "Create new software specification and add created package extention to it. ", "_____no_output_____" ] ], [ [ "meta_prop_sw_spec = {\n client.software_specifications.ConfigurationMetaNames.NAME: \"Mitigated AutoAI bases on scikit spec\",\n client.software_specifications.ConfigurationMetaNames.DESCRIPTION: \"Software specification for scikt with autoai-libs\",\n client.software_specifications.ConfigurationMetaNames.BASE_SOFTWARE_SPECIFICATION: {\"guid\": base_sw_spec_uid}\n}\n\nsw_spec_details = client.software_specifications.store(meta_props=meta_prop_sw_spec)\nsw_spec_uid = client.software_specifications.get_uid(sw_spec_details)\n\n\nstatus = client.software_specifications.add_package_extension(sw_spec_uid, pkg_extn_uid)", "SUCCESS\n" ] ], [ [ "You can get details of created software specification using `client.software_specifications.get_details(sw_spec_uid)`", "_____no_output_____" ], [ "### Store the model", "_____no_output_____" ] ], [ [ "model_props = {\n client.repository.ModelMetaNames.NAME: \"Fairer AutoAI model\",\n client.repository.ModelMetaNames.TYPE: 'scikit-learn_0.23',\n client.repository.ModelMetaNames.SOFTWARE_SPEC_UID: sw_spec_uid\n \n}\nfeature_vector = list(X.columns)", "_____no_output_____" ], [ "published_model = client.repository.store_model(\n model=best_pipeline.export_to_sklearn_pipeline(), \n meta_props=model_props,\n training_data=X_train.values,\n training_target=y_train.values,\n feature_names=feature_vector,\n label_column_names=['Risk']\n)", "_____no_output_____" ], [ "published_model_uid = client.repository.get_model_id(published_model)", "_____no_output_____" ] ], [ [ "### Deployment creation", "_____no_output_____" ] ], [ [ "metadata = {\n client.deployments.ConfigurationMetaNames.NAME: \"Deployment of fairer model\",\n client.deployments.ConfigurationMetaNames.ONLINE: {}\n}\n\ncreated_deployment = client.deployments.create(published_model_uid, meta_props=metadata)", "\n\n#######################################################################################\n\nSynchronous deployment creation for uid: '4f0eafb3-e057-4af6-afdf-c8a7075c0362' started\n\n#######################################################################################\n\n\ninitializing..........................................................................................................................................\nready\n\n\n------------------------------------------------------------------------------------------------\nSuccessfully finished deployment creation, deployment_uid='a608ee59-63fd-43ed-8eef-940b6c7d8345'\n------------------------------------------------------------------------------------------------\n\n\n" ], [ "deployment_id = client.deployments.get_uid(created_deployment)", "_____no_output_____" ] ], [ [ "#### Deployment scoring ", "_____no_output_____" ], [ "You need to pass scoring values as input data if the deployed model. Use `client.deployments.score()` method to get predictions from deployed model. ", "_____no_output_____" ] ], [ [ "values = X_test.values\n\nscoring_payload = {\n \"input_data\": [{\n 'values': values[:5]\n }]\n}", "_____no_output_____" ], [ "predictions = client.deployments.score(deployment_id, scoring_payload)\npredictions", "_____no_output_____" ] ], [ [ "<a id=\"cleanup\"></a>\n## 5. Clean up", "_____no_output_____" ], [ "If you want to clean up all created assets:\n- experiments\n- trainings\n- pipelines\n- model definitions\n- models\n- functions\n- deployments\n\nplease follow up this sample [notebook](https://github.com/IBM/watson-machine-learning-samples/blob/master/cloud/notebooks/python_sdk/instance-management/Machine%20Learning%20artifacts%20management.ipynb).", "_____no_output_____" ], [ "<a id=\"summary\"></a>\n## 6. Summary and next steps", "_____no_output_____" ], [ " You successfully completed this notebook!.\n\nCheck out used packeges domuntations:\n- `ibm-watson-machine-learning` [Online Documentation](https://www.ibm.com/cloud/watson-studio/autoai)\n- `lale`: https://github.com/IBM/lale\n- `aif360`: https://aif360.mybluemix.net/", "_____no_output_____" ], [ "### Authors \n\n**Dorota Dydo-Rożniecka**, Intern in Watson Machine Learning at IBM", "_____no_output_____" ], [ "Copyright © 2020, 2021 IBM. This notebook and its source code are released under the terms of the MIT License.", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "raw", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown", "markdown", "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "raw" ], [ "markdown", "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown", "markdown", "markdown", "markdown", "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Trade-off between classification accuracy and reconstruction error during dimensionality reduction\n\n- Low-dimensional LSTM representations are excellent at dimensionality reduction, but are poor at reconstructing the original data\n- On the other hand, PCs are excellent at reconstructing the original data but these high-variance components do not preserve class information", "_____no_output_____" ] ], [ [ "import numpy as np\nimport pandas as pd\nimport scipy as sp\nimport pickle\nimport os\nimport random\nimport sys\n\n# visualizations\nfrom _plotly_future_ import v4_subplots\nimport plotly.offline as py\npy.init_notebook_mode(connected=True)\nimport plotly.graph_objs as go\nimport plotly.subplots as tls\nimport plotly.figure_factory as ff\nimport plotly.io as pio\nimport plotly.express as px\npio.templates.default = 'plotly_white'\npio.orca.config.executable = '/home/joyneelm/fire/bin/orca'\ncolors = px.colors.qualitative.Plotly", "_____no_output_____" ], [ "class ARGS():\n roi = 300\n net = 7\n subnet = 'wb'\n train_size = 100\n batch_size = 32\n num_epochs = 50\n zscore = 1\n \n #gru\n k_hidden = 32\n k_layers = 1\n dims = [3, 4, 5, 10]", "_____no_output_____" ], [ "args = ARGS()", "_____no_output_____" ], [ "def _get_results(k_dim):\n \n RES_DIR = 'results/clip_gru_recon'\n load_path = (RES_DIR + \n '/roi_%d_net_%d' %(args.roi, args.net) + \n '_trainsize_%d' %(args.train_size) +\n '_k_hidden_%d' %(args.k_hidden) +\n '_kdim_%d' %(k_dim) +\n '_k_layers_%d' %(args.k_layers) +\n '_batch_size_%d' %(args.batch_size) +\n '_num_epochs_45' +\n '_z_%d.pkl' %(args.zscore))\n \n with open(load_path, 'rb') as f:\n results = pickle.load(f)\n# print(results.keys())\n return results", "_____no_output_____" ], [ "r = {}\nfor k_dim in args.dims:\n r[k_dim] = _get_results(k_dim)", "_____no_output_____" ], [ "def _plot_fig(ss):\n \n title_text = ss\n if ss=='var':\n ss = 'mse'\n invert = True\n else:\n invert = False\n \n subplot_titles = ['train', 'test']\n fig = tls.make_subplots(rows=1, \n cols=2, \n subplot_titles=subplot_titles,\n print_grid=False)\n\n for ii, x in enumerate(['train', 'test']):\n gru_score = {'mean':[], 'ste':[]}\n pca_score = {'mean':[], 'ste':[]}\n for k_dim in args.dims:\n\n a = r[k_dim]\n \n # gru decoder\n y = np.mean(a['%s_%s'%(x, ss)])\n gru_score['mean'].append(y)\n \n # pca decoder\n y = np.mean(a['%s_pca_%s'%(x, ss)])\n pca_score['mean'].append(y)\n \n x = np.arange(len(args.dims))\n if invert:\n y = 1 - np.array(gru_score['mean'])\n else:\n y = gru_score['mean']\n error_y = gru_score['ste']\n trace = go.Bar(x=x, y=y,\n name='lstm decoder',\n marker_color=colors[0])\n fig.add_trace(trace, 1, ii+1)\n\n if invert:\n y = 1 - np.array(pca_score['mean'])\n else:\n y = pca_score['mean']\n error_y = pca_score['ste']\n trace = go.Bar(x=x, y=y,\n name='pca recon',\n marker_color=colors[1])\n fig.add_trace(trace, 1, ii+1)\n\n fig.update_xaxes(tickvals=np.arange(len(args.dims)),\n ticktext=args.dims)\n fig.update_layout(height=350, width=700,\n title_text=title_text)\n \n return fig", "_____no_output_____" ] ], [ [ "## Mean-squared error vs number of dimensions", "_____no_output_____" ] ], [ [ "'''\nmse\n'''\nss = 'mse'\nfig = _plot_fig(ss)\nfig.show()", "_____no_output_____" ] ], [ [ "## Variance captured vs number of dimensions", "_____no_output_____" ] ], [ [ "'''\nvariance\n'''\nss = 'var'\nfig = _plot_fig(ss)\nfig.show()", "_____no_output_____" ] ], [ [ "## R-squared vs number of dimensions", "_____no_output_____" ] ], [ [ "'''\nr2\n'''\nss = 'r2'\nfig = _plot_fig(ss)\nfig.show()", "_____no_output_____" ], [ "results = r[10]\n\n# variance not captured by pca recon\npca_not = 1 - np.sum(results['pca_var'])\nprint('percent variance captured by pca components = %0.3f' %(1 - pca_not))\n# this is proportional to pca mse\npca_mse = results['test_pca_mse']\n\n# variance not captured by lstm decoder?\nlstm_mse = results['test_mse']\n\nlstm_not = lstm_mse*(pca_not/pca_mse)\nprint('percent variance captured by lstm recon = %0.3f' %(1 - lstm_not))", "percent variance captured by pca components = 0.611\npercent variance captured by lstm recon = 0.067\n" ], [ "def _plot_fig_ext(ss):\n \n title_text = ss\n if ss=='var':\n ss = 'mse'\n invert = True\n else:\n invert = False\n \n subplot_titles = ['train', 'test']\n fig = go.Figure()\n\n x = 'test'\n \n lstm_score = {'mean':[], 'ste':[]}\n pca_score = {'mean':[], 'ste':[]}\n lstm_acc = {'mean':[], 'ste':[]}\n pc_acc = {'mean':[], 'ste':[]}\n for k_dim in args.dims:\n\n a = r[k_dim]\n # lstm encoder\n k_sub = len(a['test'])\n y = np.mean(a['test'])\n error_y = 3/np.sqrt(k_sub)*np.std(a['test'])\n lstm_acc['mean'].append(y)\n lstm_acc['ste'].append(error_y)\n\n # lstm decoder\n y = np.mean(a['%s_%s'%(x, ss)])\n lstm_score['mean'].append(y)\n lstm_score['ste'].append(error_y)\n\n # pca encoder\n b = r_pc[k_dim]\n y = np.mean(b['test'])\n error_y = 3/np.sqrt(k_sub)*np.std(b['test'])\n pc_acc['mean'].append(y)\n pc_acc['ste'].append(error_y)\n\n # pca decoder\n y = np.mean(a['%s_pca_%s'%(x, ss)])\n pca_score['mean'].append(y)\n pca_score['ste'].append(error_y)\n\n x = np.arange(len(args.dims))\n\n y = lstm_acc['mean']\n error_y = lstm_acc['ste']\n trace = go.Bar(x=x, y=y,\n name='GRU Accuracy',\n error_y=dict(type='data',\n array=error_y),\n marker_color=colors[3])\n fig.add_trace(trace)\n\n y = pc_acc['mean']\n error_y = pc_acc['ste']\n trace = go.Bar(x=x, y=y,\n name='PCA Accuracy',\n error_y=dict(type='data',\n array=error_y),\n marker_color=colors[4])\n fig.add_trace(trace)\n \n if invert:\n y = 1 - np.array(lstm_score['mean'])\n else:\n y = lstm_score['mean']\n error_y = lstm_score['ste']\n trace = go.Bar(x=x, y=y,\n name='GRU Reconstruction',\n error_y=dict(type='data',\n array=error_y),\n marker_color=colors[5])\n fig.add_trace(trace)\n\n if invert:\n y = 1 - np.array(pca_score['mean'])\n else:\n y = pca_score['mean']\n error_y = pca_score['ste']\n trace = go.Bar(x=x, y=y,\n name='PCA Reconstruction',\n error_y=dict(type='data', \n array=error_y),\n marker_color=colors[2])\n fig.add_trace(trace)\n fig.update_yaxes(title=dict(text='Accuracy or % variance',\n font_size=20),\n gridwidth=1, gridcolor='#bfbfbf',\n tickfont=dict(size=20))\n fig.update_xaxes(title=dict(text='Number of dimensions',\n font_size=20),\n tickvals=np.arange(len(args.dims)),\n ticktext=args.dims,\n tickfont=dict(size=20))\n fig.update_layout(height=470, width=570,\n font_color='black',\n legend_orientation='h',\n legend_font_size=20,\n legend_x=-0.1,\n legend_y=-0.3)\n \n return fig", "_____no_output_____" ], [ "def _get_pc_results(PC_DIR, k_dim):\n load_path = (PC_DIR + \n '/roi_%d_net_%d' %(args.roi, args.net) + \n '_nw_%s' %(args.subnet) +\n '_trainsize_%d' %(args.train_size) +\n '_kdim_%d_batch_size_%d' %(k_dim, args.batch_size) +\n '_num_epochs_%d_z_%d.pkl' %(args.num_epochs, args.zscore))\n\n with open(load_path, 'rb') as f:\n results = pickle.load(f)\n print(results.keys()) \n return results", "_____no_output_____" ] ], [ [ "## Comparison of LSTM and PCA: classification accuracy and variance captured", "_____no_output_____" ] ], [ [ "'''\nvariance\n'''\nr_pc = {}\nPC_DIR = 'results/clip_pca'\nfor k_dim in args.dims:\n r_pc[k_dim] = _get_pc_results(PC_DIR, k_dim)\ncolors = px.colors.qualitative.Set3\n#colors = [\"#D55E00\", \"#009E73\", \"#56B4E9\", \"#E69F00\"]\nss = 'var'\nfig = _plot_fig_ext(ss)\nfig.show()\nfig.write_image('figures/fig3c.png')", "dict_keys(['train', 'val', 't_train', 't_test', 'test'])\ndict_keys(['train', 'val', 't_train', 't_test', 'test'])\ndict_keys(['train', 'val', 't_train', 't_test', 'test'])\ndict_keys(['train', 'val', 't_train', 't_test', 'test'])\n" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "%pylab inline\nfrom jax.scipy.ndimage import map_coordinates\nfrom constant import *\nimport warnings\nfrom jax import jit, partial, vmap\nfrom tqdm import tqdm\nwarnings.filterwarnings(\"ignore\")", "Populating the interactive namespace from numpy and matplotlib\n" ] ], [ [ "### State \n$$x = [w,n,m,s,e,o]$$ \n$w$: wealth level size: 20 \n$n$: 401k level size: 10 \n$m$: mortgage level size: 10 \n$s$: economic state size: 8 \n$e$: employment state size: 2 \n$o$: housing state: size: 2 \n\n### Action\n$c$: consumption amount size: 20 \n$b$: bond investment size: 20 \n$k$: stock investment derived from budget constrain once $c$ and $b$ are determined. \n$h$: housing consumption size, related to housing status and consumption level \n\nIf $O = 1$, the agent owns a house: \n$A = [c, b, k, h=H, action = 1]$ sold the house \n$A = [c, b, k, h=H, action = 0]$ keep the house \n\nIf $O = 0$, the agent do not own a house: \n$A = [c, b, k, h= \\frac{c}{\\alpha} \\frac{1-\\alpha}{pr}, action = 0]$ keep renting the house \n$A = [c, b, k, h= \\frac{c}{\\alpha} \\frac{1-\\alpha}{pr}, action = 1]$ buy a housing with H unit \n\n### Housing\n20% down payment of mortgage, fix mortgage rate, single housing unit available, from age between 20 and 50, agents could choose to buy a house, and could choose to sell the house at any moment. $H = 1000$ ", "_____no_output_____" ] ], [ [ "%%time\nfor t in tqdm(range(T_max-1,T_min-1, -1)):\n if t == T_max-1:\n v,cbkha = vmap(partial(V,t,Vgrid[:,:,:,:,:,:,t]))(Xs)\n else:\n v,cbkha = vmap(partial(V,t,Vgrid[:,:,:,:,:,:,t+1]))(Xs)\n Vgrid[:,:,:,:,:,:,t] = v.reshape(dim)\n cgrid[:,:,:,:,:,:,t] = cbkha[:,0].reshape(dim)\n bgrid[:,:,:,:,:,:,t] = cbkha[:,1].reshape(dim)\n kgrid[:,:,:,:,:,:,t] = cbkha[:,2].reshape(dim)\n hgrid[:,:,:,:,:,:,t] = cbkha[:,3].reshape(dim)\n agrid[:,:,:,:,:,:,t] = cbkha[:,4].reshape(dim)", "100%|██████████| 60/60 [54:21<00:00, 54.36s/it]" ], [ "np.save(\"Value00\",Vgrid)", "_____no_output_____" ] ] ]

