Datasets:

---

bigcode

/

jupyter-parsed

like 4

Follow

BigCode 1.35k

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# Part 2: Intro to Private Training with Remote Execution\n\nIn the last section, we learned about PointerTensors, which create the underlying infrastructure we need for privacy preserving Deep Learning. In this section, we're going to see how to use these basic tools to train our first deep learning model using remote execution.\n\nAuthors:\n- Yann Dupis - Twitter: [@YannDupis](https://twitter.com/YannDupis)\n- Andrew Trask - Twitter: [@iamtrask](https://twitter.com/iamtrask)\n\n### Why use remote execution?\n\nLet's say you are an AI startup who wants to build a deep learning model to detect [diabetic retinopathy (DR)](https://ai.googleblog.com/2016/11/deep-learning-for-detection-of-diabetic.html), which is the fastest growing cause of blindness. Before training your model, the first step would be to acquire a dataset of retinopathy images with signs of DR. One approach could be to work with a hospital and ask them to send you a copy of this dataset. However because of the sensitivity of the patients' data, the hospital might be exposed to liability risks.\n\n\nThat's where remote execution comes into the picture. Instead of bringing training data to the model (a central server), you bring the model to the training data (wherever it may live). In this case, it would be the hospital.\n\nThe idea is that this allows whoever is creating the data to own the only permanent copy, and thus maintain control over who ever has access to it. Pretty cool, eh?", "_____no_output_____" ], [ "# Section 2.1 - Private Training on MNIST\n\nFor this tutorial, we will train a model on the [MNIST dataset](http://yann.lecun.com/exdb/mnist/) to classify digits based on images.\n\nWe can assume that we have a remote worker named Bob who owns the data.", "_____no_output_____" ] ], [ [ "import tensorflow as tf\nimport syft as sy\n\nhook = sy.TensorFlowHook(tf)\nbob = sy.VirtualWorker(hook, id=\"bob\")", "_____no_output_____" ] ], [ [ "Let's download the MNIST data from `tf.keras.datasets`. Note that we are converting the data from numpy to `tf.Tensor` in order to have the PySyft functionalities.", "_____no_output_____" ] ], [ [ "mnist = tf.keras.datasets.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\nx_train, y_train = tf.convert_to_tensor(x_train), tf.convert_to_tensor(y_train)\nx_test, y_test = tf.convert_to_tensor(x_test), tf.convert_to_tensor(y_test)", "_____no_output_____" ] ], [ [ "As decribed in Part 1, we can send this data to Bob with the `send` method on the `tf.Tensor`. ", "_____no_output_____" ] ], [ [ "x_train_ptr = x_train.send(bob)\ny_train_ptr = y_train.send(bob)", "_____no_output_____" ] ], [ [ "Excellent! We have everything to start experimenting. To train our model on Bob's machine, we just have to perform the following steps:\n\n- Define a model, including optimizer and loss\n- Send the model to Bob\n- Start the training process\n- Get the trained model back\n\nLet's do it!", "_____no_output_____" ] ], [ [ "# Define the model\nmodel = tf.keras.models.Sequential([\n tf.keras.layers.Flatten(input_shape=(28, 28)),\n tf.keras.layers.Dense(128, activation='relu'),\n tf.keras.layers.Dropout(0.2),\n tf.keras.layers.Dense(10, activation='softmax')\n])\n\n# Compile with optimizer, loss and metrics\nmodel.compile(optimizer='adam',\n loss='sparse_categorical_crossentropy',\n metrics=['accuracy'])", "_____no_output_____" ] ], [ [ "Once you have defined your model, you can simply send it to Bob calling the `send` method. It's the exact same process as sending a tensor.", "_____no_output_____" ] ], [ [ "model_ptr = model.send(bob)", "_____no_output_____" ], [ "model_ptr", "_____no_output_____" ] ], [ [ "Now, we have a pointer pointing to the model on Bob's machine. We can validate that's the case by inspecting the attribute `_objects` on the virtual worker. ", "_____no_output_____" ] ], [ [ "bob._objects[model_ptr.id_at_location]", "_____no_output_____" ] ], [ [ "Everything is ready to start training our model on this remote dataset. You can call `fit` and pass `x_train_ptr` `y_train_ptr` which are pointing to Bob's data. Note that's the exact same interface as normal `tf.keras`.", "_____no_output_____" ] ], [ [ "model_ptr.fit(x_train_ptr, y_train_ptr, epochs=2, validation_split=0.2)", "_____no_output_____" ] ], [ [ "Fantastic! you have trained your model acheiving an accuracy greater than 95%.\n\nYou can get your trained model back by just calling `get` on it. ", "_____no_output_____" ] ], [ [ "model_gotten = model_ptr.get()\n\nmodel_gotten", "_____no_output_____" ] ], [ [ "It's good practice to see if your model can generalize by assessing its accuracy on an holdout dataset. You can simply call `evaluate`.", "_____no_output_____" ] ], [ [ "model_gotten.evaluate(x_test, y_test, verbose=2)", "_____no_output_____" ] ], [ [ "Boom! The model remotely trained on Bob's data is more than 95% accurate on this holdout dataset.", "_____no_output_____" ], [ "If your model doesn't fit into the Sequential paradigm, you can use Keras's functional API, or even subclass [tf.keras.Model](https://www.tensorflow.org/guide/keras/custom_layers_and_models#building_models) to create custom models.", "_____no_output_____" ] ], [ [ "class CustomModel(tf.keras.Model):\n\n def __init__(self, num_classes=10):\n super(CustomModel, self).__init__(name='custom_model')\n self.num_classes = num_classes\n \n self.flatten = tf.keras.layers.Flatten(input_shape=(28, 28))\n self.dense_1 = tf.keras.layers.Dense(128, activation='relu')\n self.dropout = tf.keras.layers.Dropout(0.2)\n self.dense_2 = tf.keras.layers.Dense(num_classes, activation='softmax')\n\n def call(self, inputs, training=False):\n x = self.flatten(inputs)\n x = self.dense_1(x)\n x = self.dropout(x, training=training)\n return self.dense_2(x)\n \nmodel = CustomModel(10)\n\n# need to call the model on dummy data before sending it\n# in order to set the input shape (required when saving to SavedModel)\nmodel.predict(tf.ones([1, 28, 28]))\n\nmodel.compile(optimizer='adam',\n loss='sparse_categorical_crossentropy',\n metrics=['accuracy'])\n\nmodel_ptr = model.send(bob)\n\nmodel_ptr.fit(x_train_ptr, y_train_ptr, epochs=2, validation_split=0.2)", "_____no_output_____" ] ], [ [ "## Well Done!\n\nAnd voilà! We have trained a Deep Learning model on Bob's data by sending the model to him. Never in this process do we ever see or request access to the underlying training data! We preserve the privacy of Bob!!!", "_____no_output_____" ], [ "# Congratulations!!! - Time to Join the Community!\n\nCongratulations on completing this notebook tutorial! If you enjoyed this and would like to join the movement toward privacy preserving, decentralized ownership of AI and the AI supply chain (data), you can do so in the following ways!\n\n### Star PySyft on GitHub\n\nThe easiest way to help our community is just by starring the Repos! This helps raise awareness of the cool tools we're building.\n\n- Star PySyft on GitHub! - [https://github.com/OpenMined/PySyft](https://github.com/OpenMined/PySyft)\n- Star PySyft-TensorFlow on GitHub! - [https://github.com/OpenMined/PySyft-TensorFlow]\n\n### Join our Slack!\n\nThe best way to keep up to date on the latest advancements is to join our community! You can do so by filling out the form at [http://slack.openmined.org](http://slack.openmined.org)\n\n### Join a Code Project!\n\nThe best way to contribute to our community is to become a code contributor! At any time you can go to PySyft GitHub Issues page and filter for \"Projects\". This will show you all the top level Tickets giving an overview of what projects you can join! If you don't want to join a project, but you would like to do a bit of coding, you can also look for more \"one off\" mini-projects by searching for GitHub issues marked \"good first issue\".\n\n- [PySyft Projects](https://github.com/OpenMined/PySyft/issues?q=is%3Aopen+is%3Aissue+label%3AProject)\n- [Good First Issue Tickets](https://github.com/OpenMined/PySyft/issues?q=is%3Aopen+is%3Aissue+label%3A%22good+first+issue%22)\n\n### Donate\n\nIf you don't have time to contribute to our codebase, but would still like to lend support, you can also become a Backer on our Open Collective. All donations go toward our web hosting and other community expenses such as hackathons and meetups!\n\n[OpenMined's Open Collective Page](https://opencollective.com/openmined)", "_____no_output_____" ] ] ]

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "<h1> Training on Cloud ML Engine </h1>\n\nThis notebook illustrates distributed training and hyperparameter tuning on Cloud ML Engine.", "_____no_output_____" ] ], [ [ "# change these to try this notebook out\nBUCKET = 'cloud-training-demos-ml'\nPROJECT = 'cloud-training-demos'\nREGION = 'us-central1'", "_____no_output_____" ], [ "import os\nos.environ['BUCKET'] = BUCKET\nos.environ['PROJECT'] = PROJECT\nos.environ['REGION'] = REGION", "_____no_output_____" ], [ "%%bash\nif ! gsutil ls | grep -q gs://${BUCKET}/; then\n gsutil mb -l ${REGION} gs://${BUCKET}\n # copy canonical set of preprocessed files if you didn't do previous notebook\n gsutil -m cp -R gs://cloud-training-demos/babyweight gs://${BUCKET}\nfi", "_____no_output_____" ], [ "%bash\ngsutil ls gs://${BUCKET}/babyweight/preproc/*-00000*", "gs://cloud-training-demos-ml/babyweight/preproc/eval.csv-00000-of-00012\ngs://cloud-training-demos-ml/babyweight/preproc/train.csv-00000-of-00043\n" ] ], [ [ "Now that we have the TensorFlow code working on a subset of the data, we can package the TensorFlow code up as a Python module and train it on Cloud ML Engine.\n<p>\n<h2> Train on Cloud ML Engine </h2>\n<p>\nTraining on Cloud ML Engine requires:\n<ol>\n<li> Making the code a Python package\n<li> Using gcloud to submit the training code to Cloud ML Engine\n</ol>\n<p>\nThe code in model.py is the same as in the TensorFlow notebook. I just moved it to a file so that I could package it up as a module.\n(explore the <a href=\"babyweight/trainer\">directory structure</a>).", "_____no_output_____" ] ], [ [ "%bash\ngrep \"^def\" babyweight/trainer/model.py", "def read_dataset(prefix, pattern, batch_size=512):\ndef get_wide_deep():\ndef serving_input_fn():\ndef experiment_fn(output_dir):\ndef train_and_evaluate(output_dir):\n" ] ], [ [ "After moving the code to a package, make sure it works standalone. (Note the --pattern and --train_examples lines so that I am not trying to boil the ocean on my laptop). Even then, this takes about <b>a minute</b> in which you won't see any output ...", "_____no_output_____" ] ], [ [ "%bash\necho \"bucket=${BUCKET}\"\nrm -rf babyweight_trained\nexport PYTHONPATH=${PYTHONPATH}:${PWD}/babyweight\npython -m trainer.task \\\n --bucket=${BUCKET} \\\n --output_dir=babyweight_trained \\\n --job-dir=./tmp \\\n --pattern=\"00000-of-\" --train_examples=500", "_____no_output_____" ] ], [ [ "Once the code works in standalone mode, you can run it on Cloud ML Engine. Because this is on the entire dataset, it will take a while. The training run took about <b> an hour </b> for me. You can monitor the job from the GCP console in the Cloud Machine Learning Engine section.", "_____no_output_____" ] ], [ [ "%bash\nOUTDIR=gs://${BUCKET}/babyweight/trained_model\nJOBNAME=babyweight_$(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=trainer.task \\\n --package-path=$(pwd)/babyweight/trainer \\\n --job-dir=$OUTDIR \\\n --staging-bucket=gs://$BUCKET \\\n --scale-tier=STANDARD_1 \\\n --runtime-version=1.4 \\\n -- \\\n --bucket=${BUCKET} \\\n --output_dir=${OUTDIR} \\\n --train_examples=200000", "_____no_output_____" ] ], [ [ "When I ran it, training finished, and the evaluation happened three times (filter in Stackdriver on the word \"dict\"):\n<pre>\nSaving dict for global step 390632: average_loss = 1.06578, global_step = 390632, loss = 545.55\n</pre>\nThe final RMSE was 1.066 pounds.", "_____no_output_____" ] ], [ [ "from google.datalab.ml import TensorBoard\nTensorBoard().start('gs://{}/babyweight/trained_model'.format(BUCKET))", "_____no_output_____" ], [ "for pid in TensorBoard.list()['pid']:\n TensorBoard().stop(pid)\n print 'Stopped TensorBoard with pid {}'.format(pid)", "Stopped TensorBoard with pid 10437\n" ] ], [ [ "<h2> Hyperparameter tuning </h2>\n<p>\nAll of these are command-line parameters to my program. To do hyperparameter tuning, create hyperparam.xml and pass it as --configFile.\nThis step will take <b>1 hour</b> -- you can increase maxParallelTrials or reduce maxTrials to get it done faster. Since maxParallelTrials is the number of initial seeds to start searching from, you don't want it to be too large; otherwise, all you have is a random search.\n", "_____no_output_____" ] ], [ [ "%writefile hyperparam.yaml\ntrainingInput:\n scaleTier: STANDARD_1\n hyperparameters:\n hyperparameterMetricTag: average_loss\n goal: MINIMIZE\n maxTrials: 30\n maxParallelTrials: 3\n params:\n - parameterName: batch_size\n type: INTEGER\n minValue: 8\n maxValue: 512\n scaleType: UNIT_LOG_SCALE\n - parameterName: nembeds\n type: INTEGER\n minValue: 3\n maxValue: 30\n scaleType: UNIT_LINEAR_SCALE\n - parameterName: nnsize\n type: INTEGER\n minValue: 64\n maxValue: 512\n scaleType: UNIT_LOG_SCALE", "Overwriting hyperparam.yaml\n" ] ], [ [ "In reality, you would hyper-parameter tune over your entire dataset, and not on a smaller subset (see --pattern). But because this is a demo, I wanted it to finish quickly.", "_____no_output_____" ] ], [ [ "%bash\nOUTDIR=gs://${BUCKET}/babyweight/hyperparam\nJOBNAME=babyweight_$(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=trainer.task \\\n --package-path=$(pwd)/babyweight/trainer \\\n --job-dir=$OUTDIR \\\n --staging-bucket=gs://$BUCKET \\\n --scale-tier=STANDARD_1 \\\n --config=hyperparam.yaml \\\n --runtime-version=1.4 \\\n -- \\\n --bucket=${BUCKET} \\\n --output_dir=${OUTDIR} \\\n --pattern=\"00000-of-\" --train_examples=5000", "_____no_output_____" ], [ "%bash\ngcloud ml-engine jobs describe babyweight_180123_202458", "_____no_output_____" ] ], [ [ "<h2> Repeat training </h2>\n<p>\nThis time with tuned parameters (note last line)", "_____no_output_____" ] ], [ [ "%bash\nOUTDIR=gs://${BUCKET}/babyweight/trained_model\nJOBNAME=babyweight_$(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=trainer.task \\\n --package-path=$(pwd)/babyweight/trainer \\\n --job-dir=$OUTDIR \\\n --staging-bucket=gs://$BUCKET \\\n --scale-tier=STANDARD_1 \\\n -- \\\n --bucket=${BUCKET} \\\n --output_dir=${OUTDIR} \\\n --train_examples=200000 --batch_size=35 --nembeds=16 --nnsize=281", "_____no_output_____" ] ], [ [ "Copyright 2017 Google Inc. Licensed under the Apache License, Version 2.0 (the \"License\"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License", "_____no_output_____" ] ] ]

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

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/Mengxue12/tensorflow-1-public/blob/main/C4/W4/ungraded_labs/C4_W4_Lab_1_LSTM.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ] ], [ [ "#@title 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# 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_____" ] ], [ [ "**Note:** This notebook can run using TensorFlow 2.5.0", "_____no_output_____" ] ], [ [ "#!pip install tensorflow==2.5.0", "_____no_output_____" ], [ "import tensorflow as tf\nimport numpy as np\nimport matplotlib.pyplot as plt\nprint(tf.__version__)", "2.8.0\n" ], [ "def plot_series(time, series, format=\"-\", start=0, end=None):\n plt.plot(time[start:end], series[start:end], format)\n plt.xlabel(\"Time\")\n plt.ylabel(\"Value\")\n plt.grid(True)\n\ndef trend(time, slope=0):\n return slope * time\n\ndef seasonal_pattern(season_time):\n \"\"\"Just an arbitrary pattern, you can change it if you wish\"\"\"\n return np.where(season_time < 0.4,\n np.cos(season_time * 2 * np.pi),\n 1 / np.exp(3 * season_time))\n\ndef seasonality(time, period, amplitude=1, phase=0):\n \"\"\"Repeats the same pattern at each period\"\"\"\n season_time = ((time + phase) % period) / period\n return amplitude * seasonal_pattern(season_time)\n\ndef noise(time, noise_level=1, seed=None):\n rnd = np.random.RandomState(seed)\n return rnd.randn(len(time)) * noise_level\n\ntime = np.arange(4 * 365 + 1, dtype=\"float32\")\nbaseline = 10\nseries = trend(time, 0.1) \nbaseline = 10\namplitude = 40\nslope = 0.05\nnoise_level = 5\n\n# Create the series\nseries = baseline + trend(time, slope) + seasonality(time, period=365, amplitude=amplitude)\n# Update with noise\nseries += noise(time, noise_level, seed=42)\n\nsplit_time = 1000\ntime_train = time[:split_time]\nx_train = series[:split_time]\ntime_valid = time[split_time:]\nx_valid = series[split_time:]\n\nwindow_size = 20\nbatch_size = 32\nshuffle_buffer_size = 1000", "_____no_output_____" ], [ "def windowed_dataset(series, window_size, batch_size, shuffle_buffer):\n series = tf.expand_dims(series, axis=-1)\n ds = tf.data.Dataset.from_tensor_slices(series)\n ds = ds.window(window_size + 1, shift=1, drop_remainder=True)\n ds = ds.flat_map(lambda w: w.batch(window_size + 1))\n ds = ds.shuffle(shuffle_buffer)\n ds = ds.map(lambda w: (w[:-1], w[1:]))\n return ds.batch(batch_size).prefetch(1)", "_____no_output_____" ], [ "def model_forecast(model, series, window_size):\n ds = tf.data.Dataset.from_tensor_slices(series)\n ds = ds.window(window_size, shift=1, drop_remainder=True)\n ds = ds.flat_map(lambda w: w.batch(window_size))\n ds = ds.batch(32).prefetch(1)\n forecast = model.predict(ds)\n return forecast", "_____no_output_____" ], [ "tf.keras.backend.clear_session()\ntf.random.set_seed(51)\nnp.random.seed(51)\n\nwindow_size = 30\ntrain_set = windowed_dataset(x_train, window_size, batch_size=128, shuffle_buffer=shuffle_buffer_size)\n\nmodel = tf.keras.models.Sequential([\n tf.keras.layers.Conv1D(filters=32, kernel_size=5,\n strides=1, padding=\"causal\",\n activation=\"relu\",\n input_shape=[None, 1]),\n tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(32, return_sequences=True)),\n tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(32, return_sequences=True)),\n tf.keras.layers.Dense(1),\n tf.keras.layers.Lambda(lambda x: x * 200)\n])\nlr_schedule = tf.keras.callbacks.LearningRateScheduler(\n lambda epoch: 1e-8 * 10**(epoch / 20))\noptimizer = tf.keras.optimizers.SGD(learning_rate=1e-8, momentum=0.9)\nmodel.compile(loss=tf.keras.losses.Huber(),\n optimizer=optimizer,\n metrics=[\"mae\"])\nhistory = model.fit(train_set, epochs=100, callbacks=[lr_schedule])", "Epoch 1/100\n8/8 [==============================] - 24s 86ms/step - loss: 73.1912 - mae: 73.6904 - lr: 1.0000e-08\nEpoch 2/100\n8/8 [==============================] - 0s 33ms/step - loss: 72.4799 - mae: 72.9791 - lr: 1.1220e-08\nEpoch 3/100\n8/8 [==============================] - 0s 34ms/step - loss: 71.3446 - mae: 71.8437 - lr: 1.2589e-08\nEpoch 4/100\n8/8 [==============================] - 0s 33ms/step - loss: 69.9230 - mae: 70.4223 - lr: 1.4125e-08\nEpoch 5/100\n8/8 [==============================] - 0s 31ms/step - loss: 68.2678 - mae: 68.7669 - lr: 1.5849e-08\nEpoch 6/100\n8/8 [==============================] - 0s 34ms/step - loss: 66.3621 - mae: 66.8609 - lr: 1.7783e-08\nEpoch 7/100\n8/8 [==============================] - 0s 33ms/step - loss: 64.2432 - mae: 64.7424 - lr: 1.9953e-08\nEpoch 8/100\n8/8 [==============================] - 0s 33ms/step - loss: 61.8749 - mae: 62.3738 - lr: 2.2387e-08\nEpoch 9/100\n8/8 [==============================] - 0s 33ms/step - loss: 59.2491 - mae: 59.7479 - lr: 2.5119e-08\nEpoch 10/100\n8/8 [==============================] - 0s 31ms/step - loss: 56.3267 - mae: 56.8254 - lr: 2.8184e-08\nEpoch 11/100\n8/8 [==============================] - 0s 32ms/step - loss: 53.0701 - mae: 53.5686 - lr: 3.1623e-08\nEpoch 12/100\n8/8 [==============================] - 0s 32ms/step - loss: 49.3867 - mae: 49.8849 - lr: 3.5481e-08\nEpoch 13/100\n8/8 [==============================] - 0s 33ms/step - loss: 45.1182 - mae: 45.6165 - lr: 3.9811e-08\nEpoch 14/100\n8/8 [==============================] - 0s 33ms/step - loss: 41.6538 - mae: 42.1518 - lr: 4.4668e-08\nEpoch 15/100\n8/8 [==============================] - 0s 33ms/step - loss: 40.6341 - mae: 41.1322 - lr: 5.0119e-08\nEpoch 16/100\n8/8 [==============================] - 1s 40ms/step - loss: 39.5119 - mae: 40.0100 - lr: 5.6234e-08\nEpoch 17/100\n8/8 [==============================] - 0s 31ms/step - loss: 38.1009 - mae: 38.5985 - lr: 6.3096e-08\nEpoch 18/100\n8/8 [==============================] - 0s 34ms/step - loss: 36.5304 - mae: 37.0279 - lr: 7.0795e-08\nEpoch 19/100\n8/8 [==============================] - 0s 32ms/step - loss: 34.9672 - mae: 35.4641 - lr: 7.9433e-08\nEpoch 20/100\n8/8 [==============================] - 0s 33ms/step - loss: 33.5379 - mae: 34.0347 - lr: 8.9125e-08\nEpoch 21/100\n8/8 [==============================] - 0s 33ms/step - loss: 32.2358 - mae: 32.7321 - lr: 1.0000e-07\nEpoch 22/100\n8/8 [==============================] - 0s 33ms/step - loss: 31.0509 - mae: 31.5469 - lr: 1.1220e-07\nEpoch 23/100\n8/8 [==============================] - 0s 33ms/step - loss: 29.9243 - mae: 30.4202 - lr: 1.2589e-07\nEpoch 24/100\n8/8 [==============================] - 0s 33ms/step - loss: 28.8317 - mae: 29.3274 - lr: 1.4125e-07\nEpoch 25/100\n8/8 [==============================] - 0s 32ms/step - loss: 27.7314 - mae: 28.2272 - lr: 1.5849e-07\nEpoch 26/100\n8/8 [==============================] - 0s 32ms/step - loss: 26.6107 - mae: 27.1066 - lr: 1.7783e-07\nEpoch 27/100\n8/8 [==============================] - 0s 32ms/step - loss: 25.4565 - mae: 25.9519 - lr: 1.9953e-07\nEpoch 28/100\n8/8 [==============================] - 0s 32ms/step - loss: 24.2668 - mae: 24.7617 - lr: 2.2387e-07\nEpoch 29/100\n8/8 [==============================] - 0s 33ms/step - loss: 23.0459 - mae: 23.5406 - lr: 2.5119e-07\nEpoch 30/100\n8/8 [==============================] - 0s 33ms/step - loss: 21.8057 - mae: 22.3000 - lr: 2.8184e-07\nEpoch 31/100\n8/8 [==============================] - 0s 32ms/step - loss: 20.5264 - mae: 21.0202 - lr: 3.1623e-07\nEpoch 32/100\n8/8 [==============================] - 0s 32ms/step - loss: 19.2341 - mae: 19.7269 - lr: 3.5481e-07\nEpoch 33/100\n8/8 [==============================] - 0s 33ms/step - loss: 17.9646 - mae: 18.4568 - lr: 3.9811e-07\nEpoch 34/100\n8/8 [==============================] - 0s 36ms/step - loss: 16.9362 - mae: 17.4279 - lr: 4.4668e-07\nEpoch 35/100\n8/8 [==============================] - 0s 34ms/step - loss: 16.2657 - mae: 16.7576 - lr: 5.0119e-07\nEpoch 36/100\n8/8 [==============================] - 0s 32ms/step - loss: 15.6281 - mae: 16.1192 - lr: 5.6234e-07\nEpoch 37/100\n8/8 [==============================] - 0s 33ms/step - loss: 15.0540 - mae: 15.5447 - lr: 6.3096e-07\nEpoch 38/100\n8/8 [==============================] - 0s 33ms/step - loss: 14.5219 - mae: 15.0126 - lr: 7.0795e-07\nEpoch 39/100\n8/8 [==============================] - 0s 33ms/step - loss: 13.9836 - mae: 14.4747 - lr: 7.9433e-07\nEpoch 40/100\n8/8 [==============================] - 0s 34ms/step - loss: 13.4414 - mae: 13.9318 - lr: 8.9125e-07\nEpoch 41/100\n8/8 [==============================] - 0s 32ms/step - loss: 12.8781 - mae: 13.3677 - lr: 1.0000e-06\nEpoch 42/100\n8/8 [==============================] - 0s 32ms/step - loss: 12.2855 - mae: 12.7749 - lr: 1.1220e-06\nEpoch 43/100\n8/8 [==============================] - 0s 32ms/step - loss: 11.6690 - mae: 12.1582 - lr: 1.2589e-06\nEpoch 44/100\n8/8 [==============================] - 0s 32ms/step - loss: 11.0181 - mae: 11.5063 - lr: 1.4125e-06\nEpoch 45/100\n8/8 [==============================] - 0s 32ms/step - loss: 10.3387 - mae: 10.8260 - lr: 1.5849e-06\nEpoch 46/100\n8/8 [==============================] - 0s 33ms/step - loss: 9.6894 - mae: 10.1757 - lr: 1.7783e-06\nEpoch 47/100\n8/8 [==============================] - 0s 34ms/step - loss: 9.0956 - mae: 9.5809 - lr: 1.9953e-06\nEpoch 48/100\n8/8 [==============================] - 0s 33ms/step - loss: 8.5688 - mae: 9.0536 - lr: 2.2387e-06\nEpoch 49/100\n8/8 [==============================] - 0s 32ms/step - loss: 8.1153 - mae: 8.5996 - lr: 2.5119e-06\nEpoch 50/100\n8/8 [==============================] - 0s 34ms/step - loss: 7.7040 - mae: 8.1884 - lr: 2.8184e-06\nEpoch 51/100\n8/8 [==============================] - 0s 33ms/step - loss: 7.3262 - mae: 7.8096 - lr: 3.1623e-06\nEpoch 52/100\n8/8 [==============================] - 0s 32ms/step - loss: 6.9725 - mae: 7.4548 - lr: 3.5481e-06\nEpoch 53/100\n8/8 [==============================] - 0s 33ms/step - loss: 6.6218 - mae: 7.1033 - lr: 3.9811e-06\nEpoch 54/100\n8/8 [==============================] - 0s 33ms/step - loss: 6.2898 - mae: 6.7712 - lr: 4.4668e-06\nEpoch 55/100\n8/8 [==============================] - 0s 32ms/step - loss: 5.9775 - mae: 6.4581 - lr: 5.0119e-06\nEpoch 56/100\n8/8 [==============================] - 0s 32ms/step - loss: 5.7096 - mae: 6.1898 - lr: 5.6234e-06\nEpoch 57/100\n8/8 [==============================] - 0s 33ms/step - loss: 5.4204 - mae: 5.8992 - lr: 6.3096e-06\nEpoch 58/100\n8/8 [==============================] - 0s 33ms/step - loss: 5.2126 - mae: 5.6904 - lr: 7.0795e-06\nEpoch 59/100\n8/8 [==============================] - 0s 33ms/step - loss: 4.9788 - mae: 5.4558 - lr: 7.9433e-06\nEpoch 60/100\n8/8 [==============================] - 0s 33ms/step - loss: 4.8674 - mae: 5.3438 - lr: 8.9125e-06\nEpoch 61/100\n8/8 [==============================] - 0s 32ms/step - loss: 4.7943 - mae: 5.2711 - lr: 1.0000e-05\nEpoch 62/100\n8/8 [==============================] - 0s 32ms/step - loss: 4.6056 - mae: 5.0814 - lr: 1.1220e-05\nEpoch 63/100\n8/8 [==============================] - 0s 33ms/step - loss: 4.7894 - mae: 5.2667 - lr: 1.2589e-05\nEpoch 64/100\n8/8 [==============================] - 0s 33ms/step - loss: 5.3932 - mae: 5.8745 - lr: 1.4125e-05\nEpoch 65/100\n8/8 [==============================] - 0s 33ms/step - loss: 5.8706 - mae: 6.3540 - lr: 1.5849e-05\nEpoch 66/100\n8/8 [==============================] - 0s 33ms/step - loss: 5.9400 - mae: 6.4234 - lr: 1.7783e-05\nEpoch 67/100\n8/8 [==============================] - 0s 33ms/step - loss: 4.9884 - mae: 5.4690 - lr: 1.9953e-05\nEpoch 68/100\n8/8 [==============================] - 0s 32ms/step - loss: 4.9481 - mae: 5.4273 - lr: 2.2387e-05\nEpoch 69/100\n8/8 [==============================] - 0s 33ms/step - loss: 5.5174 - mae: 5.9999 - lr: 2.5119e-05\nEpoch 70/100\n8/8 [==============================] - 0s 33ms/step - loss: 6.2505 - mae: 6.7357 - lr: 2.8184e-05\nEpoch 71/100\n8/8 [==============================] - 0s 34ms/step - loss: 5.7693 - mae: 6.2527 - lr: 3.1623e-05\nEpoch 72/100\n8/8 [==============================] - 0s 33ms/step - loss: 6.7677 - mae: 7.2529 - lr: 3.5481e-05\nEpoch 73/100\n8/8 [==============================] - 0s 31ms/step - loss: 7.3180 - mae: 7.8042 - lr: 3.9811e-05\nEpoch 74/100\n8/8 [==============================] - 0s 32ms/step - loss: 7.7811 - mae: 8.2695 - lr: 4.4668e-05\nEpoch 75/100\n8/8 [==============================] - 0s 32ms/step - loss: 7.6519 - mae: 8.1401 - lr: 5.0119e-05\nEpoch 76/100\n8/8 [==============================] - 0s 33ms/step - loss: 8.5045 - mae: 8.9957 - lr: 5.6234e-05\nEpoch 77/100\n8/8 [==============================] - 0s 32ms/step - loss: 9.3928 - mae: 9.8854 - lr: 6.3096e-05\nEpoch 78/100\n8/8 [==============================] - 0s 33ms/step - loss: 17.8352 - mae: 18.3274 - lr: 7.0795e-05\nEpoch 79/100\n8/8 [==============================] - 0s 33ms/step - loss: 32.4062 - mae: 32.9038 - lr: 7.9433e-05\nEpoch 80/100\n8/8 [==============================] - 0s 32ms/step - loss: 12.8659 - mae: 13.3580 - lr: 8.9125e-05\nEpoch 81/100\n8/8 [==============================] - 0s 33ms/step - loss: 9.9744 - mae: 10.4638 - lr: 1.0000e-04\nEpoch 82/100\n8/8 [==============================] - 0s 33ms/step - loss: 11.3594 - mae: 11.8533 - lr: 1.1220e-04\nEpoch 83/100\n8/8 [==============================] - 0s 33ms/step - loss: 10.1744 - mae: 10.6637 - lr: 1.2589e-04\nEpoch 84/100\n8/8 [==============================] - 0s 34ms/step - loss: 35.6556 - mae: 36.1541 - lr: 1.4125e-04\nEpoch 85/100\n8/8 [==============================] - 0s 37ms/step - loss: 23.5352 - mae: 24.0322 - lr: 1.5849e-04\nEpoch 86/100\n8/8 [==============================] - 1s 36ms/step - loss: 16.0095 - mae: 16.5044 - lr: 1.7783e-04\nEpoch 87/100\n8/8 [==============================] - 0s 38ms/step - loss: 14.1260 - mae: 14.6192 - lr: 1.9953e-04\nEpoch 88/100\n8/8 [==============================] - 1s 37ms/step - loss: 12.9790 - mae: 13.4720 - lr: 2.2387e-04\nEpoch 89/100\n8/8 [==============================] - 0s 38ms/step - loss: 15.1838 - mae: 15.6791 - lr: 2.5119e-04\nEpoch 90/100\n8/8 [==============================] - 1s 38ms/step - loss: 10.3792 - mae: 10.8697 - lr: 2.8184e-04\nEpoch 91/100\n8/8 [==============================] - 0s 33ms/step - loss: 9.4659 - mae: 9.9564 - lr: 3.1623e-04\nEpoch 92/100\n8/8 [==============================] - 0s 32ms/step - loss: 9.9145 - mae: 10.4040 - lr: 3.5481e-04\nEpoch 93/100\n8/8 [==============================] - 0s 33ms/step - loss: 20.9796 - mae: 21.4755 - lr: 3.9811e-04\nEpoch 94/100\n8/8 [==============================] - 0s 32ms/step - loss: 15.7421 - mae: 16.2352 - lr: 4.4668e-04\nEpoch 95/100\n8/8 [==============================] - 0s 34ms/step - loss: 21.3826 - mae: 21.8791 - lr: 5.0119e-04\nEpoch 96/100\n8/8 [==============================] - 0s 32ms/step - loss: 17.6574 - mae: 18.1528 - lr: 5.6234e-04\nEpoch 97/100\n8/8 [==============================] - 0s 32ms/step - loss: 19.3204 - mae: 19.8147 - lr: 6.3096e-04\nEpoch 98/100\n8/8 [==============================] - 0s 33ms/step - loss: 19.1665 - mae: 19.6608 - lr: 7.0795e-04\nEpoch 99/100\n8/8 [==============================] - 0s 34ms/step - loss: 18.3976 - mae: 18.8912 - lr: 7.9433e-04\nEpoch 100/100\n8/8 [==============================] - 0s 32ms/step - loss: 23.3824 - mae: 23.8778 - lr: 8.9125e-04\n" ], [ "plt.semilogx(history.history[\"lr\"], history.history[\"loss\"])\nplt.axis([1e-8, 1e-4, 0, 30])", "_____no_output_____" ], [ "tf.keras.backend.clear_session()\ntf.random.set_seed(51)\nnp.random.seed(51)\n#batch_size = 16\ndataset = windowed_dataset(x_train, window_size, batch_size, shuffle_buffer_size)\n\nmodel = tf.keras.models.Sequential([\n tf.keras.layers.Conv1D(filters=32, kernel_size=3,\n strides=1, padding=\"causal\",\n activation=\"relu\",\n input_shape=[None, 1]),\n tf.keras.layers.LSTM(32, return_sequences=True),\n tf.keras.layers.LSTM(32, return_sequences=True),\n tf.keras.layers.Dense(1),\n tf.keras.layers.Lambda(lambda x: x * 200)\n])\n\noptimizer = tf.keras.optimizers.SGD(learning_rate=1e-5, momentum=0.9)\nmodel.compile(loss=tf.keras.losses.Huber(),\n optimizer=optimizer,\n metrics=[\"mae\"])\nhistory = model.fit(dataset,epochs=500)", "Epoch 1/500\n31/31 [==============================] - 4s 18ms/step - loss: 20.0714 - mae: 20.5646\nEpoch 2/500\n31/31 [==============================] - 1s 16ms/step - loss: 7.9198 - mae: 8.4048\nEpoch 3/500\n31/31 [==============================] - 1s 16ms/step - loss: 6.6716 - mae: 7.1533\nEpoch 4/500\n31/31 [==============================] - 1s 17ms/step - loss: 6.1672 - mae: 6.6477\nEpoch 5/500\n31/31 [==============================] - 1s 16ms/step - loss: 5.7151 - mae: 6.1946\nEpoch 6/500\n31/31 [==============================] - 1s 16ms/step - loss: 5.6396 - mae: 6.1195\nEpoch 7/500\n31/31 [==============================] - 1s 16ms/step - loss: 5.4245 - mae: 5.9043\nEpoch 8/500\n31/31 [==============================] - 1s 17ms/step - loss: 5.3815 - mae: 5.8605\nEpoch 9/500\n31/31 [==============================] - 1s 17ms/step - loss: 5.3289 - mae: 5.8083\nEpoch 10/500\n31/31 [==============================] - 1s 16ms/step - loss: 5.2220 - mae: 5.7013\nEpoch 11/500\n31/31 [==============================] - 1s 17ms/step - loss: 5.1164 - mae: 5.5949\nEpoch 12/500\n31/31 [==============================] - 1s 17ms/step - loss: 5.0854 - mae: 5.5640\nEpoch 13/500\n31/31 [==============================] - 1s 17ms/step - loss: 5.0866 - mae: 5.5649\nEpoch 14/500\n31/31 [==============================] - 1s 17ms/step - loss: 5.1031 - mae: 5.5815\nEpoch 15/500\n31/31 [==============================] - 1s 17ms/step - loss: 5.0004 - mae: 5.4785\nEpoch 16/500\n31/31 [==============================] - 1s 18ms/step - loss: 5.1024 - mae: 5.5813\nEpoch 17/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.8976 - mae: 5.3766\nEpoch 18/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.8884 - mae: 5.3659\nEpoch 19/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.9353 - mae: 5.4138\nEpoch 20/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.8289 - mae: 5.3068\nEpoch 21/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.7756 - mae: 5.2546\nEpoch 22/500\n31/31 [==============================] - 1s 16ms/step - loss: 4.8007 - mae: 5.2782\nEpoch 23/500\n31/31 [==============================] - 1s 16ms/step - loss: 4.7320 - mae: 5.2094\nEpoch 24/500\n31/31 [==============================] - 1s 16ms/step - loss: 4.7392 - mae: 5.2168\nEpoch 25/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.6977 - mae: 5.1754\nEpoch 26/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.6399 - mae: 5.1180\nEpoch 27/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.6403 - mae: 5.1176\nEpoch 28/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.6991 - mae: 5.1765\nEpoch 29/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.6425 - mae: 5.1203\nEpoch 30/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.6097 - mae: 5.0867\nEpoch 31/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.5860 - mae: 5.0638\nEpoch 32/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.6052 - mae: 5.0816\nEpoch 33/500\n31/31 [==============================] - 1s 16ms/step - loss: 4.5302 - mae: 5.0077\nEpoch 34/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.5866 - mae: 5.0633\nEpoch 35/500\n31/31 [==============================] - 1s 16ms/step - loss: 4.5268 - mae: 5.0038\nEpoch 36/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.5295 - mae: 5.0062\nEpoch 37/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.5230 - mae: 5.0002\nEpoch 38/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.5310 - mae: 5.0076\nEpoch 39/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.5192 - mae: 4.9959\nEpoch 40/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.4530 - mae: 4.9293\nEpoch 41/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.5626 - mae: 5.0395\nEpoch 42/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.4388 - mae: 4.9146\nEpoch 43/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.4231 - mae: 4.8997\nEpoch 44/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.4870 - mae: 4.9628\nEpoch 45/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.5363 - mae: 5.0123\nEpoch 46/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.4089 - mae: 4.8853\nEpoch 47/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.4374 - mae: 4.9137\nEpoch 48/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.4259 - mae: 4.9016\nEpoch 49/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.4815 - mae: 4.9576\nEpoch 50/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.4074 - mae: 4.8841\nEpoch 51/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.4222 - mae: 4.8972\nEpoch 52/500\n31/31 [==============================] - 1s 16ms/step - loss: 4.6779 - mae: 5.1554\nEpoch 53/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.4200 - mae: 4.8956\nEpoch 54/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3926 - mae: 4.8681\nEpoch 55/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.5861 - mae: 5.0627\nEpoch 56/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3860 - mae: 4.8612\nEpoch 57/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.4013 - mae: 4.8770\nEpoch 58/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3656 - mae: 4.8413\nEpoch 59/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.3336 - mae: 4.8084\nEpoch 60/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3968 - mae: 4.8728\nEpoch 61/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3750 - mae: 4.8510\nEpoch 62/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3245 - mae: 4.7999\nEpoch 63/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3585 - mae: 4.8339\nEpoch 64/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3428 - mae: 4.8177\nEpoch 65/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3970 - mae: 4.8722\nEpoch 66/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3287 - mae: 4.8041\nEpoch 67/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.4945 - mae: 4.9710\nEpoch 68/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.5074 - mae: 4.9848\nEpoch 69/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2903 - mae: 4.7653\nEpoch 70/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2879 - mae: 4.7627\nEpoch 71/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3093 - mae: 4.7842\nEpoch 72/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3740 - mae: 4.8492\nEpoch 73/500\n31/31 [==============================] - 1s 16ms/step - loss: 4.3155 - mae: 4.7904\nEpoch 74/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2954 - mae: 4.7697\nEpoch 75/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2535 - mae: 4.7282\nEpoch 76/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2578 - mae: 4.7324\nEpoch 77/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2641 - mae: 4.7391\nEpoch 78/500\n31/31 [==============================] - 1s 16ms/step - loss: 4.3166 - mae: 4.7919\nEpoch 79/500\n31/31 [==============================] - 1s 16ms/step - loss: 4.3015 - mae: 4.7761\nEpoch 80/500\n31/31 [==============================] - 1s 16ms/step - loss: 4.3159 - mae: 4.7911\nEpoch 81/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2583 - mae: 4.7325\nEpoch 82/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3669 - mae: 4.8429\nEpoch 83/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3183 - mae: 4.7933\nEpoch 84/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.4260 - mae: 4.9020\nEpoch 85/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3776 - mae: 4.8540\nEpoch 86/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2718 - mae: 4.7465\nEpoch 87/500\n31/31 [==============================] - 1s 16ms/step - loss: 4.2969 - mae: 4.7725\nEpoch 88/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2208 - mae: 4.6950\nEpoch 89/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2229 - mae: 4.6976\nEpoch 90/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2355 - mae: 4.7100\nEpoch 91/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2440 - mae: 4.7182\nEpoch 92/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2490 - mae: 4.7241\nEpoch 93/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2247 - mae: 4.6989\nEpoch 94/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3162 - mae: 4.7904\nEpoch 95/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2872 - mae: 4.7617\nEpoch 96/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2412 - mae: 4.7167\nEpoch 97/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2377 - mae: 4.7118\nEpoch 98/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3287 - mae: 4.8046\nEpoch 99/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2687 - mae: 4.7435\nEpoch 100/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1908 - mae: 4.6639\nEpoch 101/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2381 - mae: 4.7121\nEpoch 102/500\n31/31 [==============================] - 1s 16ms/step - loss: 4.2119 - mae: 4.6860\nEpoch 103/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1838 - mae: 4.6583\nEpoch 104/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1888 - mae: 4.6623\nEpoch 105/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1775 - mae: 4.6511\nEpoch 106/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2577 - mae: 4.7324\nEpoch 107/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2734 - mae: 4.7483\nEpoch 108/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2263 - mae: 4.7003\nEpoch 109/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2019 - mae: 4.6761\nEpoch 110/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2340 - mae: 4.7081\nEpoch 111/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.2067 - mae: 4.6807\nEpoch 112/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2053 - mae: 4.6794\nEpoch 113/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1835 - mae: 4.6578\nEpoch 114/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2368 - mae: 4.7113\nEpoch 115/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2423 - mae: 4.7172\nEpoch 116/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1502 - mae: 4.6236\nEpoch 117/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1814 - mae: 4.6548\nEpoch 118/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.1873 - mae: 4.6612\nEpoch 119/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2185 - mae: 4.6927\nEpoch 120/500\n31/31 [==============================] - 1s 16ms/step - loss: 4.2738 - mae: 4.7481\nEpoch 121/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1680 - mae: 4.6416\nEpoch 122/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2055 - mae: 4.6790\nEpoch 123/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1662 - mae: 4.6390\nEpoch 124/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1390 - mae: 4.6121\nEpoch 125/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1783 - mae: 4.6522\nEpoch 126/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1774 - mae: 4.6508\nEpoch 127/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1499 - mae: 4.6232\nEpoch 128/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2690 - mae: 4.7433\nEpoch 129/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2381 - mae: 4.7130\nEpoch 130/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2117 - mae: 4.6856\nEpoch 131/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2289 - mae: 4.7037\nEpoch 132/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.3141 - mae: 4.7891\nEpoch 133/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3430 - mae: 4.8187\nEpoch 134/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1583 - mae: 4.6326\nEpoch 135/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1595 - mae: 4.6333\nEpoch 136/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1510 - mae: 4.6245\nEpoch 137/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1462 - mae: 4.6205\nEpoch 138/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2075 - mae: 4.6820\nEpoch 139/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2070 - mae: 4.6812\nEpoch 140/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1926 - mae: 4.6666\nEpoch 141/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1382 - mae: 4.6122\nEpoch 142/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1264 - mae: 4.6000\nEpoch 143/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1489 - mae: 4.6213\nEpoch 144/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2078 - mae: 4.6816\nEpoch 145/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1355 - mae: 4.6084\nEpoch 146/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1552 - mae: 4.6293\nEpoch 147/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1374 - mae: 4.6106\nEpoch 148/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2007 - mae: 4.6748\nEpoch 149/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1296 - mae: 4.6030\nEpoch 150/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1173 - mae: 4.5900\nEpoch 151/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1339 - mae: 4.6071\nEpoch 152/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3179 - mae: 4.7940\nEpoch 153/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.1523 - mae: 4.6258\nEpoch 154/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2748 - mae: 4.7497\nEpoch 155/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2117 - mae: 4.6868\nEpoch 156/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2009 - mae: 4.6755\nEpoch 157/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1249 - mae: 4.5978\nEpoch 158/500\n31/31 [==============================] - 1s 16ms/step - loss: 4.2036 - mae: 4.6776\nEpoch 159/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1062 - mae: 4.5794\nEpoch 160/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.1498 - mae: 4.6229\nEpoch 161/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1104 - mae: 4.5839\nEpoch 162/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1100 - mae: 4.5838\nEpoch 163/500\n31/31 [==============================] - 1s 16ms/step - loss: 4.1024 - mae: 4.5756\nEpoch 164/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1946 - mae: 4.6690\nEpoch 165/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1544 - mae: 4.6286\nEpoch 166/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1219 - mae: 4.5954\nEpoch 167/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1054 - mae: 4.5786\nEpoch 168/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1068 - mae: 4.5811\nEpoch 169/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1135 - mae: 4.5872\nEpoch 170/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2602 - mae: 4.7351\nEpoch 171/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1001 - mae: 4.5732\nEpoch 172/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1382 - mae: 4.6119\nEpoch 173/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1085 - mae: 4.5817\nEpoch 174/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1143 - mae: 4.5876\nEpoch 175/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1490 - mae: 4.6238\nEpoch 176/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1097 - mae: 4.5821\nEpoch 177/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1086 - mae: 4.5821\nEpoch 178/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1263 - mae: 4.6004\nEpoch 179/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2819 - mae: 4.7575\nEpoch 180/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1013 - mae: 4.5738\nEpoch 181/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1223 - mae: 4.5965\nEpoch 182/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1351 - mae: 4.6089\nEpoch 183/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2165 - mae: 4.6913\nEpoch 184/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0685 - mae: 4.5411\nEpoch 185/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0841 - mae: 4.5567\nEpoch 186/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1285 - mae: 4.6017\nEpoch 187/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2946 - mae: 4.7690\nEpoch 188/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1584 - mae: 4.6327\nEpoch 189/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1481 - mae: 4.6229\nEpoch 190/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1161 - mae: 4.5896\nEpoch 191/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1000 - mae: 4.5722\nEpoch 192/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1213 - mae: 4.5945\nEpoch 193/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0729 - mae: 4.5463\nEpoch 194/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1752 - mae: 4.6494\nEpoch 195/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1275 - mae: 4.6014\nEpoch 196/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.2276 - mae: 4.7023\nEpoch 197/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0873 - mae: 4.5607\nEpoch 198/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0630 - mae: 4.5360\nEpoch 199/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0887 - mae: 4.5619\nEpoch 200/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1107 - mae: 4.5841\nEpoch 201/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0687 - mae: 4.5408\nEpoch 202/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0655 - mae: 4.5383\nEpoch 203/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1284 - mae: 4.6022\nEpoch 204/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1501 - mae: 4.6246\nEpoch 205/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0926 - mae: 4.5669\nEpoch 206/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1157 - mae: 4.5900\nEpoch 207/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0555 - mae: 4.5283\nEpoch 208/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1092 - mae: 4.5827\nEpoch 209/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0578 - mae: 4.5300\nEpoch 210/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0854 - mae: 4.5591\nEpoch 211/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0821 - mae: 4.5554\nEpoch 212/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0435 - mae: 4.5159\nEpoch 213/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0618 - mae: 4.5346\nEpoch 214/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0506 - mae: 4.5235\nEpoch 215/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0641 - mae: 4.5368\nEpoch 216/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0760 - mae: 4.5496\nEpoch 217/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0867 - mae: 4.5596\nEpoch 218/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0357 - mae: 4.5074\nEpoch 219/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0535 - mae: 4.5263\nEpoch 220/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0590 - mae: 4.5319\nEpoch 221/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0376 - mae: 4.5100\nEpoch 222/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0582 - mae: 4.5313\nEpoch 223/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1867 - mae: 4.6614\nEpoch 224/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0941 - mae: 4.5676\nEpoch 225/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1245 - mae: 4.5978\nEpoch 226/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0773 - mae: 4.5496\nEpoch 227/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0314 - mae: 4.5038\nEpoch 228/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0593 - mae: 4.5323\nEpoch 229/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0608 - mae: 4.5336\nEpoch 230/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.1776 - mae: 4.6525\nEpoch 231/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.3184 - mae: 4.7943\nEpoch 232/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0747 - mae: 4.5483\nEpoch 233/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0752 - mae: 4.5488\nEpoch 234/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1668 - mae: 4.6410\nEpoch 235/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.1110 - mae: 4.5844\nEpoch 236/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0289 - mae: 4.5022\nEpoch 237/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0427 - mae: 4.5152\nEpoch 238/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.1794 - mae: 4.6534\nEpoch 239/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1095 - mae: 4.5844\nEpoch 240/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1538 - mae: 4.6275\nEpoch 241/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1868 - mae: 4.6614\nEpoch 242/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0266 - mae: 4.4989\nEpoch 243/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1063 - mae: 4.5803\nEpoch 244/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0558 - mae: 4.5283\nEpoch 245/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0461 - mae: 4.5187\nEpoch 246/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0614 - mae: 4.5347\nEpoch 247/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0499 - mae: 4.5228\nEpoch 248/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0336 - mae: 4.5061\nEpoch 249/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0433 - mae: 4.5158\nEpoch 250/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0965 - mae: 4.5703\nEpoch 251/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0898 - mae: 4.5635\nEpoch 252/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0431 - mae: 4.5160\nEpoch 253/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0376 - mae: 4.5102\nEpoch 254/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0289 - mae: 4.5017\nEpoch 255/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1242 - mae: 4.5986\nEpoch 256/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.1310 - mae: 4.6046\nEpoch 257/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1470 - mae: 4.6207\nEpoch 258/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0667 - mae: 4.5394\nEpoch 259/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0347 - mae: 4.5079\nEpoch 260/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0884 - mae: 4.5612\nEpoch 261/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1398 - mae: 4.6139\nEpoch 262/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0773 - mae: 4.5509\nEpoch 263/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0825 - mae: 4.5560\nEpoch 264/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0590 - mae: 4.5326\nEpoch 265/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0964 - mae: 4.5696\nEpoch 266/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0893 - mae: 4.5633\nEpoch 267/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1185 - mae: 4.5932\nEpoch 268/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0686 - mae: 4.5425\nEpoch 269/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0088 - mae: 4.4808\nEpoch 270/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0459 - mae: 4.5192\nEpoch 271/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0033 - mae: 4.4753\nEpoch 272/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0092 - mae: 4.4812\nEpoch 273/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0390 - mae: 4.5118\nEpoch 274/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0105 - mae: 4.4830\nEpoch 275/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0268 - mae: 4.4993\nEpoch 276/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0879 - mae: 4.5616\nEpoch 277/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0750 - mae: 4.5476\nEpoch 278/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0338 - mae: 4.5070\nEpoch 279/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0377 - mae: 4.5105\nEpoch 280/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0541 - mae: 4.5277\nEpoch 281/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0213 - mae: 4.4940\nEpoch 282/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0325 - mae: 4.5051\nEpoch 283/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1083 - mae: 4.5815\nEpoch 284/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0448 - mae: 4.5180\nEpoch 285/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0672 - mae: 4.5407\nEpoch 286/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0627 - mae: 4.5368\nEpoch 287/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0055 - mae: 4.4784\nEpoch 288/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0259 - mae: 4.4991\nEpoch 289/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0592 - mae: 4.5319\nEpoch 290/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.1093 - mae: 4.5831\nEpoch 291/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1137 - mae: 4.5886\nEpoch 292/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0494 - mae: 4.5224\nEpoch 293/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0447 - mae: 4.5173\nEpoch 294/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0563 - mae: 4.5302\nEpoch 295/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0331 - mae: 4.5060\nEpoch 296/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.1633 - mae: 4.6385\nEpoch 297/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1048 - mae: 4.5787\nEpoch 298/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0742 - mae: 4.5484\nEpoch 299/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.2062 - mae: 4.6814\nEpoch 300/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1141 - mae: 4.5880\nEpoch 301/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0195 - mae: 4.4925\nEpoch 302/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0413 - mae: 4.5158\nEpoch 303/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0142 - mae: 4.4862\nEpoch 304/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9888 - mae: 4.4607\nEpoch 305/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0065 - mae: 4.4787\nEpoch 306/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9820 - mae: 4.4540\nEpoch 307/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0182 - mae: 4.4907\nEpoch 308/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9915 - mae: 4.4643\nEpoch 309/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9878 - mae: 4.4599\nEpoch 310/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9855 - mae: 4.4579\nEpoch 311/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.1139 - mae: 4.5882\nEpoch 312/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0765 - mae: 4.5504\nEpoch 313/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0497 - mae: 4.5223\nEpoch 314/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9943 - mae: 4.4674\nEpoch 315/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9922 - mae: 4.4647\nEpoch 316/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9805 - mae: 4.4520\nEpoch 317/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0384 - mae: 4.5110\nEpoch 318/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0343 - mae: 4.5068\nEpoch 319/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9795 - mae: 4.4513\nEpoch 320/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0015 - mae: 4.4739\nEpoch 321/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0888 - mae: 4.5631\nEpoch 322/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0912 - mae: 4.5650\nEpoch 323/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0825 - mae: 4.5568\nEpoch 324/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9830 - mae: 4.4549\nEpoch 325/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9992 - mae: 4.4714\nEpoch 326/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9704 - mae: 4.4418\nEpoch 327/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9917 - mae: 4.4642\nEpoch 328/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0238 - mae: 4.4971\nEpoch 329/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0390 - mae: 4.5122\nEpoch 330/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9753 - mae: 4.4478\nEpoch 331/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9614 - mae: 4.4340\nEpoch 332/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9800 - mae: 4.4522\nEpoch 333/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9739 - mae: 4.4461\nEpoch 334/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9790 - mae: 4.4501\nEpoch 335/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0874 - mae: 4.5614\nEpoch 336/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0188 - mae: 4.4914\nEpoch 337/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0644 - mae: 4.5383\nEpoch 338/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9820 - mae: 4.4546\nEpoch 339/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9606 - mae: 4.4324\nEpoch 340/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0385 - mae: 4.5115\nEpoch 341/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0387 - mae: 4.5129\nEpoch 342/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0044 - mae: 4.4768\nEpoch 343/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9836 - mae: 4.4561\nEpoch 344/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0132 - mae: 4.4865\nEpoch 345/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0031 - mae: 4.4762\nEpoch 346/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9695 - mae: 4.4412\nEpoch 347/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9887 - mae: 4.4602\nEpoch 348/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0083 - mae: 4.4813\nEpoch 349/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0540 - mae: 4.5282\nEpoch 350/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9682 - mae: 4.4406\nEpoch 351/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9823 - mae: 4.4552\nEpoch 352/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9985 - mae: 4.4710\nEpoch 353/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9872 - mae: 4.4603\nEpoch 354/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9514 - mae: 4.4234\nEpoch 355/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9753 - mae: 4.4480\nEpoch 356/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9707 - mae: 4.4429\nEpoch 357/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0264 - mae: 4.4997\nEpoch 358/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9961 - mae: 4.4690\nEpoch 359/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9649 - mae: 4.4377\nEpoch 360/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9490 - mae: 4.4207\nEpoch 361/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9861 - mae: 4.4582\nEpoch 362/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0901 - mae: 4.5638\nEpoch 363/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0130 - mae: 4.4860\nEpoch 364/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0036 - mae: 4.4766\nEpoch 365/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0381 - mae: 4.5116\nEpoch 366/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0459 - mae: 4.5196\nEpoch 367/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9874 - mae: 4.4596\nEpoch 368/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9789 - mae: 4.4516\nEpoch 369/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0141 - mae: 4.4876\nEpoch 370/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9451 - mae: 4.4173\nEpoch 371/500\n31/31 [==============================] - 1s 17ms/step - loss: 4.0267 - mae: 4.5002\nEpoch 372/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0030 - mae: 4.4757\nEpoch 373/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9497 - mae: 4.4217\nEpoch 374/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9696 - mae: 4.4414\nEpoch 375/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9427 - mae: 4.4138\nEpoch 376/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9445 - mae: 4.4167\nEpoch 377/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9822 - mae: 4.4553\nEpoch 378/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9655 - mae: 4.4374\nEpoch 379/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9578 - mae: 4.4296\nEpoch 380/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9719 - mae: 4.4443\nEpoch 381/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9896 - mae: 4.4622\nEpoch 382/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9716 - mae: 4.4438\nEpoch 383/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9712 - mae: 4.4435\nEpoch 384/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9550 - mae: 4.4266\nEpoch 385/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9527 - mae: 4.4244\nEpoch 386/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9366 - mae: 4.4082\nEpoch 387/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0084 - mae: 4.4814\nEpoch 388/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9709 - mae: 4.4437\nEpoch 389/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9856 - mae: 4.4577\nEpoch 390/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0217 - mae: 4.4947\nEpoch 391/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9714 - mae: 4.4434\nEpoch 392/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9639 - mae: 4.4360\nEpoch 393/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9747 - mae: 4.4473\nEpoch 394/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0125 - mae: 4.4852\nEpoch 395/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9535 - mae: 4.4249\nEpoch 396/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0907 - mae: 4.5653\nEpoch 397/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9837 - mae: 4.4563\nEpoch 398/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.1708 - mae: 4.6452\nEpoch 399/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9770 - mae: 4.4489\nEpoch 400/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0197 - mae: 4.4927\nEpoch 401/500\n31/31 [==============================] - 1s 19ms/step - loss: 3.9949 - mae: 4.4676\nEpoch 402/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9822 - mae: 4.4556\nEpoch 403/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9819 - mae: 4.4546\nEpoch 404/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9565 - mae: 4.4284\nEpoch 405/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0755 - mae: 4.5493\nEpoch 406/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9904 - mae: 4.4642\nEpoch 407/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9435 - mae: 4.4154\nEpoch 408/500\n31/31 [==============================] - 1s 17ms/step - loss: 3.9561 - mae: 4.4280\nEpoch 409/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9673 - mae: 4.4394\nEpoch 410/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0533 - mae: 4.5271\nEpoch 411/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9292 - mae: 4.4007\nEpoch 412/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9370 - mae: 4.4094\nEpoch 413/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9438 - mae: 4.4154\nEpoch 414/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9687 - mae: 4.4408\nEpoch 415/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9450 - mae: 4.4168\nEpoch 416/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9394 - mae: 4.4110\nEpoch 417/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0320 - mae: 4.5054\nEpoch 418/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0215 - mae: 4.4944\nEpoch 419/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9382 - mae: 4.4105\nEpoch 420/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9463 - mae: 4.4180\nEpoch 421/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9267 - mae: 4.3977\nEpoch 422/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9311 - mae: 4.4027\nEpoch 423/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9270 - mae: 4.3980\nEpoch 424/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9330 - mae: 4.4049\nEpoch 425/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9569 - mae: 4.4293\nEpoch 426/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9457 - mae: 4.4175\nEpoch 427/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9746 - mae: 4.4471\nEpoch 428/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9477 - mae: 4.4194\nEpoch 429/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9296 - mae: 4.4011\nEpoch 430/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0002 - mae: 4.4728\nEpoch 431/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9383 - mae: 4.4103\nEpoch 432/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9333 - mae: 4.4049\nEpoch 433/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9395 - mae: 4.4107\nEpoch 434/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9541 - mae: 4.4256\nEpoch 435/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0317 - mae: 4.5046\nEpoch 436/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9338 - mae: 4.4053\nEpoch 437/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0508 - mae: 4.5254\nEpoch 438/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9907 - mae: 4.4628\nEpoch 439/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9787 - mae: 4.4511\nEpoch 440/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0298 - mae: 4.5026\nEpoch 441/500\n31/31 [==============================] - 1s 19ms/step - loss: 3.9849 - mae: 4.4575\nEpoch 442/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9680 - mae: 4.4410\nEpoch 443/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9610 - mae: 4.4335\nEpoch 444/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9891 - mae: 4.4620\nEpoch 445/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9160 - mae: 4.3870\nEpoch 446/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9340 - mae: 4.4056\nEpoch 447/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9336 - mae: 4.4056\nEpoch 448/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9826 - mae: 4.4552\nEpoch 449/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0334 - mae: 4.5063\nEpoch 450/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9566 - mae: 4.4288\nEpoch 451/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9345 - mae: 4.4059\nEpoch 452/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9131 - mae: 4.3844\nEpoch 453/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9263 - mae: 4.3976\nEpoch 454/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0225 - mae: 4.4957\nEpoch 455/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9243 - mae: 4.3954\nEpoch 456/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9108 - mae: 4.3814\nEpoch 457/500\n31/31 [==============================] - 1s 19ms/step - loss: 3.9141 - mae: 4.3853\nEpoch 458/500\n31/31 [==============================] - 1s 19ms/step - loss: 3.9405 - mae: 4.4122\nEpoch 459/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9756 - mae: 4.4480\nEpoch 460/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9513 - mae: 4.4220\nEpoch 461/500\n31/31 [==============================] - 1s 19ms/step - loss: 3.9710 - mae: 4.4436\nEpoch 462/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9282 - mae: 4.3998\nEpoch 463/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9297 - mae: 4.4012\nEpoch 464/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9267 - mae: 4.3977\nEpoch 465/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9565 - mae: 4.4288\nEpoch 466/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9587 - mae: 4.4304\nEpoch 467/500\n31/31 [==============================] - 1s 19ms/step - loss: 3.9408 - mae: 4.4131\nEpoch 468/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9225 - mae: 4.3939\nEpoch 469/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9172 - mae: 4.3893\nEpoch 470/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9596 - mae: 4.4321\nEpoch 471/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9748 - mae: 4.4470\nEpoch 472/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9533 - mae: 4.4251\nEpoch 473/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9126 - mae: 4.3841\nEpoch 474/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9419 - mae: 4.4140\nEpoch 475/500\n31/31 [==============================] - 1s 19ms/step - loss: 3.9511 - mae: 4.4231\nEpoch 476/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9539 - mae: 4.4251\nEpoch 477/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9757 - mae: 4.4487\nEpoch 478/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9368 - mae: 4.4084\nEpoch 479/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9969 - mae: 4.4695\nEpoch 480/500\n31/31 [==============================] - 1s 19ms/step - loss: 3.9865 - mae: 4.4588\nEpoch 481/500\n31/31 [==============================] - 1s 19ms/step - loss: 3.9963 - mae: 4.4687\nEpoch 482/500\n31/31 [==============================] - 1s 19ms/step - loss: 3.9440 - mae: 4.4150\nEpoch 483/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9368 - mae: 4.4078\nEpoch 484/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9241 - mae: 4.3952\nEpoch 485/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9485 - mae: 4.4211\nEpoch 486/500\n31/31 [==============================] - 1s 19ms/step - loss: 4.0004 - mae: 4.4729\nEpoch 487/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9479 - mae: 4.4194\nEpoch 488/500\n31/31 [==============================] - 1s 19ms/step - loss: 3.9322 - mae: 4.4038\nEpoch 489/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9969 - mae: 4.4701\nEpoch 490/500\n31/31 [==============================] - 1s 19ms/step - loss: 3.9317 - mae: 4.4028\nEpoch 491/500\n31/31 [==============================] - 1s 19ms/step - loss: 3.9075 - mae: 4.3781\nEpoch 492/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9379 - mae: 4.4093\nEpoch 493/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9110 - mae: 4.3827\nEpoch 494/500\n31/31 [==============================] - 1s 18ms/step - loss: 4.0001 - mae: 4.4732\nEpoch 495/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9891 - mae: 4.4608\nEpoch 496/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9279 - mae: 4.3992\nEpoch 497/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9355 - mae: 4.4072\nEpoch 498/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9500 - mae: 4.4215\nEpoch 499/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9207 - mae: 4.3924\nEpoch 500/500\n31/31 [==============================] - 1s 18ms/step - loss: 3.9123 - mae: 4.3831\n" ], [ "rnn_forecast = model_forecast(model, series[..., np.newaxis], window_size)\nrnn_forecast = rnn_forecast[split_time - window_size:-1, -1, 0]", "_____no_output_____" ], [ "plt.figure(figsize=(10, 6))\nplot_series(time_valid, x_valid)\nplot_series(time_valid, rnn_forecast)", "_____no_output_____" ], [ "tf.keras.metrics.mean_absolute_error(x_valid, rnn_forecast).numpy()", "_____no_output_____" ], [ "import matplotlib.image as mpimg\nimport matplotlib.pyplot as plt\n\n#-----------------------------------------------------------\n# Retrieve a list of list results on training and test data\n# sets for each training epoch\n#-----------------------------------------------------------\nmae=history.history['mae']\nloss=history.history['loss']\n\nepochs=range(len(loss)) # Get number of epochs\n\n#------------------------------------------------\n# Plot MAE and Loss\n#------------------------------------------------\nplt.plot(epochs, mae, 'r')\nplt.plot(epochs, loss, 'b')\nplt.title('MAE and Loss')\nplt.xlabel(\"Epochs\")\nplt.ylabel(\"Accuracy\")\nplt.legend([\"MAE\", \"Loss\"])\n\nplt.figure()\n\nepochs_zoom = epochs[200:]\nmae_zoom = mae[200:]\nloss_zoom = loss[200:]\n\n#------------------------------------------------\n# Plot Zoomed MAE and Loss\n#------------------------------------------------\nplt.plot(epochs_zoom, mae_zoom, 'r')\nplt.plot(epochs_zoom, loss_zoom, 'b')\nplt.title('MAE and Loss')\nplt.xlabel(\"Epochs\")\nplt.ylabel(\"Accuracy\")\nplt.legend([\"MAE\", \"Loss\"])\n\nplt.figure()", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# MIDAS Examples\n\nIf you're reading this you probably already know that MIDAS stands for Mixed Data Sampling, and it is a technique for creating time-series forecast models that allows you to mix series of different frequencies (ie, you can use monthly data as predictors for a quarterly series, or daily data as predictors for a monthly series, etc.). The general approach has been described in a series of papers by Ghysels, Santa-Clara, Valkanov and others.\n\nThis notebook attempts to recreate some of the examples from the paper [_Forecasting with Mixed Frequencies_](https://research.stlouisfed.org/publications/review/2010/11/01/forecasting-with-mixed-frequencies/) by Michelle T. Armesto, Kristie M. Engemann, and Michael T. Owyang.", "_____no_output_____" ] ], [ [ "%matplotlib inline\n\nimport datetime\nimport numpy as np\nimport pandas as pd\n\nfrom midas.mix import mix_freq\nfrom midas.adl import estimate, forecast, midas_adl, rmse", "_____no_output_____" ] ], [ [ "# MIDAS ADL\n\nThis package currently implements the MIDAS ADL (autoregressive distributed lag) method. We'll start with an example using quarterly GDP and monthly payroll data. We'll then show the basic steps in setting up and fitting this type of model, although in practice you'll probably used the top-level __midas_adl__ function to do forecasts.\n\nTODO: MIDAS equation and discussion\n\n\n# Example 1: GDP vs Non-Farm Payroll", "_____no_output_____" ] ], [ [ "gdp = pd.read_csv('../tests/data/gdp.csv', parse_dates=['DATE'], index_col='DATE')\npay = pd.read_csv('../tests/data/pay.csv', parse_dates=['DATE'], index_col='DATE')\ngdp.tail()", "_____no_output_____" ], [ "pay.tail()", "_____no_output_____" ] ], [ [ "## Figure 1\n\nThis is a variation of Figure 1 from the paper comparing year-over-year growth of GDP and employment.", "_____no_output_____" ] ], [ [ "gdp_yoy = ((1. + (np.log(gdp.GDP) - np.log(gdp.GDP.shift(3)))) ** 4) - 1.\nemp_yoy = ((1. + (np.log(pay.PAY) - np.log(pay.PAY.shift(1)))) ** 12) - 1.\ndf = pd.concat([gdp_yoy, emp_yoy], axis=1)\ndf.columns = ['gdp_yoy', 'emp_yoy']\ndf[['gdp_yoy','emp_yoy']].loc['1980-1-1':].plot(figsize=(15,4), style=['o','-'])", "_____no_output_____" ] ], [ [ "## Mixing Frequencies\n\nThe first step is to do the actual frequency mixing. In this case we're mixing monthly data (employment) with quarterly data (GDP). This may sometimes be useful to do directly, but again you'll probably used __midas_adl__ to do forecasting.", "_____no_output_____" ] ], [ [ "gdp['gdp_growth'] = (np.log(gdp.GDP) - np.log(gdp.GDP.shift(1))) * 100.\npay['emp_growth'] = (np.log(pay.PAY) - np.log(pay.PAY.shift(1))) * 100.\n\ny, yl, x, yf, ylf, xf = mix_freq(gdp.gdp_growth, pay.emp_growth, \"3m\", 1, 3,\n start_date=datetime.datetime(1985,1,1),\n end_date=datetime.datetime(2009,1,1))\nx.head()", "_____no_output_____" ] ], [ [ "The arguments here are as follows:\n- First, the dependent (low frequency) and independent (high-frequency) data are given as Pandas series, and they are assumed to be indexed by date.\n- xlag The number of lags for the high-frequency variable\n- ylag The number of lags for the low-frequency variable (the autoregressive part)\n- horizon: How much the high-frequency data is lagged before frequency mixing\n- start_date, end_date: The start and end date over which the model is fitted. If these are outside the range of the low-frequency data, they will be adjusted\n\n\nThe _horizon_ argument is a little tricky (the argument name was retained from the MatLab version). This is used both the align the data and to do _nowcasting_ (more on that later). For example, if it's September 2017 then the latest GDP data from FRED will be for Q2 and this will be dated 2017-04-01. The latest monthly data from non-farm payroll will be for August, which will be dated 2017-08-01. If we aligned just on dates, the payroll data for April (04-01), March (03-01), and February(02-01) would be aligned with Q2 (since xlag = \"3m\"), but what we want is June, May, and April, so here the horizon argument is 3 indicating that the high-frequency data should be lagged three months before being mixed with the quarterly data.", "_____no_output_____" ], [ "### Fitting the Model\n\nBecause of the form of the MIDAS model, fitting the model requires using non-linear least squares. For now, if you call the __estimate__ function directly, you'll get back a results of type scipy.optimize.optimize.OptimizeResult\n", "_____no_output_____" ] ], [ [ "res = estimate(y, yl, x, poly='beta')\nres.x", "_____no_output_____" ] ], [ [ "You can also call __forecast__ directly. This will use the optimization results returned from __eatimate__ to produce a forecast for every date in the index of the forecast inputs (here xf and ylf):", "_____no_output_____" ] ], [ [ "fc = forecast(xf, ylf, res, poly='beta')\nforecast_df = fc.join(yf)\nforecast_df['gap'] = forecast_df.yfh - forecast_df.gdp_growth\nforecast_df", "_____no_output_____" ], [ "gdp.join(fc)[['gdp_growth','yfh']].loc['2005-01-01':].plot(style=['-o','-+'], figsize=(12, 4))", "_____no_output_____" ] ], [ [ "### Comparison against univariate ARIMA model", "_____no_output_____" ] ], [ [ "import statsmodels.tsa.api as sm\n\nm = sm.AR(gdp['1975-01-01':'2011-01-01'].gdp_growth,)\nr = m.fit(maxlag=1)\nr.params\nfc_ar = r.predict(start='2005-01-01')\nfc_ar.name = 'xx'", "_____no_output_____" ], [ "df_p = gdp.join(fc)[['gdp_growth','yfh']]\ndf_p.join(fc_ar)[['gdp_growth','yfh','xx']].loc['2005-01-01':].plot(style=['-o','-+'], figsize=(12, 4))", "_____no_output_____" ] ], [ [ "## The midas_adl function\n\nThe __midas\\_adl__ function wraps up frequency-mixing, fitting, and forecasting into one process. The default mode of forecasting is _fixed_, which means that the data between start_date and end_date will be used to fit the model, and then any data in the input beyond end_date will be used for forecasting. For example, here we're fitting from the beginning of 1985 to the end of 2008, but the gdp data extends to Q1 of 2011 so we get nine forecast points. Three monthly lags of the high-frequency data are specified along with one quarterly lag of GDP.\n", "_____no_output_____" ] ], [ [ "rmse_fc, fc = midas_adl(gdp.gdp_growth, pay.emp_growth,\n start_date=datetime.datetime(1985,1,1),\n end_date=datetime.datetime(2009,1,1),\n xlag=\"3m\",\n ylag=1,\n horizon=3)\nrmse_fc", "_____no_output_____" ] ], [ [ "You can also change the polynomial used to weight the MIDAS coefficients. The default is 'beta', but you can also specify exponential Almom weighting ('expalmon') or beta with non-zero last term ('betann')", "_____no_output_____" ] ], [ [ "rmse_fc, fc = midas_adl(gdp.gdp_growth, pay.emp_growth,\n start_date=datetime.datetime(1985,1,1),\n end_date=datetime.datetime(2009,1,1),\n xlag=\"3m\",\n ylag=1,\n horizon=3,\n poly='expalmon')\nrmse_fc", "_____no_output_____" ] ], [ [ "### Rolling and Recursive Forecasting\n\nAs mentioned above the default forecasting method is fixed where the model is fit once and then all data after end_date is used for forecasting. Two other methods are supported _rolling window_ and _recursive_. The _rolling window_ method is just what it sounds like. The start_date and end_date are used for the initial window, and then each new forecast moves that window forward by one period so that you're always doing one step ahead forecasts. Of course, to do anything useful this also assumes that the date range of the dependent data extends beyond end_date accounting for the lags implied by _horizon_. Generally, you'll get lower RMSE values here since the forecasts are always one step ahead.", "_____no_output_____" ] ], [ [ "results = {h: midas_adl(gdp.gdp_growth, pay.emp_growth,\n start_date=datetime.datetime(1985,10,1),\n end_date=datetime.datetime(2009,1,1),\n xlag=\"3m\",\n ylag=1,\n horizon=3,\n forecast_horizon=h,\n poly='beta',\n method='rolling') for h in (1, 2, 5)}\nresults[1][0]", "_____no_output_____" ] ], [ [ "The _recursive_ method is similar except that the start date does not change, so the range over which the fitting happens increases for each new forecast.", "_____no_output_____" ] ], [ [ "results = {h: midas_adl(gdp.gdp_growth, pay.emp_growth,\n start_date=datetime.datetime(1985,10,1),\n end_date=datetime.datetime(2009,1,1),\n xlag=\"3m\",\n ylag=1,\n horizon=3,\n forecast_horizon=h,\n poly='beta',\n method='recursive') for h in (1, 2, 5)}\nresults[1][0]", "_____no_output_____" ] ], [ [ "## Nowcasting\n\nPer the manual for the MatLab Matlab Toolbox Version 1.0, you can do _nowcasting_ (or MIDAS with leads) basically by adjusting the _horizon_ parameter. For example, below we change the _horizon_ paremter to 1, we're now forecasting with a one month horizon rather than a one quarter horizon:", "_____no_output_____" ] ], [ [ "rmse_fc, fc = midas_adl(gdp.gdp_growth, pay.emp_growth,\n start_date=datetime.datetime(1985,1,1),\n end_date=datetime.datetime(2009,1,1),\n xlag=\"3m\",\n ylag=1,\n horizon=1)\nrmse_fc", "_____no_output_____" ] ], [ [ "Not surprisingly the RMSE drops considerably.", "_____no_output_____" ], [ "## CPI vs. Federal Funds Rate\n\n__UNDER CONSTRUCTION: Note that these models take considerably longer to fit__", "_____no_output_____" ] ], [ [ "cpi = pd.read_csv('CPIAUCSL.csv', parse_dates=['DATE'], index_col='DATE')\nffr = pd.read_csv('DFF_2_Vintages_Starting_2009_09_28.txt', sep='\\t', parse_dates=['observation_date'],\n index_col='observation_date')", "_____no_output_____" ], [ "cpi.head()", "_____no_output_____" ], [ "ffr.head(10)", "_____no_output_____" ], [ "cpi_yoy = ((1. + (np.log(cpi.CPIAUCSL) - np.log(cpi.CPIAUCSL.shift(1)))) ** 12) - 1.\ncpi_yoy.head()\ndf = pd.concat([cpi_yoy, ffr.DFF_20090928 / 100.], axis=1)\ndf.columns = ['cpi_growth', 'dff']\ndf.loc['1980-1-1':'2010-1-1'].plot(figsize=(15,4), style=['-+','-.'])", "_____no_output_____" ], [ "cpi_growth = (np.log(cpi.CPIAUCSL) - np.log(cpi.CPIAUCSL.shift(1))) * 100.\n\ny, yl, x, yf, ylf, xf = mix_freq(cpi_growth, ffr.DFF_20090928, \"1m\", 1, 1,\n start_date=datetime.datetime(1975,10,1),\n end_date=datetime.datetime(1991,1,1))", "_____no_output_____" ], [ "x.head()", "_____no_output_____" ], [ "res = estimate(y, yl, x)", "_____no_output_____" ], [ "fc = forecast(xf, ylf, res)\nfc.join(yf).head()", "_____no_output_____" ], [ "pd.concat([cpi_growth, fc],axis=1).loc['2008-01-01':'2010-01-01'].plot(style=['-o','-+'], figsize=(12, 4))", "_____no_output_____" ], [ "\nresults = {h: midas_adl(cpi_growth, ffr.DFF_20090928,\n start_date=datetime.datetime(1975,7,1),\n end_date=datetime.datetime(1990,11,1),\n xlag=\"1m\",\n ylag=1,\n horizon=1,\n forecast_horizon=h,\n method='rolling') for h in (1, 2, 5)}\n(results[1][0], results[2][0], results[5][0])", "/Users/mikemull/Documents/Dev/midaspy/midas/weights.py:34: RuntimeWarning: overflow encountered in power\n beta_vals = u ** (self.theta1 - 1) * (1 - u) ** (self.theta2 - 1)\n/Users/mikemull/Documents/Dev/midaspy/midas/weights.py:36: RuntimeWarning: invalid value encountered in true_divide\n beta_vals = beta_vals / sum(beta_vals)\n" ], [ "results[1][1].plot(figsize=(12,4))", "_____no_output_____" ], [ "results = {h: midas_adl(cpi_growth, ffr.DFF_20090928,\n start_date=datetime.datetime(1975,10,1),\n end_date=datetime.datetime(1991,1,1),\n xlag=\"1m\",\n ylag=1,\n horizon=1,\n forecast_horizon=h,\n method='recursive') for h in (1, 2, 5)}\nresults[1][0]", "/Users/mikemull/Documents/Dev/midaspy/midas/weights.py:34: RuntimeWarning: overflow encountered in power\n beta_vals = u ** (self.theta1 - 1) * (1 - u) ** (self.theta2 - 1)\n/Users/mikemull/Documents/Dev/midaspy/midas/weights.py:36: RuntimeWarning: invalid value encountered in true_divide\n beta_vals = beta_vals / sum(beta_vals)\n" ], [ "results[1][1].plot()", "_____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", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import sys, os\nif 'google.colab' in sys.modules and not os.path.exists('.setup_complete'):\n !wget -q https://raw.githubusercontent.com/yandexdataschool/Practical_RL/spring20/setup_colab.sh -O- | bash\n\n !wget -q https://raw.githubusercontent.com/yandexdataschool/Practical_RL/coursera/grading.py -O ../grading.py\n !wget -q https://raw.githubusercontent.com/yandexdataschool/Practical_RL/coursera/week1_intro/submit.py\n\n !touch .setup_complete\n\n# This code creates a virtual display to draw game images on.\n# It will have no effect if your machine has a monitor.\nif type(os.environ.get(\"DISPLAY\")) is not str or len(os.environ.get(\"DISPLAY\")) == 0:\n !bash ../xvfb start\n os.environ['DISPLAY'] = ':1'", "Selecting previously unselected package xvfb.\n(Reading database ... 144429 files and directories currently installed.)\nPreparing to unpack .../xvfb_2%3a1.19.6-1ubuntu4.4_amd64.deb ...\nUnpacking xvfb (2:1.19.6-1ubuntu4.4) ...\nSetting up xvfb (2:1.19.6-1ubuntu4.4) ...\nProcessing triggers for man-db (2.8.3-2ubuntu0.1) ...\nStarting virtual X frame buffer: Xvfb.\n" ], [ "import numpy as np\nimport matplotlib.pyplot as plt\n%matplotlib inline", "_____no_output_____" ] ], [ [ "### OpenAI Gym\n\nWe're gonna spend several next weeks learning algorithms that solve decision processes. We are then in need of some interesting decision problems to test our algorithms.\n\nThat's where OpenAI gym comes into play. It's a python library that wraps many classical decision problems including robot control, videogames and board games.\n\nSo here's how it works:", "_____no_output_____" ] ], [ [ "import gym\n\nenv = gym.make(\"MountainCar-v0\")\nenv.reset()\n\nplt.imshow(env.render('rgb_array'))\nprint(\"Observation space:\", env.observation_space)\nprint(\"Action space:\", env.action_space)", "Observation space: Box(2,)\nAction space: Discrete(3)\n" ] ], [ [ "Note: if you're running this on your local machine, you'll see a window pop up with the image above. Don't close it, just alt-tab away.", "_____no_output_____" ], [ "### Gym interface\n\nThe three main methods of an environment are\n* __reset()__ - reset environment to initial state, _return first observation_\n* __render()__ - show current environment state (a more colorful version :) )\n* __step(a)__ - commit action __a__ and return (new observation, reward, is done, info)\n * _new observation_ - an observation right after commiting the action __a__\n * _reward_ - a number representing your reward for commiting action __a__\n * _is done_ - True if the MDP has just finished, False if still in progress\n * _info_ - some auxilary stuff about what just happened. Ignore it ~~for now~~.", "_____no_output_____" ] ], [ [ "obs0 = env.reset()\nprint(\"initial observation code:\", obs0)\n\n# Note: in MountainCar, observation is just two numbers: car position and velocity", "initial observation code: [-0.51069213 0. ]\n" ], [ "print(\"taking action 2 (right)\")\nnew_obs, reward, is_done, _ = env.step(2)\n\nprint(\"new observation code:\", new_obs)\nprint(\"reward:\", reward)\nprint(\"is game over?:\", is_done)\n\n# Note: as you can see, the car has moved to the right slightly (around 0.0005)", "taking action 2 (right)\nnew observation code: [-0.50978891 0.00090322]\nreward: -1.0\nis game over?: False\n" ] ], [ [ "### Play with it\n\nBelow is the code that drives the car to the right. However, if you simply use the default policy, the car will not reach the flag at the far right due to gravity.\n\n__Your task__ is to fix it. Find a strategy that reaches the flag. \n\nYou are not required to build any sophisticated algorithms for now, feel free to hard-code :)", "_____no_output_____" ] ], [ [ "from IPython import display\n\n# Create env manually to set time limit. Please don't change this.\nTIME_LIMIT = 250\nenv = gym.wrappers.TimeLimit(\n gym.envs.classic_control.MountainCarEnv(),\n max_episode_steps=TIME_LIMIT + 1,\n)\nactions = {'left': 0, 'stop': 1, 'right': 2}", "_____no_output_____" ], [ "def policy(obs, t):\n # Write the code for your policy here. You can use the observation\n # (a tuple of position and velocity), the current time step, or both,\n # if you want.\n position, velocity = obs\n \n if velocity > 0:\n a = actions['right']\n else:\n a = actions['left']\n\n # This is an example policy. You can try running it, but it will not work.\n # Your goal is to fix that.\n return a", "_____no_output_____" ], [ "plt.figure(figsize=(4, 3))\ndisplay.clear_output(wait=True)\n\nobs = env.reset()\nfor t in range(TIME_LIMIT):\n plt.gca().clear()\n \n action = policy(obs, t) # Call your policy\n obs, reward, done, _ = env.step(action) # Pass the action chosen by the policy to the environment\n \n # We don't do anything with reward here because MountainCar is a very simple environment,\n # and reward is a constant -1. Therefore, your goal is to end the episode as quickly as possible.\n\n # Draw game image on display.\n plt.imshow(env.render('rgb_array'))\n \n display.clear_output(wait=True)\n display.display(plt.gcf())\n print(obs)\n\n if done:\n print(\"Well done!\")\n break\nelse:\n print(\"Time limit exceeded. Try again.\")\n\ndisplay.clear_output(wait=True)", "_____no_output_____" ], [ "from submit import submit_interface\nsubmit_interface(policy, <EMAIL>, <TOKEN>)", "_____no_output_____" ] ] ]