[ "code", "markdown", "code" ]

[ [ "code" ], [ "markdown" ], [ "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "imatlab_export_fig('print-png')", "_____no_output_____" ] ], [ [ "# Quadrature rules for 2.5-D resistivity modelling\n\nWe consider the evaluation of the integral\n\n$$\n\\Phi(x, y, z) = \\frac{2}{\\pi} \\int_0^\\infty \\tilde\\Phi(k, y, z) \\cos(k x)\\, dk\n$$\nwhere \n$$\n\\tilde\\Phi(k, y, z) = K_0\\left({k}{\\sqrt{y^2 + z^2}}\\right).\n$$\n\nThe function $\\tilde\\Phi$ exhibits a different asymptotic behaviour depending on the magnitude of the argument, i.e., with $u := kr$\n\n$$\nu\\to 0: K_0(u) \\to -\\ln(u)\n$$\nand\n$$\nu \\to \\infty: K_0(u) \\to \\frac{e^{-u}}{\\sqrt{u}}.\n$$\n\nFor a fixed distance $r = \\sqrt{y^2 + z^2} = 1$ and $10^{-6} \\le k \\le 10^1$, we obtain the following figure:", "_____no_output_____" ] ], [ [ "k = logspace(-6, 4, 101);\nkk = 1e-3;\nu = besselk(0, k * kk);\npadln = 65;\npadexp = 15;\nloglog(k, u, 'k', k(1:padln), -log(kk * k(1:padln)), 'r.', ...\n k(end-padexp:end), exp(-kk * k(end-padexp:end))./sqrt(kk * k(end-padexp:end)), 'b.')\n\nlegend('K_0(u)', '-ln(u)', 'exp(-u)/sqrt(u)')\nylabel('\\Phi(u)')\nxlabel('u')", "_____no_output_____" ] ], [ [ "We split the integration at $k = k_0$, $0 < k_0 < \\infty$.\nWe obtain\n$$\n\\int_0^\\infty \\tilde\\Phi(k)\\,dk = \\int_0^{k_0}\\tilde\\Phi(k)\\,dk + \\int_{k_0}^\\infty\\tilde\\Phi(k)\\,dk.\n$$\n\n### Gauss-Legendre quadrature\nTo avoid the singularity at $k \\to 0$ for the first integral, we substitute $k'=\\sqrt{k / k_0}$ and obtain with $dk = 2 k_0 k' dk'$\n$$\n\\int_0^{k_0}\\tilde\\Phi(k)\\,dk = \\int_0^1 g(k')\\,dk' \\approx \\sum_{n=1}^N w_n' g(k_n') = \\sum_{n=1}^N w_n \\tilde\\Phi(k_n)\n$$\nwith $w_n = 2 k_0 k_n' w_n' $ and $k_n = k_0 k_n'^2$.\n\n### Gauss-Laguerre quadrature\nFor the second integral, we substitute $k' = k / k_0 - 1$, define $g(k') = k_0 \\tilde\\Phi(k)e^{k'}$, and obtain\n$$\n\\int_{k_0}^\\infty\\tilde\\Phi(k)\\,dk = \\int_0^\\infty e^{-k'} g(k')\\,dk' \\approx \\sum_{n=1}^N w_n' g(k_n') = \\sum_{n=1}^N w_n \\tilde\\Phi(k_n)\n$$\nwith $w_n = k_0 e^{k_n'}w_n'$ and $k_n = k_0 (k_n'+1)$.\n\n### Choice of $k_0$\n\nThe actual value of $k_0$ depends on the smallest electrode spacing $r_{min}$.\nMore precisely, $k_0 = (2 r_{min})^{-1}$.\n\n## Numerical test\n\nIn the case of a point electrode with current $I$ located at $\\mathbf r' = (x', y', 0)^\\top$ at the surface of a homogeneous halfspace with resistivity $\\rho$, we obtain for the electric potential at point $\\mathbf r = (x, y, z)^\\top$\n$$\n\\Phi(\\mathbf r) = \\dfrac{\\rho I}{2 \\pi |\\mathbf r - \\mathbf r'|}.\n$$\n\nWe try to approximate the inverse Cosine transform\n$$\n\\Phi(x, y, z) = \\frac{2}{\\pi} \\int_0^\\infty \\tilde\\Phi(k, y, z) \\cos(k x)\\, dk\n$$\nfor the special case of $x = 0$ ($\\cos(0) = 1$) by means of the Gauss quadrature rules introduced above.\n\nFor the smallest electrode spacing of, e.g., $|\\mathbf r - \\mathbf r'| = r_{min} = 1$ we would set $k_0 = 0.5$.", "_____no_output_____" ] ], [ [ "rmin = 1;\nrp = rmin:1:100;\nrp = rp(:);\nk0 = 1 / (2 * rmin);\n[x1, w1] = gauleg(0, 1, 17);\n[x2, w2] = gaulag(7);\nkn1 = k0 * x1 .* x1;\nwn1 = 2 * k0 * x1 .* w1;\nkn2 = k0 * (x2 + 1);\nwn2 = k0 * exp(x2) .* w2;\n\nk = [kn1(:); kn2(:)];\nw = [wn1(:); wn2(:)];", "_____no_output_____" ] ], [ [ "We check the validity of the approximation by checking against the analytical solution for the homogeneous halfspace, which, in the case of $\\rho = 2 \\pi$ and $I = 1$, is simply\n$$\n\\Phi_a(r) = \\dfrac{1}{r}.\n$$", "_____no_output_____" ] ], [ [ "k(1)", "\nans =\n\n 1.1103e-05\n\n" ], [ "v = zeros(length(rp), 1);\nfor i = 1:length(rp)\n v(i) = 2 / pi * sum(w .* besselk(0, k * rp(i)));\nend\n\nplot(rp, v, 'r.-', rp, 1 ./ rp, 'b')\nxlabel('r in m')\nylabel('potential in V')\nlegend('transformed', 'analytical')", "_____no_output_____" ] ], [ [ "In the following plot, we display the relative error of the approximation\n$$\ne(r) := \\left(1 - \\dfrac{\\Phi(r)}{\\Phi_a(r)}\\right) \\cdot 100 \\%\n$$\nwith respect to the (normalized) electrode distance.", "_____no_output_____" ] ], [ [ "plot(rp / rmin, 100 * (1 - v .* rp), '.-');\ngrid();\nxlabel('r / r_{min}');\nylabel('rel. error in %');\nylim([-0.05 0.05])", "_____no_output_____" ] ] ]

[ "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# Introduction to Gym toolkit", "_____no_output_____" ], [ "## Gym Environments\n\nThe centerpiece of Gym is the environment, which defines the \"game\" in which your reinforcement algorithm will compete. An environment does not need to be a game; however, it describes the following game-like features:\n* **action space**: What actions can we take on the environment, at each step/episode, to alter the environment.\n* **observation space**: What is the current state of the portion of the environment that we can observe. Usually, we can see the entire environment.\n\nBefore we begin to look at Gym, it is essential to understand some of the terminology used by this library.\n\n* **Agent** - The machine learning program or model that controls the actions.\nStep - One round of issuing actions that affect the observation space.\n* **Episode** - A collection of steps that terminates when the agent fails to meet the environment's objective, or the episode reaches the maximum number of allowed steps.\n* **Render** - Gym can render one frame for display after each episode.\n* **Reward** - A positive reinforcement that can occur at the end of each episode, after the agent acts.\n* **Nondeterministic** - For some environments, randomness is a factor in deciding what effects actions have on reward and changes to the observation space.", "_____no_output_____" ] ], [ [ "import gym\n\ndef query_environment(name):\n env = gym.make(name)\n spec = gym.spec(name)\n print(f\"Action Space: {env.action_space}\")\n print(f\"Observation Space: {env.observation_space}\")\n print(f\"Max Episode Steps: {spec.max_episode_steps}\")\n print(f\"Nondeterministic: {spec.nondeterministic}\")\n print(f\"Reward Range: {env.reward_range}\")\n print(f\"Reward Threshold: {spec.reward_threshold}\")", "_____no_output_____" ], [ "query_environment(\"CartPole-v1\")", "Action Space: Discrete(2)\nObservation Space: Box(-3.4028234663852886e+38, 3.4028234663852886e+38, (4,), float32)\nMax Episode Steps: 500\nNondeterministic: False\nReward Range: (-inf, inf)\nReward Threshold: 475.0\n" ] ], [ [ "The CartPole-v1 environment challenges the agent to move a cart while keeping a pole balanced. The environment has an observation space of 4 continuous numbers:\n\n* Cart Position\n* Cart Velocity\n* Pole Angle\n* Pole Velocity At Tip\n\nTo achieve this goal, the agent can take the following actions:\n\n* Push cart to the left\n* Push cart to the right\n\nThere is also a continuous variant of the mountain car. This version does not simply have the motor on or off. For the continuous car the action space is a single floating point number that specifies how much forward or backward force is being applied.", "_____no_output_____" ], [ "### Simple", "_____no_output_____" ] ], [ [ "import random\nfrom typing import List\n\n\nclass Environment:\n def __init__(self):\n self.steps_left = 10\n\n def get_observation(self) -> List[float]:\n return [0.0, 0.0, 0.0]\n\n def get_actions(self) -> List[int]:\n return [0, 1]\n\n def is_done(self) -> bool:\n return self.steps_left == 0\n\n def action(self, action: int) -> float:\n if self.is_done():\n raise Exception(\"Game is over\")\n self.steps_left -= 1\n return random.random()\n\n\nclass Agent:\n def __init__(self):\n self.total_reward = 0.0\n\n def step(self, env: Environment):\n current_obs = env.get_observation()\n actions = env.get_actions()\n reward = env.action(random.choice(actions))\n self.total_reward += reward\n\n\nif __name__ == \"__main__\":\n env = Environment()\n agent = Agent()\n\n while not env.is_done():\n agent.step(env)\n\n print(\"Total reward got: %.4f\" % agent.total_reward)", "Total reward got: 4.6979\n" ] ], [ [ "### Frozenlake", "_____no_output_____" ] ], [ [ "import gym", "_____no_output_____" ], [ "env = gym.make(\"FrozenLake-v0\")", "_____no_output_____" ], [ "env.render()", "\n\u001b[41mS\u001b[0mFFF\nFHFH\nFFFH\nHFFG\n" ], [ "print(env.observation_space)", "Discrete(16)\n" ], [ "print(env.action_space)", "Discrete(4)\n" ] ], [ [ "| Number | Action |\n| ------ | ------ |\n| 0 | Left |\n| 1 | Down |\n| 2 | Right |\n| 3 | Up |", "_____no_output_____" ], [ "We can obtain the transition probability and the reward function by just typing env.P[state][action]. So, to obtain the transition probability of moving from state S to the other states by performing the action right, we can type env.P[S][right]. But we cannot just type state S and action right directly since they are encoded as numbers. We learned that state S is encoded as 0 and the action right is encoded as 2, so, to obtain the transition probability of state S by performing the action right, we type env.P[0][2]", "_____no_output_____" ] ], [ [ "print(env.P[0][2])", "[(0.3333333333333333, 4, 0.0, False), (0.3333333333333333, 1, 0.0, False), (0.3333333333333333, 0, 0.0, False)]\n" ] ], [ [ "Our output is in the form of [(transition probability, next state, reward, Is terminal state?)]", "_____no_output_____" ] ], [ [ "state = env.reset()", "_____no_output_____" ], [ "env.step(1)", "_____no_output_____" ], [ "(next_state, reward, done, info) = env.step(1)", "_____no_output_____" ] ], [ [ "- **next_state** represents the next state.\n- **reward** represents the obtained reward.\n- **done** implies whether our episode has ended. That is, if the next state is a terminal state, then our episode will end, so done will be marked as True else it will be marked as False.\n- **info** — Apart from the transition probability, in some cases, we also obtain other information saved as info, which is used for debugging purposes.", "_____no_output_____" ] ], [ [ "random_action = env.action_space.sample()", "_____no_output_____" ], [ "next_state, reward, done, info = env.step(random_action)", "_____no_output_____" ] ], [ [ "**Generating an episode**\nThe episode is the agent environment interaction starting from the initial state to the terminal state. The agent interacts with the environment by performing some action in each state. An episode ends if the agent reaches the terminal state. So, in the Frozen Lake environment, the episode will end if the agent reaches the terminal state, which is either the hole state (H) or goal state (G).", "_____no_output_____" ] ], [ [ "import gym\n\nenv = gym.make(\"FrozenLake-v0\")\nstate = env.reset()\nprint('Time Step 0 :')\nenv.render()\nnum_timesteps = 20\n\nfor t in range(num_timesteps):\n random_action = env.action_space.sample()\n new_state, reward, done, info = env.step(random_action)\n print ('Time Step {} :'.format(t+1))\n env.render()\n if done:\n break", "Time Step 0 :\n\n\u001b[41mS\u001b[0mFFF\nFHFH\nFFFH\nHFFG\nTime Step 1 :\n (Right)\n\u001b[41mS\u001b[0mFFF\nFHFH\nFFFH\nHFFG\nTime Step 2 :\n (Right)\nSFFF\n\u001b[41mF\u001b[0mHFH\nFFFH\nHFFG\nTime Step 3 :\n (Left)\nSFFF\nFHFH\n\u001b[41mF\u001b[0mFFH\nHFFG\nTime Step 4 :\n (Right)\nSFFF\n\u001b[41mF\u001b[0mHFH\nFFFH\nHFFG\nTime Step 5 :\n (Up)\nSFFF\nF\u001b[41mH\u001b[0mFH\nFFFH\nHFFG\n" ] ], [ [ "Instead of generating one episode, we can also generate a series of episodes by taking some random action in each state", "_____no_output_____" ] ], [ [ "import gym\nenv = gym.make(\"FrozenLake-v0\")\nnum_episodes = 10\nnum_timesteps = 20 \nfor i in range(num_episodes):\n \n state = env.reset()\n print('Time Step 0 :')\n env.render()\n \n for t in range(num_timesteps):\n random_action = env.action_space.sample()\n new_state, reward, done, info = env.step(random_action)\n print ('Time Step {} :'.format(t+1))\n env.render()\n if done:\n break", "_____no_output_____" ] ], [ [ "### Cartpole", "_____no_output_____" ] ], [ [ "env = gym.make(\"CartPole-v0\")", "_____no_output_____" ], [ "print(env.observation_space)", "Box(-3.4028234663852886e+38, 3.4028234663852886e+38, (4,), float32)\n" ] ], [ [ "Note that all of these values are continuous, that is:\n\n- The value of the cart position ranges from -4.8 to 4.8.