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

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "", "_____no_output_____" ], [ "[](https://colab.research.google.com/github/JohnSnowLabs/spark-nlp-workshop/blob/master/tutorials/Certification_Trainings/Healthcare/25.Date_Normalizer.ipynb)", "_____no_output_____" ], [ "## Colab Setup", "_____no_output_____" ] ], [ [ "import json\n\nfrom google.colab import files\n\nlicense_keys = files.upload()\n\nwith open(list(license_keys.keys())[0]) as f:\n license_keys = json.load(f)", "_____no_output_____" ], [ "license_keys['JSL_VERSION']", "_____no_output_____" ], [ "\n%%capture\nfor k,v in license_keys.items(): \n %set_env $k=$v\n\n!wget https://raw.githubusercontent.com/JohnSnowLabs/spark-nlp-workshop/master/jsl_colab_setup.sh\n!bash jsl_colab_setup.sh\n\n", "_____no_output_____" ], [ "\nimport json\nimport os\nfrom pyspark.ml import Pipeline, PipelineModel\nfrom pyspark.sql import SparkSession\n\nfrom sparknlp.annotator import *\nfrom sparknlp_jsl.annotator import *\nfrom sparknlp.base import *\nimport sparknlp_jsl\nimport sparknlp\n\nfrom sparknlp.util import *\nfrom sparknlp.pretrained import ResourceDownloader\nfrom pyspark.sql import functions as F\n\nimport pandas as pd\n \n", "_____no_output_____" ], [ "spark = sparknlp_jsl.start(license_keys['SECRET'])", "_____no_output_____" ], [ "spark", "_____no_output_____" ] ], [ [ "# **Date Normalizer**\n\nNew Annotator that transforms chunks Dates to a normalized Date with format YYYY/MM/DD. This annotator identifies dates in chunk annotations and transforms those dates to the format YYYY/MM/DD. \n\n", "_____no_output_____" ], [ "We going to create a chunks dates with different formats:", "_____no_output_____" ] ], [ [ "dates = [\n'08/02/2018',\n'11/2018',\n'11/01/2018',\n'12Mar2021',\n'Jan 30, 2018',\n'13.04.1999', \n'3April 2020',\n'next monday',\n'today',\n'next week'\n]\n\n", "_____no_output_____" ], [ "from pyspark.sql.types import StringType\ndf_dates = spark.createDataFrame(dates,StringType()).toDF('ner_chunk')", "_____no_output_____" ] ], [ [ "We going to transform that text to documents in spark-nlp.", "_____no_output_____" ] ], [ [ "document_assembler = DocumentAssembler().setInputCol('ner_chunk').setOutputCol('document')\ndocuments_DF = document_assembler.transform(df_dates)", "_____no_output_____" ] ], [ [ "After that we going to transform that documents to chunks.", "_____no_output_____" ] ], [ [ "from sparknlp.functions import map_annotations_col\n\nchunks_df = map_annotations_col(documents_DF.select(\"document\",\"ner_chunk\"),\n lambda x: [Annotation('chunk', a.begin, a.end, a.result, a.metadata, a.embeddings) for a in x], \"document\",\n \"chunk_date\", \"chunk\")", "_____no_output_____" ], [ "chunks_df.select('chunk_date').show(truncate=False)", "+---------------------------------------------------+\n|chunk_date |\n+---------------------------------------------------+\n|[{chunk, 0, 9, 08/02/2018, {sentence -> 0}, []}] |\n|[{chunk, 0, 6, 11/2018, {sentence -> 0}, []}] |\n|[{chunk, 0, 9, 11/01/2018, {sentence -> 0}, []}] |\n|[{chunk, 0, 8, 12Mar2021, {sentence -> 0}, []}] |\n|[{chunk, 0, 11, Jan 30, 2018, {sentence -> 0}, []}]|\n|[{chunk, 0, 9, 13.04.1999, {sentence -> 0}, []}] |\n|[{chunk, 0, 10, 3April 2020, {sentence -> 0}, []}] |\n|[{chunk, 0, 10, next monday, {sentence -> 0}, []}] |\n|[{chunk, 0, 4, today, {sentence -> 0}, []}] |\n|[{chunk, 0, 8, next week, {sentence -> 0}, []}] |\n+---------------------------------------------------+\n\n" ] ], [ [ "Now we going to normalize that chunks using the DateNormalizer.", "_____no_output_____" ] ], [ [ "date_normalizer = DateNormalizer().setInputCols('chunk_date').setOutputCol('date')\n", "_____no_output_____" ], [ "date_normaliced_df = date_normalizer.transform(chunks_df)", "_____no_output_____" ] ], [ [ "We going to show how the date is normalized.", "_____no_output_____" ] ], [ [ "dateNormalizedClean = date_normaliced_df.selectExpr(\"ner_chunk\",\"date.result as dateresult\",\"date.metadata as metadata\")\n\ndateNormalizedClean.withColumn(\"dateresult\", dateNormalizedClean[\"dateresult\"]\n .getItem(0)).withColumn(\"metadata\", dateNormalizedClean[\"metadata\"]\n .getItem(0)['normalized']).show(truncate=False)", "+------------+----------+--------+\n|ner_chunk |dateresult|metadata|\n+------------+----------+--------+\n|08/02/2018 |2018/08/02|true |\n|11/2018 |2018/11/DD|true |\n|11/01/2018 |2018/11/01|true |\n|12Mar2021 |2021/03/12|true |\n|Jan 30, 2018|2018/01/30|true |\n|13.04.1999 |1999/04/13|true |\n|3April 2020 |2020/04/03|true |\n|next monday |2021/06/19|true |\n|today |2021/06/13|true |\n|next week |2021/06/20|true |\n+------------+----------+--------+\n\n" ] ], [ [ "We can configure the `anchorDateYear`,`anchorDateMonth` and `anchorDateDay` for the relatives dates.", "_____no_output_____" ], [ "In the following example we will use as a relative date 2021/02/22, to make that possible we need to set up the `anchorDateYear` to 2020, the `anchorDateMonth` to 2 and the `anchorDateDay` to 27. I will show you the configuration with the following example.", "_____no_output_____" ] ], [ [ "date_normalizer = DateNormalizer().setInputCols('chunk_date').setOutputCol('date')\\\n .setAnchorDateDay(27)\\\n .setAnchorDateMonth(2)\\\n .setAnchorDateYear(2021)", "_____no_output_____" ], [ "date_normaliced_df = date_normalizer.transform(chunks_df)\ndateNormalizedClean = date_normaliced_df.selectExpr(\"ner_chunk\",\"date.result as dateresult\",\"date.metadata as metadata\")\ndateNormalizedClean.withColumn(\"dateresult\", dateNormalizedClean[\"dateresult\"]\n .getItem(0)).withColumn(\"metadata\", dateNormalizedClean[\"metadata\"]\n .getItem(0)['normalized']).show(truncate=False)\n", "+------------+----------+--------+\n|ner_chunk |dateresult|metadata|\n+------------+----------+--------+\n|08/02/2018 |2018/08/02|true |\n|11/2018 |2018/11/DD|true |\n|11/01/2018 |2018/11/01|true |\n|12Mar2021 |2021/03/12|true |\n|Jan 30, 2018|2018/01/30|true |\n|13.04.1999 |1999/04/13|true |\n|3April 2020 |2020/04/03|true |\n|next monday |2021/02/29|true |\n|today |2021/02/27|true |\n|next week |2021/03/03|true |\n+------------+----------+--------+\n\n" ] ], [ [ "As you see the relatives dates like `next monday` , `today` and `next week` takes the `2021/02/22` as reference date.\n", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "%config InteractiveShell.ast_node_interactivity='last_expr_or_assign'\n\nimport numpy as np", "_____no_output_____" ], [ "a = np.array([1,1])", "_____no_output_____" ], [ "diff = np.ediff1d(a, 1, 1)", "_____no_output_____" ], [ "index = np.nonzero(diff)[0]", "_____no_output_____" ], [ "counts = np.ediff1d(index)", "_____no_output_____" ], [ "vals = a[index[:-1]]", "_____no_output_____" ], [ "# best way to interleave arrays\n# https://stackoverflow.com/questions/5347065/interweaving-two-numpy-arrays\nb = np.empty((vals.size + counts.size,), dtype=vals.dtype)\nb[0::2] = counts\nb[1::2] = vals\nb", "_____no_output_____" ], [ "def look_and_say(a):\n diff = np.ediff1d(a, 1, 1)\n index = np.nonzero(diff)[0]\n counts = np.ediff1d(index)\n vals = a[index[:-1]]\n c = np.empty((vals.size + counts.size,), dtype=vals.dtype)\n c[0::2] = counts\n c[1::2] = vals\n return c", "_____no_output_____" ], [ "look_and_say(b)", "_____no_output_____" ], [ "c = a\nprint(c)\n\nfor i in range(20):\n c = look_and_say(c)\n print(c)", "[1 1]\n[2 1]\n[1 2 1 1]\n[1 1 1 2 2 1]\n[3 1 2 2 1 1]\n[1 3 1 1 2 2 2 1]\n[1 1 1 3 2 1 3 2 1 1]\n[3 1 1 3 1 2 1 1 1 3 1 2 2 1]\n[1 3 2 1 1 3 1 1 1 2 3 1 1 3 1 1 2 2 1 1]\n[1 1 1 3 1 2 2 1 1 3 3 1 1 2 1 3 2 1 1 3 2 1 2 2 2 1]\n[3 1 1 3 1 1 2 2 2 1 2 3 2 1 1 2 1 1 1 3 1 2 2 1 1 3 1 2 1 1 3 2 1 1]\n[1 3 2 1 1 3 2 1 3 2 1 1 1 2 1 3 1 2 2 1 1 2 3 1 1 3 1 1 2 2 2 1 1 3 1 1 1\n 2 2 1 1 3 1 2 2 1]\n[1 1 1 3 1 2 2 1 1 3 1 2 1 1 1 3 1 2 3 1 1 2 1 1 1 3 1 1 2 2 2 1 1 2 1 3 2\n 1 1 3 2 1 3 2 2 1 1 3 3 1 2 2 2 1 1 3 1 1 2 2 1 1]\n[3 1 1 3 1 1 2 2 2 1 1 3 1 1 1 2 3 1 1 3 1 1 1 2 1 3 2 1 1 2 3 1 1 3 2 1 3\n 2 2 1 1 2 1 1 1 3 1 2 2 1 1 3 1 2 1 1 1 3 2 2 2 1 2 3 1 1 3 2 2 1 1 3 2 1\n 2 2 2 1]\n[1 3 2 1 1 3 2 1 3 2 2 1 1 3 3 1 1 2 1 3 2 1 1 3 3 1 1 2 1 1 1 3 1 2 2 1 1\n 2 1 3 2 1 1 3 1 2 1 1 1 3 2 2 2 1 1 2 3 1 1 3 1 1 2 2 2 1 1 3 1 1 1 2 3 1\n 1 3 3 2 1 1 1 2 1 3 2 1 1 3 2 2 2 1 1 3 1 2 1 1 3 2 1 1]\n[1 1 1 3 1 2 2 1 1 3 1 2 1 1 1 3 2 2 2 1 2 3 2 1 1 2 1 1 1 3 1 2 2 1 2 3 2\n 1 1 2 3 1 1 3 1 1 2 2 2 1 1 2 1 1 1 3 1 2 2 1 1 3 1 1 1 2 3 1 1 3 3 2 2 1\n 1 2 1 3 2 1 1 3 2 1 3 2 2 1 1 3 3 1 1 2 1 3 2 1 2 3 1 2 3 1 1 2 1 1 1 3 1\n 2 2 1 1 3 3 2 2 1 1 3 1 1 1 2 2 1 1 3 1 2 2 1]\n[3 1 1 3 1 1 2 2 2 1 1 3 1 1 1 2 3 1 1 3 3 2 1 1 1 2 1 3 1 2 2 1 1 2 3 1 1\n 3 1 1 2 2 1 1 1 2 1 3 1 2 2 1 1 2 1 3 2 1 1 3 2 1 3 2 2 1 1 2 3 1 1 3 1 1\n 2 2 2 1 1 3 3 1 1 2 1 3 2 1 2 3 2 2 2 1 1 2 1 1 1 3 1 2 2 1 1 3 1 2 1 1 1\n 3 2 2 2 1 2 3 2 1 1 2 1 1 1 3 1 2 1 1 1 2 1 3 1 1 1 2 1 3 2 1 1 2 3 1 1 3\n 1 1 2 2 2 1 2 3 2 2 2 1 1 3 3 1 2 2 2 1 1 3 1 1 2 2 1 1]\n[1 3 2 1 1 3 2 1 3 2 2 1 1 3 3 1 1 2 1 3 2 1 2 3 1 2 3 1 1 2 1 1 1 3 1 1 2\n 2 2 1 1 2 1 3 2 1 1 3 2 1 2 2 3 1 1 2 1 1 1 3 1 1 2 2 2 1 1 2 1 1 1 3 1 2\n 2 1 1 3 1 2 1 1 1 3 2 2 2 1 1 2 1 3 2 1 1 3 2 1 3 2 2 1 2 3 2 1 1 2 1 1 1\n 3 1 2 1 1 1 2 1 3 3 2 2 1 1 2 3 1 1 3 1 1 2 2 2 1 1 3 1 1 1 2 3 1 1 3 3 2\n 1 1 1 2 1 3 1 2 2 1 1 2 3 1 1 3 1 1 1 2 3 1 1 2 1 1 1 3 3 1 1 2 1 1 1 3 1\n 2 2 1 1 2 1 3 2 1 1 3 2 1 3 2 1 1 1 2 1 3 3 2 2 1 2 3 1 1 3 2 2 1 1 3 2 1\n 2 2 2 1]\n[1 1 1 3 1 2 2 1 1 3 1 2 1 1 1 3 2 2 2 1 2 3 2 1 1 2 1 1 1 3 1 2 1 1 1 2 1\n 3 1 1 1 2 1 3 2 1 1 2 3 1 1 3 2 1 3 2 2 1 1 2 1 1 1 3 1 2 2 1 1 3 1 2 1 1\n 2 2 1 3 2 1 1 2 3 1 1 3 2 1 3 2 2 1 1 2 3 1 1 3 1 1 2 2 2 1 1 3 1 1 1 2 3\n 1 1 3 3 2 2 1 1 2 1 1 1 3 1 2 2 1 1 3 1 2 1 1 1 3 2 2 1 1 1 2 1 3 1 2 2 1\n 1 2 3 1 1 3 1 1 1 2 3 1 1 2 1 1 2 3 2 2 2 1 1 2 1 3 2 1 1 3 2 1 3 2 2 1 1\n 3 3 1 1 2 1 3 2 1 2 3 1 2 3 1 1 2 1 1 1 3 1 1 2 2 2 1 1 2 1 3 2 1 1 3 3 1\n 1 2 1 3 2 1 1 2 3 1 2 3 2 1 1 2 3 1 1 3 1 1 2 2 2 1 1 2 1 1 1 3 1 2 2 1 1\n 3 1 2 1 1 1 3 1 2 3 1 1 2 1 1 2 3 2 2 1 1 1 2 1 3 2 1 1 3 2 2 2 1 1 3 1 2\n 1 1 3 2 1 1]\n[3 1 1 3 1 1 2 2 2 1 1 3 1 1 1 2 3 1 1 3 3 2 1 1 1 2 1 3 1 2 2 1 1 2 3 1 1\n 3 1 1 1 2 3 1 1 2 1 1 1 3 3 1 1 2 1 1 1 3 1 2 2 1 1 2 1 3 2 1 1 3 1 2 1 1\n 1 3 2 2 2 1 1 2 3 1 1 3 1 1 2 2 2 1 1 3 1 1 1 2 2 1 2 2 1 1 1 3 1 2 2 1 1\n 2 1 3 2 1 1 3 1 2 1 1 1 3 2 2 2 1 1 2 1 3 2 1 1 3 2 1 3 2 2 1 1 3 3 1 1 2\n 1 3 2 1 2 3 2 2 2 1 1 2 3 1 1 3 1 1 2 2 2 1 1 3 1 1 1 2 3 1 1 3 2 2 3 1 1\n 2 1 1 1 3 1 1 2 2 2 1 1 2 1 3 2 1 1 3 3 1 1 2 1 3 2 1 1 2 2 1 1 2 1 3 3 2\n 2 1 1 2 1 1 1 3 1 2 2 1 1 3 1 2 1 1 1 3 2 2 2 1 2 3 2 1 1 2 1 1 1 3 1 2 1\n 1 1 2 1 3 1 1 1 2 1 3 2 1 1 2 3 1 1 3 2 1 3 2 2 1 1 2 1 1 1 3 1 2 2 1 2 3\n 2 1 1 2 1 1 1 3 1 2 2 1 1 2 1 3 1 1 1 2 1 3 1 2 2 1 1 2 1 3 2 1 1 3 2 1 3\n 2 2 1 1 2 3 1 1 3 1 1 2 2 2 1 1 3 1 1 1 2 3 1 1 3 1 1 1 2 1 3 2 1 1 2 2 1\n 1 2 1 3 2 2 3 1 1 2 1 1 1 3 1 2 2 1 1 3 3 2 2 1 1 3 1 1 1 2 2 1 1 3 1 2 2\n 1]\n[1 3 2 1 1 3 2 1 3 2 2 1 1 3 3 1 1 2 1 3 2 1 2 3 1 2 3 1 1 2 1 1 1 3 1 1 2\n 2 2 1 1 2 1 3 2 1 1 3 3 1 1 2 1 3 2 1 1 2 3 1 2 3 2 1 1 2 3 1 1 3 1 1 2 2\n 2 1 1 2 1 1 1 3 1 2 2 1 1 3 1 1 1 2 3 1 1 3 3 2 2 1 1 2 1 3 2 1 1 3 2 1 3\n 2 2 1 1 3 3 1 2 2 1 1 2 2 3 1 1 3 1 1 2 2 2 1 1 2 1 1 1 3 1 2 2 1 1 3 1 1\n 1 2 3 1 1 3 3 2 2 1 1 2 1 1 1 3 1 2 2 1 1 3 1 2 1 1 1 3 2 2 2 1 2 3 2 1 1\n 2 1 1 1 3 1 2 1 1 1 2 1 3 3 2 2 1 1 2 1 3 2 1 1 3 2 1 3 2 2 1 1 3 3 1 1 2\n 1 3 2 1 1 3 2 2 1 3 2 1 1 2 3 1 1 3 2 1 3 2 2 1 1 2 1 1 1 3 1 2 2 1 2 3 2\n 1 1 2 1 1 1 3 1 2 2 1 2 2 2 1 1 2 1 1 2 3 2 2 2 1 1 2 3 1 1 3 1 1 2 2 2 1\n 1 3 1 1 1 2 3 1 1 3 3 2 1 1 1 2 1 3 1 2 2 1 1 2 3 1 1 3 1 1 1 2 3 1 1 2 1\n 1 1 3 3 1 1 2 1 1 1 3 1 2 2 1 1 2 1 3 2 1 1 3 1 2 1 1 1 3 2 2 2 1 1 2 3 1\n 1 3 1 1 2 2 1 1 1 2 1 3 1 2 2 1 1 2 3 1 1 3 1 1 2 2 2 1 1 2 1 1 1 3 3 1 1\n 2 1 1 1 3 1 1 2 2 2 1 1 2 1 1 1 3 1 2 2 1 1 3 1 2 1 1 1 3 2 2 2 1 1 2 1 3\n 2 1 1 3 2 1 3 2 2 1 1 3 3 1 1 2 1 3 2 1 1 3 3 1 1 2 1 1 1 3 1 2 2 1 2 2 2\n 1 1 2 1 1 1 3 2 2 1 3 2 1 1 2 3 1 1 3 1 1 2 2 2 1 2 3 2 2 2 1 1 3 3 1 2 2\n 2 1 1 3 1 1 2 2 1 1]\n" ] ] ]

[ "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/gndede/python/blob/main/Python_Project1.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ] ], [ [ "import pandas as pd\n#data = pd.read_csv(filename, encoding= 'unicode_escape')\ndf = pd.read_csv(\"/content/test _123.csv\", encoding= 'unicode_escape')", "_____no_output_____" ], [ "df.head()", "_____no_output_____" ], [ "#del df[\"planType\"]", "_____no_output_____" ], [ "#del df[\"customField4\"]", "_____no_output_____" ], [ " #del\tdf[\"customField5\"]", "_____no_output_____" ], [ "#del df[\"pdfName\"]\n#del df[\"phoneNumber\"]\n", "_____no_output_____" ], [ "#del df[\"dropDate\"]", "_____no_output_____" ], [ "#df", "_____no_output_____" ], [ "#Group by the coordinatedMailingId\ngk = df.groupby(['coordinatedMailingId','asset1','addressLine1','isPrimary'])", "_____no_output_____" ], [ "gk.first()", "_____no_output_____" ], [ "#drop ALL duplicate values\n#df.drop_duplicates(subset = \"customerPersonId\", keep = False, inplace = True)", "_____no_output_____" ], [ "#drop ALL duplicate values\n#df.drop_duplicates(subset = \"asset1\", keep = False, inplace = True)", "_____no_output_____" ], [ "#drop ALL duplicate values\n#df.drop_duplicates(subset = \"asset2\", keep = False, inplace = True)", "_____no_output_____" ], [ "#df", "_____no_output_____" ], [ "df['fullName'] = df['firstName'].str.cat(df['lastName'],sep=\" \")\n#firstName\tlastName\tfullName", "_____no_output_____" ], [ "df", "_____no_output_____" ], [ "# Sorting by column \"isPrimary\"\ndf.sort_values(by=['isPrimary'], ascending=False)", "_____no_output_____" ], [ "df['phoneNumber']\n\nfor phone_no in df['phoneNumber']: \n contactphone = \"%c-%c%c%c-%c%c%c-%c%c%c%c\" % tuple(map(ord,list(str(phone_no)[:11])))\n\n print(contactphone)", "1-888-598-7545\n1-855-872-6808\n1-855-872-6808\n1-855-872-6808\n1-855-872-6808\n1-855-872-6808\n1-855-238-0490\n1-855-238-0490\n1-855-832-0978\n1-844-448-7300\n1-844-448-7300\n1-844-596-0467\n1-844-809-5244\n1-844-596-0460\n1-866-715-4673\n1-844-620-5728\n1-866-715-4673\n1-888-812-2190\n1-855-832-0971\n1-855-832-0971\n1-855-233-5513\n1-855-233-5513\n1-855-832-0971\n1-855-832-0971\n1-844-353-0772\n1-844-638-4642\n1-855-832-0976\n1-855-832-0976\n1-855-832-0976\n1-844-596-0460\n1-855-832-0976\n1-855-832-0976\n1-866-202-9691\n1-888-812-2190\n1-844-889-9143\n1-844-889-9143\n1-844-889-9143\n1-844-889-9143\n1-855-832-0976\n1-855-832-0976\n1-855-238-0490\n1-844-596-0467\n1-855-671-6086\n1-855-832-0971\n1-855-832-0971\n1-844-596-0462\n1-855-238-0490\n1-866-202-9574\n1-844-287-9945\n1-888-586-0696\n1-888-586-0696\n1-888-586-0696\n1-855-671-6086\n1-844-596-0460\n1-866-202-9574\n1-855-832-0971\n1-855-832-0971\n1-888-427-8835\n1-888-427-8835\n1-866-202-9731\n1-866-202-9731\n1-866-202-9574\n1-844-596-0462\n1-844-596-0462\n1-866-202-9574\n1-866-202-9574\n1-855-671-6086\n1-855-238-0490\n1-844-809-5244\n1-844-809-5244\n1-844-809-5244\n1-844-809-5244\n1-855-238-0490\n1-844-809-5244\n1-844-809-5244\n1-844-809-5244\n1-844-809-5244\n1-844-809-5244\n1-844-809-5244\n1-844-809-5244\n1-855-832-0971\n1-866-202-9574\n1-855-671-6086\n1-855-232-5748\n1-855-359-7380\n1-855-359-7380\n1-866-202-9574\n1-844-266-1392\n1-844-266-1392\n1-844-266-1392\n1-844-266-1392\n1-855-238-0490\n1-888-427-8835\n1-844-686-0483\n1-855-663-4227\n1-855-241-5717\n1-855-671-6086\n1-866-202-9574\n1-866-202-9574\n1-855-232-5748\n1-866-844-4340\n1-866-844-4340\n1-844-596-0462\n1-844-596-0462\n1-844-596-0462\n1-866-202-9574\n1-855-259-2314\n1-888-651-7308\n1-888-651-7308\n1-888-651-7308\n1-855-832-0978\n1-888-794-1519\n1-855-671-6086\n1-888-651-7308\n1-888-812-2190\n1-855-359-7380\n1-866-451-4616\n1-866-451-4616\n1-866-451-4616\n1-866-451-4616\n1-866-451-4616\n1-866-451-4616\n1-866-451-4616\n1-855-832-0976\n1-855-535-7140\n1-866-202-9574\n1-866-202-9678\n1-855-832-0976\n1-866-202-9574\n1-844-889-9143\n1-855-241-5717\n1-844-287-9945\n1-855-241-5717\n1-866-451-4616\n1-844-596-0460\n1-855-241-5717\n1-855-241-5717\n1-866-202-9574\n1-833-945-1110\n1-866-451-4635\n1-855-671-6086\n1-855-671-6086\n1-855-671-6086\n1-888-429-8490\n1-866-202-9731\n1-855-241-5717\n1-855-241-5717\n1-855-241-5717\n1-855-359-7380\n1-855-359-7380\n1-855-238-0488\n1-855-671-6086\n1-844-266-1392\n1-855-241-5717\n1-855-359-7380\n1-855-359-7380\n1-855-359-7380\n1-855-671-6086\n1-866-202-9574\n1-866-202-9574\n1-866-202-9574\n1-866-202-9574\n1-866-202-9574\n1-866-202-9582\n1-866-202-9582\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-596-0460\n1-844-596-0460\n1-844-638-4642\n1-844-596-0460\n1-844-596-0460\n1-844-596-0460\n1-888-812-2190\n1-844-287-9945\n1-844-287-9945\n1-844-596-0460\n1-866-259-2944\n1-844-596-0460\n1-855-535-7140\n1-844-596-0460\n1-888-354-7664\n1-844-809-5244\n1-844-809-5244\n1-844-809-5244\n1-844-809-5244\n1-844-809-5244\n1-888-812-2190\n1-888-429-8490\n1-866-356-0166\n1-844-596-0460\n1-844-596-0467\n1-844-287-9945\n1-844-287-9945\n1-844-596-0460\n1-844-596-0460\n1-888-429-8490\n1-888-598-7545\n1-888-598-7545\n1-866-451-4616\n1-866-451-4616\n1-855-241-5720\n1-855-832-0976\n1-844-448-7300\n1-844-448-7300\n1-844-596-0460\n1-877-495-6386\n1-844-448-7300\n1-844-448-7300\n1-855-671-6086\n1-844-256-0921\n1-855-651-5058\n1-855-241-5717\n1-844-596-0460\n1-866-202-9574\n1-866-356-1034\n1-844-596-0467\n1-855-832-0971\n1-844-448-7300\n1-844-448-7300\n1-888-794-1519\n1-844-596-0467\n1-844-448-7300\n1-844-686-0485\n1-888-864-0764\n1-888-864-0764\n1-844-596-0460\n1-844-596-0467\n1-855-671-6086\n1-888-264-0638\n1-844-448-7300\n1-866-259-2944\n1-844-596-0460\n1-866-766-6019\n1-855-835-3916\n1-855-241-5720\n1-844-596-0467\n1-844-596-0460\n1-844-596-0460\n1-844-448-7300\n1-866-682-4841\n1-866-682-4841\n1-855-241-5720\n1-844-287-9945\n1-888-812-2190\n1-844-596-0460\n1-866-766-6019\n1-855-881-7868\n1-844-596-0460\n1-844-596-0460\n1-844-596-0460\n1-844-596-0460\n1-844-596-0460\n1-844-596-0460\n1-844-596-0460\n1-844-596-0460\n1-844-596-0460\n1-844-596-0460\n1-844-596-0460\n1-844-596-0460\n1-844-596-0460\n1-844-596-0460\n1-844-596-0460\n1-844-596-0467\n1-844-889-9143\n1-844-889-9143\n1-844-889-9143\n1-866-508-5118\n1-855-323-8831\n1-855-323-8831\n1-855-323-8831\n1-855-323-8831\n1-855-323-8831\n1-855-323-8831\n1-855-323-8831\n1-844-809-5243\n1-844-809-5243\n1-866-249-7785\n1-888-429-8490\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-855-835-3811\n1-855-835-3811\n1-855-241-5724\n1-866-844-4340\n1-866-844-4340\n1-888-230-0074\n1-888-230-0074\n1-888-230-0074\n1-888-230-0074\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-844-287-9945\n1-866-682-4841\n1-855-835-3827\n1-855-835-3827\n1-855-835-3827\n1-855-835-3827\n1-855-663-4227\n1-855-238-0489\n1-866-682-4741\n1-844-448-7300\n1-844-448-7300\n1-844-448-7300\n1-844-448-7300\n1-844-448-7300\n1-844-448-7300\n1-844-448-7300\n1-844-448-7300\n1-844-448-7300\n1-844-448-7300\n1-844-448-7300\n1-844-448-7300\n1-844-448-7300\n1-844-448-7300\n1-844-448-7300\n1-844-448-7300\n1-844-448-7300\n1-855-663-4228\n1-855-663-4228\n1-855-663-4228\n1-855-663-4228\n1-855-238-0485\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-844-686-0483\n1-866-201-0367\n1-866-201-0367\n1-866-201-0367\n1-888-812-2190\n1-866-766-6019\n1-855-835-3916\n1-855-241-5720\n1-844-596-0467\n1-844-596-0460\n" ], [ "df['phoneNumber']\n\n\nfor phone_no in df['phoneNumber']:\n contactphone = \"%c-%c%c-%c%c%c-%c%c%c%c\" % tuple(map(ord,list(str(phone_no)[:11])))\n print(contactphone)", "_____no_output_____" ], [ "df = df.assign(phone=[2, 2, 4, 7, 4, 1],\n Identity=[0, 1, 1, 3, 2, 5])", "_____no_output_____" ] ] ]

[ "markdown", "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Code to download The Guardian UK data and clean data for text analysis\n@Jorge de Leon \n\nThis script allows you to download news articles that match your parameters from the Guardian newspaper, https://www.theguardian.com/us.", "_____no_output_____" ], [ "## Set-up", "_____no_output_____" ] ], [ [ "import os\nimport re \nimport glob\nimport json\nimport requests\nimport pandas as pd \n\nfrom glob import glob\nfrom os import makedirs\nfrom textblob import TextBlob\nfrom os.path import join, exists\nfrom datetime import date, timedelta\n\nos.chdir(\"..\")\n\nimport nltk\nnltk.download('punkt')\nnltk.download('wordnet')\nnltk.download('stopwords')\nfrom nltk import sent_tokenize, word_tokenize\nfrom nltk.stem.snowball import SnowballStemmer\nfrom nltk.stem.wordnet import WordNetLemmatizer\nfrom nltk.corpus import stopwords", "_____no_output_____" ] ], [ [ "## API and news articles requests\n\nThis section contains the code that will be used to download articles from the Guardian website. \nthe initial variables will be determined as user-defined parameters.", "_____no_output_____" ] ], [ [ "#Enter API and parameters - these parameters can be obtained by playing around with the Guardian API tool:\n# https://open-platform.theguardian.com/explore/\n\n# Set up initial and end date \n\nstart_date_global = date(2000, 1, 1)\nend_date_global = date(2020, 5, 17)\nquery = \"JPMorgan\"\nterm = ('stock')\n\n#Enter API key, endpoint and parameters\nmy_api_key = open(\"..\\\\input files\\\\creds_guardian.txt\").read().strip()\napi_endpoint = \"http://content.guardianapis.com/search?\"\nmy_params = {\n 'from-date': '',\n 'to-date': '',\n 'show-fields': 'bodyText',\n 'q': query,\n 'page-size': 200,\n 'api-key': my_api_key\n}", "_____no_output_____" ], [ "articles_dir = join('theguardian','jpmorgan')\nmakedirs(articles_dir, exist_ok=True)", "_____no_output_____" ], [ "# day iteration from here:\n# http://stackoverflow.com/questions/7274267/print-all-day-dates-between-two-dates\nstart_date = start_date_global\nend_date = end_date_global\ndayrange = range((end_date - start_date).days + 1)\nfor daycount in dayrange:\n dt = start_date + timedelta(days=daycount)\n datestr = dt.strftime('%Y-%m-%d')\n fname = join(articles_dir, datestr + '.json')\n if not exists(fname):\n # then let's download it\n print(\"Downloading\", datestr)\n all_results = []\n my_params['from-date'] = datestr\n my_params['to-date'] = datestr\n current_page = 1\n total_pages = 1\n while current_page <= total_pages:\n print(\"...page\", current_page)\n my_params['page'] = current_page\n resp = requests.get(api_endpoint, my_params)\n data = resp.json()\n all_results.extend(data['response']['results'])\n # if there is more than one page\n current_page += 1\n total_pages = data['response']['pages']\n\n with open(fname, 'w') as f:\n print(\"Writing to\", fname)\n\n # re-serialize it for pretty indentation\n f.write(json.dumps(all_results, indent=2))", "_____no_output_____" ], [ "#Read all json files that will be concatenated\ntest_files = sorted(glob('theguardian/jpmorgan/*.json'))", "_____no_output_____" ], [ "#intialize empty list that we will append dataframes to\nall_files = []\n \n#write a for loop that will go through each of the file name through globbing and the end result will be the list \n#of dataframes\nfor file in test_files:\n try:\n articles = pd.read_json(file)\n all_files.append(articles)\n except pd.errors.EmptyDataError:\n print('Note: filename.csv ws empty. Skipping')\n continue #will skip the rest of the bloc and move to next file\n\n#create dataframe with data from json files\ntheguardian_rawdata = pd.concat(all_files, axis=0, ignore_index=True) ", "_____no_output_____" ] ], [ [ "## Text Analysis", "_____no_output_____" ] ], [ [ "#Drop empty columns\ntheguardian_rawdata = theguardian_rawdata.iloc[:,0:12]", "_____no_output_____" ], [ "#show types of media that was downloaded by type\ntheguardian_rawdata['type'].unique()", "_____no_output_____" ], [ "#filter only for articles\ntheguardian_rawdata = theguardian_rawdata[theguardian_rawdata['type'].str.match('article',na=False)]", "_____no_output_____" ], [ "#remove columns that do not contain relevant information for analysis\ntheguardian_dataset = theguardian_rawdata.drop(['apiUrl','id', 'isHosted', 'pillarId', 'pillarName',\n 'sectionId', 'sectionName', 'type','webTitle', 'webUrl'], axis=1)", "_____no_output_____" ], [ "#Modify the column webPublicationDate to Date and the fields to string and lower case\ntheguardian_dataset[\"date\"] = pd.to_datetime(theguardian_dataset[\"webPublicationDate\"]).dt.strftime('%Y-%m-%d')\ntheguardian_dataset['fields'] = theguardian_dataset['fields'].astype(str).str.lower()", "_____no_output_____" ], [ "#Clean the articles from URLS, remove punctuaction and numbers. \ntheguardian_dataset['fields'] = theguardian_dataset['fields'].str.replace('<.*?>','') # remove HTML tags\ntheguardian_dataset['fields'] = theguardian_dataset['fields'].str.replace('[^\\w\\s]','') # remove punc.", "_____no_output_____" ], [ "#Generate sentiment analysis for each article\n#Using TextBlob obtain polarity\ntheguardian_dataset['sentiment_polarity'] = theguardian_dataset['fields'].apply(lambda row: TextBlob(row).sentiment.polarity)\n#Using TextBlob obtain subjectivity\ntheguardian_dataset['sentiment_subjectivity'] = theguardian_dataset['fields'].apply(lambda row: TextBlob(row).sentiment.subjectivity)", "_____no_output_____" ], [ "#Remove numbers from text\ntheguardian_dataset['fields'] = theguardian_dataset['fields'].str.replace('\\d+','') # remove numbers\n\n#Then I will tokenize each word and remover stop words\ntheguardian_dataset['tokenized_fields'] = theguardian_dataset.apply(lambda row: nltk.word_tokenize(row['fields']), axis=1)", "_____no_output_____" ], [ "#Stop words\nstop_words=set(stopwords.words(\"english\"))", "_____no_output_____" ], [ "#Remove stop words\ntheguardian_dataset['tokenized_fields'] = theguardian_dataset['tokenized_fields'].apply(lambda x: [item for item in x if item not in stop_words])", "_____no_output_____" ], [ "#Count number of words and create a column with the most common 5 words per article\nfrom collections import Counter\ntheguardian_dataset['high_recurrence'] = theguardian_dataset['tokenized_fields'].apply(lambda x: [k for k, v in Counter(x).most_common(5)])", "_____no_output_____" ], [ "#Create a word count for the word \"stock\"\ntheguardian_dataset['word_ocurrence'] = theguardian_dataset['tokenized_fields'].apply(lambda x: [w for w in x if re.search(term, w)])\ntheguardian_dataset['word_count'] = theguardian_dataset['word_ocurrence'].apply(len)", "_____no_output_____" ], [ "#Create a count of the total number of words\ntheguardian_dataset['total_words'] = theguardian_dataset['tokenized_fields'].apply(len)", "_____no_output_____" ], [ "#Create new table with average polarity, subjectivity, count of the word \"stock\" per day\nguardian_microsoft = theguardian_dataset.groupby('date')['sentiment_polarity','sentiment_subjectivity','word_count','total_words'].agg('mean')", "_____no_output_____" ], [ "#Create a variable for the number of articles per day\ncount_articles = theguardian_dataset\ncount_articles['no_articles'] = count_articles.groupby(['date'])['fields'].transform('count')\ncount_articles = count_articles[[\"date\",\"no_articles\"]]\ncount_articles_df = count_articles.drop_duplicates(subset = \"date\", \n keep = \"first\", inplace=False) ", "_____no_output_____" ], [ "#Join tables by date\nguardian_microsoft = guardian_microsoft.merge(count_articles_df, on='date', how ='left')", "_____no_output_____" ], [ "#Save dataframes into CSV\ntheguardian_dataset.to_csv('theguardian/jpmorgan/theguardian_jpmorgan_text.csv', encoding='utf-8')\nguardian_microsoft.to_csv('theguardian/jpmorgan/theguardian_jpmorgan_data.csv', encoding='utf-8')", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import torch\nimport matplotlib.pyplot as plt\nimport numpy as np\nimport seaborn as sns\nfrom pathlib import Path\n\nsns.color_palette(\"tab10\")\nsns.set(rc={\n \"figure.dpi\": 150,\n \"text.usetex\": True,\n \"xtick.labelsize\": \"small\",\n \"ytick.labelsize\": \"small\",\n \"axes.labelsize\": \"small\",\n \"axes.titlesize\": \"small\",\n \"figure.titlesize\": \"medium\",\n \"axes.titlepad\": 2.0,\n \"xtick.major.pad\": -4.0,\n #\"figure.subplot.hspace\": 0.0,\n \"figure.constrained_layout.use\": True,\n})", "_____no_output_____" ], [ "def attribute_to_matrix(sequences, attribute):\n attribute_list = []\n for seq in sequences:\n attribute_list.append(seq[attribute])\n return np.array(attribute_list)\n\ndef cold_starts(init_times):\n num_activations = init_times.size\n num_cold_starts = np.sum(init_times > 0)\n return num_activations, num_cold_starts\n\ndef app_name(filename: str) -> str:\n app_name = filename[filename.find(\":\")+4:filename.find(\"_fetched_\")]\n if \"_rand\" in filename:\n app_name += \"_rand\"\n return app_name\n\ndef get_index(applist: list, filename: str) -> int:\n return applist.index(app_name(filename))", "_____no_output_____" ], [ "# Number of activations vs number of cold starts.\n\n# Configuration\ndir = \"final_high_load_n_1000\"\n#dir = \"final_batched_high_load_n_400\"\n\nfiles = Path(f\"../data/{dir}\").glob('*.pkl')\n\nfor f in sorted(files):\n print(f)\n with open(f, \"rb\") as stream:\n data = torch.load(stream)\n num_activations, num_cold_starts = cold_starts(attribute_to_matrix(data[\"sequences\"], \"init_times\"))\n #print(f\"Number of activations: {num_activations}\")\n #print(f\"Number of cold starts: {num_cold_starts}\")\n print(f\"Cold start to activation ratio: {round(num_cold_starts / num_activations, 4)}\")", "../data/final_high_load_n_1000/injected_20210610_23:14_fanout_large_fetched_397_n_400_l_0.01_u_0.01_b_50_w_120.pkl\nCold start to activation ratio: 0.3002\n../data/final_high_load_n_1000/injected_20210611_00:04_parallel_large_fetched_390_n_400_l_0.01_u_0.01_b_50_w_120.pkl\nCold start to activation ratio: 0.3008\n../data/final_high_load_n_1000/injected_20210611_00:53_tree_large_fetched_399_n_400_l_0.01_u_0.01_b_50_w_120.pkl\nCold start to activation ratio: 0.3007\n../data/final_high_load_n_1000/injected_20210611_01:43_fanout_large_fetched_391_n_400_l_0.01_u_0.01_b_50_w_120_rand.pkl\nCold start to activation ratio: 0.3002\n../data/final_high_load_n_1000/injected_20210611_02:33_parallel_large_fetched_383_n_400_l_0.01_u_0.01_b_50_w_120_rand.pkl\nCold start to activation ratio: 0.3005\n../data/final_high_load_n_1000/injected_20210611_03:23_tree_large_fetched_398_n_400_l_0.01_u_0.01_b_50_w_120_rand.pkl\nCold start to activation ratio: 0.3006\n../data/final_high_load_n_1000/injected_20210611_04:13_fanout_small_fetched_384_n_400_l_0.0001_u_0.0001_b_50_w_120.pkl\nCold start to activation ratio: 0.3008\n../data/final_high_load_n_1000/injected_20210611_07:23_fanout_small_fetched_394_n_400_l_0.0001_u_0.0001_b_50_w_120_rand.pkl\nCold start to activation ratio: 0.3002\n../data/final_high_load_n_1000/injected_20210611_11:32_parallel_small_fetched_397_n_400_l_0.0001_u_0.0001_b_50_w_120.pkl\nCold start to activation ratio: 0.3006\n../data/final_high_load_n_1000/injected_20210611_12:23_tree_small_fetched_399_n_400_l_0.0001_u_0.0001_b_50_w_120.pkl\nCold start to activation ratio: 0.3004\n../data/final_high_load_n_1000/injected_20210611_12:49_parallel_small_fetched_398_n_400_l_0.0001_u_0.0001_b_50_w_120_rand.pkl\nCold start to activation ratio: 0.3003\n../data/final_high_load_n_1000/injected_20210611_13:40_tree_small_fetched_394_n_400_l_0.0001_u_0.0001_b_50_w_120_rand.pkl\nCold start to activation ratio: 0.3002\n../data/final_high_load_n_1000/injected_20210611_16:55_sequence_fetched_500_n_500_l_0.0001_u_0.0001_b_50_w_120.pkl\nCold start to activation ratio: 0.3001\n../data/final_high_load_n_1000/injected_20210611_17:16_sequence_fetched_500_n_500_l_0.0001_u_0.0001_b_50_w_120_rand.pkl\nCold start to activation ratio: 0.3002\n" ], [ "# Average inter-event time for each step.\n\n# Configuration\ndir = \"final_low_load_n_1000\"\n#dir = \"final_high_load_n_1000\"\nsave = False\n\ndef boxplot(applist: list, file2data: dict, title: str, ylabel: str) -> plt.Figure:\n fig, ax = plt.subplots(len(applist), 1)\n fig.set_size_inches(5.5, 6.8)\n fig.suptitle(title)\n for filename, data in file2data.items():\n if app_name(filename) in applist:\n sns.boxplot(ax=ax[get_index(applist, filename)], data=data, width=0.5, showfliers=False, linewidth=0.9)\n ax[get_index(applist, filename)].set_title(app_name(filename).replace(\"_\", \"\\_\"))\n ax[get_index(applist, filename)].set_ylabel(ylabel)\n ax[get_index(applist, filename)].set_xticklabels(list(range(1, data.shape[-1] + 1)))\n ax[-1].set_xlabel(\"i\")\n return fig\n\nprestr = \"ll_\"\nif \"high_load\" in dir: prestr = \"hl_\"\n\npoststr = \" (no cold starts)\"\nif \"high_load\" in dir: poststr = \" (30\\% cold starts)\"\n\napps = [\"sequence\", \"parallel_small\", \"tree_small\", \"fanout_small\", \"parallel_large\", \"tree_large\", \"fanout_large\"]\napps_rand = [\"sequence_rand\", \"parallel_small_rand\", \"tree_small_rand\", \"fanout_small_rand\", \"parallel_large_rand\", \"tree_large_rand\", \"fanout_large_rand\"]\n\ninter_dict = {}\ninit_dict = {}\nwait_dict = {}\n\nfiles = Path(f\"../data/{dir}\").glob('*.pkl')\nfor f in sorted(files):\n with open(f, \"rb\") as stream:\n data = torch.load(stream)\n inter_dict[str(f)] = np.diff(attribute_to_matrix(data[\"sequences\"], \"arrival_times\"), axis=-1)\n init_dict[str(f)] = attribute_to_matrix(data[\"sequences\"], \"init_times\")\n wait_dict[str(f)] = attribute_to_matrix(data[\"sequences\"], \"wait_times\") / 1000\n\ntitle = fr\"Distribution of inter-event time $\\tau_i${poststr}\"\nylabel = \"ms\"\nfig1 = boxplot(apps, inter_dict, title, ylabel)\nif save: plt.savefig(f\"data_plots/{prestr}dist_inter.pdf\")\nfig2 = boxplot(apps_rand, inter_dict, title, ylabel)\nif save: plt.savefig(f\"data_plots/{prestr}dist_inter_rand.pdf\")\n\nif \"high_load\" in dir:\n title = fr\"Distribution of initTime $i_i${poststr}\"\n ylabel = \"ms\"\n fig1 = boxplot(apps, init_dict, title, ylabel)\n if save: plt.savefig(f\"data_plots/{prestr}dist_init.pdf\")\n fig2 = boxplot(apps_rand, init_dict, title, ylabel)\n if save: plt.savefig(f\"data_plots/{prestr}dist_init_rand.pdf\")\n\n title = fr\"Distribution of waitTime $w_i${poststr}\"\n ylabel = \"sec\"\n fig1 = boxplot(apps, wait_dict, title, ylabel)\n if save: plt.savefig(f\"data_plots/{prestr}dist_wait.pdf\")\n fig2 = boxplot(apps_rand, wait_dict, title, ylabel)\n if save: plt.savefig(f\"data_plots/{prestr}dist_wait_rand.pdf\") ", "_____no_output_____" ], [ "# Distribution of inter-event, init and wait times.\n\n# Configuration\ndir = \"final_low_load_n_1000\"\n#dir = \"final_high_load_n_1000\"\n#dir = \"final_batched_high_load_n_400\"\nplot_init_times = True\nplot_wait_times = True\nsave = False\n\ndef compose_title(filename):\n def app_name(filename):\n app_name = filename[filename.find(\":\")+4:filename.find(\"_fetched_\")]\n return app_name.replace(\"_\", \"\\_\")\n title = app_name(filename)\n if \"_rand\" in filename:\n title += \"\\_rand\"\n if \"_b_\" in filename:\n title += \" (30% cold starts)\"\n return title\n\napps = [\"sequence\", \"parallel_small\", \"tree_small\", \"fanout_small\", \"parallel_large\", \n \"tree_large\", \"fanout_large\", \"sequence_rand\", \"parallel_small_rand\", \"tree_small_rand\",\n \"fanout_small_rand\", \"parallel_large_rand\", \"tree_large_rand\", \"fanout_large_rand\"]\nfiles = Path(f\"../data/{dir}\").glob('*.pkl')\n\nfig_inter, ax_inter = plt.subplots(14, 1)\nfig_inter.set_size_inches(6.4, 18)\nfig_init, ax_init = plt.subplots(14, 1)\nfig_init.set_size_inches(6.4, 18)\nfig_wait, ax_wait = plt.subplots(14, 1)\nfig_wait.set_size_inches(6.4, 18)\n\nfor i, f in enumerate(sorted(files)):\n with open(f, \"rb\") as stream:\n data = torch.load(stream)\n \n if \"small\" in str(f): \n numb_traces = 200\n else:\n numb_traces = 100\n \n inter_event_times = np.diff(attribute_to_matrix(data[\"sequences\"], \"arrival_times\"), axis=-1)[:numb_traces].reshape(-1)\n init_times = attribute_to_matrix(data[\"sequences\"], \"init_times\")[:numb_traces].reshape(-1)\n wait_times = attribute_to_matrix(data[\"sequences\"], \"wait_times\")[:numb_traces].reshape(-1) / 1000\n \n max_range = int(np.quantile(inter_event_times, 0.99))\n ax_inter[get_index(apps, str(f))].hist(inter_event_times, bins=300, range=(0.0, max_range), log=True)\n ax_inter[get_index(apps, str(f))].set_xlabel(\"milliseconds\")\n ax_inter[get_index(apps, str(f))].set_ylabel(\"count\")\n ax_inter[get_index(apps, str(f))].title.set_text(f\"Distribution over all inter-event times - {compose_title(str(f))}\")\n \n if plot_init_times:\n max_range = int(np.quantile(init_times, 0.99))\n ax_init[get_index(apps, str(f))].hist(init_times, bins=300, range=(0.0, max_range), log=True)\n ax_init[get_index(apps, str(f))].set_xlabel(\"milliseconds\")\n ax_init[get_index(apps, str(f))].set_ylabel(\"count\")\n ax_init[get_index(apps, str(f))].title.set_text(f\"Distribution over all initTimes - {compose_title(str(f))}\")\n \n if plot_wait_times:\n max_range = int(np.quantile(wait_times, 0.99))\n ax_wait[get_index(apps, str(f))].hist(wait_times, bins=300, range=(0.0, max_range), log=True)\n ax_wait[get_index(apps, str(f))].set_xlabel(\"seconds\")\n ax_wait[get_index(apps, str(f))].set_ylabel(\"count\")\n ax_wait[get_index(apps, str(f))].title.set_text(f\"Distribution over all waitTimes - {compose_title(str(f))}\")\n \n#fig_inter.tight_layout()\n#fig_init.tight_layout()\n#fig_wait.tight_layout()\n \nif save:\n filename = f\"{str(f)[str(f).rfind('/')+1:str(f).rfind('.pkl')]}.png\"\n plt.savefig(filename)", "_____no_output_____" ], [ "# Standard deviation of inter-event, init and wait time\n\n# Configuration\ndir = \"final_low_load_n_1000\"\n\nfiles = Path(f\"../data/{dir}\").glob('*.pkl')\n\ndef app_name(filename):\n app_name = filename[filename.find(\":\")+4:filename.find(\"_fetched_\")]\n if \"rand\" in filename:\n app_name += \"_rand\"\n return app_name\n\nfor f in sorted(files):\n with open(f, \"rb\") as stream:\n data = torch.load(stream)\n inter_event_times = np.diff(attribute_to_matrix(data[\"sequences\"], \"arrival_times\"), axis=-1)\n init_times = attribute_to_matrix(data[\"sequences\"], \"init_times\")\n wait_times = attribute_to_matrix(data[\"sequences\"], \"wait_times\") #/ 1000\n \n print(f\"app={app_name(str(f))}\\n\"\n f\"std_inter_event={np.std(inter_event_times)}\\n\"\n f\"std_init={np.std(init_times)}\\n\"\n f\"std_wait={np.std(wait_times)}\\n\")", "app=fanout_large\nstd_inter_event=151.05177762236863\nstd_init=8.144257456732442\nstd_wait=114.77229044370371\n\napp=parallel_large\nstd_inter_event=109.44044903032881\nstd_init=9.297179170864\nstd_wait=127.57963467870893\n\napp=tree_large\nstd_inter_event=131.23154306078888\nstd_init=16.78905591489502\nstd_wait=291.31143757231473\n\napp=fanout_large_rand\nstd_inter_event=151.65812343240964\nstd_init=8.883851129119272\nstd_wait=130.1648441444701\n\napp=parallel_large_rand\nstd_inter_event=109.44652378874352\nstd_init=10.602357329970584\nstd_wait=138.1014849145728\n\napp=tree_large_rand\nstd_inter_event=97.52263492725262\nstd_init=11.27306997866161\nstd_wait=179.97986456871658\n\napp=fanout_small\nstd_inter_event=239.70282899756836\nstd_init=13.732588276619522\nstd_wait=234.60153714089307\n\napp=parallel_small\nstd_inter_event=161.9196585169324\nstd_init=11.742867522581214\nstd_wait=153.98688379668323\n\napp=sequence\nstd_inter_event=224.63991208489645\nstd_init=11.105719767848612\nstd_wait=139.50763483666051\n\napp=tree_small\nstd_inter_event=104.72267932207933\nstd_init=11.101776254275707\nstd_wait=157.67831093003883\n\napp=fanout_small_rand\nstd_inter_event=198.97358421057848\nstd_init=10.779445649697989\nstd_wait=167.92800845212872\n\napp=parallel_small_rand\nstd_inter_event=204.9385673453389\nstd_init=13.69195516467261\nstd_wait=199.91791532708436\n\napp=sequence_rand\nstd_inter_event=249.20868229207025\nstd_init=12.790173494387357\nstd_wait=164.10402365230425\n\napp=tree_small_rand\nstd_inter_event=111.61791923038606\nstd_init=9.332272280104133\nstd_wait=130.05539527063073\n\n" ] ] ]