\n- The value of the cart velocity ranges from -Inf to Inf ( to ).\n- The value of the pole angle ranges from -0.418 radians to 0.418 radians.\n- The value of the pole velocity at the tip ranges from -Inf to Inf.", "_____no_output_____" ] ], [ [ "print(env.reset())", "[-0.03805974 -0.00851157 -0.00346854 -0.03263184]\n" ], [ "print(env.observation_space.high)", "[4.8000002e+00 3.4028235e+38 4.1887903e-01 3.4028235e+38]\n" ] ], [ [ "It implies that:\n\n1. The maximum value of the cart position is 4.8.\n2. We learned that the maximum value of the cart velocity is +Inf, and we know that infinity is not really a number, so it is represented using the largest positive real value 3.4028235e+38.\n3. The maximum value of the pole angle is 0.418 radians.\n4. The maximum value of the pole velocity at the tip is +Inf, so it is represented using the largest positive real value 3.4028235e+38.", "_____no_output_____" ] ], [ [ "print(env.observation_space.low)", "[-4.8000002e+00 -3.4028235e+38 -4.1887903e-01 -3.4028235e+38]\n" ] ], [ [ "It states that:\n\n1. The minimum value of the cart position is -4.8.\n2. We learned that the minimum value of the cart velocity is -Inf, and we know that infinity is not really a number, so it is represented using the largest negative real value -3.4028235e+38.\n3. The minimum value of the pole angle is -0.418 radians.\n4. The minimum value of the pole velocity at the tip is -Inf, so it is represented using the largest negative real value -3.4028235e+38.", "_____no_output_____" ] ], [ [ "print(env.action_space)", "Discrete(2)\n" ] ], [ [ "| Number | Action |\n| ------ | ------ |\n| 0 | Push cart to the left |\n| 1 | Push cart to the right |", "_____no_output_____" ] ], [ [ "import gym\n\n\nif __name__ == \"__main__\":\n env = gym.make(\"CartPole-v0\")\n\n total_reward = 0.0\n total_steps = 0\n obs = env.reset()\n\n while True:\n action = env.action_space.sample()\n obs, reward, done, _ = env.step(action)\n total_reward += reward\n total_steps += 1\n if done:\n break\n\n print(\"Episode done in %d steps, total reward %.2f\" % (\n total_steps, total_reward))", "Episode done in 20 steps, total reward 20.00\n" ] ], [ [ "## Wrappers\n\nVery frequently, you will want to extend the environment's functionality in some generic way. For example, imagine an environment gives you some observations, but you want to accumulate them in some buffer and provide to the agent the N last observations. This is a common scenario for dynamic computer games, when one single frame is just not enough to get the full information about the game state. Another example is when you want to be able to crop or preprocess an image's pixels to make it more convenient for the agent to digest, or if you want to normalize reward scores somehow. There are many such situations that have the same structure – you want to \"wrap\" the existing environment and add some extra logic for doing something. Gym provides a convenient framework for these situations – the Wrapper class.", "_____no_output_____" ], [ "**Random action wrapper**", "_____no_output_____" ] ], [ [ "import gym\nfrom typing import TypeVar\nimport random\n\nAction = TypeVar('Action')\n\n\nclass RandomActionWrapper(gym.ActionWrapper):\n def __init__(self, env, epsilon=0.1):\n super(RandomActionWrapper, self).__init__(env)\n self.epsilon = epsilon\n\n def action(self, action: Action) -> Action:\n if random.random() < self.epsilon:\n print(\"Random!\")\n return self.env.action_space.sample()\n return action\n\n\nif __name__ == \"__main__\":\n env = RandomActionWrapper(gym.make(\"CartPole-v0\"))\n\n obs = env.reset()\n total_reward = 0.0\n\n while True:\n obs, reward, done, _ = env.step(0)\n total_reward += reward\n if done:\n break\n\n print(\"Reward got: %.2f\" % total_reward)", "Reward got: 9.00\n" ] ], [ [ "## Atari GAN", "_____no_output_____" ] ], [ [ "! wget http://www.atarimania.com/roms/Roms.rar\n! mkdir /content/ROM/\n! unrar e /content/Roms.rar /content/ROM/\n! python -m atari_py.import_roms /content/ROM/", "_____no_output_____" ] ], [ [ "### Normal", "_____no_output_____" ] ], [ [ "import random\nimport argparse\nimport cv2\n\nimport torch\nimport torch.nn as nn\nimport torch.optim as optim\nfrom torch.utils.tensorboard import SummaryWriter\n\nimport torchvision.utils as vutils\n\nimport gym\nimport gym.spaces\n\nimport numpy as np\n\nlog = gym.logger\nlog.set_level(gym.logger.INFO)\n\nLATENT_VECTOR_SIZE = 100\nDISCR_FILTERS = 64\nGENER_FILTERS = 64\nBATCH_SIZE = 16\n\n# dimension input image will be rescaled\nIMAGE_SIZE = 64\n\nLEARNING_RATE = 0.0001\nREPORT_EVERY_ITER = 100\nSAVE_IMAGE_EVERY_ITER = 1000\n\n\nclass InputWrapper(gym.ObservationWrapper):\n \"\"\"\n Preprocessing of input numpy array:\n 1. resize image into predefined size\n 2. move color channel axis to a first place\n \"\"\"\n def __init__(self, *args):\n super(InputWrapper, self).__init__(*args)\n assert isinstance(self.observation_space, gym.spaces.Box)\n old_space = self.observation_space\n self.observation_space = gym.spaces.Box(\n self.observation(old_space.low),\n self.observation(old_space.high),\n dtype=np.float32)\n\n def observation(self, observation):\n # resize image\n new_obs = cv2.resize(\n observation, (IMAGE_SIZE, IMAGE_SIZE))\n # transform (210, 160, 3) -> (3, 210, 160)\n new_obs = np.moveaxis(new_obs, 2, 0)\n return new_obs.astype(np.float32)\n\n\nclass Discriminator(nn.Module):\n def __init__(self, input_shape):\n super(Discriminator, self).__init__()\n # this pipe converges image into the single number\n self.conv_pipe = nn.Sequential(\n nn.Conv2d(in_channels=input_shape[0], out_channels=DISCR_FILTERS,\n kernel_size=4, stride=2, padding=1),\n nn.ReLU(),\n nn.Conv2d(in_channels=DISCR_FILTERS, out_channels=DISCR_FILTERS*2,\n kernel_size=4, stride=2, padding=1),\n nn.BatchNorm2d(DISCR_FILTERS*2),\n nn.ReLU(),\n nn.Conv2d(in_channels=DISCR_FILTERS * 2, out_channels=DISCR_FILTERS * 4,\n kernel_size=4, stride=2, padding=1),\n nn.BatchNorm2d(DISCR_FILTERS * 4),\n nn.ReLU(),\n nn.Conv2d(in_channels=DISCR_FILTERS * 4, out_channels=DISCR_FILTERS * 8,\n kernel_size=4, stride=2, padding=1),\n nn.BatchNorm2d(DISCR_FILTERS * 8),\n nn.ReLU(),\n nn.Conv2d(in_channels=DISCR_FILTERS * 8, out_channels=1,\n kernel_size=4, stride=1, padding=0),\n nn.Sigmoid()\n )\n\n def forward(self, x):\n conv_out = self.conv_pipe(x)\n return conv_out.view(-1, 1).squeeze(dim=1)\n\n\nclass Generator(nn.Module):\n def __init__(self, output_shape):\n super(Generator, self).__init__()\n # pipe deconvolves input vector into (3, 64, 64) image\n self.pipe = nn.Sequential(\n nn.ConvTranspose2d(in_channels=LATENT_VECTOR_SIZE, out_channels=GENER_FILTERS * 8,\n kernel_size=4, stride=1, padding=0),\n nn.BatchNorm2d(GENER_FILTERS * 8),\n nn.ReLU(),\n nn.ConvTranspose2d(in_channels=GENER_FILTERS * 8, out_channels=GENER_FILTERS * 4,\n kernel_size=4, stride=2, padding=1),\n nn.BatchNorm2d(GENER_FILTERS * 4),\n nn.ReLU(),\n nn.ConvTranspose2d(in_channels=GENER_FILTERS * 4, out_channels=GENER_FILTERS * 2,\n kernel_size=4, stride=2, padding=1),\n nn.BatchNorm2d(GENER_FILTERS * 2),\n nn.ReLU(),\n nn.ConvTranspose2d(in_channels=GENER_FILTERS * 2, out_channels=GENER_FILTERS,\n kernel_size=4, stride=2, padding=1),\n nn.BatchNorm2d(GENER_FILTERS),\n nn.ReLU(),\n nn.ConvTranspose2d(in_channels=GENER_FILTERS, out_channels=output_shape[0],\n kernel_size=4, stride=2, padding=1),\n nn.Tanh()\n )\n\n def forward(self, x):\n return self.pipe(x)\n\n\ndef iterate_batches(envs, batch_size=BATCH_SIZE):\n batch = [e.reset() for e in envs]\n env_gen = iter(lambda: random.choice(envs), None)\n\n while True:\n e = next(env_gen)\n obs, reward, is_done, _ = e.step(e.action_space.sample())\n if np.mean(obs) > 0.01:\n batch.append(obs)\n if len(batch) == batch_size:\n # Normalising input between -1 to 1\n batch_np = np.array(batch, dtype=np.float32) * 2.0 / 255.0 - 1.0\n yield torch.tensor(batch_np)\n batch.clear()\n if is_done:\n e.reset()\n\n\nif __name__ == \"__main__\":\n parser = argparse.ArgumentParser()\n parser.add_argument(\n \"--cuda\", default=False, action='store_true',\n help=\"Enable cuda computation\")\n args = parser.parse_args(args={})\n\n device = torch.device(\"cuda\" if args.cuda else \"cpu\")\n envs = [\n InputWrapper(gym.make(name))\n for name in ('Breakout-v0', 'AirRaid-v0', 'Pong-v0')\n ]\n input_shape = envs[0].observation_space.shape\n\n net_discr = Discriminator(input_shape=input_shape).to(device)\n net_gener = Generator(output_shape=input_shape).to(device)\n\n objective = nn.BCELoss()\n gen_optimizer = optim.Adam(\n params=net_gener.parameters(), lr=LEARNING_RATE,\n betas=(0.5, 0.999))\n dis_optimizer = optim.Adam(\n params=net_discr.parameters(), lr=LEARNING_RATE,\n betas=(0.5, 0.999))\n writer = SummaryWriter()\n\n gen_losses = []\n dis_losses = []\n iter_no = 0\n\n true_labels_v = torch.ones(BATCH_SIZE, device=device)\n fake_labels_v = torch.zeros(BATCH_SIZE, device=device)\n\n for batch_v in iterate_batches(envs):\n # fake samples, input is 4D: batch, filters, x, y\n gen_input_v = torch.FloatTensor(\n BATCH_SIZE, LATENT_VECTOR_SIZE, 1, 1)\n gen_input_v.normal_(0, 1)\n gen_input_v = gen_input_v.to(device)\n batch_v = batch_v.to(device)\n gen_output_v = net_gener(gen_input_v)\n\n # train discriminator\n dis_optimizer.zero_grad()\n dis_output_true_v = net_discr(batch_v)\n dis_output_fake_v = net_discr(gen_output_v.detach())\n dis_loss = objective(dis_output_true_v, true_labels_v) + \\\n objective(dis_output_fake_v, fake_labels_v)\n dis_loss.backward()\n dis_optimizer.step()\n dis_losses.append(dis_loss.item())\n\n # train generator\n gen_optimizer.zero_grad()\n dis_output_v = net_discr(gen_output_v)\n gen_loss_v = objective(dis_output_v, true_labels_v)\n gen_loss_v.backward()\n gen_optimizer.step()\n gen_losses.append(gen_loss_v.item())\n\n iter_no += 1\n if iter_no % REPORT_EVERY_ITER == 0:\n log.info(\"Iter %d: gen_loss=%.3e, dis_loss=%.3e\",\n iter_no, np.mean(gen_losses),\n np.mean(dis_losses))\n writer.add_scalar(\n \"gen_loss\", np.mean(gen_losses), iter_no)\n writer.add_scalar(\n \"dis_loss\", np.mean(dis_losses), iter_no)\n gen_losses = []\n dis_losses = []\n if iter_no % SAVE_IMAGE_EVERY_ITER == 0:\n writer.add_image(\"fake\", vutils.make_grid(\n gen_output_v.data[:64], normalize=True), iter_no)\n writer.add_image(\"real\", vutils.make_grid(\n batch_v.data[:64], normalize=True), iter_no)", "INFO: Making new env: Breakout-v0\nINFO: Making new env: AirRaid-v0\nINFO: Making new env: Pong-v0\nINFO: Iter 100: gen_loss=5.454e+00, dis_loss=5.009e-02\nINFO: Iter 200: gen_loss=7.054e+00, dis_loss=4.306e-03\nINFO: Iter 300: gen_loss=7.568e+00, dis_loss=2.140e-03\nINFO: Iter 400: gen_loss=7.842e+00, dis_loss=1.272e-03\nINFO: Iter 500: gen_loss=8.155e+00, dis_loss=1.019e-03\nINFO: Iter 600: gen_loss=8.442e+00, dis_loss=6.918e-04\nINFO: Iter 700: gen_loss=8.560e+00, dis_loss=5.483e-04\nINFO: Iter 800: gen_loss=9.014e+00, dis_loss=4.792e-04\nINFO: Iter 900: gen_loss=7.517e+00, dis_loss=2.132e-01\nINFO: Iter 1000: gen_loss=7.375e+00, dis_loss=1.050e-01\nINFO: Iter 1100: gen_loss=6.722e+00, dis_loss=1.718e-02\nINFO: Iter 1200: gen_loss=6.346e+00, dis_loss=6.303e-03\nINFO: Iter 1300: gen_loss=6.636e+00, dis_loss=6.348e-03\nINFO: Iter 1400: gen_loss=6.612e+00, dis_loss=7.664e-02\nINFO: Iter 1500: gen_loss=6.028e+00, dis_loss=7.801e-03\nINFO: Iter 1600: gen_loss=6.665e+00, dis_loss=3.651e-03\nINFO: Iter 1700: gen_loss=7.290e+00, dis_loss=5.616e-02\nINFO: Iter 1800: gen_loss=6.314e+00, dis_loss=7.723e-02\nINFO: Iter 1900: gen_loss=5.940e+00, dis_loss=3.784e-01\nINFO: Iter 2000: gen_loss=5.053e+00, dis_loss=2.623e-01\nINFO: Iter 2100: gen_loss=5.465e+00, dis_loss=9.114e-02\nINFO: Iter 2200: gen_loss=5.480e+00, dis_loss=3.963e-01\nINFO: Iter 2300: gen_loss=4.549e+00, dis_loss=2.361e-01\nINFO: Iter 2400: gen_loss=5.407e+00, dis_loss=1.310e-01\nINFO: Iter 2500: gen_loss=5.766e+00, dis_loss=5.550e-02\nINFO: Iter 2600: gen_loss=5.816e+00, dis_loss=1.418e-01\nINFO: Iter 2700: gen_loss=6.737e+00, dis_loss=5.231e-02\nINFO: Iter 2800: gen_loss=7.147e+00, dis_loss=1.491e-01\nINFO: Iter 2900: gen_loss=6.541e+00, dis_loss=2.155e-02\nINFO: Iter 3000: gen_loss=7.072e+00, dis_loss=1.127e-01\nINFO: Iter 3100: gen_loss=6.137e+00, dis_loss=6.138e-02\nINFO: Iter 3200: gen_loss=7.406e+00, dis_loss=3.540e-02\nINFO: Iter 3300: gen_loss=7.850e+00, dis_loss=5.691e-03\nINFO: Iter 3400: gen_loss=8.614e+00, dis_loss=7.228e-03\nINFO: Iter 3500: gen_loss=8.885e+00, dis_loss=3.191e-03\nINFO: Iter 3600: gen_loss=5.367e+00, dis_loss=5.296e-01\nINFO: Iter 3700: gen_loss=4.176e+00, dis_loss=3.335e-01\nINFO: Iter 3800: gen_loss=5.174e+00, dis_loss=2.732e-01\nINFO: Iter 3900: gen_loss=5.492e+00, dis_loss=1.298e-01\nINFO: Iter 4000: gen_loss=6.570e+00, dis_loss=1.961e-02\nINFO: Iter 4100: gen_loss=7.011e+00, dis_loss=2.517e-02\nINFO: Iter 4200: gen_loss=8.362e+00, dis_loss=4.330e-03\nINFO: Iter 4300: gen_loss=6.908e+00, dis_loss=2.161e-01\nINFO: Iter 4400: gen_loss=5.226e+00, dis_loss=2.762e-01\nINFO: Iter 4500: gen_loss=4.998e+00, dis_loss=2.893e-01\nINFO: Iter 4600: gen_loss=5.078e+00, dis_loss=3.962e-01\nINFO: Iter 4700: gen_loss=4.886e+00, dis_loss=1.932e-01\nINFO: Iter 4800: gen_loss=6.110e+00, dis_loss=7.615e-02\nINFO: Iter 4900: gen_loss=5.402e+00, dis_loss=1.634e-01\nINFO: Iter 5000: gen_loss=5.336e+00, dis_loss=1.919e-01\nINFO: Iter 5100: gen_loss=5.749e+00, dis_loss=8.817e-02\nINFO: Iter 5200: gen_loss=5.879e+00, dis_loss=1.182e-01\nINFO: Iter 5300: gen_loss=5.417e+00, dis_loss=1.651e-01\nINFO: Iter 5400: gen_loss=6.747e+00, dis_loss=3.846e-02\nINFO: Iter 5500: gen_loss=5.133e+00, dis_loss=1.996e-01\nINFO: Iter 5600: gen_loss=6.116e+00, dis_loss=2.946e-01\nINFO: Iter 5700: gen_loss=5.858e+00, dis_loss=2.152e-02\n" ] ], [ [ "### Ignite", "_____no_output_____" ] ], [ [ "import random\nimport argparse\nimport cv2\n\nimport torch\nimport torch.nn as nn\nimport torch.optim as optim\nfrom ignite.engine import Engine, Events\nfrom ignite.metrics import RunningAverage\nfrom ignite.contrib.handlers import tensorboard_logger as tb_logger\n\nimport torchvision.utils as vutils\n\nimport gym\nimport gym.spaces\n\nimport numpy as np\n\nlog = gym.logger\nlog.set_level(gym.logger.INFO)\n\nLATENT_VECTOR_SIZE = 100\nDISCR_FILTERS = 64\nGENER_FILTERS = 64\nBATCH_SIZE = 16\n\n# dimension input image will be rescaled\nIMAGE_SIZE = 64\n\nLEARNING_RATE = 0.0001\nREPORT_EVERY_ITER = 100\nSAVE_IMAGE_EVERY_ITER = 1000\n\n\nclass InputWrapper(gym.ObservationWrapper):\n \"\"\"\n Preprocessing of input numpy array:\n 1. resize image into predefined size\n 2. move color channel axis to a first place\n \"\"\"\n def __init__(self, *args):\n super(InputWrapper, self).__init__(*args)\n assert isinstance(self.observation_space, gym.spaces.Box)\n old_space = self.observation_space\n self.observation_space = gym.spaces.Box(self.observation(old_space.low), self.observation(old_space.high),\n dtype=np.float32)\n\n def observation(self, observation):\n # resize image\n new_obs = cv2.resize(observation, (IMAGE_SIZE, IMAGE_SIZE))\n # transform (210, 160, 3) -> (3, 210, 160)\n new_obs = np.moveaxis(new_obs, 2, 0)\n return new_obs.astype(np.float32)\n\n\nclass Discriminator(nn.Module):\n def __init__(self, input_shape):\n super(Discriminator, self).__init__()\n # this pipe converges image into the single number\n self.conv_pipe = nn.Sequential(\n nn.Conv2d(in_channels=input_shape[0], out_channels=DISCR_FILTERS,\n kernel_size=4, stride=2, padding=1),\n nn.ReLU(),\n nn.Conv2d(in_channels=DISCR_FILTERS, out_channels=DISCR_FILTERS*2,\n kernel_size=4, stride=2, padding=1),\n nn.BatchNorm2d(DISCR_FILTERS*2),\n nn.ReLU(),\n nn.Conv2d(in_channels=DISCR_FILTERS * 2, out_channels=DISCR_FILTERS * 4,\n kernel_size=4, stride=2, padding=1),\n nn.BatchNorm2d(DISCR_FILTERS * 4),\n nn.ReLU(),\n nn.Conv2d(in_channels=DISCR_FILTERS * 4, out_channels=DISCR_FILTERS * 8,\n kernel_size=4, stride=2, padding=1),\n nn.BatchNorm2d(DISCR_FILTERS * 8),\n nn.ReLU(),\n nn.Conv2d(in_channels=DISCR_FILTERS * 8, out_channels=1,\n kernel_size=4, stride=1, padding=0),\n nn.Sigmoid()\n )\n\n def forward(self, x):\n conv_out = self.conv_pipe(x)\n return conv_out.view(-1, 1).squeeze(dim=1)\n\n\nclass Generator(nn.Module):\n def __init__(self, output_shape):\n super(Generator, self).__init__()\n # pipe deconvolves input vector into (3, 64, 64) image\n self.pipe = nn.Sequential(\n nn.ConvTranspose2d(in_channels=LATENT_VECTOR_SIZE, out_channels=GENER_FILTERS * 8,\n kernel_size=4, stride=1, padding=0),\n nn.BatchNorm2d(GENER_FILTERS * 8),\n nn.ReLU(),\n nn.ConvTranspose2d(in_channels=GENER_FILTERS * 8, out_channels=GENER_FILTERS * 4,\n kernel_size=4, stride=2, padding=1),\n nn.BatchNorm2d(GENER_FILTERS * 4),\n nn.ReLU(),\n nn.ConvTranspose2d(in_channels=GENER_FILTERS * 4, out_channels=GENER_FILTERS * 2,\n kernel_size=4, stride=2, padding=1),\n nn.BatchNorm2d(GENER_FILTERS * 2),\n nn.ReLU(),\n nn.ConvTranspose2d(in_channels=GENER_FILTERS * 2, out_channels=GENER_FILTERS,\n kernel_size=4, stride=2, padding=1),\n nn.BatchNorm2d(GENER_FILTERS),\n nn.ReLU(),\n nn.ConvTranspose2d(in_channels=GENER_FILTERS, out_channels=output_shape[0],\n kernel_size=4, stride=2, padding=1),\n nn.Tanh()\n )\n\n def forward(self, x):\n return self.pipe(x)\n\n\ndef iterate_batches(envs, batch_size=BATCH_SIZE):\n batch = [e.reset() for e in envs]\n env_gen = iter(lambda: random.choice(envs), None)\n\n while True:\n e = next(env_gen)\n obs, reward, is_done, _ = e.step(e.action_space.sample())\n if np.mean(obs) > 0.01:\n batch.append(obs)\n if len(batch) == batch_size:\n # Normalising input between -1 to 1\n batch_np = np.array(batch, dtype=np.float32) * 2.0 / 255.0 - 1.0\n yield torch.tensor(batch_np)\n batch.clear()\n if is_done:\n e.reset()\n\n\nif __name__ == \"__main__\":\n parser = argparse.ArgumentParser()\n parser.add_argument(\"--cuda\", default=False, action='store_true', help=\"Enable cuda computation\")\n args = parser.parse_args(args={})\n\n device = torch.device(\"cuda\" if args.cuda else \"cpu\")\n # envs = [InputWrapper(gym.make(name)) for name in ('Breakout-v0', 'AirRaid-v0', 'Pong-v0')]\n envs = [InputWrapper(gym.make(name)) for name in ['Breakout-v0']]\n input_shape = envs[0].observation_space.shape\n\n net_discr = Discriminator(input_shape=input_shape).to(device)\n net_gener = Generator(output_shape=input_shape).to(device)\n\n objective = nn.BCELoss()\n gen_optimizer = optim.Adam(params=net_gener.parameters(), lr=LEARNING_RATE, betas=(0.5, 0.999))\n dis_optimizer = optim.Adam(params=net_discr.parameters(), lr=LEARNING_RATE, betas=(0.5, 0.999))\n\n true_labels_v = torch.ones(BATCH_SIZE, device=device)\n fake_labels_v = torch.zeros(BATCH_SIZE, device=device)\n\n def process_batch(trainer, batch):\n gen_input_v = torch.FloatTensor(\n BATCH_SIZE, LATENT_VECTOR_SIZE, 1, 1)\n gen_input_v.normal_(0, 1)\n gen_input_v = gen_input_v.to(device)\n batch_v = batch.to(device)\n gen_output_v = net_gener(gen_input_v)\n\n # train discriminator\n dis_optimizer.zero_grad()\n dis_output_true_v = net_discr(batch_v)\n dis_output_fake_v = net_discr(gen_output_v.detach())\n dis_loss = objective(dis_output_true_v, true_labels_v) + \\\n objective(dis_output_fake_v, fake_labels_v)\n dis_loss.backward()\n dis_optimizer.step()\n\n # train generator\n gen_optimizer.zero_grad()\n dis_output_v = net_discr(gen_output_v)\n gen_loss = objective(dis_output_v, true_labels_v)\n gen_loss.backward()\n gen_optimizer.step()\n\n if trainer.state.iteration % SAVE_IMAGE_EVERY_ITER == 0:\n fake_img = vutils.make_grid(\n gen_output_v.data[:64], normalize=True)\n trainer.tb.writer.add_image(\n \"fake\", fake_img, trainer.state.iteration)\n real_img = vutils.make_grid(\n batch_v.data[:64], normalize=True)\n trainer.tb.writer.add_image(\n \"real\", real_img, trainer.state.iteration)\n trainer.tb.writer.flush()\n return dis_loss.item(), gen_loss.item()\n\n engine = Engine(process_batch)\n tb = tb_logger.TensorboardLogger(log_dir=None)\n engine.tb = tb\n RunningAverage(output_transform=lambda out: out[1]).\\\n attach(engine, \"avg_loss_gen\")\n RunningAverage(output_transform=lambda out: out[0]).\\\n attach(engine, \"avg_loss_dis\")\n\n handler = tb_logger.OutputHandler(tag=\"train\",\n metric_names=['avg_loss_gen', 'avg_loss_dis'])\n tb.attach(engine, log_handler=handler,\n event_name=Events.ITERATION_COMPLETED)\n\n @engine.on(Events.ITERATION_COMPLETED)\n def log_losses(trainer):\n if trainer.state.iteration % REPORT_EVERY_ITER == 0:\n log.info(\"%d: gen_loss=%f, dis_loss=%f\",\n trainer.state.iteration,\n trainer.state.metrics['avg_loss_gen'],\n trainer.state.metrics['avg_loss_dis'])\n\n engine.run(data=iterate_batches(envs))", "INFO: Making new env: Breakout-v0\nINFO: 100: gen_loss=5.327549, dis_loss=0.200626\nINFO: 200: gen_loss=6.850880, dis_loss=0.028281\nINFO: 300: gen_loss=7.435633, dis_loss=0.004672\nINFO: 400: gen_loss=7.708136, dis_loss=0.001331\nINFO: 500: gen_loss=8.000729, dis_loss=0.000699\nINFO: 600: gen_loss=8.314868, dis_loss=0.000474\nINFO: 700: gen_loss=8.620416, dis_loss=0.000328\nINFO: 800: gen_loss=8.779677, dis_loss=0.000286\nINFO: 900: gen_loss=8.907359, dis_loss=0.000267\nINFO: 1000: gen_loss=6.822098, dis_loss=0.558477\nINFO: 1100: gen_loss=6.491067, dis_loss=0.079029\nINFO: 1200: gen_loss=6.794054, dis_loss=0.012632\nINFO: 1300: gen_loss=7.230944, dis_loss=0.002747\nINFO: 1400: gen_loss=7.698738, dis_loss=0.000962\nINFO: 1500: gen_loss=8.162801, dis_loss=0.000497\nINFO: 1600: gen_loss=8.546710, dis_loss=0.000326\nINFO: 1700: gen_loss=8.939020, dis_loss=0.000224\nINFO: 1800: gen_loss=9.027502, dis_loss=0.000216\nINFO: 1900: gen_loss=9.230650, dis_loss=0.000172\nINFO: 2000: gen_loss=9.495270, dis_loss=0.000129\nINFO: 2100: gen_loss=9.700210, dis_loss=0.000104\nINFO: 2200: gen_loss=9.862649, dis_loss=0.000086\nINFO: 2300: gen_loss=10.042667, dis_loss=0.000075\nINFO: 2400: gen_loss=10.333560, dis_loss=0.000052\nINFO: 2500: gen_loss=10.437976, dis_loss=0.000045\nINFO: 2600: gen_loss=10.592011, dis_loss=0.000040\nINFO: 2700: gen_loss=10.633485, dis_loss=0.000039\nINFO: 2800: gen_loss=10.627324, dis_loss=0.000036\nINFO: 2900: gen_loss=10.665850, dis_loss=0.000036\nINFO: 3000: gen_loss=10.712931, dis_loss=0.000036\nINFO: 3100: gen_loss=10.853663, dis_loss=0.000030\nINFO: 3200: gen_loss=10.868406, dis_loss=0.000030\nINFO: 3300: gen_loss=10.904878, dis_loss=0.000027\nINFO: 3400: gen_loss=11.031057, dis_loss=0.000022\nINFO: 3500: gen_loss=11.114413, dis_loss=0.000022\nINFO: 3600: gen_loss=11.334650, dis_loss=0.000018\nINFO: 3700: gen_loss=11.537755, dis_loss=0.000013\nINFO: 3800: gen_loss=11.573673, dis_loss=0.000015\nINFO: 3900: gen_loss=11.594438, dis_loss=0.000013\nINFO: 4000: gen_loss=11.650991, dis_loss=0.000012\nINFO: 4100: gen_loss=11.350557, dis_loss=0.000023\nINFO: 4200: gen_loss=11.715774, dis_loss=0.000012\nINFO: 4300: gen_loss=11.970108, dis_loss=0.000008\nINFO: 4400: gen_loss=12.142686, dis_loss=0.000007\nINFO: 4500: gen_loss=12.200508, dis_loss=0.000007\nINFO: 4600: gen_loss=12.209455, dis_loss=0.000006\nINFO: 4700: gen_loss=12.215595, dis_loss=0.000007\nINFO: 4800: gen_loss=12.352226, dis_loss=0.000006\nINFO: 4900: gen_loss=12.434466, dis_loss=0.000006\nINFO: 5000: gen_loss=12.517082, dis_loss=0.000005\nINFO: 5100: gen_loss=12.604175, dis_loss=0.000005\nINFO: 5200: gen_loss=12.744095, dis_loss=0.000004\nINFO: 5300: gen_loss=12.880165, dis_loss=0.000004\nINFO: 5400: gen_loss=12.999031, dis_loss=0.000003\n" ] ], [ [ "## Render environments in Colab", "_____no_output_____" ], [ "### Alternative 1\n\nIt is possible to visualize the game your agent is playing, even on CoLab. This section provides information on how to generate a video in CoLab that shows you an episode of the game your agent is playing. This video process is based on suggestions found [here](https://colab.research.google.com/drive/1flu31ulJlgiRL1dnN2ir8wGh9p7Zij2t).\n\nBegin by installing **pyvirtualdisplay** and **python-opengl**.", "_____no_output_____" ] ], [ [ "!pip install gym pyvirtualdisplay > /dev/null 2>&1\n!apt-get install -y xvfb python-opengl ffmpeg > /dev/null 2>&1\n\n!apt-get update > /dev/null 2>&1\n!apt-get install cmake > /dev/null 2>&1\n!pip install --upgrade setuptools 2>&1\n!pip install ez_setup > /dev/null 2>&1\n!pip install gym[atari] > /dev/null 2>&1\n\n!wget http://www.atarimania.com/roms/Roms.rar\n!mkdir /content/ROM/\n!unrar e /content/Roms.rar /content/ROM/\n!python -m atari_py.import_roms /content/ROM/", "_____no_output_____" ], [ "import gym\nfrom gym.wrappers import Monitor\nimport glob\nimport io\nimport base64\nfrom IPython.display import HTML\nfrom pyvirtualdisplay import Display\nfrom IPython import display as ipythondisplay\n\ndisplay = Display(visible=0, size=(1400, 900))\ndisplay.start()\n\n\"\"\"\nUtility functions to enable video recording of gym environment \nand displaying it.