[ "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "Neuroon cross-validation\n------------------------\n\n\nNeuroon and PSG recordings were simultanously collected over the course of two nights. This analysis will show whether Neuroon is able to accurately classify sleep stages. The PSG classification will be a benchmark against which Neuroon performance will be tested. \"The AASM Manual for te Scoring of Sleep ad Associated Events\" identifies 5 sleep stages: \n\n* Stage W (Wakefulness)\n* Stage N1 (NREM 1)\n* Stage N1 (NREM 2)\n* Stage N1 (NREM 3)\n* Stage R (REM)\n<img src=\"images/sleep_stages.png\">\n\n\nThese stages can be identified following the rules guidelines in [1] either visually or digitally using combined information from EEG, EOG and EMG. Extensive research is beeing conducted on developing automated and simpler methods for sleep stage classification suitable for everyday home use (for a review see [2]). Automatic methods based on single channel EEG, which is the Neuroon category, were shown to work accurately when compared to PSG scoring [3]. \n\n[1] Berry RB BR, Gamaldo CE, Harding SM, Lloyd RM, Marcus CL, Vaughn BV; for the American Academy of Sleep Medicine. The AASM Manual for the Scoring of Sleep and Associated Events: Rules, Terminology and Technical Specifications.,Version 2.0.3. Darien, IL: American Academy of Sleep Medicine; 2014.\n\n[2] Van De Water, A. T. M., Holmes, A., & Hurley, D. a. (2011). Objective measurements of sleep for non-laboratory settings as alternatives to polysomnography - a systematic review. Journal of Sleep Research, 20, 183–200. \n\n[3] Berthomier, C., Drouot, X., Herman-Stoïca, M., Berthomier, P., Prado, J., Bokar-Thire, D. d’Ortho, M.P. (2007). Automatic analysis of single-channel sleep EEG: validation in healthy individuals. Sleep, 30(11), 1587–1595.\n", "_____no_output_____" ], [ "Signals time-synchronization using crosscorelation\n--------------------------------------------------\n\nNeuroon and PSG were recorded on devices with (probably) unsycnhronized clocks. First we will use a cross-correlation method [4] to find the time offset between the two recordings.\n\n[4] Fridman, L., Brown, D. E., Angell, W., Abdić, I., Reimer, B., & Noh, H. Y. (2016). Automated synchronization of driving data using vibration and steering events. Pattern Recognition Letters, 75, 9-15.\n\nDefine cross correlation function - code from: (http://lexfridman.com/blogs/research/2015/09/18/fast-cross-correlation-and-time-series-synchronization-in-python/)\n\nfor other examlpes see: (http://stackoverflow.com/questions/4688715/find-time-shift-between-two-similar-waveforms)\n\n", "_____no_output_____" ] ], [ [ "\n%matplotlib inline\n%load_ext autoreload\n%autoreload 2\n\nimport numpy as np\nimport matplotlib.pyplot as plt\nfrom itertools import tee\nimport pandas as pd\nimport seaborn as sns\nfrom numpy.fft import fft, ifft, fft2, ifft2, fftshift\nfrom collections import OrderedDict \nfrom datetime import timedelta\n\nplt.rcParams['figure.figsize'] = (9.0, 5.0)\n\n\n\n", "The autoreload extension is already loaded. To reload it, use:\n %reload_ext autoreload\n" ], [ "from parse_signal import load_psg, load_neuroon\n\n# Cross-correlation function. Equivalent to numpy.correlate(x,y mode = 'full') but faster for large arrays \n# This function was tested against other cross correlation methods in -- LINK TO OTHER NOTEBOOK\ndef cross_correlation_using_fft(x, y):\n f1 = fft(x)\n f2 = fft(np.flipud(y))\n cc = np.real(ifft(f1 * f2))\n return fftshift(cc)\n \n# shift < 0 means that y starts 'shift' time steps before x # shift > 0 means that y starts 'shift' time steps after x\ndef compute_shift(x, y):\n assert len(x) == len(y)\n c = cross_correlation_using_fft(x, y)\n assert len(c) == len(x)\n zero_index = int(len(x) / 2) - 1\n shift = zero_index - np.argmax(c)\n return shift,c\n\n\ndef cross_correlate():\n # Load the signal from hdf database and parse it to pandas series with datetime index\n psg_signal = load_psg('F3-A2')\n neuroon_signal = load_neuroon()\n # Resample the signal to 100hz, to have the same length for cross correlation\n psg_10 = psg_signal.resample('10ms').mean()\n neuroon_10 = neuroon_signal.resample('10ms').mean()\n \n # Create ten minute intervals\n dates_range = pd.date_range(psg_signal.head(1).index.get_values()[0], neuroon_signal.tail(1).index.get_values()[0], freq=\"10min\")\n \n # Convert datetime interval boundaries to string with only hours, minutes and seconds\n dates_range = [d.strftime('%H:%M:%S') for d in dates_range]\n \n \n all_coefs = []\n \n # iterate over overlapping pairs of 10 minutes boundaries \n for start, end in pairwise(dates_range):\n # cut 10 minutes piece of signal\n neuroon_cut = neuroon_10.between_time(start, end)\n psg_cut = psg_10.between_time(start, end)\n # Compute the correlation using fft convolution\n shift, coeffs = compute_shift(neuroon_cut, psg_cut)\n #normalize the coefficients because they will be shown on the same heatmap and need a common color scale\n all_coefs.append((coeffs - coeffs.mean()) / coeffs.std())\n \n #print('max corr at shift %s is at sample %i'%(start, shift))\n\n all_coefs = np.array(all_coefs)\n return all_coefs, dates_range\n\n# This function is used to iterate over a list, taking two consecutive items at each iteration\ndef pairwise(iterable):\n \"s -> (s0,s1), (s1,s2), (s2, s3), ...\"\n a, b = tee(iterable)\n next(b, None)\n return zip(a, b)", "_____no_output_____" ], [ "# Construct a matrix where each row represents a 10 minute window from the recording\n# and each column represent correlation coefficient between neuroon and psg signals offset by samples number.\n# 0 samples offset coefficient is stored at the middle column -1. Negative offset and positive offset span left and right from the center.\n# offset < 0 means that psg starts 'shift' time steps before neuroon\n# offset > 0 means that psg starts 'shift' time steps after neuroon\n\ncoeffs_matrix, dates = cross_correlate()", "_____no_output_____" ], [ "from plotting_collection import plot_crosscorrelation_heatmap\n\n#Plot part of the coefficients matrix centered around the max average correlation for all 10 minute windows\nplot_crosscorrelation_heatmap(coeffs_matrix, dates)", "_____no_output_____" ] ], [ [ "Hipnogram time-delay\n--------------------\nFrom the crosscorrelation of the eeg signals we can see the two devices are off by 2 minutes 41 seconds. Now we'll see if there is a point in time where the hipnograms are most simmilar. The measure of hipnogram simmilarity will be the sum of times when two devices classified the same sleep stage.\n", "_____no_output_____" ] ], [ [ "import parse_hipnogram as ph\n\ndef get_hipnogram_intersection(neuroon_hipnogram, psg_hipnogram, time_shift):\n\n neuroon_hipnogram.index = neuroon_hipnogram.index + timedelta(seconds = int(time_shift))\n\n \n combined = psg_hipnogram.join(neuroon_hipnogram, how = 'outer', lsuffix = '_psg', rsuffix = '_neuro')\n \n combined.loc[:, ['stage_num_psg', 'stage_name_psg', 'stage_num_neuro', 'stage_name_neuro', 'event_number_psg', 'event_number_neuro']] = combined.loc[:, ['stage_num_psg', 'stage_name_psg', 'stage_num_neuro', 'stage_name_neuro', 'event_number_psg', 'event_number_neuro']].fillna( method = 'bfill') \n \n combined.loc[:, ['stage_shift_psg', 'stage_shift_neuro']] = combined.loc[:, ['stage_shift_psg', 'stage_shift_neuro']].fillna( value = 'inside') \n \n # From the occupied room number subtract the room occupied by another mouse.\n combined['overlap'] = combined['stage_num_psg'] - combined['stage_num_neuro']\n \n same_stage = combined.loc[combined['overlap'] == 0]\n same_stage.loc[:, 'event_union'] = same_stage['event_number_psg'] + same_stage['event_number_neuro']\n\n\n# common_window = np.array([neuroon_hipnogram.tail(1).index.get_values()[0] - psg_hipnogram.head(1).index.get_values()[0]],dtype='timedelta64[m]').astype(int)[0]\n\n all_durations = OrderedDict()\n\n for stage_name, intersection in same_stage.groupby('event_union'):\n # Subtract the first row timestamp from the last to get the duration. Store as the duration in milliseconds.\n duration = (intersection.index.to_series().iloc[-1]- intersection.index.to_series().iloc[0]).total_seconds()\n \n stage_id = intersection.iloc[0, intersection.columns.get_loc('stage_name_neuro')] \n # Keep appending results to a list stored in a dict. Check if the list exists, if not create it.\n if stage_id not in all_durations.keys():\n all_durations[stage_id] = [duration]\n \n else: \n all_durations[stage_id].append(duration)\n \n\n \n means = OrderedDict()\n stds = OrderedDict()\n sums = OrderedDict()\n stages_sum = 0\n #Adding it here so its first in ordered dict and leftmost on the plot\n sums['stages_sum'] = 0\n for key, value in all_durations.items():\n #if key != 'wake':\n means[key] = np.array(value).mean()\n stds[key] = np.array(value).std()\n sums[key] = np.array(value).sum()\n \n stages_sum += np.array(value).sum() \n \n sums['stages_sum'] = stages_sum\n # Divide total seconds by 60 to get minutes \n #return stages_sum\n return sums, means, stds", "_____no_output_____" ], [ "def intersect_with_shift():\n psg_hipnogram = ph.parse_psg_stages()\n neuroon_hipnogram = ph.parse_neuroon_stages()\n \n intersection = OrderedDict([('wake', []), ('rem',[]), ('N1',[]), ('N2',[]), ('N3', []), ('stages_sum', [])])\n\n shift_range = np.arange(-500, 100, 10)\n for shift in shift_range:\n sums, _, _ = get_hipnogram_intersection(neuroon_hipnogram.copy(), psg_hipnogram.copy(), shift)\n for stage, intersect_dur in sums.items():\n intersection[stage].append(intersect_dur)\n \n return intersection, shift_range\n", "_____no_output_____" ], [ "\ndef plot_intersection(intersection, shift_range):\n \n psg_hipnogram = ph.parse_psg_stages()\n neuroon_hipnogram = ph.parse_neuroon_stages()\n \n stage_color_dict = {'N1' : 'royalblue', 'N2' :'forestgreen', 'N3' : 'coral', 'rem' : 'plum', 'wake' : 'lightgrey', 'stages_sum': 'dodgerblue'}\n\n \n fig, axes = plt.subplots(2)\n zscore_ax = axes[0].twinx()\n \n for stage in ['rem', 'N2', 'N3', 'wake']:\n intersect_sum = np.array(intersection[stage])\n z_scored = (intersect_sum - intersect_sum.mean()) / intersect_sum.std()\n zscore_ax.plot(shift_range, z_scored, color = stage_color_dict[stage], label = stage, alpha = 0.5, linestyle = '--')\n \n\n max_overlap = shift_range[np.argmax(intersection['stages_sum'])]\n fig.suptitle('max overlap at %i seconds offset'%max_overlap)\n axes[0].plot(shift_range, intersection['stages_sum'], label = 'stages sum', color = 'dodgerblue')\n axes[0].axvline(max_overlap, color='k', linestyle='--')\n\n axes[0].set_ylabel('time in the same sleep stage')\n axes[0].set_xlabel('offset in seconds')\n \n axes[0].legend(loc = 'center right')\n zscore_ax.grid(b=False)\n zscore_ax.legend()\n \n sums0, means0, stds0 = get_hipnogram_intersection(neuroon_hipnogram.copy(), psg_hipnogram.copy(), 0)\n#\n width = 0.35 \n ind = np.arange(5)\n colors_inorder = ['dodgerblue', 'lightgrey', 'forestgreen', 'coral', 'plum']\n #Plot the non shifted overlaps \n axes[1].bar(left = ind, height = list(sums0.values()),width = width, alpha = 0.8, \n tick_label =list(sums0.keys()), edgecolor = 'black', color= colors_inorder)\n\n sumsMax, meansMax, stdsMax = get_hipnogram_intersection(neuroon_hipnogram.copy(), psg_hipnogram.copy(), max_overlap)\n # Plot the shifted overlaps\n axes[1].bar(left = ind +width, height = list(sumsMax.values()),width = width, alpha = 0.8,\n tick_label =list(sumsMax.keys()), edgecolor = 'black', color = colors_inorder)\n \n axes[1].set_xticks(ind + width)\n \n plt.tight_layout()\n", "_____no_output_____" ], [ "intersection, shift_range = intersect_with_shift()", "_____no_output_____" ], [ "plot_intersection(intersection, shift_range)", "_____no_output_____" ] ], [ [ "hipnogram analysis indicates the same direction of time delay - psg is 4 minutes 10 seconds before neuroon. The time delay is larger for the hipnograms than for the signals by 1 minute 30 seconds.\n\nTodo: \n\n* add second axis with percentages\n* see if the overlap increased in proportion with offset\n* plot parts of time corrected signals and hipnograms\n* add different correlation tests notebook\n* add spectral and pca analysis\n\n", "_____no_output_____" ] ] ]