\nTo enable video, just do \"env = wrap_env(env)\"\"\n\"\"\"\n\ndef show_video():\n mp4list = glob.glob('video/*.mp4')\n if len(mp4list) > 0:\n mp4 = mp4list[0]\n video = io.open(mp4, 'r+b').read()\n encoded = base64.b64encode(video)\n ipythondisplay.display(HTML(data='''<video alt=\"test\" autoplay \n loop controls style=\"height: 400px;\">\n <source src=\"data:video/mp4;base64,{0}\" type=\"video/mp4\" />\n </video>'''.format(encoded.decode('ascii'))))\n else: \n print(\"Could not find video\")\n \n\ndef wrap_env(env):\n env = Monitor(env, './video', force=True)\n return env", "_____no_output_____" ], [ "#env = wrap_env(gym.make(\"MountainCar-v0\"))\nenv = wrap_env(gym.make(\"Atlantis-v0\"))\n\nobservation = env.reset()\n\nwhile True:\n env.render()\n #your agent goes here\n action = env.action_space.sample() \n observation, reward, done, info = env.step(action)\n if done: \n break;\n \nenv.close()\nshow_video()", "_____no_output_____" ] ], [ [ "### Alternative 2", "_____no_output_____" ] ], [ [ "!apt-get install -y xvfb python-opengl ffmpeg > /dev/null 2>&1\n!pip install colabgymrender", "_____no_output_____" ], [ "import gym\nfrom colabgymrender.recorder import Recorder\n\nenv = gym.make(\"Breakout-v0\")\ndirectory = './video'\nenv = Recorder(env, directory)\n\nobservation = env.reset()\nterminal = False\nwhile not terminal:\n action = env.action_space.sample()\n observation, reward, terminal, info = env.step(action)\n\nenv.play()", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Implementation of Stack\n\n## Stack Attributes and Methods\n\nBefore we implement our own Stack class, let's review the properties and methods of a Stack.\n\nThe stack abstract data type is defined by the following structure and operations. A stack is structured, as described above, as an ordered collection of items where items are added to and removed from the end called the “top.” Stacks are ordered LIFO. The stack operations are given below.\n\n* Stack() creates a new stack that is empty. It needs no parameters and returns an empty stack.\n* push(item) adds a new item to the top of the stack. It needs the item and returns nothing.\n* pop() removes the top item from the stack. It needs no parameters and returns the item. The stack is modified.\n* peek() returns the top item from the stack but does not remove it. It needs no parameters. The stack is not modified.\n* isEmpty() tests to see whether the stack is empty. It needs no parameters and returns a boolean value.\n* size() returns the number of items on the stack. It needs no parameters and returns an integer.", "_____no_output_____" ], [ "____\n\n## Stack Implementation", "_____no_output_____" ] ], [ [ "class Stack:\n \n \n def __init__(self):\n self.items = []\n\n def isEmpty(self):\n return self.items == []\n\n def push(self, item):\n self.items.append(item)\n\n def pop(self):\n return self.items.pop()\n\n def peek(self):\n return self.items[len(self.items)-1]\n\n def size(self):\n return len(self.items)", "_____no_output_____" ] ], [ [ "Let's try it out!", "_____no_output_____" ] ], [ [ "s = Stack()", "_____no_output_____" ], [ "print s.isEmpty()", "True\n" ], [ "s.push(1)", "_____no_output_____" ], [ "s.push('two')", "_____no_output_____" ], [ "s.peek()", "_____no_output_____" ], [ "s.push(True)", "_____no_output_____" ], [ "s.size()", "_____no_output_____" ], [ "s.isEmpty()", "_____no_output_____" ], [ "s.pop()", "_____no_output_____" ], [ "s.pop()", "two\n" ], [ "s.size()", "_____no_output_____" ], [ "s.pop()", "_____no_output_____" ], [ "s.isEmpty()", "_____no_output_____" ] ], [ [ "## Good Job!", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ] ]

[ "BSD-2-Clause" ]

[ "BSD-2-Clause" ]

[ "BSD-2-Clause" ]

[ [ [ "| Name | Description | Date\n| :- |-------------: | :-:\n|<font color=red>__Reza Hashemi__</font>| __Function approximation by linear model and deep network LOOP test__. | __On 10th of August 2019__", "_____no_output_____" ], [ "# Function approximation with linear models and neural network\n* Are Linear models sufficient for approximating transcedental functions? What about polynomial functions?\n* Do neural networks perform better in those cases?\n* Does the depth of the neural network matter?", "_____no_output_____" ], [ "### Import basic libraries", "_____no_output_____" ] ], [ [ "import numpy as np\nimport pandas as pd\nimport math\nimport matplotlib.pyplot as plt\n%matplotlib inline", "_____no_output_____" ] ], [ [ "### Global variables for the program", "_____no_output_____" ] ], [ [ "N_points = 100 # Number of points for constructing function\nx_min = 1 # Min of the range of x (feature)\nx_max = 25 # Max of the range of x (feature)\nnoise_mean = 0 # Mean of the Gaussian noise adder\nnoise_sd = 10 # Std.Dev of the Gaussian noise adder\ntest_set_fraction = 0.2", "_____no_output_____" ] ], [ [ "### Generate feature and output vector for a non-linear function with transcedental terms\nThe ground truth or originating function is as follows:\n\n$$ y=f(x)= (20x+3x^2+0.1x^3).sin(x).e^{-0.1x}+\\psi(x) $$\n\n$$ {OR} $$\n\n$$ y=f(x)= (20x+3x^2+0.1x^3)+\\psi(x) $$\n\n$${where,}\\ \\psi(x) : {\\displaystyle f(x\\;|\\;\\mu ,\\sigma ^{2})={\\frac {1}{\\sqrt {2\\pi \\sigma ^{2}}}}\\;e^{-{\\frac {(x-\\mu )^{2}}{2\\sigma ^{2}}}}} $$", "_____no_output_____" ] ], [ [ "# Definition of the function with exponential and sinusoidal terms\ndef func_trans(x):\n result = (20*x+3*x**2+0.1*x**3)*np.sin(x)*np.exp(-0.1*x)\n return (result)", "_____no_output_____" ], [ "# Definition of the function without exponential and sinusoidal terms i.e. just the polynomial\ndef func_poly(x):\n result = 20*x+3*x**2+0.1*x**3\n return (result)", "_____no_output_____" ], [ "# Densely spaced points for generating the ideal functional curve\nx_smooth = np.array(np.linspace(x_min,x_max,501))\n\n# Use one of the following\ny_smooth = func_trans(x_smooth)\n#y_smooth = func_poly(x_smooth)", "_____no_output_____" ], [ "# Linearly spaced sample points\nX=np.array(np.linspace(x_min,x_max,N_points))", "_____no_output_____" ], [ "# Added observational/measurement noise\nnoise_x = np.random.normal(loc=noise_mean,scale=noise_sd,size=N_points)", "_____no_output_____" ], [ "# Observed output after adding the noise\ny = func_trans(X)+noise_x", "_____no_output_____" ], [ "# Store the values in a DataFrame\ndf = pd.DataFrame(data=X,columns=['X'])\ndf['Ideal y']=df['X'].apply(func_trans)\ndf['Sin_X']=df['X'].apply(math.sin)\ndf['y']=y\ndf.head()", "_____no_output_____" ] ], [ [ "### Plot the function(s), both the ideal characteristic and the observed output (with process and observation noise)", "_____no_output_____" ] ], [ [ "df.plot.scatter('X','y',title='True process and measured samples\\n',\n grid=True,edgecolors=(0,0,0),c='blue',s=60,figsize=(10,6))\nplt.plot(x_smooth,y_smooth,'k')", "_____no_output_____" ] ], [ [ "### Import scikit-learn librares and prepare train/test splits", "_____no_output_____" ] ], [ [ "from sklearn.linear_model import LinearRegression\nfrom sklearn.linear_model import LassoCV\nfrom sklearn.linear_model import RidgeCV\nfrom sklearn.ensemble import AdaBoostRegressor\nfrom sklearn.preprocessing import PolynomialFeatures\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.pipeline import make_pipeline", "_____no_output_____" ], [ "X_train, X_test, y_train, y_test = train_test_split(df[['X','Sin_X']], df['y'], test_size=test_set_fraction)\n\n#X_train=X_train.reshape(X_train.size,1)\n#y_train=y_train.reshape(y_train.size,1)\n#X_test=X_test.reshape(X_test.size,1)\n#y_test=y_test.reshape(y_test.size,1)\n\n#X_train=X_train.reshape(-1,1)\n#y_train=y_train.reshape(-1,1)\n#X_test=X_test.reshape(-1,1)\n#y_test=y_test.reshape(-1,1)\n\nfrom sklearn import preprocessing\nX_scaled = preprocessing.scale(X_train)\ny_scaled = preprocessing.scale(y_train)", "_____no_output_____" ] ], [ [ "### Polynomial model with LASSO/Ridge regularization (pipelined) with lineary spaced samples\n** This is an advanced machine learning method which prevents over-fitting by penalizing high-valued coefficients i.e. keep them bounded **", "_____no_output_____" ] ], [ [ "# Regression model parameters\nridge_alpha = tuple([10**(x) for x in range(-3,0,1) ]) # Alpha (regularization strength) of ridge regression\n# Alpha (regularization strength) of LASSO regression\nlasso_eps = 0.0001\nlasso_nalpha=20\nlasso_iter=5000\n\n# Min and max degree of polynomials features to consider\ndegree_min = 2\ndegree_max = 8", "_____no_output_____" ], [ "linear_sample_score = []\npoly_degree = []\nrmse=[]\nt_linear=[]\nimport time\nfor degree in range(degree_min,degree_max+1):\n t1=time.time()\n #model = make_pipeline(PolynomialFeatures(degree), RidgeCV(alphas=ridge_alpha,normalize=True,cv=5))\n model = make_pipeline(PolynomialFeatures(degree), LassoCV(eps=lasso_eps,n_alphas=lasso_nalpha, \n max_iter=lasso_iter,normalize=True,cv=5))\n #model = make_pipeline(PolynomialFeatures(degree), LinearRegression(normalize=True))\n model.fit(X_train, y_train)\n t2=time.time()\n t = t2-t1\n t_linear.append(t)\n test_pred = np.array(model.predict(X_test))\n RMSE=np.sqrt(np.sum(np.square(test_pred-y_test)))\n test_score = model.score(X_test,y_test)\n linear_sample_score.append(test_score)\n rmse.append(RMSE)\n poly_degree.append(degree)\n #print(\"Test score of model with degree {}: {}\\n\".format(degree,test_score))\n \n plt.figure()\n plt.title(\"Predicted vs. actual for polynomial of degree {}\".format(degree),fontsize=15)\n plt.xlabel(\"Actual values\")\n plt.ylabel(\"Predicted values\")\n plt.scatter(y_test,test_pred)\n plt.plot(y_test,y_test,'r',lw=2)", "_____no_output_____" ], [ "linear_sample_score", "_____no_output_____" ], [ "plt.figure(figsize=(8,5))\nplt.grid(True)\nplt.plot(poly_degree,rmse,lw=3,c='red')\nplt.title(\"Model complexity (highest polynomial degree) vs. test score\\n\",fontsize=20)\nplt.xlabel (\"\\nDegree of polynomial\",fontsize=20)\nplt.ylabel (\"Root-mean-square error on test set\",fontsize=15)", "_____no_output_____" ], [ "df_score = pd.DataFrame(data={'degree':[d for d in range(degree_min,degree_max+1)],\n 'Linear sample score':linear_sample_score})", "_____no_output_____" ], [ "# Save the best R^2 score\nr2_linear = max(linear_sample_score)\nprint(\"Best R^2 score for linear polynomial degree models:\",r2_linear)", "Best R^2 score for linear polynomial degree models: 0.995693773025\n" ], [ "plt.figure(figsize=(8,5))\nplt.grid(True)\nplt.plot(poly_degree,linear_sample_score,lw=3,c='red')\nplt.xlabel (\"\\nModel Complexity: Degree of polynomial\",fontsize=20)\nplt.ylabel (\"R^2 score on test set\",fontsize=15)", "_____no_output_____" ] ], [ [ "## 1-hidden layer (Shallow) network", "_____no_output_____" ] ], [ [ "import tensorflow as tf\nlearning_rate = 1e-6\ntraining_epochs = 150000\n\nn_input = 1 # Number of features\nn_output = 1 # Regression output is a number only\n\nn_hidden_layer = 100 # layer number of features", "_____no_output_____" ], [ "weights = {\n 'hidden_layer': tf.Variable(tf.random_normal([n_input, n_hidden_layer])),\n 'out': tf.Variable(tf.random_normal([n_hidden_layer, n_output]))\n}\nbiases = {\n 'hidden_layer': tf.Variable(tf.random_normal([n_hidden_layer])),\n 'out': tf.Variable(tf.random_normal([n_output]))\n}", "_____no_output_____" ], [ "# tf Graph input\nx = tf.placeholder(\"float32\", [None,n_input])\ny = tf.placeholder(\"float32\", [None,n_output])", "_____no_output_____" ], [ "# Hidden layer with RELU activation\nlayer_1 = tf.add(tf.matmul(x, weights['hidden_layer']),biases['hidden_layer'])\nlayer_1 = tf.sin(layer_1)\n\n# Output layer with linear activation\nops = tf.add(tf.matmul(layer_1, weights['out']), biases['out'])", "_____no_output_____" ], [ "# Define loss and optimizer\ncost = tf.reduce_mean(tf.squared_difference(ops,y))\noptimizer = tf.train.GradientDescentOptimizer(learning_rate=learning_rate).minimize(cost)", "_____no_output_____" ], [ "from tqdm import tqdm\nimport time\n\n# Initializing the variables\ninit = tf.global_variables_initializer()\n\n# Empty lists for book-keeping purpose\nepoch=0\nlog_epoch = []\nepoch_count=[]\nacc=[]\nloss_epoch=[]\n\nX_train, X_test, y_train, y_test = train_test_split(df['X'], df['y'], \n test_size=test_set_fraction)\n\nX_train=X_train.reshape(X_train.size,1)\ny_train=y_train.reshape(y_train.size,1)\nX_test=X_test.reshape(X_test.size,1)\ny_test=y_test.reshape(y_test.size,1)\n\n# Launch the graph and time the session\nt1=time.time()\nwith tf.Session() as sess:\n sess.run(init) \n # Loop over epochs\n for epoch in tqdm(range(training_epochs)):\n # Run optimization process (backprop) and cost function (to get loss value)\n _,l=sess.run([optimizer,cost], feed_dict={x: X_train, y: y_train})\n loss_epoch.append(l) # Save the loss for every epoch \n epoch_count.append(epoch+1) #Save the epoch count\n \n # print(\"Epoch {}/{} finished. Loss: {}, Accuracy: {}\".format(epoch+1,training_epochs,round(l,4),round(accu,4)))\n #print(\"Epoch {}/{} finished. Loss: {}\".format(epoch+1,training_epochs,round(l,4)))\n w=sess.run(weights)\n b = sess.run(biases)\n yhat=sess.run(ops,feed_dict={x:X_test})\nt2=time.time()\n\ntime_SNN = t2-t1", "C:\\Users\\Tirtha\\Python\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:17: FutureWarning: reshape is deprecated and will raise in a subsequent release. Please use .values.reshape(...) instead\nC:\\Users\\Tirtha\\Python\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:18: FutureWarning: reshape is deprecated and will raise in a subsequent release. Please use .values.reshape(...) instead\nC:\\Users\\Tirtha\\Python\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:19: FutureWarning: reshape is deprecated and will raise in a subsequent release. Please use .values.reshape(...) instead\nC:\\Users\\Tirtha\\Python\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:20: FutureWarning: reshape is deprecated and will raise in a subsequent release. Please use .values.reshape(...) instead\n100%|████████████████████████████████████████████████████████████████████████| 150000/150000 [02:10<00:00, 1150.13it/s]\n" ], [ "plt.plot(loss_epoch)", "_____no_output_____" ], [ "# Total variance\nSSt_SNN = np.sum(np.square(y_test-np.mean(y_test)))\n# Residual sum of squares\nSSr_SNN = np.sum(np.square(yhat-y_test))\n# Root-mean-square error\nRMSE_SNN = np.sqrt(np.sum(np.square(yhat-y_test)))\n# R^2 coefficient\nr2_SNN = 1-(SSr_SNN/SSt_SNN)\n\nprint(\"RMSE error of the shallow neural network:\",RMSE_SNN)\nprint(\"R^2 value of the shallow neural network:\",r2_SNN)", "RMSE error of the shallow neural network: 94.9837638167\nR^2 value of the shallow neural network: 0.983809646773\n" ], [ "plt.figure(figsize=(10,6))\nplt.title(\"Predicted vs. actual (test set) for shallow (1-hidden layer) neural network\\n\",fontsize=15)\nplt.xlabel(\"Actual values (test set)\")\nplt.ylabel(\"Predicted values\")\nplt.scatter(y_test,yhat,edgecolors='k',s=100,c='green')\nplt.grid(True)\nplt.plot(y_test,y_test,'r',lw=2)", "_____no_output_____" ] ], [ [ "## Deep Neural network for regression", "_____no_output_____" ], [ "### Import and declaration of variables", "_____no_output_____" ] ], [ [ "import tensorflow as tf\nlearning_rate = 1e-6\ntraining_epochs = 15000\n\nn_input = 1 # Number of features\nn_output = 1 # Regression output is a number only\n\nn_hidden_layer_1 = 30 # Hidden layer 1\nn_hidden_layer_2 = 30 # Hidden layer 2", "_____no_output_____" ] ], [ [ "### Weights and bias variable", "_____no_output_____" ] ], [ [ "# Store layers weight & bias as Variables classes in dictionaries\nweights = {\n 'hidden_layer_1': tf.Variable(tf.random_normal([n_input, n_hidden_layer_1])),\n 'hidden_layer_2': tf.Variable(tf.random_normal([n_hidden_layer_1, n_hidden_layer_2])),\n 'out': tf.Variable(tf.random_normal([n_hidden_layer_2, n_output]))\n}\nbiases = {\n 'hidden_layer_1': tf.Variable(tf.random_normal([n_hidden_layer_1])),\n 'hidden_layer_2': tf.Variable(tf.random_normal([n_hidden_layer_2])),\n 'out': tf.Variable(tf.random_normal([n_output]))\n}", "_____no_output_____" ] ], [ [ "### Input data as placeholder", "_____no_output_____" ] ], [ [ "# tf Graph input\nx = tf.placeholder(\"float32\", [None,n_input])\ny = tf.placeholder(\"float32\", [None,n_output])", "_____no_output_____" ] ], [ [ "### Hidden and output layers definition (using TensorFlow mathematical functions)", "_____no_output_____" ] ], [ [ "# Hidden layer with activation\nlayer_1 = tf.add(tf.matmul(x, weights['hidden_layer_1']),biases['hidden_layer_1'])\nlayer_1 = tf.sin(layer_1)\n\nlayer_2 = tf.add(tf.matmul(layer_1, weights['hidden_layer_2']),biases['hidden_layer_2'])\nlayer_2 = tf.nn.relu(layer_2)\n\n# Output layer with linear activation\nops = tf.add(tf.matmul(layer_2, weights['out']), biases['out'])", "_____no_output_____" ] ], [ [ "### Gradient descent optimizer for training (backpropagation):\nFor the training of the neural network we need to perform __backpropagation__ i.e. propagate the errors, calculated by this cost function, backwards through the layers all the way up to the input weights and bias in order to adjust them accordingly (minimize the error). This involves taking first-order derivatives of the activation functions and applying chain-rule to ___'multiply'___ the effect of various layers as the error propagates back.\n\nYou can read more on this here: [Backpropagation in Neural Network](https://en.wikipedia.org/wiki/Backpropagation)\n\nFortunately, TensorFlow already implicitly implements this step i.e. takes care of all the chained differentiations for us. All we need to do is to specify an Optimizer object and pass on the cost function. Here, we are using a Gradient Descent Optimizer.\n\nGradient descent is a first-order iterative optimization algorithm for finding the minimum of a function. To find a local minimum of a function using gradient descent, one takes steps proportional to the negative of the gradient (or of the approximate gradient) of the function at the current point.\n\nYou can read more on this: [Gradient Descent](https://en.wikipedia.org/wiki/Gradient_descent)\n", "_____no_output_____" ] ], [ [ "# Define loss and optimizer\ncost = tf.reduce_mean(tf.squared_difference(ops,y))\noptimizer = tf.train.GradientDescentOptimizer(learning_rate=learning_rate).minimize(cost)", "_____no_output_____" ] ], [ [ "### TensorFlow Session for training and loss estimation", "_____no_output_____" ] ], [ [ "from tqdm import tqdm\nimport time\n# Initializing the variables\ninit = tf.global_variables_initializer()\n\n# Empty lists for book-keeping purpose\nepoch=0\nlog_epoch = []\nepoch_count=[]\nacc=[]\nloss_epoch=[]\nr2_DNN = []\ntest_size = []\n\nfor i in range(5):\n X_train, X_test, y_train, y_test = train_test_split(df['X'], df['y'], \n test_size=test_set_fraction)\n\n X_train=X_train.reshape(X_train.size,1)\n y_train=y_train.reshape(y_train.size,1)\n X_test=X_test.reshape(X_test.size,1)\n y_test=y_test.reshape(y_test.size,1)\n # Launch the graph and time the session\n with tf.Session() as sess:\n sess.run(init) \n # Loop over epochs\n for epoch in tqdm(range(training_epochs)):\n # Run optimization process (backprop) and cost function (to get loss value)\n #r1 = int(epoch/10000)\n #learning_rate = learning_rate-r1*3e-6\n #optimizer = tf.train.GradientDescentOptimizer(learning_rate=learning_rate).minimize(cost)\n _,l=sess.run([optimizer,cost], feed_dict={x: X_train, y: y_train})\n \n yhat=sess.run(ops,feed_dict={x:X_test})\n\n #test_size.append(0.5-(i*0.04))\n # Total variance\n SSt_DNN = np.sum(np.square(y_test-np.mean(y_test)))\n # Residual sum of squares\n SSr_DNN = np.sum(np.square(yhat-y_test))\n # Root-mean-square error\n RMSE_DNN = np.sqrt(np.sum(np.square(yhat-y_test)))\n # R^2 coefficient\n r2 = 1-(SSr_DNN/SSt_DNN)\n r2_DNN.append(r2)\n print(\"Run: {} finished. Score: {}\".format(i+1,r2))\n", "C:\\Users\\Tirtha\\Python\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:19: FutureWarning: reshape is deprecated and will raise in a subsequent release. Please use .values.reshape(...) instead\nC:\\Users\\Tirtha\\Python\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:20: FutureWarning: reshape is deprecated and will raise in a subsequent release. Please use .values.reshape(...) instead\nC:\\Users\\Tirtha\\Python\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:21: FutureWarning: reshape is deprecated and will raise in a subsequent release. Please use .values.reshape(...) instead\nC:\\Users\\Tirtha\\Python\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:22: FutureWarning: reshape is deprecated and will raise in a subsequent release. Please use .values.reshape(...) instead\n100%|██████████████████████████████████████████████████████████████████████████| 15000/15000 [00:12<00:00, 1184.93it/s]\n" ] ], [ [ "### Plot R2 score corss-validation results", "_____no_output_____" ] ], [ [ "plt.figure(figsize=(10,6))\nplt.title(\"\\nR2-score for cross-validation runs of \\ndeep (2-layer) neural network\\n\",fontsize=25)\nplt.xlabel(\"\\nCross-validation run with random test/train split #\",fontsize=15)\nplt.ylabel(\"R2 score (test set)\\n\",fontsize=15)\nplt.scatter([i+1 for i in range(5)],r2_DNN,edgecolors='k',s=100,c='green')\nplt.grid(True)", "_____no_output_____" ], [ "", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "import os\nimport requests", "_____no_output_____" ], [ "os.getcwd()", "_____no_output_____" ], [ "os.chdir('dags/tableau')", "_____no_output_____" ], [ "from subscription_classes import *\nfrom config import tableau_server_config, subscription_email_config", "_____no_output_____" ], [ "conn = TableauServer(config_json=tableau_server_config)", "_____no_output_____" ], [ "conn.sign_in()", "_____no_output_____" ], [ "conn.get_user_details()", "_____no_output_____" ], [ "conn.get_email_users()", "_____no_output_____" ], [ "conn.get_distribution_list('InterWorks - Airflow Demo')", "_____no_output_____" ], [ "base_url = conn.base_get_url", "_____no_output_____" ], [ "base_url", "_____no_output_____" ], [ "conn.get_group_by_name(group_name='InterWorks - Airflow Demo')", "_____no_output_____" ], [ "test_group = '0194f591-9d29-4a4b-af50-a5c606b9dc5f'", "_____no_output_____" ], [ "test_url = \"{0}/groups/{1}/users\".format(base_url, test_group)", "_____no_output_____" ], [ "test_url", "_____no_output_____" ], [ "response = requests.get(test_url, headers=conn.default_headers)", "_____no_output_____" ], [ "response", "_____no_output_____" ], [ "response.json()", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Controlling Flow with Conditional Statements\n\nNow that you've learned how to create conditional statements, let's learn how to use them to control the flow of our programs. This is done with `if`, `elif`, and `else` statements. ", "_____no_output_____" ], [ "## The `if` Statement\n\n What if we wanted to check if a number was divisible by 2 and if so then print that number out. Let's diagram that out.", "_____no_output_____" ], [ "\n\n- Check to see if A is even\n- If yes, then print our message: \"A is even\"", "_____no_output_____" ], [ "This use case can be translated into a \"if\" statement. I'm going to write this out in pseudocode which looks very similar to Python.\n\n```text\nif A is even:\n print \"A is even\"\n```", "_____no_output_____" ] ], [ [ "# Let's translate this into Python code\ndef check_evenness(A):\n if A % 2 == 0:\n print(f\"A ({A:02}) is even!\")\n\nfor i in range(1, 11):\n check_evenness(i)", "A (02) is even!\nA (04) is even!\nA (06) is even!\nA (08) is even!\nA (10) is even!\n" ], [ "# You can do multiple if statements and they're executed sequentially\n\nA = 10\n\nif A > 0:\n print('A is positive')\nif A % 2 == 0:\n print('A is even!')", "A is positive\nA is even!\n" ] ], [ [ "## The `else` Statement\nBut what if we wanted to know if the number was even OR odd? Let's diagram that out:\n\n", "_____no_output_____" ], [ "Again, translating this to pseudocode, we're going to use the 'else' statement:\n\n```text\nif A is even:\n print \"A is even\"\nelse:\n print \"A is odd\"\n```", "_____no_output_____" ] ], [ [ "# Let's translate this into Python code\ndef check_evenness(A):\n if A % 2 == 0:\n print(f\"A ({A:02}) is even!\") \n else:\n print(f'A ({A:02}) is odd!')\n\nfor i in range(1, 11):\n check_evenness(i)", "A (01) is odd!\nA (02) is even!\nA (03) is odd!\nA (04) is even!\nA (05) is odd!\nA (06) is even!\nA (07) is odd!\nA (08) is even!\nA (09) is odd!\nA (10) is even!\n" ] ], [ [ "# The 'else if' or `elif` Statement\nWhat if we wanted to check if A is divisible by 2 or 3? Let's diagram that out:", "_____no_output_____" ], [ "", "_____no_output_____" ], [ "Again, translating this into psuedocode, we're going to use the 'else if' statement.\n\n```text\nif A is divisible by 2:\n print \"2 divides A\"\nelse if A is divisible by 3:\n print \"3 divides A\"\nelse\n print \"2 and 3 don't divide A\"\n```", "_____no_output_____" ] ], [ [ "# Let's translate this into Python code\ndef check_divisible_by_2_and_3(A):\n if A % 2 == 0:\n print(f\"2 divides A ({A:02})!\") \n \n # else if in Python is elif\n elif A % 3 == 0:\n print(f'3 divides A ({A:02})!')\n else:\n print(f'A ({A:02}) is not divisible by 2 or 3)')\n \nfor i in range(1, 11):\n check_divisible_by_2_and_3(i)", "A (01) is not divisible by 2 or 3)\n2 divides A (02)!\n3 divides A (03)!\n2 divides A (04)!\nA (05) is not divisible by 2 or 3)\n2 divides A (06)!\nA (07) is not divisible by 2 or 3)\n2 divides A (08)!\n3 divides A (09)!\n2 divides A (10)!\n" ] ], [ [ "## Order Matters\n\nWhen chaining conditionals, you need to be careful how you order them. For example, what if we wanted te check if a number is divisible by 2, 3, or both:", "_____no_output_____" ], [ "", "_____no_output_____" ] ], [ [ "# Let's translate this into Python code\ndef check_divisible_by_2_and_3(A):\n if A % 2 == 0:\n print(f\"2 divides A ({A:02})!