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

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "@author : krishan subudhi", "_____no_output_____" ], [ "## create a list of 50 random numbers", "_____no_output_____" ] ], [ [ "import random\narr = [random.randint(0,1000) for i in range(50)]\narr[:10]", "_____no_output_____" ] ], [ [ "## Calculate square root\n", "_____no_output_____" ] ], [ [ "import math\n%timeit [math.sqrt(a) for a in arr]", "20.9 µs ± 177 ns per loop (mean ± std. dev. of 7 runs, 10000 loops each)\n" ] ], [ [ "# numpy \n1. Faster\n1. Easy to use\n2. Memory effecient\n3. Rich in mathematical functions\n4. Easy matrix operations", "_____no_output_____" ] ], [ [ "import numpy as np\nnumpy_arr = np.array(arr)\nnumpy_arr[:10]", "_____no_output_____" ], [ "%timeit np.sqrt(numpy_arr)", "2.82 µs ± 196 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)\n" ] ], [ [ "## Matrix operations", "_____no_output_____" ] ], [ [ "matrix_2d = np.random.randint(0,100,(2,3))\nmatrix_2d", "_____no_output_____" ], [ "matrix_2d.sum(axis = 0)", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# PI-ICR analysis\n\nCreated on 17 July 2019 for the ISOLTRAP experiment\n- V1.1 (24 June 2020): Maximum likelihood estimation was simplified based on SciPy PDF's and the CERN-ROOT6 minimizer via the iminuit package (→ great performance)\n- V1.2 (20 February 2021): Preparations for scientific publication and iminuit v2 update integration\n\n@author: Jonas Karthein<br>\n@contact: [email protected]<br>\n@license: MIT license\n\n### References\n[1]: https://doi.org/10.1007/s00340-013-5621-0\n[2]: https://doi.org/10.1103/PhysRevLett.110.082501\n[3]: https://doi.org/10.1007/s10751-019-1601-z\n[4]: https://doi.org/10.1103/PhysRevLett.124.092502\n\n[1] S. Eliseev, _et al._ Appl. Phys. B (2014) 114: 107.<br>\n[2] S. Eliseev, _et al._ Phys. Rev. Lett. 110, 082501 (2013).<br>\n[3] J. Karthein, _et al._ Hyperfine Interact (2019) 240: 61.<br>\n\n### Application\n\nThe code was used to analyse data for the following publications:\n\n[3] J. Karthein, _et al._ Hyperfine Interact (2019) 240: 61.<br>\n[4] V. Manea and J. Karthein, _et al._ Phys. Rev. Lett. 124, 092502 (2020)<br>\n[5] M. Mougeot, _et al._ in preparation (2020)<br>\n\n### Introduction\n\nThe following code was written to reconstruct raw Phase-Imaging Ion-Cyclotron-Resonance (PI-ICR) data, to fit PI-ICR position information and calculate a frequency using the patter 1/2 scheme described in Ref. [1] and to determine a frequency ratio between a measurement ion and a reference ion. Additionally, the code allows to analyze isomeric states separated in pattern 2.\n data, to fit PI-ICR position information and calculate a frequency using the patter 1/2 scheme described in Ref. [1] and to determine a frequency ratio between a measurement ion and a reference ion. Additionally, the code allows to analyze isomeric states separated in pattern 2.\n\n### Required software and libraries\n\nThe following code was written in Python 3.7. The required libraries are listed below with a rough description for their task in the code. It doesn't claim to be a full description of the library.\n* pandas (data storage and calculation)\n* numpy (calculation)\n* matplotlib (plotting)\n* scipy (PDFs, least squares estimation)\n* configparser (configuration file processing)\n* jupyter (Python notebook environment)\n* iminuit (CERN-ROOT6 minimizer)\n\nAll packages can be fetched using pip:", "_____no_output_____" ] ], [ [ "!pip3 install --user pandas numpy matplotlib scipy configparser jupyter iminuit", "_____no_output_____" ] ], [ [ "Instead of the regular jupyter environment, one can also use CERN's SWAN service or Google Colab.", "_____no_output_____" ] ], [ [ "google_colab = False\n\nif google_colab:\n try:\n from google.colab import drive\n drive.mount('/content/drive')\n %cd /content/drive/My\\ Drive/Colab/pi-icr/\n except:\n %cd ~/cernbox/Documents/Colab/pi-icr/", "_____no_output_____" ] ], [ [ "### Data files\n\nSpecify, whether the analysis involves one or two states separated in pattern 2 by commenting out the not applicable case in lines 10 or 11. Then enter the file paths for all your data files without the `*.txt` extension. In the following, `ioi` represents the Ion of interest, and `ref` the reference ion.", "_____no_output_____" ] ], [ [ "%config InlineBackend.figure_format ='retina'\n\nimport pandas as pd\nimport numpy as np\nimport matplotlib.pyplot as plt\nimport matplotlib as mpl\nimport pickle, os\n\n# analysis = {'ioi_g': {},'ref': {}}\nanalysis = {'ioi_g': {},'ioi_m': {},'ref': {}}\n\nfiles_ioi_g = ['data/ioi_ground/85Rb_c_000',\n 'data/ioi_ground/85Rb_002',\n 'data/ioi_ground/85Rb_004',\n 'data/ioi_ground/85Rb_006']\n# files_ioi_m = ['data/ioi_isomer/101In_c_000',\n# 'data/ioi_isomer/101In_005']\nfiles_ref = ['data/ref/133Cs_c_000',\n 'data/ref/133Cs_003',\n 'data/ref/133Cs_005',\n 'data/ref/133Cs_007']\n\nlatex_ioi_g = '$^{88}$Rb'\n# latex_ioi_m = '$^{101}$In$^m$'\nlatex_ref = '$^{133}$Cs'", "_____no_output_____" ] ], [ [ "### Load pre-analyzed data from file or reconstruct raw data\n\nAll files are loaded and reconstructed in one big dictionary of dictionaries. It contains besides the positions and timestamps also information about the measurement conditions (excitation frequencies, rounds etc). One can load a whole beamtime at once. Center files must be indicated by a `_c_` in the name (e.g. regular name: `101In_001.txt` $\\rightarrow$ center name `101In_c_000.txt`). All the data is at later stages saved in a `pickle` file. This enables quick loading of the data dictionary without the need of re-reconstructing the data.\n\nThe reconstruction code is parallelized and can be found in the subfolder `bin/reconstruction.py`", "_____no_output_____" ] ], [ [ "from bin.reconstruction import PIICR\npiicr = PIICR()\n\nif os.path.isfile('data/data-save.p'):\n analysis = pickle.load(open('data/data-save.p','rb'))\n print('\\nLoading finished!')\nelse:\n for file in files_ioi_g:\n analysis['ioi_g'].update({file: piicr.prepare(file)})\n if analysis['ioi_m'] != {}:\n for file in files_ioi_m:\n analysis['ioi_m'].update({file: piicr.prepare(file)})\n for file in files_ref:\n analysis['ref'].update({file: piicr.prepare(file)})\n \n print('\\nReconstruction finished!')", "\nLoading finished!\n" ] ], [ [ "### Individual file selection\n\nThe analysis dictionary contains all files. The analysis however is intended to be performed on a file-by-file basis. Please select the individual files here in the variable `file_name`.", "_____no_output_____" ] ], [ [ "# load P1, P2 and C data in panda dataframes for selected file\n\n# file_name = files_ioi_g[1]\n# file_name = files_ioi_m[1]\n# file_name = files_ref[1]\n\nfile_name = files_ioi_g[3]\n\nprint('Selected file:',file_name)\n\nif 'ground' in file_name:\n df_p1 = pd.DataFrame(analysis['ioi_g'][file_name]['p1'], columns=['event','x','y','time'])\n df_p2 = pd.DataFrame(analysis['ioi_g'][file_name]['p2'], columns=['event','x','y','time'])\n df_c = pd.DataFrame(analysis['ioi_g'][file_name.split('_0', 1)[0]+'_c_000']['c'],\n columns=['event','x','y','time'])\nelif 'isomer' in file_name:\n df_p1 = pd.DataFrame(analysis['ioi_m'][file_name]['p1'], columns=['event','x','y','time'])\n df_p2 = pd.DataFrame(analysis['ioi_m'][file_name]['p2'], columns=['event','x','y','time'])\n df_c = pd.DataFrame(analysis['ioi_m'][file_name.split('_0', 1)[0]+'_c_000']['c'],\n columns=['event','x','y','time'])\nelse:\n df_p1 = pd.DataFrame(analysis['ref'][file_name]['p1'], columns=['event','x','y','time'])\n df_p2 = pd.DataFrame(analysis['ref'][file_name]['p2'], columns=['event','x','y','time'])\n df_c = pd.DataFrame(analysis['ref'][file_name.split('_0', 1)[0]+'_c_000']['c'],\n columns=['event','x','y','time'])", "Selected file: data/ioi_ground/85Rb_006\n" ] ], [ [ "### Manual space and time cut\n\nPlease perform a rough manual space cut for each file to improve results on the automatic space cutting tool. This is necessary if one deals with two states in pattern two or if there is a lot of background. This selection will be ellipsoidal. Additionally, please perform a rough time of flight (ToF) cut.", "_____no_output_____" ] ], [ [ "# manual_space_cut = [x_peak_pos, x_peak_spread, y_peak_pos, y_peak_spread]\nmanual_space_cut = {'data/ioi_ground/85Rb_002': [150, 150, 100, 150],\n 'data/ioi_ground/85Rb_004': [150, 150, 100, 150],\n 'data/ioi_ground/85Rb_006': [150, 150, 100, 150],\n 'data/ref/133Cs_003': [120, 150, 80, 150],\n 'data/ref/133Cs_005': [120, 150, 80, 150],\n 'data/ref/133Cs_007': [120, 150, 80, 150]}\n\n# manual_tof_cut = [tof_min, tof_max]\nmanual_tof_cut = [20, 50]\n\n# manual_z_cut <= number of ions in the trap\nmanual_z_cut = 5", "_____no_output_____" ] ], [ [ "### Automatic time and space cuts based on Gaussian distribution\n\nThis section contains all cuts in time and space in different steps. \n1. In the time domain contaminants are removed by fitting a gaussian distribution via maximum likelihood estimation to the largest peak in the ToF spectrum and cutting +/- 5 $\\sigma$ (change cut range in lines 70 & 71). The ToF distribution has to be binned first before the maximum can be found, but the fit is performed on the unbinned data set.\n2. Manual space cut is applied for pattern 1 and pattern 2 (not for the center spot)\n3. Outlyers/wrongly excited ions are removed +/- 3 $\\sigma$ by measures of a simple mean in x and y after applying the manual cut (change cut range in lines).\n4. Ejections with more than `manual_z_cut` number of ions in the trap (without taking into account the detector efficiency) are rejected (= z-class cut)", "_____no_output_____" ] ], [ [ "%config InlineBackend.figure_format ='retina'\n\nimport matplotlib as mpl\nfrom scipy.stats import norm\nfrom iminuit import Minuit\n\n# Utopia LaTeX font with greek letters\nmpl.rc('font', family='serif', serif='Linguistics Pro')\nmpl.rc('text', usetex=False)\nmpl.rc('mathtext', fontset='custom',\n rm='Linguistics Pro',\n it='Linguistics Pro:italic',\n bf='Linguistics Pro:bold')\nmpl.rcParams.update({'font.size': 18})\n\ncol = ['#FFCC00', '#FF2D55', '#00A2FF', '#61D935', 'k', 'grey', 'pink'] # yellow, red, blue, green\n\ndf_list = [df_p1, df_p2, df_c]\npattern = ['p1', 'p2', 'c']\nbin_time_df = [0,0,0] # [p1,p2,c] list of dataframes containing the time-binned data\nresult_t = [0,0,0] # [p1,p2,c] list of MLE fit result dicts\ncut_df = [0,0,0] # [p1,p2,c] list of dataframes containing the time- and space-cut data\nexcludes_df = [0,0,0] # [p1,p2,c] list of dataframes containing the time- and space-cut excluded data\nfig, axes = plt.subplots(nrows=3, ncols=2, figsize=(12, 15))\n\nfor df_nr in range(len(df_list)):\n \n \n ##############################\n ### BINNING, FITTING TOF DISTR\n ##############################\n \n \n bin_time_df[df_nr] = pd.DataFrame(pd.value_counts(pd.cut(df_list[df_nr].time, bins=np.arange(manual_tof_cut[0], manual_tof_cut[1],0.02))).sort_index()).rename(index=str, columns={'time': 'counts'}).reset_index(drop=True)\n bin_time_df[df_nr]['time'] = np.arange(manual_tof_cut[0]+0.01,manual_tof_cut[1]-0.01,0.02)\n\n # fit gaussian to time distribution using unbinned maximum likelihood estimation\n def NLL_1D(mean, sig):\n '''Negative log likelihood function for (n=1)-dimensional Gaussian distribution.'''\n return( -np.sum(norm.logpdf(x=data_t,\n loc=mean,\n scale=sig)) )\n\n def Start_Par(data):\n '''Starting parameter based on simple mean of 1D numpy array.'''\n return(np.array([data.mean(), # meanx\n data.std()])) #rho\n \n # minimize negative log likelihood function first for the symmetric case\n data_t = df_list[df_nr][(df_list[df_nr].time > bin_time_df[df_nr].time[bin_time_df[df_nr].counts.idxmax()] - 1.0) &\n (df_list[df_nr].time < bin_time_df[df_nr].time[bin_time_df[df_nr].counts.idxmax()] + 1.0)].time.to_numpy()\n\n result_t[df_nr] = Minuit(NLL_1D, mean=Start_Par(data_t)[0], sig=Start_Par(data_t)[1])\n result_t[df_nr].errors = (0.1, 0.1) # initital step size\n result_t[df_nr].limits =[(None, None), (None, None)] # fit ranges\n result_t[df_nr].errordef = Minuit.LIKELIHOOD # MLE definition (instead of Minuit.LEAST_SQUARES)\n result_t[df_nr].migrad() # finds minimum of mle function\n result_t[df_nr].hesse() # computes errors\n for p in result_t[df_nr].parameters:\n print(\"{} = {:3.5f} +/- {:3.5f}\".format(p, result_t[df_nr].values[p], result_t[df_nr].errors[p]))\n \n\n ##############################\n ### VISUALIZE TOF DISTRIBUTION # kind='bar' is VERY time consuming -> use kind='line' instead!\n ##############################\n \n \n # whole distribution\n bin_time_df[df_nr].plot(x='time', y='counts', kind='line', xticks=np.arange(manual_tof_cut[0],manual_tof_cut[1]+1,5), ax=axes[df_nr,0])\n\n # reduced peak plus fit\n bin_time_df[df_nr][bin_time_df[df_nr].counts.idxmax()-50:bin_time_df[df_nr].counts.idxmax()+50].plot(x='time', y='counts', kind='line', ax=axes[df_nr,1])\n pdf_x = np.arange(bin_time_df[df_nr].time[bin_time_df[df_nr].counts.idxmax()-50],\n bin_time_df[df_nr].time[bin_time_df[df_nr].counts.idxmax()+51],\n (bin_time_df[df_nr].time[bin_time_df[df_nr].counts.idxmax()+51]\n -bin_time_df[df_nr].time[bin_time_df[df_nr].counts.idxmax()-50])/100)\n pdf_y = norm.pdf(pdf_x, result_t[df_nr].values['mean'], result_t[df_nr].values['sig'])\n axes[df_nr,0].plot(pdf_x, pdf_y/pdf_y.max()*bin_time_df[df_nr].counts.max(), 'r', label='PDF')\n axes[df_nr,1].plot(pdf_x, pdf_y/pdf_y.max()*bin_time_df[df_nr].counts.max(), 'r', label='PDF')\n\n # mark events in t that will be cut away (+/- 3 sigma = 99.73% of data)\n bin_time_df[df_nr][(bin_time_df[df_nr].time < result_t[df_nr].values['mean'] - 3*result_t[df_nr].values['sig']) |\n (bin_time_df[df_nr].time > result_t[df_nr].values['mean'] + 3*result_t[df_nr].values['sig'])].plot(x='time', y='counts', kind='scatter', ax=axes[df_nr,0], c='y', marker='x', s=50, label='excluded')\n bin_time_df[df_nr][(bin_time_df[df_nr].time < result_t[df_nr].values['mean'] - 3*result_t[df_nr].values['sig']) |\n (bin_time_df[df_nr].time > result_t[df_nr].values['mean'] + 3*result_t[df_nr].values['sig'])].plot(x='time', y='counts', kind='scatter', ax=axes[df_nr,1], c='y', marker='x', s=50, label='excluded')\n \n # legend title shows total number of events and reduced number of events\n axes[df_nr,0].legend(title='total: {}'.format(bin_time_df[df_nr].counts.sum()),loc='upper right', fontsize=16)\n axes[df_nr,1].legend(title='considered: {}'.format(bin_time_df[df_nr].counts.sum()-bin_time_df[df_nr][(bin_time_df[df_nr].time < result_t[df_nr].values['mean'] - 3*result_t[df_nr].values['sig']) |\n (bin_time_df[df_nr].time > result_t[df_nr].values['mean'] + 3*result_t[df_nr].values['sig'])].counts.sum()),loc='upper left', fontsize=16)\n\n \n ##############################\n ### APPYING ALL CUTS\n ##############################\n \n \n # cutting in t: mean +/- 5 sigma\n cut_df[df_nr] = df_list[df_nr][(df_list[df_nr].time > (result_t[df_nr].values['mean'] - 5*result_t[df_nr].values['sig']))&\n (df_list[df_nr].time < (result_t[df_nr].values['mean'] + 5*result_t[df_nr].values['sig']))]\n len1 = cut_df[df_nr].shape[0]\n \n # applying manual cut in x and y:\n if df_nr < 2: # only for p1 and p2, not for c \n cut_df[df_nr] = cut_df[df_nr][((cut_df[df_nr].x-manual_space_cut[file_name][0])**2 + (cut_df[df_nr].y-manual_space_cut[file_name][2])**2) <\n manual_space_cut[file_name][1]*manual_space_cut[file_name][3]]\n len2 = cut_df[df_nr].shape[0]\n\n # applyig automatic cut in x and y: mean +/- 3 std in an ellipsoidal cut\n cut_df[df_nr] = cut_df[df_nr][((cut_df[df_nr].x-cut_df[df_nr].x.mean())**2 + (cut_df[df_nr].y-cut_df[df_nr].y.mean())**2) <\n 3*cut_df[df_nr].x.std()*3*cut_df[df_nr].y.std()]\n len3 = cut_df[df_nr].shape[0]\n\n # applying automatic z-class-cut (= cut by number of ions per event) for z>5 ions per event to reduce space-charge effects:\n cut_df[df_nr] = cut_df[df_nr][cut_df[df_nr].event.isin(cut_df[df_nr].event.value_counts()[cut_df[df_nr].event.value_counts() <= 6].index)]\n \n # printing the reduction of the number of ions per file in each of the cut steps\n print('\\n{}: data size: {} -> time cut: {} -> manual space cut: {} -> automatic space cut: {} -> z-class-cut: {}\\n'.format(pattern[df_nr], df_list[df_nr].shape[0], len1, len2, len3, cut_df[df_nr].shape[0]))\n\n # saves excluded data (allows visual checking later)\n excludes_df[df_nr] = pd.concat([df_list[df_nr], cut_df[df_nr]]).drop_duplicates(keep=False).reset_index(drop=True)\n\nplt.savefig('{}-tof.pdf'.format(file_name))\nplt.show()", "mean = 25.54630 +/- 0.01769\nsig = 0.27570 +/- 0.01251\n\np1: data size: 248 -> time cut: 244 -> manual space cut: 239 -> automatic space cut: 235 -> z-class-cut: 235\n\nmean = 25.55908 +/- 0.01934\nsig = 0.32012 +/- 0.01367\n\np2: data size: 283 -> time cut: 281 -> manual space cut: 266 -> automatic space cut: 265 -> z-class-cut: 265\n\nmean = 25.54692 +/- 0.01284\nsig = 0.27694 +/- 0.00908\n\nc: data size: 480 -> time cut: 477 -> manual space cut: 477 -> automatic space cut: 474 -> z-class-cut: 359\n\n" ] ], [ [ "### Spot fitting\n\n2D multivariate gaussian maximum likelihood estimations of the cleaned pattern 1, pattern 2 and center spot positions are performed SciPy PDF's and ROOT's minimizer. Displayed are all uncut data with a blue-transparent point. This allows displaying a density of points by the shade of blue without the need of binning the data (= reducing the information; also: binning is much more time-consuming). The cut data is displayed with a black \"x\" at the position of the blue point. These points are not considered in the fit (represented by the red (6-$\\sigma$ band) but allow for an additional check of the cutting functions. The scale of the MCP-position plots is given in the time unit of the position-sensitive MCP data. There is no need in converting it into a mm-unit since one is only interested in the angle. ", "_____no_output_____" ] ], [ [ "%config InlineBackend.figure_format ='retina'\n# activate interactive matplotlib plot -> uncomment line below!\n# %matplotlib notebook\n\nimport pickle, os\nfrom scipy.stats import multivariate_normal, linregress, pearsonr\nfrom scipy.optimize import minimize\nimport numpy as np\nfrom iminuit import Minuit\n\n# open preanalyzed dataset if existing\nif os.path.isfile('data/data-save.p'):\n analysis = pickle.load(open('data/data-save.p','rb'))\n\ndf_list = [df_p1, df_p2, df_c]\nresult = [{},{},{}]\nroot_res = [0,0,0]\nparameters = ['meanx', 'meany', 'sigx', 'sigy', 'theta']\nfig2, axes2 = plt.subplots(nrows=3, ncols=1, figsize=(7.5, 20))\npiicr_scheme_names = ['p1','p2','c']\n\n\n##############################\n### Prepare maximum likelihood estimation\n##############################\n\n\ndef Rot(theta):\n '''Rotation (matrix) of angle theta to cartesian coordinates.'''\n return np.array([[np.cos(theta), -np.sin(theta)],\n [np.sin(theta), np.cos(theta)]])\n\ndef NLL_2D(meanx, meany, sigx, sigy, theta):\n '''Negative log likelihood function for (n=2)-dimensional Gaussian distribution for Minuit.'''\n cov = Rot(theta) @ np.array([[np.power(sigx,2),0],[0,np.power(sigy,2)]]) @ Rot(theta).T\n return( -np.sum(multivariate_normal.logpdf(x=data,\n mean=np.array([meanx, meany]),\n cov=cov,\n allow_singular=True)) )\n\ndef NLL_2D_scipy(param):\n '''Negative log likelihood function for (n=2)-dimensional Gaussian distribution for SciPy.'''\n meanx, meany, sigx, sigy, theta = param\n cov = Rot(theta) @ np.array([[np.power(sigx,2),0],[0,np.power(sigy,2)]]) @ Rot(theta).T\n return( -np.sum(multivariate_normal.logpdf(x=data,\n mean=np.array([meanx, meany]),\n cov=cov,\n allow_singular=True)) )\n\ndef Start_Par(data):\n '''Starting parameter based on simple linear regression and 2D numpy array.'''\n # simple linear regression to guess the rotation angle based on slope\n slope, intercept, r_value, p_value, std_err = linregress(data[:, 0], data[:, 1])\n theta_guess = -np.arctan(slope)\n\n # data rotated based on theta guess\n data_rotated_guess = np.dot(Rot(theta_guess), [data[:,0], data[:,1]])\n \n first_guess = np.array([data[:,0].mean()+0.2, # meanx\n data[:,1].mean()+0.2, # meany\n data_rotated_guess[1].std(), # sigma-x\n data_rotated_guess[0].std(), # sigma-y\n theta_guess]) # rot. angle based on slope of lin. reg.\n \n # based on a first guess, a minimization based on a robust simplex is performed\n start_par = minimize(NLL_2D_scipy, first_guess, method='Nelder-Mead')\n return(start_par['x'])\n\n\n##############################\n### Fitting and visualization of P1, P2, C\n##############################\n\n\nfor df_nr in range(len(df_list)):\n \n # minimize negative log likelihood function first for the symmetric case\n data = cut_df[df_nr][['x', 'y']].to_numpy()\n root_res[df_nr] = Minuit(NLL_2D, meanx=Start_Par(data)[0], meany=Start_Par(data)[1],\n sigx=Start_Par(data)[2], sigy=Start_Par(data)[3],\n theta=Start_Par(data)[4])\n root_res[df_nr].errors = (0.1, 0.1, 0.1, 0.1, 0.1) # initital step size\n root_res[df_nr].limits =[(None, None), (None, None), (None, None), (None, None), (None, None)] # fit ranges\n root_res[df_nr].errordef = Minuit.LIKELIHOOD # MLE definition (instead of Minuit.LEAST_SQUARES)\n root_res[df_nr].migrad() # finds minimum of mle function\n root_res[df_nr].hesse() # computes errors\n \n # plotting of data, excluded data, reference MCP circle, and fit results\n axes2[df_nr].plot(df_list[df_nr].x.to_numpy(),df_list[df_nr].y.to_numpy(),'o',alpha=0.15,label='data',zorder=0)\n axes2[df_nr].plot(excludes_df[df_nr].x.to_numpy(), excludes_df[df_nr].y.to_numpy(), 'x k',\n label='excluded data',zorder=1)\n\n mcp_circ = mpl.patches.Ellipse((0,0), 1500, 1500, edgecolor='k', fc='None', lw=2)\n axes2[df_nr].add_patch(mcp_circ)\n axes2[df_nr].scatter(root_res[df_nr].values['meanx'], root_res[df_nr].values['meany'], marker='o', color=col[1], linewidth=0, zorder=2)\n sig = mpl.patches.Ellipse((root_res[df_nr].values['meanx'], root_res[df_nr].values['meany']),\n 3*root_res[df_nr].values['sigx'], 3*root_res[df_nr].values['sigy'],\n np.degrees(root_res[df_nr].values['theta']),\n edgecolor=col[1], fc='None', lw=2, label='6-$\\sigma$ band (fit)', zorder=2)\n axes2[df_nr].add_patch(sig)\n\n\n axes2[df_nr].legend(title='fit(x) = {:1.0f}({:1.0f})\\nfit(y) = {:1.0f}({:1.0f})'.format(root_res[df_nr].values['meanx'],root_res[df_nr].errors['meanx'],\n root_res[df_nr].values['meany'],root_res[df_nr].errors['meany']),\n loc='lower left', fontsize=14)\n axes2[df_nr].axis([-750,750,-750,750])\n axes2[df_nr].grid(True)\n axes2[df_nr].text(-730, 660, '{}: {}'.format(file_name.split('/',1)[-1], piicr_scheme_names[df_nr]))\n plt.tight_layout()\n\n \n # save fit information for each parameter:\n # 'parameter': [fitresult, fiterror, Hesse-covariance matrix]\n for i in range(len(parameters)):\n result[df_nr].update({'{}'.format(parameters[i]): [np.array(root_res[df_nr].values)[i],\n np.array(root_res[df_nr].errors)[i],\n root_res[df_nr].covariance]})\n if 'ground' in file_name:\n analysis['ioi_g'][file_name]['fit-{}'.format(piicr_scheme_names[df_nr])] = result[df_nr]\n elif 'isomer' in file_name:\n analysis['ioi_m'][file_name]['fit-{}'.format(piicr_scheme_names[df_nr])] = result[df_nr]\n else:\n analysis['ref'][file_name]['fit-{}'.format(piicr_scheme_names[df_nr])] = result[df_nr]\n\nplt.savefig('{}-fit.pdf'.format(file_name))\nplt.show()\n\n# save all data using pickle\npickle.dump(analysis, open('data/data-save.p','wb'))", "'LinguisticsPro-Italic.otf' can not be subsetted into a Type 3 font. The entire font will be embedded in the output.\n'LinguisticsPro-Regular.otf' can not be subsetted into a Type 3 font. The entire font will be embedded in the output.\n" ] ], [ [ "\n---\n# !!! <font color='red'>REPEAT</font> CODE ABOVE FOR ALL INDIVIDUAL FILES !!!\n---\n<br>\n<br>", "_____no_output_____" ], [ "### Save fit data to dataframe and *.csv file\n\n<br>Continue here after analyzing all files individually. The following command saves all necessary data and fit information in a `*.csv` file.", "_____no_output_____" ] ], [ [ "calc_df = pd.DataFrame()\nfor key in analysis.keys():\n for subkey in analysis[key].keys():\n if '_c_' not in subkey:\n calc_df = calc_df.append(pd.DataFrame({'file': subkey,\n 'p1_x': analysis[key][subkey]['fit-p1']['meanx'][0],\n 'p1_y': analysis[key][subkey]['fit-p1']['meany'][0],\n 'p2_x': analysis[key][subkey]['fit-p2']['meanx'][0],\n 'p2_y': analysis[key][subkey]['fit-p2']['meany'][0],\n 'c_x': analysis[key][subkey]['fit-c']['meanx'][0],\n 'c_y': analysis[key][subkey]['fit-c']['meany'][0],\n 'p1_x_unc': analysis[key][subkey]['fit-p1']['meanx'][1],\n 'p1_y_unc': analysis[key][subkey]['fit-p1']['meany'][1],\n 'p2_x_unc': analysis[key][subkey]['fit-p2']['meanx'][1],\n 'p2_y_unc': analysis[key][subkey]['fit-p2']['meany'][1],\n 'c_x_unc': analysis[key][subkey]['fit-c']['meanx'][1],\n 'c_y_unc': analysis[key][subkey]['fit-c']['meany'][1],\n 'cyc_freq_guess': analysis[key][subkey]['cyc_freq'],\n 'red_cyc_freq': analysis[key][subkey]['red_cyc_freq'],\n 'mag_freq': analysis[key][subkey]['mag_freq'],\n 'cyc_acc_time': analysis[key][subkey]['cyc_acc_time'],\n 'n_acc': analysis[key][subkey]['n_acc'],\n 'time_start': pd.to_datetime('{} {}'.format(analysis[key][subkey]['time-info'][0], analysis[key][subkey]['time-info'][1]), format='%m/%d/%Y %H:%M:%S', errors='ignore'),\n 'time_end': pd.to_datetime('{} {}'.format(analysis[key][subkey]['time-info'][2], analysis[key][subkey]['time-info'][3]), format='%m/%d/%Y %H:%M:%S', errors='ignore')}, index=[0]), ignore_index=True)\ncalc_df.to_csv('data/analysis-summary.csv')\ncalc_df", "_____no_output_____" ] ], [ [ "### Calculate $\\nu_c$ from position fits\n\n[1]: https://doi.org/10.1007/s00340-013-5621-0\n[2]: https://doi.org/10.1103/PhysRevLett.110.082501\n[3]: https://doi.org/10.1007/s10751-019-1601-z\n\nCan be run independently from everything above by loading the `analysis-summary.csv` file!<br> A detailed description of the $\\nu_c$ calculation can be found in Ref. [1], [2] and [3].", "_____no_output_____" ] ], [ [ "import pandas as pd\nimport numpy as np\n\n# load fit-data file, datetime has to be converted\ncalc_df = pd.read_csv('data/analysis-summary.csv', header=0, index_col=0)\n\n# calculate angle between the P1-vector (P1_x/y - C_x/y) and the P2-vector (P2_x/y - C_x/y)\ncalc_df['p1p2_angle'] = np.arctan2(calc_df.p1_y - calc_df.c_y, calc_df.p1_x - calc_df.c_x) \\\n - np.arctan2(calc_df.p2_y - calc_df.c_y, calc_df.p2_x - calc_df.c_x)\n \n# calculate the uncertainty on the angle between the P1/P2 vectors\n# see https://en.wikipedia.org/wiki/Atan2\ncalc_df['p1p2_angle_unc'] = np.sqrt(\n ( calc_df.p1_x_unc * (calc_df.c_y - calc_df.p1_y) / ( (calc_df.p1_x - calc_df.c_x)**2 + (calc_df.p1_y - calc_df.c_y)**2 ) )**2\n + ( calc_df.p1_y_unc * (calc_df.p1_x - calc_df.c_x) / ( (calc_df.p1_x - calc_df.c_x)**2 + (calc_df.p1_y - calc_df.c_y)**2 ) )**2\n + ( calc_df.p2_x_unc * (calc_df.c_y - calc_df.p2_y) / ( (calc_df.p2_x - calc_df.c_x)**2 + (calc_df.p2_y - calc_df.c_y)**2 ) )**2\n + ( calc_df.p2_y_unc * (calc_df.p2_x - calc_df.c_x) / ( (calc_df.p2_x - calc_df.c_x)**2 + (calc_df.p2_y - calc_df.c_y)**2 ) )**2\n + ( calc_df.c_x_unc *\n ( -(calc_df.c_y - calc_df.p1_y) / ( (calc_df.p1_x - calc_df.c_x)**2 + (calc_df.p1_y - calc_df.c_y)**2 ) \n -(calc_df.c_y - calc_df.p2_y) / ( (calc_df.p2_x - calc_df.c_x)**2 + (calc_df.p2_y - calc_df.c_y)**2 ) ) )**2\n + ( calc_df.c_y_unc *\n ( (calc_df.p1_x - calc_df.c_x) / ( (calc_df.p1_x - calc_df.c_x)**2 + (calc_df.p1_y - calc_df.c_y)**2 ) \n +(calc_df.p2_x - calc_df.c_x) / ( (calc_df.p2_x - calc_df.c_x)**2 + (calc_df.p2_y - calc_df.c_y)**2 ) ) )**2 )\n\n# calculate cyc freq: total phase devided by total time\ncalc_df['cyc_freq'] = (calc_df.p1p2_angle + 2*np.pi * calc_df.n_acc) / (2*np.pi * calc_df.cyc_acc_time * 0.000001)\ncalc_df['cyc_freq_unc'] = calc_df.p1p2_angle_unc / (2*np.pi * calc_df.cyc_acc_time * 0.000001)\n\ncalc_df.to_csv('data/analysis-summary.csv')", "_____no_output_____" ], [ "calc_df.head()", "_____no_output_____" ] ], [ [ "### Frequency-ratio calculation\n\n[1]: https://doi.org/10.1007/s00340-013-5621-0\n[2]: https://doi.org/10.1103/PhysRevLett.110.082501\n[3]: https://doi.org/10.1007/s10751-019-1601-z\n\nIn order to determine the frequency ratio between the ioi and the ref, simultaneous fits of all for the data set possible polynomial degrees are performed. The code calculates the reduced $\\chi^2_{red}$ for each fit and returns only the one with a $\\chi^2_{red}$ closest to 1. A detailed description of the procedure can be found in Ref. [3]. If problems in the fitting occur, please try to vary the starting parameter section in lines 125-135 of `~/bin/freq_ratio.py`", "_____no_output_____" ] ], [ [ "import pandas as pd\nimport numpy as np\nfrom bin.freq_ratio import Freq_ratio\nfreq = Freq_ratio()\n\n# load fit-data file\ncalc_df = pd.read_csv('data/analysis-summary.csv', header=0, index_col=0)\n\n# save average time of measurement: t_start+(t_end-t_start)/2\ncalc_df.time_start = pd.to_datetime(calc_df.time_start)\ncalc_df.time_end = pd.to_datetime(calc_df.time_end)\ncalc_df['time'] = calc_df.time_start + (calc_df.time_end - calc_df.time_start)/2\ncalc_df.to_csv('data/analysis-summary.csv')\n\n# convert avg.time to difference in minutes from first measurement -> allows fitting with small number as x value\ncalc_df['time_delta'] = ((calc_df['time']-calc_df['time'].min())/np.timedelta64(1, 's')/60)\n\n# selecting data for isotopes\ndf_ioi_g = calc_df[calc_df.file.str.contains('ground')][['time_delta','cyc_freq','cyc_freq_unc','time','file']]\ndf_ioi_m = calc_df[calc_df.file.str.contains('isomer')][['time_delta','cyc_freq','cyc_freq_unc','time','file']]\n\n# allows to define a subset of reference frequencies for ground and isomer\ndf_ref_g = calc_df[calc_df.file.str.contains('ref')][['time_delta','cyc_freq','cyc_freq_unc','time','file']]\ndf_ref_m = calc_df[calc_df.file.str.contains('ref')][['time_delta','cyc_freq','cyc_freq_unc','time','file']]\n\n# simultaneous polynomial fit, see https://doi.org/10.1007/s10751-019-1601-z\nfit1, fit2, ratio1, ratio_unc1, chi_sq1 = freq.ratio_sim_fit(['ref', 'ioi_g'],\n df_ref_g.time_delta.tolist(),\n df_ref_g.cyc_freq.tolist(),\n df_ref_g.cyc_freq_unc.tolist(),\n df_ioi_g.time_delta.tolist(),\n df_ioi_g.cyc_freq.tolist(),\n df_ioi_g.cyc_freq_unc.tolist())\nif len(df_ioi_m) > 0:\n fit3, fit4, ratio2, ratio_unc2, chi_sq2 = freq.ratio_sim_fit(['ref', 'ioi_m'],\n df_ref_m.time_delta.tolist(),\n df_ref_m.cyc_freq.tolist(),\n df_ref_m.cyc_freq_unc.tolist(),\n df_ioi_m.time_delta.tolist(),\n df_ioi_m.cyc_freq.tolist(),\n df_ioi_m.cyc_freq_unc.tolist())", "[-10, 685181.1403450029, 1]\nPoly-degree: 2\n\nRed.Chi.Sq.: 0.6548141372439115\nCorellation: 2.7799756249708774e-12\nRatio fit parameter: 0.6388872125439365 +/- 2.1557485311110636e-09\n[1, -10, 685181.1403450029, 1]\nPoly-degree: 3\n\nRed.Chi.Sq.: 0.7550362063128235\n" ] ], [ [ "### Frequency-ratio plotting", "_____no_output_____" ] ], [ [ "%config InlineBackend.figure_format ='retina'\nimport matplotlib.pyplot as plt\nimport matplotlib as mpl\nimport pandas as pd\nimport numpy as np\nmpl.rc('font', family='serif', serif='Linguistics Pro') # open source Utopia LaTeX font with greek letters\nmpl.rc('text', usetex=False)\nmpl.rc('mathtext', fontset='custom',\n rm='Linguistics Pro',\n it='Linguistics Pro:italic',\n bf='Linguistics Pro:bold')\nmpl.rcParams.update({'font.size': 18})\n\n# prepare fit data\nx1 = np.linspace(min([df_ioi_g.time_delta.min(),df_ref_g.time_delta.min()]),max([df_ioi_g.time_delta.max(),df_ref_g.time_delta.max()]),500)\nt1 = pd.date_range(pd.Series([df_ioi_g.time.min(),df_ref_g.time.min()]).min(),pd.Series([df_ioi_g.time.max(),df_ref_g.time.max()]).max(),periods=500)\nif len(df_ioi_m) > 0:\n x2 = np.linspace(min([df_ioi_m.time_delta.min(),df_ref_m.time_delta.min()]),max([df_ioi_m.time_delta.max(),df_ref_m.time_delta.max()]),500)\n t2 = pd.date_range(pd.Series([df_ioi_m.time.min(),df_ref_m.time.min()]).min(),pd.Series([df_ioi_m.time.max(),df_ref_m.time.max()]).max(),periods=500)\n\nfit1_y = [np.polyval(fit1, i) for i in x1]\nfit2_y = [np.polyval(fit2, i) for i in x1]\nif len(df_ioi_m) > 0:\n fit3_y = [np.polyval(fit3, i) for i in x2]\n fit4_y = [np.polyval(fit4, i) for i in x2]\n\n\n#########################\n### PLOTTING ground state\n#########################\n\nif len(df_ioi_m) > 0:\n fig, (ax1, ax3) = plt.subplots(figsize=(9,12),nrows=2, ncols=1)\nelse:\n fig, ax1 = plt.subplots(figsize=(9,6),nrows=1, ncols=1)\n\nax1.errorbar(df_ref_g.time, df_ref_g.cyc_freq, yerr=df_ref_g.cyc_freq_unc, fmt='o', label='{}'.format(latex_ref), marker='d', c='#1E77B4', ms=10, elinewidth=2.5)\nax1.set_xlabel('Time', fontsize=24, fontweight='bold')\n# Make the y-axis label, ticks and tick labels match the line color.\nax1.set_ylabel('Frequency (Hz)', fontsize=24, fontweight='bold')\nax1.tick_params('y', colors='#1E77B4')\nax1.plot(t1, fit1_y, ls=(5.5, (5, 1, 1, 1, 1, 1, 1, 1)),c='#1E77B4', label='poly-fit')\n# Allowing two axes in one subplot\nax2 = ax1.twinx()\nax2.errorbar(df_ioi_g.time, df_ioi_g.cyc_freq, yerr=df_ioi_g.cyc_freq_unc, fmt='o', color='#D62728', label='{}'.format(latex_ioi_g), fillstyle='none', ms=10, elinewidth=2.5) # green: #2ca02c\nax2.tick_params('y', colors='#D62728')\nax2.plot(t1, fit2_y, ls=(0, (5, 3, 1, 3)),c='#D62728', label='poly-fit')\n\n# adjust the y axes to be the same height\nmiddle_y1 = df_ref_g.cyc_freq.min() + (df_ref_g.cyc_freq.max() - df_ref_g.cyc_freq.min())/2\nmiddle_y2 = df_ioi_g.cyc_freq.min() + (df_ioi_g.cyc_freq.max() - df_ioi_g.cyc_freq.min())/2\nrange_y1 = df_ref_g.cyc_freq.max() - df_ref_g.cyc_freq.min() + 2 * df_ref_g.cyc_freq_unc.max()\nrange_y2 = df_ioi_g.cyc_freq.max() - df_ioi_g.cyc_freq.min() + 2 * df_ioi_g.cyc_freq_unc.max()\nax1.set_ylim(middle_y1 - 1.3 * max([range_y1, middle_y1*range_y2/middle_y2])/2, middle_y1 + 1.1 * max([range_y1, middle_y1*range_y2/middle_y2])/2) # outliers only\nax2.set_ylim(middle_y2 - 1.1 * max([middle_y2*range_y1/middle_y1, range_y2])/2, middle_y2 + 1.3 * max([middle_y2*range_y1/middle_y1, range_y2])/2) # most of the data\n\n# plotting only hours without the date\nax2.xaxis.set_major_formatter(mpl.dates.DateFormatter('%H:%M'))\nax2.xaxis.set_minor_locator(mpl.dates.HourLocator())\n\nhandles1, labels1 = ax1.get_legend_handles_labels()\nhandles2, labels2 = ax2.get_legend_handles_labels()\nhandles_g = [handles1[1], handles2[1], (handles1[0], handles2[0])]\nlabels_g = [labels1[1], labels2[1], labels1[0]]\nplt.legend(handles=handles_g, labels=labels_g,fontsize=18,title='Ratio: {:1.10f}\\n $\\\\pm${:1.10f}'.format(ratio1, ratio_unc1), loc='upper right')\nplt.text(0.03,0.03,'poly-{}: $\\chi^2_{{red}}$ {:3.2f}'.format(len(fit1)-1, chi_sq1),transform=ax1.transAxes)\n\n\n###########################\n### PLOTTING isomeric state\n###########################\n\nif len(df_ioi_m) > 0:\n ax3.errorbar(df_ref_m.time, df_ref_m.cyc_freq, yerr=df_ref_m.cyc_freq_unc, fmt='o', label='{}'.format(latex_ref), marker='d', c='#1E77B4', ms=10, elinewidth=2.5)\n ax3.set_xlabel('Time', fontsize=24, fontweight='bold')\n # Make the y-axis label, ticks and tick labels match the line color.\n ax3.set_ylabel('Frequency (Hz)', fontsize=24, fontweight='bold')\n ax3.tick_params('y', colors='#1E77B4')\n ax3.plot(t2, fit3_y, ls=(5.5, (5, 1, 1, 1, 1, 1, 1, 1)),c='#1E77B4', label='poly-fit')\n # Allowing two axes in one subplot\n ax4 = ax3.twinx()\n ax4.errorbar(df_ioi_m.time, df_ioi_m.cyc_freq, yerr=df_ioi_m.cyc_freq_unc, fmt='o', color='#D62728', label='{}'.format(latex_ioi_m), fillstyle='none', ms=10, elinewidth=2.5) # green: #2ca02c\n ax4.tick_params('y', colors='#D62728')\n ax4.plot(t2, fit4_y, ls=(0, (5, 3, 1, 3)),c='#D62728', label='poly-fit')\n\n # adjust the y axes to be the same height\n middle_y3 = df_ref_m.cyc_freq.min() + (df_ref_m.cyc_freq.max() - df_ref_m.cyc_freq.min())/2\n middle_y4 = df_ioi_m.cyc_freq.min() + (df_ioi_m.cyc_freq.max() - df_ioi_m.cyc_freq.min())/2\n range_y3 = df_ref_m.cyc_freq.max() - df_ref_m.cyc_freq.min() + 2 * df_ref_m.cyc_freq_unc.max()\n range_y4 = df_ioi_m.cyc_freq.max() - df_ioi_m.cyc_freq.min() + 2 * df_ioi_m.cyc_freq_unc.max()\n ax3.set_ylim(middle_y3 - 1.3 * max([range_y3, middle_y3*range_y4/middle_y4])/2, middle_y3 + 1.1 * max([range_y3, middle_y3*range_y4/middle_y4])/2) # outliers only\n ax4.set_ylim(middle_y4 - 1.1 * max([middle_y4*range_y3/middle_y3, range_y4])/2, middle_y4 + 1.3 * max([middle_y4*range_y3/middle_y3, range_y4])/2) # most of the data\n\n # plotting only hours without the date\n ax4.xaxis.set_major_formatter(mpl.dates.DateFormatter('%H:%M'))\n ax4.xaxis.set_minor_locator(mpl.dates.HourLocator())\n\n handles3, labels3 = ax3.get_legend_handles_labels()\n handles4, labels4 = ax4.get_legend_handles_labels()\n handles_m = [handles3[1], handles4[1], (handles3[0], handles4[0])]\n labels_m = [labels3[1], labels4[1], labels3[0]]\n plt.legend(handles=handles_m, labels=labels_m, fontsize=18,title='Ratio: {:1.10f}\\n $\\\\pm${:1.10f}'.format(ratio2, ratio_unc2), loc='upper right')\n plt.text(0.03,0.03,'poly-{}: $\\chi^2_{{red}}$ {:3.2f}'.format(len(fit3)-1, chi_sq2),transform=ax3.transAxes)\n\nplt.tight_layout()\nplt.savefig('data/freq-ratios.pdf')\nplt.show()", "'LinguisticsPro-Bold.otf' can not be subsetted into a Type 3 font. The entire font will be embedded in the output.\n'LinguisticsPro-Italic.otf' can not be subsetted into a Type 3 font. The entire font will be embedded in the output.\n'LinguisticsPro-Regular.otf' can not be subsetted into a Type 3 font. The entire font will be embedded in the output.\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" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Introdution to Jupyter Notebooks and Text Processing in Python\nThis 'document' is a Jupyter notebook. It allows you to combine explanatory **text** and **code** that executes to produce results you can see on the same page.\n\n## Notebook Basics\n\n### Text cells\n\nThe box this text is written in is called a *cell*. It is a *text cell* written in a very simple markup language called 'Markdown'. Here is a useful [Markdown cheatsheet](https://github.com/adam-p/markdown-here/wiki/Markdown-Cheatsheet). You can edit and then run cells to produce a result. Running this text cell produces formatted text.\n\n### Code cells\n\nThe other main kind of cell is a *code cell*. The cell immediately below this one is a code cell. Running a code cell runs the code in the cell and produces a result.", "_____no_output_____" ] ], [ [ "# This is a comment in a code cell. Comments start with a # symbol. They are ignored and do not do anything.", "_____no_output_____" ], [ "# This box is a code cell. When this cell is run, the code below will execute and produce a result\n3 + 4", "_____no_output_____" ] ], [ [ "## Simple String Manipulation in Python\nThis section introduces some very basic things you can do in Python to create and manipulate *strings*. A string is a simple sequence of characters, like `flabbergast`. This introduction is limited to those things that may be useful to know in order to understand the *Bughunt!* data mining in the following two notebooks.", "_____no_output_____" ], [ "### Creating and Storing Strings in Variables\nStrings are simple to create in Python. You can simply write some characters in quote marks.", "_____no_output_____" ] ], [ [ "'Butterflies are important as pollinators.'", "_____no_output_____" ] ], [ [ "In order to do something useful with this string, other than print it out, we need to store in a *variable* by using the assignment operator `=` (equals sign). Whatever is on the right-hand side of the `=` is stored into a variable with the name on the left-hand side.", "_____no_output_____" ] ], [ [ "# my_variable is the variable on the left\n# 'manuscripts' is the string on the right that is stored in the variable my_variable\n\nmy_variable = 'Butterflies are important as pollinators.'", "_____no_output_____" ] ], [ [ "Notice that nothing is printing to the screen. That's because the string is stored in the variable `my_variable`. In order to see what is inside the variable `my_variable` we can simply write `my_variable` in a code cell, run it, and the interpreter will print it out for us.", "_____no_output_____" ] ], [ [ "my_variable", "_____no_output_____" ] ], [ [ "### Manipulating Bits of Strings\n\n#### Accessing Individual Characters\nA strings is just a sequence (or list) of characters. You can access **individual characters** in a string by specifying which ones you want in square brackets. If you want the first character you specify `1`.", "_____no_output_____" ] ], [ [ "my_variable[1]", "_____no_output_____" ] ], [ [ "Hang on a minute! Why did it give us `u` instead of `B`?\n\nIn programming, everything tends to be *zero indexed*, which means that things are counted from 0 rather than 1. Thus, in the example above, `1` gives us the *second* character in the string.\n\nIf you want the first character in the string, you need to specify the index `0`! ", "_____no_output_____" ] ], [ [ "my_variable[0]", "_____no_output_____" ] ], [ [ "#### Accessing a Range of Characters\n\nYou can also pick out a **range of characters** from within a string, by giving the *start index* followed by the *end index* with a semi-colon (`:`) in between.\n\nThe example below gives us the character at index `0` all the way up to, *but not including*, the character at index `20`.", "_____no_output_____" ] ], [ [ "my_variable[0:20]", "_____no_output_____" ] ], [ [ "### Changing Whole Strings with Functions\nPython has some built-in *functions* that allow you to change a whole string at once. You can change all characters to lowercase or uppercase:", "_____no_output_____" ] ], [ [ "my_variable.lower()", "_____no_output_____" ], [ "my_variable.upper()", "_____no_output_____" ] ], [ [ "NB: These functions do not change the original string but create a new one. Our original string is still the same as it was before:", "_____no_output_____" ] ], [ [ "my_variable", "_____no_output_____" ] ], [ [ "### Testing Strings\n\nYou can also test a string to see if it is passes some test, e.g. is the string all alphabetic characters only?", "_____no_output_____" ] ], [ [ "my_variable.isalpha()", "_____no_output_____" ] ], [ [ "Does the string have the letter `p` in it?", "_____no_output_____" ] ], [ [ "'p' in my_variable", "_____no_output_____" ] ], [ [ "### Lists of Strings\nAnother important thing we can do with strings is creating a list of strings by listing them inside square brackets `[]`:", "_____no_output_____" ] ], [ [ "my_list = ['Butterflies are important as pollinators',\n 'Butterflies feed primarily on nectar from flowers',\n 'Butterflies are widely used in objects of art']\nmy_list", "_____no_output_____" ] ], [ [ "### Manipulating Lists of Strings\nJust like with strings, we can access individual items inside a list by index number:", "_____no_output_____" ] ], [ [ "my_list[0]", "_____no_output_____" ] ], [ [ "And we can access a range of items inside a list by *slicing*:", "_____no_output_____" ] ], [ [ "my_list[0:2]", "_____no_output_____" ] ], [ [ "### Advanced: Creating Lists of Strings with List Comprehensions\nWe can create new lists in an elegant way by combining some of the things we have covered above. Here is an example where we have taken our original list `my_list` and created a new list `new_list` by going over each string in the list:", "_____no_output_____" ] ], [ [ "new_list = [string for string in my_list]\nnew_list", "_____no_output_____" ] ], [ [ "Why do this? If we combine it with a test, we can have a list that only contains strings with the letter `p` in them:", "_____no_output_____" ] ], [ [ "new_list_p = [string for string in my_list if 'p' in string]\nnew_list_p", "_____no_output_____" ] ], [ [ "This is a very powerful way to quickly create lists. We can even change all the strings to uppercase at the same time!", "_____no_output_____" ] ], [ [ "new_list_p_upper = [string.upper() for string in my_list if 'p' in string]\nnew_list_p_upper", "_____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", "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" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "from src import helpers, utils, DMRG", "_____no_output_____" ], [ "DMRG.conv_FHH_SU2_n", "_____no_output_____" ], [ "helpers.mps_nm(p1)", "_____no_output_____" ], [ "utils.", "_____no_output_____" ] ] ]

[ "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<font size=\"+1\">This notebook will illustrate how to access DeepLabCut(DLC) results for IBL sessions and how to create short videos with DLC labels printed onto, as well as wheel angle, starting by downloading data from the IBL flatiron server. It requires ibllib, a ONE account and the following script: https://github.com/int-brain-lab/iblapps/blob/master/DLC_labeled_video.py</font>", "_____no_output_____" ] ], [ [ "run '/home/mic/Dropbox/scripts/IBL/DLC_labeled_video.py'", "_____no_output_____" ], [ "one = ONE()", "Connected to https://alyx.internationalbrainlab.org as michael.schartner\n" ] ], [ [ "Let's first find IBL ephys sessions with DLC results:", "_____no_output_____" ] ], [ [ "eids= one.search(task_protocol='ephysChoiceworld', dataset_types=['camera.dlc'], details=False)", "_____no_output_____" ], [ "len(eids)", "_____no_output_____" ] ], [ [ "For a particular session, we can create a short labeled video by calling the function Viewer, specifying the eid of the desired session, the video type (there's 'left', 'right' and 'body' videos), and a range of trials for which the video should be created. Most sesions have around 700 trials. In the following, this is illustrated with session '3663d82b-f197-4e8b-b299-7b803a155b84', video type 'left', trials range [10,13] and without a zoom for the eye, such that nose, paw and tongue tracking is visible. The eye-zoom option shows only the four points delineating the pupil edges, which are too small to be visible in the normal view. Note that this automatically starts the download of the video from flatiron (in case it is not locally stored already), which may take a while since these videos are about 8 GB large. ", "_____no_output_____" ] ], [ [ "eid = eids[6]", "_____no_output_____" ], [ "Viewer(eid, 'left', [10,13], save_video=True, eye_zoom=False)", "Connected to https://alyx.internationalbrainlab.org as michael.schartner\nConnected to https://alyx.internationalbrainlab.org as michael.schartner\n" ] ], [ [ "As usual when downloading IBL data from flatiron, the dimensions are listed. Below is one frame of the video for illustration. One can see one point for each paw, two points for the edges of the tongue, one point for the nose and there are 4 points close together around the pupil edges. All points for which the DLC network had a confidence probability of below 0.9 are hidden. For instance when the mouse is not licking, there is no tongue and so the network cannot detect it, and no points are shown. \n\nThe script will display and save the short video in your local folder. ", "_____no_output_____" ], [ "![alt text](video_frame.png \"Example frame of video with DLC labels\")", "_____no_output_____" ], [ "Sections of the script <code>DLC_labeled_video.py</code> can be recycled to analyse DLC traces. For example let's plot the x coordinate for the right paw in a <code>'left'</code> cam video for a given trial. ", "_____no_output_____" ] ], [ [ "one = ONE()", "Connected to https://alyx.internationalbrainlab.org as michael.schartner\n" ], [ "dataset_types = ['camera.times','trials.intervals','camera.dlc']\nvideo_type = 'left'", "_____no_output_____" ], [ "# get paths to load in data\nD = one.load('3663d82b-f197-4e8b-b299-7b803a155b84',dataset_types=dataset_types, dclass_output=True)\nalf_path = Path(D.local_path[0]).parent.parent / 'alf'\nvideo_data = alf_path.parent / 'raw_video_data'", "_____no_output_____" ], [ "# get trials start and end times, camera time stamps (one for each frame, synced with DLC trace)\ntrials = alf.io.load_object(alf_path, '_ibl_trials')\ncam0 = alf.io.load_object(alf_path, '_ibl_%sCamera' % video_type)\ncam1 = alf.io.load_object(video_data, '_ibl_%sCamera' % video_type)\ncam = {**cam0,**cam1}", "Inconsistent dimensions for object:_ibl_trials\n(893,), itiDuration\n(767,), stimOn_times\n(767, 2), intervals\n(893,), feedbackType\n(893,), choice\n(893,), rewardVolume\n(767,), feedback_times\n(893,), contrastRight\n(893,), probabilityLeft\n(893,), contrastLeft\n(767,), goCue_times\nInconsistent dimensions for object:_ibl_leftCamera\n(379271,), times\n(379265,), nose_tip_x\n(379265,), nose_tip_y\n(379265,), nose_tip_likelihood\n(379265,), pupil_top_r_x\n(379265,), pupil_top_r_y\n(379265,), pupil_top_r_likelihood\n(379265,), pupil_right_r_x\n(379265,), pupil_right_r_y\n(379265,), pupil_right_r_likelihood\n(379265,), pupil_bottom_r_x\n(379265,), pupil_bottom_r_y\n(379265,), pupil_bottom_r_likelihood\n(379265,), pupil_left_r_x\n(379265,), pupil_left_r_y\n(379265,), pupil_left_r_likelihood\n(379265,), paw_l_x\n(379265,), paw_l_y\n(379265,), paw_l_likelihood\n(379265,), paw_r_x\n(379265,), paw_r_y\n(379265,), paw_r_likelihood\n(379265,), tube_top_x\n(379265,), tube_top_y\n(379265,), tube_top_likelihood\n(379265,), tube_bottom_x\n(379265,), tube_bottom_y\n(379265,), tube_bottom_likelihood\n(379265,), tongue_end_l_x\n(379265,), tongue_end_l_y\n(379265,), tongue_end_l_likelihood\n(379265,), tongue_end_r_x\n(379265,), tongue_end_r_y\n(379265,), tongue_end_r_likelihood\n" ], [ "# for each tracked point there's x,y in [px] in the frame and a likelihood that indicates the network's confidence\ncam.keys()", "_____no_output_____" ] ], [ [ "There is also <code>'times'</code> in this dictionary, the time stamps for each frame that we'll use to sync it with other events in the experiment. Let's get rid of it briefly to have only DLC points and set coordinates to nan when the likelihood is below 0.9. ", "_____no_output_____" ] ], [ [ "Times = cam['times'] \ndel cam['times']\npoints = np.unique(['_'.join(x.split('_')[:-1]) for x in cam.keys()])\ncam['times'] = Times", "_____no_output_____" ], [ "# A helper function to find closest time stamps\ndef find_nearest(array, value):\n array = np.asarray(array)\n idx = (np.abs(array - value)).argmin()\n return idx", "_____no_output_____" ] ], [ [ "Let's pick say the 5th trial and find all DLC traces for it.", "_____no_output_____" ] ], [ [ "frame_start = find_nearest(cam['times'], trials['intervals'][4][0])\nframe_stop = find_nearest(cam['times'], trials['intervals'][4][1])", "_____no_output_____" ], [ "XYs = {}\nfor point in points:\n x = np.ma.masked_where(\n cam[point + '_likelihood'] < 0.9, cam[point + '_x'])\n x = x.filled(np.nan)\n y = np.ma.masked_where(\n cam[point + '_likelihood'] < 0.9, cam[point + '_y'])\n y = y.filled(np.nan)\n XYs[point] = np.array(\n [x[frame_start:frame_stop], y[frame_start:frame_stop]])", "_____no_output_____" ], [ "import matplotlib.pyplot as plt", "_____no_output_____" ], [ "plt.plot(cam['times'][frame_start:frame_stop],XYs['paw_r'][0])\nplt.xlabel('time [sec]')\nplt.ylabel('x location of right paw [px]')", "_____no_output_____" ] ] ]