\") \n elif A % 3 == 0:\n print(f'3 divides A ({A:02})!')\n elif A % 2 == 0 and A % 3 == 0:\n print(f'2 and 3 divides A ({A:02})!')\n else:\n print(f\"2 or 3 doesn't divide A ({A:02})\")\n \nfor i in range(1, 11):\n check_divisible_by_2_and_3(i)", "2 or 3 doesn't divide A (01)\n2 divides A (02)!\n3 divides A (03)!\n2 divides A (04)!\n2 or 3 doesn't divide A (05)\n2 divides A (06)!\n2 or 3 doesn't divide A (07)\n2 divides A (08)!\n3 divides A (09)!\n2 divides A (10)!\n" ] ], [ [ "Wait! we would expect that 6, which is divisible by both 2 and 3 to show that! Looking back at the graphic, we can see that the flow is checking for 2 first, and since that's true we follow that path first. Let's make a correction to our diagram to fix this:", "_____no_output_____" ], [ "", "_____no_output_____" ] ], [ [ "# Let's translate this into Python code\ndef check_divisible_by_2_and_3(A):\n if A % 2 == 0 and A % 3 == 0:\n print(f'2 and 3 divides A ({A:02})!')\n elif A % 3 == 0:\n print(f'3 divides A ({A:02})!')\n elif A % 2 == 0:\n print(f\"2 divides A ({A:02})!\") \n else:\n print(f\"2 or 3 doesn't divide A ({A:02})\")\n \nfor i in range(1, 11):\n check_divisible_by_2_and_3(i)", "2 or 3 doesn't divide A (01)\n2 divides A (02)!\n3 divides A (03)!\n2 divides A (04)!\n2 or 3 doesn't divide A (05)\n2 and 3 divides A (06)!\n2 or 3 doesn't divide A (07)\n2 divides A (08)!\n3 divides A (09)!\n2 divides A (10)!\n" ] ], [ [ "**NOTE:** Always put your most restrictive conditional at the top of your if statements and then work your way down to the least restrictive.\n\n", "_____no_output_____" ], [ "## In-Class Assignments\n\n- Create a funcition that takes two inputs variables `A` and `divisor`. Check if `divisor` divides into `A`. If it does, print `\"<value of A> is divided by <value of divisor>\"`. Don't forget about the `in` operator that checks if a substring is in another string.\n- Create a function that takes an input variable `A` which is a string. Check if `A` has the substring `apple`, `peach`, or `blueberry` in it. Print out which of these are found within the string. Note: you could do this using just if/elif/else statements, but is there a better way using lists, for loops, and if/elif/else statements?", "_____no_output_____" ], [ "## Solutions", "_____no_output_____" ] ], [ [ "def is_divisible(A, divisor):\n if A % divisor == 0:\n print(f'{A} is divided by {divisor}')\n \nA = 37\n\n# this is actually a crude way to find if the number is prime\nfor i in range(2, int(A / 2)):\n is_divisible(A, i)\n \n# notice that nothing was printed? That's because 37 is prime\n \nB = 27\nfor i in range(2, int(B / 2)):\n is_divisible(B, i)", "27 is divided by 3\n27 is divided by 9\n" ], [ "# this is ONE solution. There are more out there and probably better\n# one too\ndef check_for_fruit(A):\n found_fruit = []\n if 'apple' in A:\n found_fruit.append('apple')\n if 'peach' in A:\n found_fruit.append('peach')\n if 'blueberry' in A:\n found_fruit.append('blueberry')\n \n found_fruit_str = ''\n for fruit in found_fruit:\n found_fruit_str += fruit\n found_fruit_str += ', '\n \n if len(found_fruit) > 0:\n print(found_fruit_str + ' is found within the string')\n else:\n print('No fruit found in the string')", "_____no_output_____" ], [ "check_for_fruit('there are apples and peaches in this pie')", "apple, peach, is found within the string\n" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown", "markdown", "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown", "markdown" ], [ "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "## Example 4.1\n\nLet \\{0,1,2,3\\} denote the actions \\{up, right, down, left\\} respectively.", "_____no_output_____" ] ], [ [ "import numpy as np\nclass gridworld:\n def __init__(self):\n self.terminal_state = [0,15]\n self.action = [0,1,2,3]\n self.value = np.zeros(16)\n self.reward = -1 \n \n def next_state(self, s, a):\n if s in self.terminal_state:\n return s\n if a == 0:\n s = s - 4 if s >= 4 else s \n elif a == 1:\n s = s + 1 if (s+1)%4 != 0 else s\n elif a == 2:\n s = s + 4 if (s+4) < 16 else s\n elif a == 3:\n s = s - 1 if s % 4 !=0 else s\n return s\n \n def policy_evaluation(self):\n k=0\n while True:\n delta = 0\n v_new = np.zeros(16)\n for s in range(16):\n v = 0\n if s in self.terminal_state:\n v_new[s] = 0\n else:\n v = self.value[s]\n temp = 0\n for a in range(4):\n v_new[s] += 0.25*(-1 + self.value[self.next_state(s,a)])\n delta = max(delta, abs(v-v_new[s]))\n self.value = v_new\n \n # greedy policy\n policy = np.zeros((16,4))\n for s in range(1,15):\n vmax = -30\n nmax = 0\n for a in range(4):\n v = self.value[self.next_state(s,a)]\n if v > vmax:\n vmax = v\n nmax = 1\n elif vmax - v < 0.05:\n nmax += 1\n for a in range(4):\n v = self.value[self.next_state(s,a)]\n if v == vmax:\n policy[s,a] = 1/nmax\n \n k += 1\n if k == 1 or k==2 or k==3 or k==10 or k==131:\n print('The state value for %ith iteration:\\n' %k, v_new.reshape((4,4)))\n print('The greedy policy for %ith iteration:\\n' %k, policy)\n if delta < 0.001:\n return\n\n \n \n\n\n \n", "_____no_output_____" ], [ "a = gridworld()", "_____no_output_____" ], [ "a.policy_evaluation()", "The state value for 1th iteration:\n [[ 0. -1. -1. -1.]\n [-1. -1. -1. -1.]\n [-1. -1. -1. -1.]\n [-1. -1. -1. 0.]]\nThe greedy policy for 1th iteration:\n [[0. 0. 0. 0. ]\n [0. 0. 0. 1. ]\n [0.25 0.25 0.25 0.25]\n [0.25 0.25 0.25 0.25]\n [1. 0. 0. 0. ]\n [0.25 0.25 0.25 0.25]\n [0.25 0.25 0.25 0.25]\n [0.25 0.25 0.25 0.25]\n [0.25 0.25 0.25 0.25]\n [0.25 0.25 0.25 0.25]\n [0.25 0.25 0.25 0.25]\n [0. 0. 1. 0. ]\n [0.25 0.25 0.25 0.25]\n [0.25 0.25 0.25 0.25]\n [0. 1. 0. 0. ]\n [0. 0. 0. 0. ]]\nThe state value for 2th iteration:\n [[ 0. -1.75 -2. -2. ]\n [-1.75 -2. -2. -2. ]\n [-2. -2. -2. -1.75]\n [-2. -2. -1.75 0. ]]\nThe greedy policy for 2th iteration:\n [[0. 0. 0. 0. ]\n [0. 0. 0. 1. ]\n [0. 0. 0. 1. ]\n [0.25 0.25 0.25 0.25]\n [1. 0. 0. 0. ]\n [0.5 0. 0. 0.5 ]\n [0.25 0.25 0.25 0.25]\n [0. 0. 1. 0. ]\n [1. 0. 0. 0. ]\n [0.25 0.25 0.25 0.25]\n [0. 0.5 0.5 0. ]\n [0. 0. 1. 0. ]\n [0.25 0.25 0.25 0.25]\n [0. 1. 0. 0. ]\n [0. 1. 0. 0. ]\n [0. 0. 0. 0. ]]\nThe state value for 3th iteration:\n [[ 0. -2.4375 -2.9375 -3. ]\n [-2.4375 -2.875 -3. -2.9375]\n [-2.9375 -3. -2.875 -2.4375]\n [-3. -2.9375 -2.4375 0. ]]\nThe greedy policy for 3th iteration:\n [[0. 0. 0. 0. ]\n [0. 0. 0. 1. ]\n [0. 0. 0. 1. ]\n [0. 0. 0.5 0.5]\n [1. 0. 0. 0. ]\n [0.5 0. 0. 0.5]\n [0. 0. 0.5 0.5]\n [0. 0. 1. 0. ]\n [1. 0. 0. 0. ]\n [0.5 0.5 0. 0. ]\n [0. 0.5 0.5 0. ]\n [0. 0. 1. 0. ]\n [0.5 0.5 0. 0. ]\n [0. 1. 0. 0. ]\n [0. 1. 0. 0. ]\n [0. 0. 0. 0. ]]\nThe state value for 10th iteration:\n [[ 0. -6.13796997 -8.35235596 -8.96731567]\n [-6.13796997 -7.73739624 -8.42782593 -8.35235596]\n [-8.35235596 -8.42782593 -7.73739624 -6.13796997]\n [-8.96731567 -8.35235596 -6.13796997 0. ]]\nThe greedy policy for 10th iteration:\n [[0. 0. 0. 0. ]\n [0. 0. 0. 1. ]\n [0. 0. 0. 1. ]\n [0. 0. 0.5 0.5]\n [1. 0. 0. 0. ]\n [0.5 0. 0. 0.5]\n [0. 0. 0.5 0.5]\n [0. 0. 1. 0. ]\n [1. 0. 0. 0. ]\n [0.5 0.5 0. 0. ]\n [0. 0.5 0.5 0. ]\n [0. 0. 1. 0. ]\n [0.5 0.5 0. 0. ]\n [0. 1. 0. 0. ]\n [0. 1. 0. 0. ]\n [0. 0. 0. 0. ]]\nThe state value for 131th iteration:\n [[ 0. -13.98945772 -19.98437823 -21.98251832]\n [-13.98945772 -17.98623815 -19.98448273 -19.98437823]\n [-19.98437823 -19.98448273 -17.98623815 -13.98945772]\n [-21.98251832 -19.98437823 -13.98945772 0. ]]\nThe greedy policy for 131th iteration:\n [[0. 0. 0. 0. ]\n [0. 0. 0. 1. ]\n [0. 0. 0. 1. ]\n [0. 0. 0.5 0.5]\n [1. 0. 0. 0. ]\n [0.5 0. 0. 0. ]\n [0. 0. 0.5 0.5]\n [0. 0. 1. 0. ]\n [1. 0. 0. 0. ]\n [0.5 0.5 0. 0. ]\n [0. 0.5 0.5 0. ]\n [0. 0. 1. 0. ]\n [0.5 0.5 0. 0. ]\n [0. 1. 0. 0. ]\n [0. 1. 0. 0. ]\n [0. 0. 0. 0. ]]\n" ] ] ]

[ "markdown", "code" ]

[ [ "markdown" ], [ "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Laboratorio 8", "_____no_output_____" ] ], [ [ "import pandas as pd\nimport matplotlib.pyplot as plt\n\nfrom sklearn import datasets\nfrom sklearn.neighbors import KNeighborsClassifier\nfrom sklearn.linear_model import LogisticRegression\nfrom sklearn.metrics import plot_confusion_matrix \n\n%matplotlib inline", "_____no_output_____" ], [ "digits_X, digits_y = datasets.load_digits(return_X_y=True, as_frame=True)\ndigits = pd.concat([digits_X, digits_y], axis=1)\ndigits.head()", "_____no_output_____" ] ], [ [ "## Ejercicio 1\n\n(1 pto.)\n\nUtilizando todos los datos, ajusta un modelo de regresión logística a los datos de dígitos. No agregues intercepto y define un máximo de iteraciones de 400.\n\nObtén el _score_ y explica el tan buen resultado.", "_____no_output_____" ] ], [ [ "logistic = LogisticRegression(solver=\"lbfgs\", max_iter=400, fit_intercept=False)\nfit=logistic.fit(digits_X, digits_y)\n\n\nprint(f\"El score del modelo de regresión logística es {fit.score(digits_X, digits_y)}\")", "El score del modelo de regresión logística es 1.0\n" ] ], [ [ "__Respuesta:__ Supongo que es porque no estamos usando los datos originales, y no otros datos predecidos a partir de los originales", "_____no_output_____" ], [ "## Ejercicio 2\n\n(1 pto.)\n\nUtilizando todos los datos, ¿Cuál es la mejor elección del parámetro $k$ al ajustar un modelo kNN a los datos de dígitos? Utiliza valores $k=2, ..., 10$.", "_____no_output_____" ] ], [ [ "for k in range(2, 11):\n kNN = KNeighborsClassifier(n_neighbors=k)\n fit=kNN.fit(digits_X, digits_y)\n print(f\"El score del modelo de kNN con k={k} es {fit.score(digits_X, digits_y)}\")", "El score del modelo de kNN con k=2 es 0.9910962715637173\nEl score del modelo de kNN con k=3 es 0.993322203672788\nEl score del modelo de kNN con k=4 es 0.9922092376182526\nEl score del modelo de kNN con k=5 es 0.9905397885364496\nEl score del modelo de kNN con k=6 es 0.989983305509182\nEl score del modelo de kNN con k=7 es 0.9905397885364496\nEl score del modelo de kNN con k=8 es 0.9894268224819143\nEl score del modelo de kNN con k=9 es 0.9888703394546466\nEl score del modelo de kNN con k=10 es 0.9855314412910406\n" ] ], [ [ "__Respuesta:__ El caso k=3, porque es el mas cercano a 1.", "_____no_output_____" ], [ "## Ejercicio 3\n\n(1 pto.)\n\nGrafica la matriz de confusión normalizada por predicción de ambos modelos (regresión logística y kNN con la mejor elección de $k$).\n\n¿Qué conclusión puedes sacar?\n\nHint: Revisa el argumento `normalize` de la matriz de confusión.", "_____no_output_____" ] ], [ [ "plot_confusion_matrix(logistic, digits_X, digits_y, normalize='true');", "_____no_output_____" ], [ "best_knn = KNeighborsClassifier(n_neighbors=3)\nB_kNN = best_knn.fit(digits_X, digits_y)\nplot_confusion_matrix(B_kNN, digits_X, digits_y, normalize='true');", "_____no_output_____" ] ], [ [ "__Respuesta:__ Que la primera matriz es una mejor prediccion que la segunda, esto porque se pudo obtener una matriz diagonal con, asumo, menor cantidad de errores comparado a los valores que no se encuentran en la diagonal y que son distintos de 0 en el segundo caso.", "_____no_output_____" ], [ "## Ejercicio 4\n\n(1 pto.)\n\nEscoge algún registro donde kNN se haya equivocado, _plotea_ la imagen y comenta las razones por las que el algoritmo se pudo haber equivocado.", "_____no_output_____" ] ], [ [ "neigh_tt = KNeighborsClassifier(n_neighbors=5)\nneigh_tt.fit(digits_X, digits_y)", "_____no_output_____" ] ], [ [ "El valor real del registro seleccionado es", "_____no_output_____" ] ], [ [ "i = 5\nneigh_tt.predict(digits_X.iloc[[i], :])", "_____no_output_____" ] ], [ [ "Mentras que la predicción dada por kNN es", "_____no_output_____" ] ], [ [ "neigh_tt.predict_proba(digits_X.iloc[[i], :])", "_____no_output_____" ] ], [ [ "A continuación la imagen", "_____no_output_____" ] ], [ [ "plt.imshow(digits_X.loc[[i], :].to_numpy().reshape(8, 8), cmap=plt.cm.gray_r, interpolation='nearest');", "_____no_output_____" ] ], [ [ "__Respuesta:__ Se me viene a la cabeza la forma de los numeros de los relojes digitales tipo : https://i.linio.com/p/b6286f5db6ae58cdd0aef38e070a51b5-product.jpg\nDonde las figuras se parecen bastante, sobre todo porque tienen un formato reducido para mostrar los numeritos (hay una base donde detras que se ilumina dependiendo del numero), entonces como la matriz es chikita, igual quedan figuras similares y con no tan buena resolucion, por lo que lleva a errores, en este caso, confundiendo un 5 con un 9.", "_____no_output_____" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown", "code", "markdown" ]

[ [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# **Spit some [tensor] flow**\n\nWe need to learn the intricacies of tensorflow to master deep learning\n\n`Let's get this over with`\n\n", "_____no_output_____" ] ], [ [ "import numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nimport tensorflow as tf\nimport cv2\nprint(tf.__version__)", "2.2.0\n" ], [ "def evaluation_tf(report, y_test, y_pred, classes):\n plt.plot(report.history['loss'], label = 'training_loss')\n plt.plot(report.history['val_loss'], label = 'validation_loss')\n plt.legend()\n plt.show()\n\n plt.plot(report.history['accuracy'], label = 'training_accuracy')\n plt.plot(report.history['val_accuracy'], label = 'validation_accuracy')\n plt.legend()\n plt.show()\n\n from sklearn.metrics import confusion_matrix\n import itertools\n cm = confusion_matrix(y_test, y_pred)\n\n plt.figure(figsize=(10,10))\n plt.imshow(cm, cmap=plt.cm.Blues)\n for i,j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):\n plt.text(j, i, format(cm[i,j], 'd'),\n horizontalalignment = 'center',\n color='black')\n plt.xlabel(\"Predicted labels\")\n plt.ylabel(\"True labels\")\n plt.xticks(range(0,classes))\n plt.yticks(range(0,classes))\n plt.title('Confusion matrix')\n plt.colorbar()\n plt.show()", "_____no_output_____" ], [ "# Taken from https://www.cs.toronto.edu/~kriz/cifar.html\nlabels = \"airplane,automobile,bird,cat,deer,dog,frog,horse,ship,truck\".split(\",\")", "_____no_output_____" ] ], [ [ "## As a rule of thumb\n\nRemember that the pooling operation decreases the size of the image, and we lose information.\n\nHowever, the number of features generally increases and we get more features extracted from the images.\n\nThe choices of hyperparams bother us sometimes, because DL has a lot of trial and error involved, we can choose the \n\n- learning rate\n\n- number of layers\n\n- number of neurons per layer \n\n- feature size \n\n- feature number \n\n- pooling size \n\n- stride \n\nOn a side note, if you use strided convolution layers, they will decrease the size of the image as well\n\n\nIf we have images with different sizes as inputs; for example; H1 x W1 x 3 and H2 x W2 x 3, then the output will be flatten-ed to different sizes, this won't work for DENSE layers as they do not have change-able input sizes, so we use global max pooling to make a vector of size 1 x 1 x (#_Of_Feature_Maps_)", "_____no_output_____" ] ], [ [ "from tensorflow.keras.layers import Input, Conv2D, Dropout, Dense, Flatten, BatchNormalization, MaxPooling2D\nfrom tensorflow.keras.models import Model", "_____no_output_____" ], [ "from tensorflow.keras.datasets import cifar10\n(X_train, y_train), (X_test, y_test) = cifar10.load_data()", "_____no_output_____" ], [ "X_train, X_test = X_train / 255.0 , X_test / 255.0 \nprint(X_train.shape)\nprint(X_test.shape)\nprint(y_train.shape)\nprint(y_test.shape)", "(50000, 32, 32, 3)\n(10000, 32, 32, 3)\n(50000, 1)\n(10000, 1)\n" ], [ "y_train, y_test = y_train.flatten(), y_test.flatten() \nprint(y_train.shape)\nprint(y_test.shape)", "(50000,)\n(10000,)\n" ], [ "classes = len(set(y_train))\nprint(classes)", "10\n" ], [ "input_shape = X_train[0].shape", "_____no_output_____" ], [ "i_layer = Input(shape = input_shape)\nh_layer = Conv2D(32, (3,3),activation='relu', padding='same')(i_layer)\nh_layer = BatchNormalization()(h_layer)\nh_layer = Conv2D(64, (3,3), activation='relu', padding='same')(h_layer)\nh_layer = BatchNormalization()(h_layer)\nh_layer = Conv2D(128, (3,3), activation='relu', padding='same')(h_layer)\nh_layer = BatchNormalization()(h_layer)\nh_layer = MaxPooling2D((2,2))(h_layer)\nh_layer = Conv2D(128, (3,3), activation='relu', padding='same')(h_layer)\nh_layer = BatchNormalization()(h_layer)\nh_layer = MaxPooling2D((2,2))(h_layer)\nh_layer = Flatten()(h_layer)\nh_layer = Dropout(0.5)(h_layer)\nh_layer = Dense(512, activation='relu')(h_layer)\nh_layer = Dropout(0.5)(h_layer)\no_layer = Dense(classes, activation='softmax')(h_layer)\n\nmodel = Model(i_layer, o_layer)\n", "_____no_output_____" ], [ "model.compile(optimizer='adam', \n loss = 'sparse_categorical_crossentropy',\n metrics = ['accuracy'])\n\nreport = model.fit(X_train, y_train, validation_data=(X_test, y_test), epochs=50)", "Epoch 1/50\n1563/1563 [==============================] - 13s 8ms/step - loss: 0.4014 - accuracy: 0.8681 - val_loss: 0.3888 - val_accuracy: 0.8753\nEpoch 2/50\n1354/1563 [========================>.....] - ETA: 1s - loss: 0.3561 - accuracy: 0.8822" ], [ "y_pred = model.predict(X_test).argmax(axis=1) \n# only for sparse categorical crossentropy", "_____no_output_____" ], [ "evaluation_tf(report, y_test, y_pred, classes)", "_____no_output_____" ], [ "misshits = np.where(y_pred!=y_test)[0]\nprint(\"total Mishits = \" + str(len(misshits)))\nindex = np.random.choice(misshits)\nplt.imshow(X_test[index])\nplt.title(\"Predicted = \" + str(labels[y_pred[index]]) + \", Real = \" + str(labels[y_test[index]]))", "total Mishits = 1715\n" ] ], [ [ "## LET'S ADD SOME DATA AUGMENTATION FROM KERAS \n\ntaken from https://keras.io/api/preprocessing/image/", "_____no_output_____" ] ], [ [ "batch_size = 32\ndata_generator = tf.keras.preprocessing.image.ImageDataGenerator(width_shift_range = 0.1, \n height_shift_range = 0.1, \n horizontal_flip=True)\n", "_____no_output_____" ], [ "model_dg = Model(i_layer, o_layer)\nmodel_dg.compile(optimizer='adam', \n loss = 'sparse_categorical_crossentropy',\n metrics = ['accuracy'])\ntrain_data_generator = data_generator.flow(X_train, y_train, batch_size)\nspe = X_train.shape[0] // batch_size\n\nreport = model_dg.fit_generator(train_data_generator, validation_data=(X_test, y_test), steps_per_epoch=spe, epochs=50)", "Epoch 1/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3662 - accuracy: 0.8841 - val_loss: 0.3942 - val_accuracy: 0.8808\nEpoch 2/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3575 - accuracy: 0.8838 - val_loss: 0.4664 - val_accuracy: 0.8674\nEpoch 3/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3647 - accuracy: 0.8821 - val_loss: 0.4152 - val_accuracy: 0.8667\nEpoch 4/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3581 - accuracy: 0.8831 - val_loss: 0.4398 - val_accuracy: 0.8765\nEpoch 5/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3571 - accuracy: 0.8836 - val_loss: 0.4192 - val_accuracy: 0.8642\nEpoch 6/50\n1562/1562 [==============================] - 26s 17ms/step - loss: 0.3556 - accuracy: 0.8843 - val_loss: 0.4226 - val_accuracy: 0.8768\nEpoch 7/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3532 - accuracy: 0.8846 - val_loss: 0.3996 - val_accuracy: 0.8882\nEpoch 8/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3642 - accuracy: 0.8826 - val_loss: 0.4041 - val_accuracy: 0.8841\nEpoch 9/50\n1562/1562 [==============================] - 26s 17ms/step - loss: 0.3549 - accuracy: 0.8851 - val_loss: 0.4054 - val_accuracy: 0.8825\nEpoch 10/50\n1562/1562 [==============================] - 26s 17ms/step - loss: 0.3473 - accuracy: 0.8871 - val_loss: 0.4510 - val_accuracy: 0.8660\nEpoch 11/50\n1562/1562 [==============================] - 26s 17ms/step - loss: 0.3530 - accuracy: 0.8852 - val_loss: 0.3841 - val_accuracy: 0.8832\nEpoch 12/50\n1562/1562 [==============================] - 26s 17ms/step - loss: 0.3525 - accuracy: 0.8860 - val_loss: 0.4192 - val_accuracy: 0.8818\nEpoch 13/50\n1562/1562 [==============================] - 26s 17ms/step - loss: 0.3485 - accuracy: 0.8867 - val_loss: 0.4487 - val_accuracy: 0.8713\nEpoch 14/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3538 - accuracy: 0.8846 - val_loss: 0.4198 - val_accuracy: 0.8740\nEpoch 15/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3530 - accuracy: 0.8847 - val_loss: 0.4317 - val_accuracy: 0.8783\nEpoch 16/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3524 - accuracy: 0.8852 - val_loss: 0.3986 - val_accuracy: 0.8811\nEpoch 17/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3550 - accuracy: 0.8875 - val_loss: 0.4242 - val_accuracy: 0.8808\nEpoch 18/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3521 - accuracy: 0.8851 - val_loss: 0.4596 - val_accuracy: 0.8725\nEpoch 19/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3545 - accuracy: 0.8863 - val_loss: 0.4020 - val_accuracy: 0.8744\nEpoch 20/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3433 - accuracy: 0.8883 - val_loss: 0.4245 - val_accuracy: 0.8859\nEpoch 21/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3497 - accuracy: 0.8866 - val_loss: 0.4546 - val_accuracy: 0.8720\nEpoch 22/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3409 - accuracy: 0.8908 - val_loss: 0.4262 - val_accuracy: 0.8759\nEpoch 23/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3454 - accuracy: 0.8880 - val_loss: 0.4293 - val_accuracy: 0.8871\nEpoch 24/50\n1562/1562 [==============================] - 26s 17ms/step - loss: 0.3442 - accuracy: 0.8891 - val_loss: 0.4051 - val_accuracy: 0.8773\nEpoch 25/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3510 - accuracy: 0.8853 - val_loss: 0.4168 - val_accuracy: 0.8806\nEpoch 26/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3452 - accuracy: 0.8889 - val_loss: 0.4177 - val_accuracy: 0.8843\nEpoch 27/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3334 - accuracy: 0.8909 - val_loss: 0.5204 - val_accuracy: 0.8715\nEpoch 28/50\n1562/1562 [==============================] - 26s 17ms/step - loss: 0.3448 - accuracy: 0.8890 - val_loss: 0.4044 - val_accuracy: 0.8764\nEpoch 29/50\n1562/1562 [==============================] - 26s 17ms/step - loss: 0.3351 - accuracy: 0.8916 - val_loss: 0.3904 - val_accuracy: 0.8850\nEpoch 30/50\n1562/1562 [==============================] - 27s 18ms/step - loss: 0.3397 - accuracy: 0.8906 - val_loss: 0.3834 - val_accuracy: 0.8852\nEpoch 31/50\n1562/1562 [==============================] - 28s 18ms/step - loss: 0.3373 - accuracy: 0.8913 - val_loss: 0.4081 - val_accuracy: 0.8782\nEpoch 32/50\n1562/1562 [==============================] - 28s 18ms/step - loss: 0.3416 - accuracy: 0.8902 - val_loss: 0.4038 - val_accuracy: 0.8877\nEpoch 33/50\n1562/1562 [==============================] - 28s 18ms/step - loss: 0.3417 - accuracy: 0.8891 - val_loss: 0.4396 - val_accuracy: 0.8863\nEpoch 34/50\n1562/1562 [==============================] - 28s 18ms/step - loss: 0.3329 - accuracy: 0.8917 - val_loss: 0.3944 - val_accuracy: 0.8823\nEpoch 35/50\n1562/1562 [==============================] - 28s 18ms/step - loss: 0.3389 - accuracy: 0.8901 - val_loss: 0.4085 - val_accuracy: 0.8782\nEpoch 36/50\n1562/1562 [==============================] - 27s 18ms/step - loss: 0.3405 - accuracy: 0.8886 - val_loss: 0.4484 - val_accuracy: 0.8664\nEpoch 37/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3353 - accuracy: 0.8922 - val_loss: 0.4940 - val_accuracy: 0.8784\nEpoch 38/50\n1562/1562 [==============================] - 28s 18ms/step - loss: 0.3432 - accuracy: 0.8905 - val_loss: 0.4127 - val_accuracy: 0.8928\nEpoch 39/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3335 - accuracy: 0.8922 - val_loss: 0.4257 - val_accuracy: 0.8786\nEpoch 40/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3358 - accuracy: 0.8921 - val_loss: 0.3832 - val_accuracy: 0.8832\nEpoch 41/50\n1562/1562 [==============================] - 28s 18ms/step - loss: 0.3302 - accuracy: 0.8921 - val_loss: 0.4106 - val_accuracy: 0.8854\nEpoch 42/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3327 - accuracy: 0.8925 - val_loss: 0.4659 - val_accuracy: 0.8833\nEpoch 43/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3337 - accuracy: 0.8928 - val_loss: 0.4045 - val_accuracy: 0.8772\nEpoch 44/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3336 - accuracy: 0.8932 - val_loss: 0.4963 - val_accuracy: 0.8734\nEpoch 45/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3385 - accuracy: 0.8915 - val_loss: 0.4232 - val_accuracy: 0.8957\nEpoch 46/50\n1562/1562 [==============================] - 28s 18ms/step - loss: 0.3366 - accuracy: 0.8929 - val_loss: 0.3889 - val_accuracy: 0.8842\nEpoch 47/50\n1562/1562 [==============================] - 28s 18ms/step - loss: 0.3258 - accuracy: 0.8941 - val_loss: 0.4424 - val_accuracy: 0.8709\nEpoch 48/50\n1562/1562 [==============================] - 27s 18ms/step - loss: 0.3416 - accuracy: 0.8898 - val_loss: 0.4321 - val_accuracy: 0.8874\nEpoch 49/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3275 - accuracy: 0.8963 - val_loss: 0.4120 - val_accuracy: 0.8833\nEpoch 50/50\n1562/1562 [==============================] - 27s 17ms/step - loss: 0.3278 - accuracy: 0.8954 - val_loss: 0.3850 - val_accuracy: 0.8807\n" ], [ "y_pred = model.predict(X_test).argmax(axis=1) \n# only for sparse categorical crossentropy", "_____no_output_____" ], [ "evaluation_tf(report, y_test, y_pred, classes)", "_____no_output_____" ], [ "misshits = np.where(y_pred!=y_test)[0]\nprint(\"total Mishits = \" + str(len(misshits)))\nindex = np.random.choice(misshits)\nplt.imshow(X_test[index])\nplt.title(\"Predicted = \" + str(labels[y_pred[index]]) + \", Real = \" + str(labels[y_test[index]]))", "total Mishits = 1193\n" ] ] ]

[ "markdown", "code", "markdown", "code", "markdown", "code" ]

[ [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ] ]

[ "CC-BY-3.0" ]

[ "CC-BY-3.0" ]