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

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "import keras", "Using TensorFlow backend.\n" ], [ "from keras.datasets import fashion_mnist", "_____no_output_____" ], [ "(X_train, y_train), (X_test, y_test) = fashion_mnist.load_data()", "Downloading data from http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-labels-idx1-ubyte.gz\n32768/29515 [=================================] - 0s 6us/step\nDownloading data from http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-images-idx3-ubyte.gz\n26427392/26421880 [==============================] - 21s 1us/step\nDownloading data from http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-labels-idx1-ubyte.gz\n8192/5148 [===============================================] - 0s 0us/step\nDownloading data from http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-images-idx3-ubyte.gz\n4423680/4422102 [==============================] - 4s 1us/step\n" ], [ "type(X_train)", "_____no_output_____" ], [ "import matplotlib.pyplot as plt", "_____no_output_____" ], [ "plt.imshow(X_train[1,:,:])", "_____no_output_____" ], [ "plt.imsave('sample_tshirt.png', X_train[1,:,:])", "_____no_output_____" ] ] ]

[ "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Week 3 - Functions", "_____no_output_____" ], [ "The real power in any programming language is the **Function**.\n\nA function is:\n\n* a little block of script (one line or many) that performs specific task or a series of tasks.\n* reusable and helps us make our code DRY.\n* triggered when something \"invokes\" or \"calls\" it.\n* ideally modular – it performs a narrow task and you call several functions to perform more complex tasks.\n", "_____no_output_____" ], [ "### What we'll cover today:\n\n* Simple function\n* Return statements\n*\n", "_____no_output_____" ] ], [ [ "## Build a function called myFunction that adds 2 numbers together\n## it should print \"The total is (whatever the number is)!\"", "_____no_output_____" ], [ "## build it here\n\n", "_____no_output_____" ], [ "## Call myFunction using 4 and 5 as the arguments\n", "_____no_output_____" ], [ "## Call myFunction using 10 and 2 as the arguments\n", "_____no_output_____" ], [ "## you might forget what arguments are needed for the function to work.\n## you can add notes that appear on shift-tab as you call the function.\n## write it here\n", "_____no_output_____" ], [ "## test it on 3 and 4\n", "_____no_output_____" ] ], [ [ "### To use or not use functions?\n\nLet's compare the two options with a simple example:", "_____no_output_____" ] ], [ [ "## You have a list of numbers. \nmylist1 = [1, -5, 22, -44.2, 33, -45]", "_____no_output_____" ], [ "## Turn each number into an absolute number.\n## a for loop works perfectly fine here.\n", "_____no_output_____" ], [ "## The problem is that your project keeps generating more lists.\n## Each list of numbers has to be turned into absolute numbers\nmylist2 = [-56, -34, -75, -111, -22]\nmylist3 = [-100, -200, 100, -300, -100]\nmylist4 = [-23, -89, -11, -45, -27]\nmylist5 = [0, 1, 2, 3, 4, 5]", "_____no_output_____" ] ], [ [ "## DRY\n\n\n### Do you keep writing for loops for each list?\n\n### No, that's a lot of repetition!\n### DRY stands for \"Don't Repeat Yourself\"", "_____no_output_____" ] ], [ [ "## Instead we write a function that takes a list,\n## converts each list item to an absolute number,\n## and prints out the number\n\n", "_____no_output_____" ], [ "## Try swapping out different lists into the function:\n", "_____no_output_____" ] ], [ [ "## Timesaver \n### Imagine for a moment that your editor tells you that the calculation needs to be updated. Instead of needing the absolute number, you need the absolute number minus 5.\n\n### Having used multiple for loops, you'd have to change each one. What if you miss one or two? Either way, it's a chore.\n\n### With functions, you just revise the function and the update runs everywhere.\n\n", "_____no_output_____" ] ], [ [ "## So if an editor says to actually multiply the absolute number by 1_000_000,\n\n", "_____no_output_____" ], [ "## Try swapping out different lists into the function:\n", "_____no_output_____" ] ], [ [ "## Return Statements\n\n### So far we have only printed out values processed by a function. \n\n### But we really want to retain the value the function creates. \n\n### We can then pass that value to other parts of our calculations and code.", "_____no_output_____" ] ], [ [ "## Simple example\n## A function that adds two numbers together and prints the value:\n", "_____no_output_____" ], [ "## call the function with the numbers 2 and 4\n", "_____no_output_____" ], [ "## let's try to save it in a variable called myCalc\n", "_____no_output_____" ], [ "## Print myCalc. What does it hold?\n", "_____no_output_____" ] ], [ [ "### The return Statement", "_____no_output_____" ] ], [ [ "## Tweak our function by adding return statement\n## instead of printing a value we want to return a value(or values).\n", "_____no_output_____" ], [ "## call the function add_numbers_ret \n## and store in variable called myCalc\n", "_____no_output_____" ], [ "## print myCalc\n", "_____no_output_____" ], [ "## What type is myCalc?\n", "_____no_output_____" ] ], [ [ "## Return multiple values", "_____no_output_____" ] ], [ [ "## demo function\n\n\n", "_____no_output_____" ], [ "name,age,country = getPerson(\"David\", 35, \"France\")\n", "_____no_output_____" ] ], [ [ "### Let's revise our earlier absolute values converter with a return statement\n#### Here is the earlier version:\n<img src=\"https://github.com/sandeepmj/fall20-student-practical-python/blob/master/support_files/abs-function.png?raw=true\" style=\"width: 100%;\">", "_____no_output_____" ] ], [ [ "## Here it is revised with a return statement\n\n \n \n ", "_____no_output_____" ], [ "## Let's actually make that a list comprehension version of the function:\n", "_____no_output_____" ], [ "## Let's test it by storing the return value in variable x\n", "_____no_output_____" ], [ "## What type of data object is it?\n", "_____no_output_____" ] ], [ [ "# Make a function more flexible and universal\n\n* Currently, we have a function that takes ONLY a list as an argument.\n* We'd have to write another one for a single number argument.", "_____no_output_____" ] ], [ [ "## try using return_absolutes_lc on a single number like -10\n## it will break\n\n", "_____no_output_____" ] ], [ [ "# Universalize our absolute numbers function", "_____no_output_____" ] ], [ [ "## call the function make_abs\n", "_____no_output_____" ], [ "## try it on -10\n", "_____no_output_____" ], [ "## Try it on mylist3 - it will break!\n", "_____no_output_____" ] ], [ [ "## We can use the ```map()``` function to tackle this problem.\n\n```map()``` takes 2 arguments: a ```function``` and ```iterable like a list```.", "_____no_output_____" ] ], [ [ "## try it on make_abs and mylist3\n", "_____no_output_____" ], [ "## save it into a list\n", "_____no_output_____" ] ], [ [ "## ```map()``` also works for multiple iterables\n\n#### remember our ```add_numbers_ret``` function.", "_____no_output_____" ] ], [ [ "## here it is again:\n\ndef add_numbers_ret(number1, number2):\n return (number1 + number2)", "_____no_output_____" ], [ "## two lists\na_even = [2, 4, 6, 8]\na_odd = [1, 3, 5, 7, 9] ## note this has one more item in the list.", "_____no_output_____" ], [ "## run map on a_even and a_odd\nb = list(map(add_numbers_ret, a_even, a_odd))\nb", "_____no_output_____" ] ], [ [ "## Functions that call other funcions", "_____no_output_____" ] ], [ [ "## let's create a function that returns the square of a number\n\n", "_____no_output_____" ], [ "## what is 9 squared?\n", "_____no_output_____" ] ], [ [ "### Making a point here with a simple example\nLet's say we want to add 2 numbers together and then square that result.\nInstead of writing one \"complex\" function, we can call on our modular functions.", "_____no_output_____" ] ], [ [ "## a function that calls our modular functions\n\n ", "_____no_output_____" ], [ "## call make_point() on 2 and 5\nmake_point(2,5)", "_____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", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# piston example with explicit Euler scheme", "_____no_output_____" ] ], [ [ "%matplotlib inline\n\nimport matplotlib\nimport matplotlib.pyplot as plt\nimport matplotlib.animation as anim\n\nimport numpy as np\n\nimport sys\nsys.path.insert(0, './code')\n\nimport ideal_gas", "_____no_output_____" ] ], [ [ "### physical parameters", "_____no_output_____" ] ], [ [ "# length of cylinder\nl = 0.1\n\n# radius of cylinder\nr = 0.05 \n\n# thickness of wall\nw = 0.006\n\n# derived geometrical data\nr2 = 2 * r # diameter of cylinder\nw2 = w / 2 # halved thickness of wall\nl2 = l - w2\nA = r**2 * np.pi # cross-sectional area", "_____no_output_____" ], [ "def get_v_1(q):\n \"\"\"first volume\"\"\"\n return A * (q - w2)\n \ndef get_v_2(q):\n \"\"\"second volume\"\"\"\n return A * (l2 - q)", "_____no_output_____" ], [ "# density of aluminium\nm_Al = 2700.0\nm_Cu = 8960.0\n\n# mass of piston\nm = m_Cu * A * w\n\n# thermal conductivity of aluminium\nκ_Al = 237.0\nκ_Cu = 401.0\n\n# thermal conduction coefficient\nα = κ_Cu * A / w\n\nm_inv = 1 / m", "_____no_output_____" ] ], [ [ "### initial conditions\n\ndetermine $n_1$, $n_2$, $s_1$, $s_2$", "_____no_output_____" ] ], [ [ "# wanted conditions\nv_1 = v_2 = get_v_1(l/2)\nθ_1 = 273.15 + 25.0\nπ_1 = 1.5 * 1e5\nθ_2 = 273.15 + 20.0\nπ_2 = 1.0 * 1e5", "_____no_output_____" ], [ "from scipy.optimize import fsolve", "_____no_output_____" ], [ "n_1 = fsolve(lambda n : ideal_gas.S_π(ideal_gas.U2(θ_1, n), v_1, n) - π_1, x0=2e22)[0]\ns_1 = ideal_gas.S(ideal_gas.U2(θ_1, n_1), v_1, n_1)", "_____no_output_____" ], [ "# check temperature\nideal_gas.U_θ(s_1, v_1, n_1) - 273.15", "_____no_output_____" ], [ "# check pressure\nideal_gas.U_π(s_1, v_1, n_1) * 1e-5", "_____no_output_____" ], [ "n_2 = fsolve(lambda n : ideal_gas.S_π(ideal_gas.U2(θ_2, n), v_2, n) - π_2, x0=2e22)[0]\ns_2 = ideal_gas.S(ideal_gas.U2(θ_2, n_2), v_2, n_2)", "_____no_output_____" ], [ "# check temperature\nideal_gas.U_θ(s_2, v_2, n_2) - 273.15", "_____no_output_____" ], [ "# check pressure\nideal_gas.U_π(s_2, v_2, n_2) * 1e-5", "_____no_output_____" ], [ "x_0 = l/2, 0, s_1, s_2", "_____no_output_____" ] ], [ [ "### simulation", "_____no_output_____" ] ], [ [ "def set_state(data, i, x):\n q, p, s_1, s_2 = x\n\n data[i, 0] = q\n \n data[i, 1] = p\n data[i, 2] = v = m_inv * p\n \n data[i, 3] = v_1 = get_v_1(q)\n data[i, 4] = π_1 = ideal_gas.U_π(s_1, v_1, n_1)\n \n data[i, 5] = s_1\n data[i, 6] = θ_1 = ideal_gas.U_θ(s_1, v_1, n_1)\n \n data[i, 7] = v_2 = get_v_2(q)\n data[i, 8] = π_2 = ideal_gas.U_π(s_2, v_2, n_2)\n\n data[i, 9] = s_2\n data[i, 10] = θ_2 = ideal_gas.U_θ(s_2, v_2, n_2)\n \n data[i, 11] = E_kin = 0.5 * m_inv * p**2\n data[i, 12] = u_1 = ideal_gas.U(s_1, v_1, n_1)\n data[i, 13] = u_2 = ideal_gas.U(s_2, v_2, n_2)\n data[i, 14] = E = E_kin + u_1 + u_2\n data[i, 15] = S = s_1 + s_2\n\n \ndef get_state(data, i):\n return data[i, (0, 1, 5, 9)]", "_____no_output_____" ], [ "def rhs(x):\n \"\"\"right hand side of the explicit system\n of differential equations\n \"\"\"\n q, p, s_1, s_2 = x\n \n v_1 = get_v_1(q)\n v_2 = get_v_2(q)\n \n π_1 = ideal_gas.U_π(s_1, v_1, n_1)\n π_2 = ideal_gas.U_π(s_2, v_2, n_2)\n \n θ_1 = ideal_gas.U_θ(s_1, v_1, n_1)\n θ_2 = ideal_gas.U_θ(s_2, v_2, n_2)\n \n return np.array((m_inv*p, A*(π_1-π_2), α*(θ_2-θ_1)/θ_1, α*(θ_1-θ_2)/θ_2))", "_____no_output_____" ], [ "t_f = 1.0\ndt = 1e-4\n\nsteps = int(t_f // dt)\nprint(f'steps={steps}')\n\nt = np.linspace(0, t_f, num=steps)\ndt = t[1] - t[0]\n\ndata = np.empty((steps, 16), dtype=float)\nset_state(data, 0, x_0)\n\nx_old = get_state(data, 0)\nfor i in range(1, steps):\n x_new = x_old + dt * rhs(x_old)\n set_state(data, i, x_new)\n x_old = x_new\n\nθ_min = np.min(data[:, (6,10)])\nθ_max = np.max(data[:, (6,10)])", "steps=9999\n" ], [ "# plot transient\nfig, ax = plt.subplots(dpi=200)\nax.set_title(\"piston position q\")\nax.plot(t, data[:, 0]);", "_____no_output_____" ], [ "fig, ax = plt.subplots(dpi=200)\nax.set_title(\"total entropy S\")\nax.plot(t, data[:, 15]);", "_____no_output_____" ], [ "fig, ax = plt.subplots(dpi=200)\nax.set_title(\"total energy E\")\nax.plot(t, data[:, 14]);", "_____no_output_____" ] ], [ [ "the total energy is not conserved well", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "설치하기", "_____no_output_____" ] ], [ [ "!pip install git+https://github.com/gbolmier/funk-svd", "_____no_output_____" ], [ "from funk_svd.dataset import fetch_ml_ratings\nfrom funk_svd import SVD\nfrom sklearn.metrics import mean_absolute_error\nimport pandas as pd", "_____no_output_____" ], [ "ds_ratings = pd.read_csv(\"../ml-latest-small/ratings.csv\")\nds_movies = pd.read_csv(\"../ml-latest-small/movies.csv\")\n# funk-svd 는 유저 칼럼이 u_id, 아이템 칼럼이 i_id여서 이름을 변경해줍니다\ndf = ds_ratings.rename(\n {\n \"userId\": \"u_id\",\n \"movieId\": \"i_id\"\n },\n axis=1\n)\ndf", "_____no_output_____" ], [ "# 80%를 학습에 사용하고 10%를 validation set, 10%를 test_set으로 사용합니다\ntrain = df.sample(frac=0.8, random_state=7)\nval = df.drop(train.index.tolist()).sample(frac=0.5, random_state=8)\ntest = df.drop(train.index.tolist()).drop(val.index.tolist())", "_____no_output_____" ], [ "# SVD를 학습합니다. n_factors가 SVD의 k를 의미합니다\nsvd = SVD(lr=0.001, reg=0.005, n_epochs=100, n_factors=15, early_stopping=True,\n shuffle=False, min_rating=1, max_rating=5)\nsvd.fit(X=train, X_val=val)\n\n# 학습 결과를 test set 으로 평가합니다.\npred = svd.predict(test)\nmae = mean_absolute_error(test['rating'], pred)\n\nprint(f'Test MAE: {mae:.2f}')", "Preprocessing data...\n\nPreprocessing data...\n\nEpoch 1/100 | val_loss: 0.95 - val_rmse: 0.98 - val_mae: 0.78 - took 0.0 sec\nEpoch 2/100 | val_loss: 0.91 - val_rmse: 0.95 - val_mae: 0.76 - took 0.0 sec\nEpoch 3/100 | val_loss: 0.88 - val_rmse: 0.94 - val_mae: 0.75 - took 0.0 sec\nEpoch 4/100 | val_loss: 0.87 - val_rmse: 0.93 - val_mae: 0.74 - took 0.0 sec\nEpoch 5/100 | val_loss: 0.85 - val_rmse: 0.92 - val_mae: 0.73 - took 0.0 sec\nEpoch 6/100 | val_loss: 0.84 - val_rmse: 0.92 - val_mae: 0.72 - took 0.0 sec\nEpoch 7/100 | val_loss: 0.84 - val_rmse: 0.91 - val_mae: 0.72 - took 0.0 sec\nEpoch 8/100 | val_loss: 0.83 - val_rmse: 0.91 - val_mae: 0.71 - took 0.0 sec\nEpoch 9/100 | val_loss: 0.82 - val_rmse: 0.91 - val_mae: 0.71 - took 0.0 sec\nEpoch 10/100 | val_loss: 0.82 - val_rmse: 0.90 - val_mae: 0.71 - took 0.0 sec\nEpoch 11/100 | val_loss: 0.81 - val_rmse: 0.90 - val_mae: 0.70 - took 0.0 sec\nEpoch 12/100 | val_loss: 0.81 - val_rmse: 0.90 - val_mae: 0.70 - took 0.0 sec\nEpoch 13/100 | val_loss: 0.81 - val_rmse: 0.90 - val_mae: 0.70 - took 0.0 sec\nEpoch 14/100 | val_loss: 0.80 - val_rmse: 0.90 - val_mae: 0.70 - took 0.0 sec\nEpoch 15/100 | val_loss: 0.80 - val_rmse: 0.90 - val_mae: 0.70 - took 0.0 sec\nEpoch 16/100 | val_loss: 0.80 - val_rmse: 0.89 - val_mae: 0.69 - took 0.0 sec\nEpoch 17/100 | val_loss: 0.80 - val_rmse: 0.89 - val_mae: 0.69 - took 0.0 sec\nEpoch 18/100 | val_loss: 0.79 - val_rmse: 0.89 - val_mae: 0.69 - took 0.0 sec\nEpoch 19/100 | val_loss: 0.79 - val_rmse: 0.89 - val_mae: 0.69 - took 0.0 sec\nEpoch 20/100 | val_loss: 0.79 - val_rmse: 0.89 - val_mae: 0.69 - took 0.0 sec\n\nTraining took 0 sec\nTest MAE: 0.68\n" ], [ "user_ratings = ds_ratings.pivot(index=\"userId\", columns=\"movieId\", values=\"rating\")\ndef get_user_real_score(user_id, item_id):\n return user_ratings.loc[user_id, item_id]", "_____no_output_____" ], [ "\ndef get_user_unseen_ranks(user_id, max_rank=100):\n # predict_pair 메서드를 이용해서\n # 유저의 모든 영화 예측 평점을 계산합니다.\n movie_ids = df.i_id.unique()\n rec = pd.DataFrame(\n [{\n \"id\": id, \n \"recommendation_score\": svd.predict_pair(user_id, id), \n \"real_score\": get_user_real_score(user_id, id)\n } \n for id in movie_ids\n ]\n )\n \n # 유저가 본 영화는 제외합니다\n user_seen_movies = train[train.u_id == user_id]\n rec = rec[~rec.id.isin(user_seen_movies.i_id)]\n rec.sort_values(\"recommendation_score\", ascending=False, inplace=True)\n \n # max_rank 개만 보여줍니다\n if max_rank is not None:\n rec = rec.head(max_rank)\n \n # 순위를 컬럼에 추가합니다\n rec[\"rank\"] = range(1, len(rec) + 1)\n \n # train 에는 포함되지 않았지만 실제로 유저가 봤던 영화 ID를 가져옵니다\n # 이후 추천 결과에서 해당 영화들만 필터링해서 몇 위로 추천됬는지 확인합니다\n user_unseen_movies = pd.concat([val, test], axis=0)\n user_unseen_movies = user_unseen_movies[user_unseen_movies.u_id == user_id].i_id\n rec = rec[rec.id.isin(user_unseen_movies)]\n rec.index = rec.id\n del rec[\"id\"]\n \n # 실제 추천할 영화 정보와 join합니다.\n rec = ds_movies.merge(rec, left_on=\"movieId\", right_index=True)\n rec.sort_values(\"rank\", inplace=True)\n \n top_k_accuracy = len(rec) / len(user_unseen_movies)\n return rec, top_k_accuracy\n\nfrom IPython.display import display\n\nuser_ids = df.u_id.unique()[:10]\ntotal_acc = 0\nfor uid in user_ids:\n top100, acc = get_user_unseen_ranks(uid)\n total_acc += acc\n print(\"User: \", uid, \"TOP 100 accuracy: \", round(acc, 2))\n display(top100)\n\ntotal_acc / len(user_ids)", "User: 1 TOP 100 accuracy: 0.27\n" ] ] ]

[ "markdown", "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Import libraries\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nimport sqlite3\nfrom sklearn.pipeline import Pipeline\n\n# used for train/test splits\nfrom sklearn.cross_validation import train_test_split\n\n# used to impute mean for data\nfrom sklearn.preprocessing import Imputer\n\n# logistic regression is our model of choice\nfrom sklearn.linear_model import LogisticRegression\nfrom sklearn.linear_model import LogisticRegressionCV\n\n# used to calculate AUROC/accuracy\nfrom sklearn import metrics\n\n# used to create confusion matrix\nfrom sklearn.metrics import confusion_matrix\n\nfrom sklearn.cross_validation import cross_val_score\n\n\n%matplotlib inline", "/Users/danielcphelps/anaconda/lib/python2.7/site-packages/sklearn/cross_validation.py:44: DeprecationWarning: This module was deprecated in version 0.18 in favor of the model_selection module into which all the refactored classes and functions are moved. Also note that the interface of the new CV iterators are different from that of this module. This module will be removed in 0.20.\n \"This module will be removed in 0.20.\", DeprecationWarning)\n" ], [ "# Connect to MIMIC\n# be sure to add the password as appropriate!\ncon = psycopg2.connect(dbname='MIMIC', user='workshop', password=''\n , host='<xxxxx>.amazonaws.com'\n , port=5432)\ncur = con.cursor()\ncur.execute('SET search_path to ''mimiciii_workshop''')", "_____no_output_____" ], [ "query = \"\"\"\nwith ce as\n(\n select\n icustay_id, charttime, itemid, valuenum\n from chartevents\n -- specify what data we want from chartevents\n where itemid in\n (\n 211, -- Heart Rate\n 618, --\tRespiratory Rate\n 615 --\tResp Rate (Total)\n )\n -- how did we know heart rates were stored using ITEMID 211? Simple, we looked in D_ITEMS!\n -- Try it for yourself: select * from d_items where lower(label) like '%heart rate%'\n)\nselect\n -- ICUSTAY_ID identifies each unique patient ICU stay\n -- note that if the same person stays in the ICU more than once, each stay would have a *different* ICUSTAY_ID\n -- however, since it's the same person, all those stays would have the same SUBJECT_ID\n ie.icustay_id\n\n -- this is the outcome of interest: in-hospital mortality\n , max(adm.HOSPITAL_EXPIRE_FLAG) as OUTCOME\n\n -- this is a case statement - essentially an \"if, else\" clause\n , min(\n case\n -- if the itemid is 211\n when itemid = 211\n -- then return the actual value stored in VALUENUM\n then valuenum\n -- otherwise, return 'null', which is SQL standard for an empty value\n else null\n -- end the case statement\n end\n ) as HeartRate_Min\n\n -- note we wrapped the above in \"min()\"\n -- this takes the minimum of all values inside, and *ignores* nulls\n -- by calling this on our case statement, we are ignoring all values except those with ITEMID = 211\n -- since ITEMID 211 are heart rates, we take the minimum of only heart rates\n\n , max(case when itemid = 211 then valuenum else null end) as HeartRate_Max\n , min(case when itemid in (615,618) then valuenum else null end) as RespRate_Min\n , max(case when itemid in (615,618) then valuenum else null end) as RespRate_Max\nfrom icustays ie\n\n-- join to the admissions table to get hospital outcome\ninner join admissions adm\n on ie.hadm_id = adm.hadm_id\n\n-- join to the chartevents table to get the observations\nleft join ce\n -- match the tables on the patient identifier\n on ie.icustay_id = ce.icustay_id\n -- and require that the observation be made after the patient is admitted to the ICU\n and ce.charttime >= ie.intime\n -- and *before* their admission time + 1 day, i.e. the observation must be made on their first day in the ICU\n and ce.charttime <= ie.intime + interval '1' day\ngroup by ie.icustay_id\norder by ie.icustay_id\n\"\"\"\nconn = sqlite3.connect('/Users/danielcphelps/Boxcryptor/Google Drive/daniel.c.phelps/Research/HealthcareAnalytics/mimic-workshop/data/mimicdata.sqlite')\ndata = pd.read_sql_query(query,conn)\nprint(data.head())", "_____no_output_____" ], [ "# close the connection as we are done loading data from server\ncur.close()\ncon.close()", "_____no_output_____" ], [ "# move from a data frame into a numpy array\nX = data.values\ny = X[:,1]\n\n# delete first 2 columns: the ID and the outcome\nX = np.delete(X,0,axis=1)\nX = np.delete(X,0,axis=1)", "_____no_output_____" ], [ "# evaluate a logistic regression model using an 80%-20% training/test split\nX_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)\n\n# impute mean for missing values\nimp = Imputer(missing_values='NaN', strategy='mean', axis=0)\nimp.fit(X_train)\n\nX_train = imp.transform(X_train)\nX_test = imp.transform(X_test)\n\nmodel = LogisticRegression(fit_intercept=True)\nmodel = model.fit(X_train, y_train)\n\n# predict class labels for the test set\ny_pred = model.predict(X_test)\n\n# generate class probabilities\ny_prob = model.predict_proba(X_test)\n\n# generate evaluation metrics\nprint('Accuracy = {}'.format(metrics.accuracy_score(y_test, y_pred)))\nprint('AUROC = {}'.format(metrics.roc_auc_score(y_test, y_prob[:, 1])))\n\nprint('\\nConfusion matrix')\nprint(metrics.confusion_matrix(y_test, y_pred))\nprint('\\nClassification report')\nprint(metrics.classification_report(y_test, y_pred))", "Accuracy = 0.784267912773\nAUROC = 0.642288212031\n\nConfusion matrix\n[[977 17]\n [260 30]]\n\nClassification report\n precision recall f1-score support\n\n 0.0 0.79 0.98 0.88 994\n 1.0 0.64 0.10 0.18 290\n\navg / total 0.76 0.78 0.72 1284\n\n" ], [ "# evaluate a logistic regression with L1 regularization\n\n# evaluate the model using 5-fold cross-validation\n# see: http://scikit-learn.org/stable/modules/model_evaluation.html#scoring-parameter\n# for list of scoring parameters\n\nestimator = Pipeline([(\"imputer\", Imputer(missing_values='NaN',\n strategy=\"mean\",\n axis=0)),\n (\"regression\", LogisticRegressionCV(penalty='l1',\n cv=5,\n scoring='roc_auc',\n solver='liblinear'))])\n\nscores = cross_val_score(estimator\n , X, y\n , scoring='roc_auc', cv=5)\n\n\nprint('AUROC for all folds:')\nprint(scores)\nprint('Average AUROC across folds:')\nprint(scores.mean())", "AUROC for all folds:\n[ 0.632241 0.66711432 0.65462583 0.63505984 0.64856111]\nAverage AUROC across folds:\n0.647520418729\n" ] ] ]

[ "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "#Using our synthetic data library for today's exercise\n#pip install ydata", "/usr/local/lib/python3.6/dist-packages/statsmodels/tools/_testing.py:19: FutureWarning: pandas.util.testing is deprecated. Use the functions in the public API at pandas.testing instead.\n import pandas.util.testing as tm\n" ], [ "#Loading the census dataset from kaggle\nimport logging\nimport os\nimport requests\nimport pandas as pd\nimport seaborn as sns\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.preprocessing import OneHotEncoder\n#import ydata.synthetic.regular as synthetic\n\n#Dataset URL from kaggle\ndata_url = 'https://www.kaggle.com/uciml/adult-census-income/downloads/adult.csv'\n\n# The local path where the data set is saved.\nlocal_filename = \"adult.csv\"\n\n# Kaggle Username and Password\nkaggle_info = {'UserName': \"myusername\", 'Password': \"mypassword\"}\n\n# Attempts to download the CSV file. Gets rejected because we are not logged in.\nr = requests.get(data_url)\n\n# Login to Kaggle and retrieve the data.\nr = requests.post(r.url, data = kaggle_info)\n\n# Writes the data to a local file one chunk at a time.\nf = open(local_filename, 'wb')\nfor chunk in r.iter_content(chunk_size = 512 * 1024): # Reads 512KB at a time into memory\n\n if chunk: # filter out keep-alive new chunks\n f.write(chunk)\nf.close()", "_____no_output_____" ], [ "adult_census = pd.read_csv('adult.csv')\n\n#For the purpose of this exercise we will filter information regarding only black and white individuals. \nadult_census = adult_census[(adult_census['race']=='White') | (adult_census['race']=='Black')]\n\nincome = adult_census['income']\nadult_census = adult_census.drop('education.num', axis=1)\nadult_census = adult_census.drop('income', axis=1)\n\ntrain_adult, test_adult, income_train, income_test = train_test_split(adult_census, income, test_size=0.33, random_state=42)\n\ntrain_adult['income'] = income_train\n\ntrain_adult.head(10)", "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:12: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame.\nTry using .loc[row_indexer,col_indexer] = value instead\n\nSee the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n if sys.path[0] == '':\n" ], [ "sns.set(style=\"dark\", rc={'figure.figsize':(11.7,8.27)})\n\nsns.countplot(x=\"race\",\n palette=\"Paired\", edgecolor=\".6\",\n data=train_adult)", "_____no_output_____" ], [ "#Let's tackling the bias present in the dataset.\n#For that purpose we will need to filter the records belonging to only the black individuals. \ndef filter_fn(row):\n if row['race'] == 'Black':\n return True\n else:\n return False\n\nfilt = train_adult.apply(filter_fn, axis=1)\ntrain_adult_black = train_adult[filt]", "_____no_output_____" ] ], [ [ "", "_____no_output_____" ] ], [ [ "print(\"Number of records belonging to black individuals: {}\".format(train_adult_black.shape[0]))\n\nsns.set(style=\"dark\", rc={'figure.figsize':(11.7,8.27)})\n\nsns.countplot(x=\"sex\",\n palette=\"Paired\", edgecolor=\".6\",\n data=train_adult_black)\n#In what concerns sex, we have an equal representation of women and man for the black population of the dataset", "Number of records belonging to black individuals: 2097\n" ], [ "#Using the YData synthetic data lib to generate new 3000 individuals for the black population\nsynth_model = synthetic.SynthTabular()\nsynth_model.fit(adult_black)\n\nsynth_data = synth_model.sample(n_samples=3000)", "_____no_output_____" ], [ "synth_data = pd.read_csv('synth_data.csv', index_col=[0])\nsynth_data = synth_data.drop('education.num', axis=1)\n\nsynth_data = pd.concat([synth_data[synth_data['income']=='>50K'],synth_data[synth_data['income']=='<=50K'][:1000]])\nsynth_data.describe()", "_____no_output_____" ], [ "#Now combining both the datasets\ntest_adult['income'] = income_test\nadult_combined = synth_data.append(test_adult).sample(frac=1)\n\n#Let's check again how are we regarding the balancing of our classes for the race variable\nsns.set(style=\"dark\", rc={'figure.figsize':(11.7,8.27)})\n\nsns.countplot(x=\"race\",\n palette=\"Paired\", edgecolor=\".6\",\n data=adult_combined)", "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:2: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame.\nTry using .loc[row_indexer,col_indexer] = value instead\n\nSee the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n \n" ], [ "#Auxiliar function to encode the categorical variables\nimport numpy as np\nfrom sklearn.preprocessing import OneHotEncoder\nfrom sklearn.linear_model import LogisticRegression\nfrom sklearn.ensemble import RandomForestClassifier\nfrom sklearn.tree import DecisionTreeClassifier\nfrom sklearn.svm import SVC\nfrom sklearn.neighbors import KNeighborsClassifier\nfrom sklearn.metrics import f1_score, accuracy_score, average_precision_score\n\ndef numerical_encoding(df, cat_cols=[], ord_cols=[]):\n try:\n assert isinstance(df, pd.DataFrame)\n except AssertionError as e:\n logging.error('The df input object must a Pandas dataframe. This action will not be executed.')\n return\n ord_cols_val = None\n cat_cols_val = None\n dummies = None\n cont_cols = list(set(df.columns) - set(cat_cols+ord_cols))\n cont_vals = df[cont_cols].values\n\n if len(ord_cols) > 0:\n ord_cols_val = df[ord_cols].values\n label_encoder = LabelEncoder\n ord_encoded = label_encoder.fit_transform(ord_cols_val)\n\n if len(cat_cols) > 0:\n cat_cols_val = df[cat_cols].values\n hot_encoder = OneHotEncoder()\n cat_encoded = hot_encoder.fit_transform(cat_cols_val).toarray()\n\n dummies = []\n for i, cat in enumerate(hot_encoder.categories_):\n for j in cat:\n dummies.append(cat_cols[i]+'_'+str(j))\n\n if ord_cols_val is not None and cat_cols_val is not None:\n encoded = np.hstack([cont_vals, ord_encoded, cat_encoded])\n columns = cont_cols+ord_cols+dummies\n elif cat_cols_val is not None:\n encoded = np.hstack([cont_vals, cat_encoded])\n columns = cont_cols+ord_cols+dummies\n else:\n encoded = cont_vals\n columns = cont_cols\n\n return pd.DataFrame(encoded, columns=columns), dummies", "_____no_output_____" ], [ "#validation functions\ndef score_estimators(estimators, x_test, y_test):\n\n #f1_score average='micro'\n scores = {type(clf).__name__: f1_score(y_test, clf.predict(x_test), average='micro') for clf in estimators}\n\n return scores\n\n\ndef fit_estimators(estimators, data_train, y_train):\n estimators_fit = []\n for i, estimator in enumerate(estimators):\n estimators_fit.append(estimator.fit(data_train, y_train))\n return estimators_fit\n\ndef estimator_eval(data, y, cat_cols=[]):\n def order_cols(df):\n cols = sorted(df.columns.tolist())\n return df[cols]\n\n data,_ = numerical_encoding(data, cat_cols=cat_cols)\n\n y, uniques = pd.factorize(y)\n data = order_cols(data)\n\n x_train, x_test, y_train, y_test = train_test_split(data, y, test_size=0.33, random_state=42)\n\n # Prepare train and test datasets\n estimators = [\n LogisticRegression(multi_class='auto', solver='lbfgs', max_iter=500, random_state=42),\n RandomForestClassifier(n_estimators=10, random_state=42),\n DecisionTreeClassifier(random_state=42),\n SVC(gamma='auto'),\n KNeighborsClassifier(n_neighbors=5)\n ]\t\n \n estimators_names = [type(clf).__name__ for clf in estimators]\n\n for estimator in estimators:\n assert hasattr(estimator, 'fit')\n assert hasattr(estimator, 'score')\n\n estimators = fit_estimators(estimators, x_train, y_train) \n scores = score_estimators(estimators, x_test, y_test)\n\n return scores", "_____no_output_____" ], [ "real_scores = estimator_eval(data=test_adult.drop('income', axis=1),\n y=test_adult['income'], \n cat_cols=['workclass', 'education', 'marital.status', 'occupation', 'relationship','race', 'sex', 'native.country'])\n\nsynth_scores = estimator_eval(data=adult_combined.drop('income', axis=1),\n y=adult_combined['income'], \n cat_cols=['workclass', 'education', 'marital.status', 'occupation', 'relationship','race', 'sex', 'native.country'])", "_____no_output_____" ], [ "dict_results = {'original': real_scores, 'synthetic': synth_scores}\nresults = pd.DataFrame(dict_results).reset_index()\n\nprint(\"Mean average accuracy improvement: {}\".format((results['synthetic'] - results['original']).mean()))\n\nresults_graph = results.melt('index', var_name='data_source', value_name='accuracy')", "Mean average accuracy improvement: 0.024448394946566386\n" ], [ " pd.DataFrame(dict_results).transpose()", "_____no_output_____" ], [ "#Final results comparision\n\nsns.barplot(x=\"index\", y=\"accuracy\", hue=\"data_source\", data=results_graph, \n palette=\"Paired\", edgecolor=\".6\")", "_____no_output_____" ] ] ]

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

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# LAB 5b: Deploy and predict with Keras model on Cloud AI Platform.\n\n**Learning Objectives**\n\n1. Setup up the environment\n1. Deploy trained Keras model to Cloud AI Platform\n1. Online predict from model on Cloud AI Platform\n1. Batch predict from model on Cloud AI Platform\n\n\n## Introduction \nIn this notebook, we'll deploying our Keras model to Cloud AI Platform and creating predictions.\n\nWe will set up the environment, deploy a trained Keras model to Cloud AI Platform, online predict from deployed model on Cloud AI Platform, and batch predict from deployed model on Cloud AI Platform.\n\nEach learning objective will correspond to a __#TODO__ in this student lab notebook -- try to complete this notebook first and then review the [solution notebook](../solutions/5b_deploy_keras_ai_platform_babyweight.ipynb).", "_____no_output_____" ], [ "## Set up environment variables and load necessary libraries", "_____no_output_____" ], [ "Import necessary libraries.", "_____no_output_____" ] ], [ [ "import os", "_____no_output_____" ] ], [ [ "### Lab Task #1: Set environment variables.\n\nSet environment variables so that we can use them throughout the entire lab. We will be using our project name for our bucket, so you only need to change your project and region.", "_____no_output_____" ] ], [ [ "%%bash\nPROJECT=$(gcloud config list project --format \"value(core.project)\")\necho \"Your current GCP Project Name is: \"$PROJECT", "_____no_output_____" ], [ "# Change these to try this notebook out\nPROJECT = \"cloud-training-demos\" # TODO: Replace with your PROJECT\nBUCKET = PROJECT # defaults to PROJECT\nREGION = \"us-central1\" # TODO: Replace with your REGION", "_____no_output_____" ], [ "os.environ[\"BUCKET\"] = BUCKET\nos.environ[\"REGION\"] = REGION\nos.environ[\"TFVERSION\"] = \"2.1\"", "_____no_output_____" ], [ "%%bash\ngcloud config set compute/region $REGION\ngcloud config set ai_platform/region global", "_____no_output_____" ] ], [ [ "## Check our trained model files\n\nLet's check the directory structure of our outputs of our trained model in folder we exported the model to in our last [lab](../solutions/10_train_keras_ai_platform_babyweight.ipynb). We'll want to deploy the saved_model.pb within the timestamped directory as well as the variable values in the variables folder. Therefore, we need the path of the timestamped directory so that everything within it can be found by Cloud AI Platform's model deployment service.", "_____no_output_____" ] ], [ [ "%%bash\ngsutil ls gs://${BUCKET}/babyweight/trained_model", "_____no_output_____" ], [ "%%bash\nMODEL_LOCATION=$(gsutil ls -ld -- gs://${BUCKET}/babyweight/trained_model/2* \\\n | tail -1)\ngsutil ls ${MODEL_LOCATION}", "_____no_output_____" ] ], [ [ "## Lab Task #2: Deploy trained model.\n\nDeploying the trained model to act as a REST web service is a simple gcloud call. Complete __#TODO__ by providing location of saved_model.pb file to Cloud AI Platoform model deployment service. The deployment will take a few minutes.", "_____no_output_____" ] ], [ [ "%%bash\nMODEL_NAME=\"babyweight\"\nMODEL_VERSION=\"ml_on_gcp\"\nMODEL_LOCATION=# TODO: Add GCS path to saved_model.pb file.\necho \"Deleting and deploying $MODEL_NAME $MODEL_VERSION from $MODEL_LOCATION\"\n# gcloud ai-platform versions delete ${MODEL_VERSION} --model ${MODEL_NAME}\n# gcloud ai-platform models delete ${MODEL_NAME}\ngcloud ai-platform models create ${MODEL_NAME} --regions ${REGION}\ngcloud ai-platform versions create ${MODEL_VERSION} \\\n --model=${MODEL_NAME} \\\n --origin=${MODEL_LOCATION} \\\n --runtime-version=2.1 \\\n --python-version=3.7", "_____no_output_____" ] ], [ [ "## Lab Task #3: Use model to make online prediction.\n\nComplete __#TODO__s for both the Python and gcloud Shell API methods of calling our deployed model on Cloud AI Platform for online prediction.", "_____no_output_____" ], [ "### Python API\n\nWe can use the Python API to send a JSON request to the endpoint of the service to make it predict a baby's weight. The order of the responses are the order of the instances.", "_____no_output_____" ] ], [ [ "from oauth2client.client import GoogleCredentials\nimport requests\nimport json\n\nMODEL_NAME = # TODO: Add model name\nMODEL_VERSION = # TODO: Add model version\n\ntoken = GoogleCredentials.get_application_default().get_access_token().access_token\napi = \"https://ml.googleapis.com/v1/projects/{}/models/{}/versions/{}:predict\" \\\n .format(PROJECT, MODEL_NAME, MODEL_VERSION)\nheaders = {\"Authorization\": \"Bearer \" + token }\ndata = {\n \"instances\": [\n {\n \"is_male\": \"True\",\n \"mother_age\": 26.0,\n \"plurality\": \"Single(1)\",\n \"gestation_weeks\": 39\n },\n {\n \"is_male\": \"False\",\n \"mother_age\": 29.0,\n \"plurality\": \"Single(1)\",\n \"gestation_weeks\": 38\n },\n {\n \"is_male\": \"True\",\n \"mother_age\": 26.0,\n \"plurality\": \"Triplets(3)\",\n \"gestation_weeks\": 39\n },\n # TODO: Create another instance\n ]\n}\nresponse = requests.post(api, json=data, headers=headers)\nprint(response.content)", "_____no_output_____" ] ], [ [ "The predictions for the four instances were: 5.33, 6.09, 2.50, and 5.86 pounds respectively when I ran it (your results might be different).", "_____no_output_____" ], [ "### gcloud shell API\n\nInstead we could use the gcloud shell API. Create a newline delimited JSON file with one instance per line and submit using gcloud.", "_____no_output_____" ] ], [ [ "%%writefile inputs.json\n{\"is_male\": \"True\", \"mother_age\": 26.0, \"plurality\": \"Single(1)\", \"gestation_weeks\": 39}\n{\"is_male\": \"False\", \"mother_age\": 26.0, \"plurality\": \"Single(1)\", \"gestation_weeks\": 39}", "_____no_output_____" ] ], [ [ "Now call `gcloud ai-platform predict` using the JSON we just created and point to our deployed `model` and `version`.", "_____no_output_____" ] ], [ [ "%%bash\ngcloud ai-platform predict \\\n --model=babyweight \\\n --json-instances=inputs.json \\\n --version=# TODO: Add model version", "_____no_output_____" ] ], [ [ "## Lab Task #4: Use model to make batch prediction.\n\nBatch prediction is commonly used when you have thousands to millions of predictions. It will create an actual Cloud AI Platform job for prediction. Complete __#TODO__s so we can call our deployed model on Cloud AI Platform for batch prediction.", "_____no_output_____" ] ], [ [ "%%bash\nINPUT=gs://${BUCKET}/babyweight/batchpred/inputs.json\nOUTPUT=gs://${BUCKET}/babyweight/batchpred/outputs\ngsutil cp inputs.json $INPUT\ngsutil -m rm -rf $OUTPUT \ngcloud ai-platform jobs submit prediction babypred_$(date -u +%y%m%d_%H%M%S) \\\n --data-format=TEXT \\\n --region ${REGION} \\\n --input-paths=$INPUT \\\n --output-path=$OUTPUT \\\n --model=babyweight \\\n --version=# TODO: Add model version", "_____no_output_____" ] ], [ [ "## Lab Summary:\nIn this lab, we set up the environment, deployed a trained Keras model to Cloud AI Platform, online predicted from deployed model on Cloud AI Platform, and batch predicted from deployed model on Cloud AI Platform.", "_____no_output_____" ], [ "Copyright 2019 Google Inc. Licensed under the Apache License, Version 2.0 (the \"License\"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "#!/usr/bin/python3\nimport sys\nsys.path.insert(0, '../src/')\n\nfrom ntree import *\nfrom tree_vis import *\n\nimport numpy as np\nimport matplotlib.pyplot as plt", "_____no_output_____" ] ], [ [ "### Summary\nWe are given : \n* a positive integer $n$ which is the number of dimension of the space. \n* a positive integer $k$ which is the total number of discrete points we need to place inside that space. \n* each continuous point $x$ queried in that space that follows a probability density function $PDF$.\n\nAnd we are asked for: \n* the $k$ discrete points that are placed in a way that minimizes the average distance (or mean error $ME$) from those continuous points $x$ .\n\n# Proposed approach\n\nWe propose a method that: \n* is an almost optimal solution that provides a decent decrease of the $ME$.\n* works for almost any number $k$ of points.\n* works for almost any number $n$ of dimensions.\n* adapts very fast (close to $\\log k$ rate).\n* is always better than the $ME$ provided by a uniform discretization, even before it is adapted.\n* is \"model\" free. The $PDF$ of the appearance of $x$'s don't have to be known.\n\n## How it works\n\nThe core idea behind this method is the $2^n$-trees. Like quadtrees and octrees with $n$ equal to 2 and 3 respectively, n-trees have any positive integer number $n$ as spatial subdivision factor, or also known as branching factor. $n$ describes the dimensions of the space that these trees spread. The $2^n$ child branches of each parent node, expand to all the basic directions in order to fill up all the space as uniform as possible. \n\nEach node is a point in space that covers a specified region. This region is the corresponding fraction of its parent's region. In other words, the parent splits the region that is assigned to it to $2^n$ equal sub-regions with volume $\\frac{1}{2^n}$ times smaller than their parent's. The sub-regions that are produced are n-dimensional cubes too and have no overlapping except their touching surfaces. Of coursethey fully overlap with their parent because they are actually its sub-set. As the height of the tree is growing, the nodes that are created have smaller and smaller volumes. Each point is located in the middle of the cube so there will be no way for two or more points to be in the same location. So a tree with $k$ nodes will have $k$ different points inside the given space.\n\nWe assign to the root node the whole space we want to use, in our case the unit cube between the points $[0, 0, ..., 0]$ and $[1, 1, ..., 1]$. For example, lets take $n=2$. Root, the node in level zero, will have the point $[0.5,0.5]$ and determined by the vertices $[0,0],[1,1]$. Its $2^n=4$ branches, will split this area in 4 equal 2d-cubes (aka rectangles) with vertices and points: \n* cube 1 -> vertices $[0,0],[0.5,0.5]$ , point $[0.25,0.25]$ \n* cube 2 -> vertices $[0.5,0],[1,0.5]$ , point $[0.75,0.25]$ \n* cube 3 -> vertices $[0,0.5],[0.5,1]$ , point $[0.25,0.75]$ \n* cube 4 -> vertices $[0.5,0.5],[1,1]$ , point $[0.75,0.75]$ \n\nSo now we have 5 points to fill up this space. Those 4 branches can be extended to create 16 new sub-branches (4 more branches each) adding to a total of 21 points.", "_____no_output_____" ] ], [ [ "tree = Tree(2, 21)\ntree.plot()\n\npoints = tree.get_points()\n\nplt.figure()\nfor p in points:\n plt.plot(p[0], p[1], 'bo', label='Points')\nplt.plot([0, 0], [0, 1], 'g--', label='Space')\nplt.plot([0, 1], [0, 0], 'g--')\nplt.plot([1, 1], [0, 1], 'g--')\nplt.plot([1, 0], [1, 1], 'g--')\n\nplt.xticks(np.linspace(0, 1, 9))\nplt.yticks(np.linspace(0, 1, 9))\nplt.title('Points spread into space')\nplt.grid(True)\nplt.legend()\nplt.show()", "_____no_output_____" ] ], [ [ "Additionally, nodes keep some useful informations that help **evaluate** them. It's very important to know what _ME_ each node produces and what it would have produced if it wasn't there. When user queries a point $x$, we find the corresponding node by starting from root and recursively finding all the sub-branches until that one leaf, that is responsible for the region that this point belongs. This search will not necessarily give us the nearest point to $x$, but it is very useful for two other things.\n* First of all, the result may not be the nearest point, but it will be the node that is __responsible for this specific area__. This means that if we want to increase the resolution of points, near that particular $x$ we have to expand the node that will be returned from that search. Using our previous example, for $x=[0.4, 0.4]$ the result of the search will be the node with the point $[0.25, 0.25]$ even if root's point $[0.5, 0.5]$ is closer. But expanding this node will create a branch with point $[0.375, 0.375]$. \n* The second benefit is that while doing this search, __we can easily collect the information we need for the evaluation__, like the distance from the responsible point and the distance from its parent's point. Using again our previous example, searching for $[0.4, 0,4]$ will traverse root and its branch with the point $[0.25, 0.25]$. In the process we can find the Euclidean distance of $x$ to each node point, which is $dist([0.5, 0.5], [0.4, 0.4])=0.141421$ and $dist([0.25, 0.25], [0.4, 0.4])=0.212132$ respectively. This gives us useful information about each node's utility.\n\n\n## Adaption of the tree\nWe said before that the points of the nodes are have fixed locations, which make the unable to move and adapt by themselves. __By creating and deleting points we try to match the right set of points that match the PDF the most__. So the adaption is actually a fitting. And we have all the required components to do that. We can expand and prune nodes to increase and decrease resolution and we have a way of evaluating them to choose what is the best for each one of them. Also, in order to ensure that the adaption will produce stable results, and will always decrease the total %ME%, we make all the changes only to the outer level. The maximum difference between the old tree and the updated one is restricted to 1 level for each region.\n\n### Increase resolution\nIn order to decrease $ME$ we need to increase the resolution of points in regions that is needed. Obviously, the way to increase resolution in the desired regions is to expand the corresponding nodes. But there are two types of nodes. The expandable ones, that have unexpanded sub-branches, and the not expandable that are already fully expanded. \n\nTo expand an expandable node we just create all its sub-branches. New points are spread to all directions trying to fill up uniformly the space around the parent point. The goal is actually to search this region for a place of interest. Points that fall into this place will marked as useful by the evaluation process and will \"survive\" as the other will be deleted in the next update.\n\nExpanding a non-expandable node is trickier than expanding an expandable one. As we said before, this node is already fully expanded. But if our evaluation system says that we need to do that, we can't ignore it. Fully expand its closest expandable children would be a solution but it will create too many useless new nodes. We need a more targeted expansion. We can partly expand all the $2^n$ expandable nodes that are closest to the this node. By partly i mean only the node that is towards the desired point.\n\nTo generalize into a more compact solution, we create all the nodes that will approach the point we want independently of the current state of the corresponding node. This means that direct (if there are any) and more distant sub-branches will be created. So expansion always creates $2^n$ new nodes, but not necessary in the same levels. \n\n\n### Decrease resolution\nDecreasing resolution almost any time increases $ME$ but is necessary to maintain the right number of points. So we delete a node to free space for another node to be created. The goal of each prune is to increase the total $ME$ less than the amount that it will decreased by the following expansion. Pruning is done by simply deleting/cutting a leaf node from its parent. Because leaf nodes have no sub-branches, the total number of points is decreased by 1. \n\n\n### Decide which node to expand and which to cut\nOur goal is to decrease the $ME$ as much as possible so we get as close as possible to the optimal solution. The only criterion we can use is the evaluation of each node, which depends only on the distances collected while searching. We have available for each discrete point the sum of distances of each continuous point searched near it. Also, we store in every parent node, the total distance that will be collected if each of its child wasn't available. In the example we used earlier, for $x=[0.2, 0.2]$, we will add to the total distance counter of node $[0.25, 0.25]$ (which is the nearest neighbor) the number $dist([0.25, 0.25], [0.2, 0.2])=0.070711$. Also we will store on its parent node, $[0.5, 0.5]$, the information of the increase of the total distance that we would get if node $[0.25, 0.25]$ wasn't there. So the corresponding counter of node $[0.5, 0.5]$ for the node $[0.25, 0.25]$ will increased by $dist([0.5, 0.5], [0.2, 0.2])=0.424264$. With that information, we can predict how much the total $ME$ will increase if we cut node $[0.25, 0.25]$.\n\nNow that we have available the information we need, we make an update to our tree. Each update has 2 steps. One pass for pruning first to free some space and one for expanding.\n* On pruning, we take all the leaf nodes and we cut the ones that will not produce more $ME$ than the average $ME$ of all nodes.\n* On expanding, expand the nodes with the highest $ME$ until the tree reaches the desired size again. In case that there are not that many nodes to expand, the tree will stay incomplete until the next update.", "_____no_output_____" ], [ "## Comparison with other approaches\nThis solution contains many ideas and methods used to solve similar problem. Though, none of them have the same goals as this one.\n\nThe most similar approaches to this one, are quad and octrees used for spatial subdivision([AMR](https://en.wikipedia.org/wiki/Adaptive_mesh_refinement)) and collision detection. Except the fixed dimensions $n$ of these methods, there are differences on the adaption procedure too. First of all, almost none of them have a fixed number of points and they are all increasing resolution on the regions that is needed, while never decrease it on other regions to maintain their upper limit. The increase of the resolution is also different, while these methods use as points only the leaves of the tree and discard all the parent nodes inside it. This means that each single expansion results on an increase of $2^n-1$ (1 for the deleted parent) points, instead of $2^n$ of my approach. This seems like a disadvantage but it is not. When discarding the parent, you become unable (or you make it more difficult) to delete individual leaf nodes and can only merge the whole branch to recreate the parent. In the case that you need only one of the children, you are forced to keep all the other $2^n-1$ of them and this limits the maximum level that your tree can reach. In our method, the overhead is 1 node which means that the maximum level[\\*]() the tree can reach is the total number of points $k$. Of coursethis is the worst case scenario for their method and the best for ours, but in any case our method gives more freedom to the tree and reduces the overhead to minimum, as even the overheads are points in space that many times are required too.\n", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "from io import open\nimport pickle\n\nclass Pelicula:\n \n # Constructor de clase\n def __init__(self, titulo, duracion, lanzamiento):\n self.titulo = titulo\n self.duracion = duracion\n self.lanzamiento = lanzamiento\n print('Se ha creado la película:',self.titulo)\n \n def __str__(self):\n return '{} ({})'.format(self.titulo, self.lanzamiento)\n\n\nclass Catalogo:\n \n peliculas = []\n \n # Constructor de clase\n def __init__(self, peliculas=[]):\n self.peliculas = peliculas\n \n def agregar(self,p):\n self.peliculas.append(p)\n \n def mostrar(self):\n for p in self.peliculas:\n print(p)", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "## XYZ Pro Features\nThis notebook demonstrates some of the pro features for XYZ Hub API.\n\nXYZ paid features can be found here: [xyz pro features](https://www.here.xyz/xyz_pro/).\n\nXYZ plans can be found here: [xyz plans](https://developer.here.com/pricing).", "_____no_output_____" ], [ "### Virtual Space\nA virtual space is described by definition which references other existing spaces(the upstream spaces).\nQueries being done to a virtual space will return the features of its upstream spaces combined.\n\nBelow are different predefined operations of how to combine the features of the upstream spaces.\n\n- [group](#group_cell)\n- [merge](#merge_cell)\n- [override](#override_cell)\n- [custom](#custom_cell)", "_____no_output_____" ] ], [ [ "# Make necessary imports.\nimport os\nimport json\nimport warnings\n\nfrom xyzspaces.datasets import get_chicago_parks_data, get_countries_data\nfrom xyzspaces.exceptions import ApiError\nimport xyzspaces", "_____no_output_____" ] ], [ [ "<div class=\"alert alert-block alert-warning\">\n<b>Warning:</b> Before running below cells please make sure you have XYZ Token to interact with xyzspaces. \n Please see README.md in notebooks folder for more info on XYZ_TOKEN\n</div>", "_____no_output_____" ] ], [ [ "os.environ[\"XYZ_TOKEN\"] = \"MY-XYZ-TOKEN\" # Replace your token here.", "_____no_output_____" ], [ "xyz = xyzspaces.XYZ()", "_____no_output_____" ], [ "# create two spaces which will act as upstream spaces for virtual space created later.\n\ntitle1 = \"Testing xyzspaces\"\ndescription1 = \"Temporary space containing countries data.\"\nspace1 = xyz.spaces.new(title=title1, description=description1)\n\n# Add some data to it space1\ngj_countries = get_countries_data()\nspace1.add_features(features=gj_countries)\nspace_id1 = space1.info[\"id\"]\n\ntitle2 = \"Testing xyzspaces\"\ndescription2 = \"Temporary space containing Chicago parks data.\"\nspace2 = xyz.spaces.new(title=title2, description=description2)\n\n# Add some data to space2\nwith open(\"./data/chicago_parks.geo.json\", encoding=\"utf-8-sig\") as json_file:\n gj_chicago = json.load(json_file)\n\nspace2.add_features(features=gj_chicago)\nspace_id2 = space2.info[\"id\"]", "_____no_output_____" ] ], [ [ "<a id='group_cell'></a>\n#### Group\nGroup means to combine the content of the specified spaces. All objects of each space will be part of the response when the virtual space is queried by the user. The information about which object came from which space can be found in the XYZ-namespace in the properties of each feature. When writing back these objects to the virtual space they'll be written back to the upstream space from which they were actually coming.", "_____no_output_____" ] ], [ [ "# Create a new virtual space by grouping two spaces created above.\n\ntitle = \"Virtual Space for coutries and Chicago parks data\"\ndescription = \"Test group functionality of virtual space\"\n\nupstream_spaces = [space_id1, space_id2]\nkwargs = {\"virtualspace\": dict(group=upstream_spaces)}\n\nvspace = xyz.spaces.virtual(title=title, description=description, **kwargs)\nprint(json.dumps(vspace.info, indent=2))", "_____no_output_____" ], [ "# Reading a particular feature from space1 via virtual space.\n\nvfeature1 = vspace.get_feature(feature_id=\"FRA\")\nfeature1 = space1.get_feature(feature_id=\"FRA\")\nassert vfeature1 == feature1\n\n# Reading a particular feature from space2 via virtual space.\nvfeature2 = vspace.get_feature(feature_id=\"LP\")\nfeature2 = space2.get_feature(feature_id=\"LP\")\nassert vfeature2 == feature2", "_____no_output_____" ], [ "# Deleting a feature from virtual space deletes corresponding feature from upstream space.\n\nvspace.delete_feature(feature_id=\"FRA\")\ntry:\n space1.get_feature(\"FRA\")\nexcept ApiError as err:\n print(err)", "_____no_output_____" ], [ "# Delete temporary spaces created.\nvspace.delete()\nspace1.delete()\nspace2.delete()", "_____no_output_____" ] ], [ [ "<a id='merge_cell'></a>\n#### Merge\nMerge means that objects with the same ID will be merged together. If there are duplicate feature-IDs in the various data of the upstream spaces, the duplicates will be merged to build a single feature. The result will be a response that is guaranteed to have no features with duplicate IDs. The merge will happen in the order of the space-references in the specified array. That means objects coming from the second space will overwrite potentially existing property values of objects coming from the first space. The information about which object came from which space(s) can be found in the XYZ-namespace in the properties of each feature. When writing back these objects to the virtual space they'll be written back to the upstream space from which they were actually coming, or the last one in the list if none was specified.When deleting features from the virtual space a new pseudo-deleted feature is written to the last space in the list. Trying to read the feature with that ID from the virtual space is not possible afterward.", "_____no_output_____" ] ], [ [ "# create two spaces with duplicate data\n\ntitle1 = \"Testing xyzspaces\"\ndescription1 = \"Temporary space containing Chicago parks data.\"\nspace1 = xyz.spaces.new(title=title1, description=description1)\n\nwith open(\"./data/chicago_parks.geo.json\", encoding=\"utf-8-sig\") as json_file:\n gj_chicago = json.load(json_file)\n\n# Add some data to it space1\nspace1.add_features(features=gj_chicago)\nspace_id1 = space1.info[\"id\"]\n\ntitle2 = \"Testing xyzspaces duplicate\"\ndescription2 = \"Temporary space containing Chicago parks data duplicate\"\nspace2 = xyz.spaces.new(title=title1, description=description1)\n\n# Add some data to it space2\nspace2.add_features(features=gj_chicago)\nspace_id2 = space2.info[\"id\"]", "_____no_output_____" ], [ "# update a particular feature of second space so that post merge virtual space will have this feature merged\nlp = space2.get_feature(\"LP\")\nspace2.update_feature(feature_id=\"LP\", data=lp, add_tags=[\"foo\", \"bar\"])", "_____no_output_____" ], [ "# Create a new virtual space by merging two spaces created above.\n\ntitle = \"Virtual Space for coutries and Chicago parks data\"\ndescription = \"Test merge functionality of virtual space\"\n\nupstream_spaces = [space_id1, space_id2]\nkwargs = {\"virtualspace\": dict(merge=upstream_spaces)}\n\nvspace = xyz.spaces.virtual(title=title, description=description, **kwargs)\nprint(vspace.info)", "_____no_output_____" ], [ "vfeature1 = vspace.get_feature(feature_id=\"LP\")\nassert vfeature1[\"properties\"][\"@ns:com:here:xyz\"][\"tags\"] == [\"foo\", \"bar\"]", "_____no_output_____" ], [ "bp = space2.get_feature(\"BP\")\nspace2.update_feature(feature_id=\"BP\", data=lp, add_tags=[\"foo1\", \"bar1\"])", "_____no_output_____" ], [ "vfeature2 = vspace.get_feature(feature_id=\"BP\")\nassert vfeature2[\"properties\"][\"@ns:com:here:xyz\"][\"tags\"] == [\"foo1\", \"bar1\"]", "_____no_output_____" ], [ "space1.delete()\nspace2.delete()\nvspace.delete()", "_____no_output_____" ] ], [ [ "<a id='override_cell'></a>\n#### Override\nOverride means that objects with the same ID will be overridden completely. If there are duplicate feature-IDs in the various data of the upstream spaces, the duplicates will be overridden to result in a single feature. The result will be a response that is guaranteed to have no features with duplicate IDs. The override will happen in the order of the space-references in the specified array. That means objects coming from the second space one will override potentially existing features coming from the first space. The information about which object came from which space can be found in the XYZ-namespace in the properties of each feature. When writing back these objects to the virtual space they'll be written back to the upstream space from which they were actually coming. When deleting features from the virtual space the same rules as for merge apply.", "_____no_output_____" ] ], [ [ "# create two spaces with duplicate data\n\ntitle1 = \"Testing xyzspaces\"\ndescription1 = \"Temporary space containing Chicago parks data.\"\nspace1 = xyz.spaces.new(title=title1, description=description1)\n\nwith open(\"./data/chicago_parks.geo.json\", encoding=\"utf-8-sig\") as json_file:\n gj_chicago = json.load(json_file)\n\n# Add some data to it space1\nspace1.add_features(features=gj_chicago)\nspace_id1 = space1.info[\"id\"]\n\ntitle2 = \"Testing xyzspaces duplicate\"\ndescription2 = \"Temporary space containing Chicago parks data duplicate\"\nspace2 = xyz.spaces.new(title=title1, description=description1)\n\n# Add some data to it space2\nspace2.add_features(features=gj_chicago)\nspace_id2 = space2.info[\"id\"]", "_____no_output_____" ], [ "# Create a new virtual space by override operation.\n\ntitle = \"Virtual Space for coutries and Chicago parks data\"\ndescription = \"Test merge functionality of virtual space\"\n\nupstream_spaces = [space_id1, space_id2]\nkwargs = {\"virtualspace\": dict(override=upstream_spaces)}\n\nvspace = xyz.spaces.virtual(title=title, description=description, **kwargs)\nprint(vspace.info)", "_____no_output_____" ], [ "bp = space2.get_feature(\"BP\")\nspace2.update_feature(feature_id=\"BP\", data=bp, add_tags=[\"foo1\", \"bar1\"])", "_____no_output_____" ], [ "vfeature2 = vspace.get_feature(feature_id=\"BP\")\nassert vfeature2[\"properties\"][\"@ns:com:here:xyz\"][\"tags\"] == [\"foo1\", \"bar1\"]", "_____no_output_____" ], [ "space1.delete()\nspace2.delete()\nvspace.delete()", "_____no_output_____" ] ], [ [ "### Applying clustering in space", "_____no_output_____" ] ], [ [ "# create two spaces which will act as upstream spaces for virtual space created later.\n\ntitle1 = \"Testing xyzspaces\"\ndescription1 = \"Temporary space containing countries data.\"\nspace1 = xyz.spaces.new(title=title1, description=description1)\n\n# Add some data to it space1\ngj_countries = get_countries_data()\nspace1.add_features(features=gj_countries)\nspace_id1 = space1.info[\"id\"]", "_____no_output_____" ], [ "# Genereate clustering for the space\nspace1.cluster(clustering=\"hexbin\")", "_____no_output_____" ], [ "# Delete created space\nspace1.delete()", "_____no_output_____" ] ], [ [ "### Rule based Tagging\nRule based tagging makes tagging multiple features in space tagged to a particular tag, based in rules mentioned based on JSON-path expression. Users can update space with a map of rules where the key is the tag to be applied to all features matching the JSON-path expression being the value.\n\nIf multiple rules are matching, multiple tags will be applied to the according to matched sets of features. It could even happen that a feature is matched by multiple rules and thus multiple tags will get added to it.", "_____no_output_____" ] ], [ [ "# Create a new space\ntitle = \"Testing xyzspaces\"\ndescription = \"Temporary space containing Chicago parks data.\"\nspace = xyz.spaces.new(title=title, description=description)", "_____no_output_____" ], [ "# Add data to the space.\nwith open(\"./data/chicago_parks.geo.json\", encoding=\"utf-8-sig\") as json_file:\n gj_chicago = json.load(json_file)\n_ = space.add_features(features=gj_chicago)", "_____no_output_____" ], [ "# update space to add tagging rules to the above mentioned space.\ntagging_rules = {\n \"large\": \"$.features[?(@.properties.area>=500)]\",\n \"small\": \"$.features[?(@.properties.area<500)]\",\n}\n_ = space.update(tagging_rules=tagging_rules)", "_____no_output_____" ], [ "# verify that features are tagged correctly based on rules.\nlarge_parks = space.search(tags=[\"large\"])\nfor park in large_parks:\n assert park[\"id\"] in [\"LP\", \"BP\", \"JP\"]\nsmall_parks = space.search(tags=[\"small\"])\nfor park in small_parks:\n assert park[\"id\"] in [\"MP\", \"GP\", \"HP\", \"DP\", \"CP\", \"COP\"]", "_____no_output_____" ], [ "# Delete created space\nspace.delete()", "_____no_output_____" ] ], [ [ "### Activity Log\nThe Activity log will enable tracking of changes in your space.\nTo activate it, just create a space with the listener added and enable_uuid set to True.\nMore information on the activity log can be found [here](https://www.here.xyz/api/devguide/activitylogguide/).", "_____no_output_____" ] ], [ [ "title = \"Activity-Log Test\"\ndescription = \"Activity-Log Test\"\nlisteners = {\n \"id\": \"activity-log\",\n \"params\": {\"states\": 5, \"storageMode\": \"DIFF_ONLY\", \"writeInvalidatedAt\": \"true\"},\n \"eventTypes\": [\"ModifySpaceEvent.request\"],\n}\nspace = xyz.spaces.new(\n title=title,\n description=description,\n enable_uuid=True,\n listeners=listeners,\n)", "_____no_output_____" ], [ "from time import sleep\n\n# As activity log is async operation adding sleep to get info\nsleep(5)\nprint(json.dumps(space.info, indent=2))", "_____no_output_____" ], [ "space.delete()", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Pyber Ride Sharing\n\n3 observations from the data:\n* Urban drivers typically drive more frequently yet charge on average (i.e., <30) less than rural drivers.\n* Roughly two-thirds of all rides occur in Urban cities, however, roughly 80% of all drivers work in Urban areas.\n* While less rides occur in rural cities, there are on average less drivers to manage the load, creating a more favorable driver to ride ratio. \n* Rural drivers have the greatest fare distribution (i.e., roughly 40 dollars/driver) among drivers of all 3 city types.", "_____no_output_____" ] ], [ [ "import pandas as pd\nimport matplotlib.pyplot as plt\nimport numpy as np", "_____no_output_____" ], [ "# Read in City Data csv file\ncity_df = pd.read_csv('city_data.csv')\n\n# Read in Ride Data csv file\nride_df = pd.read_csv('ride_data.csv')", "_____no_output_____" ], [ "# Combine the 2 dataframes\npyber_df = pd.merge(city_df, ride_df, on=\"city\", how='left')\npyber_df.head()", "_____no_output_____" ], [ "# Find the total fare per city\ncity_fare_total = pyber_df.groupby('city')['fare'].sum().to_frame()\n\n# Find the average fare ($) per city\ncity_fare_avg = pyber_df.groupby('city')['fare'].mean().to_frame()\n\n# Find the total number of rides per city\ncity_total_rides = pyber_df.groupby('city')['ride_id'].count().to_frame()\n\n# Find the total number of drivers per city\ncity_driver_count = pyber_df.groupby('city')['driver_count'].unique().to_frame()\ncity_driver_count['driver_count'] = city_driver_count['driver_count'].str.get(0)\n\n# Find the city type (urban, suburban, rural)\ncity_type = pyber_df.groupby('city')['type'].unique().to_frame()\ncity_type['type'] = city_type['type'].str.get(0)\n", "_____no_output_____" ], [ "# Combine each dataframe\ncity_fare_avg.columns=[\"city\"]\njoin_one = city_fare_avg.join(city_total_rides, how=\"left\")\njoin_one.columns=[\"Average Fare\", \"Total Rides\"]\n\njoin_two = join_one.join(city_fare_total, how=\"inner\")\njoin_two.columns=[\"Average Fare\", \"Total Rides\", \"City Fare Total\"]\n\njoin_three = join_two.join(city_driver_count, how=\"inner\")\njoin_three.columns=[\"Average Fare\", \"Total Rides\", \"City Fare Total\", \"Driver Count\"]\n\ncity_agg = join_three.join(city_type, how='inner')\ncity_agg.columns=[\"Average Fare\", \"Total Rides\", \"City Fare Total\", \"Driver Count\", \"City Type\"]\ncity_agg.head()", "_____no_output_____" ], [ "# Separate data by City Type\nurban_data = city_agg.loc[(city_agg['City Type']=='Urban'), :]\nsuburban_data = city_agg.loc[(city_agg['City Type']=='Suburban'), :]\nrural_data = city_agg.loc[(city_agg['City Type']=='Rural'), :]", "_____no_output_____" ] ], [ [ "## Bubble Plot", "_____no_output_____" ] ], [ [ "## Bubble Plot Data\nall_urban_rides = urban_data.groupby('city')['Total Rides'].sum()\navg_urban_fare = urban_data.groupby('city')['Average Fare'].mean()\n\nall_suburban_rides = suburban_data.groupby('city')['Total Rides'].sum()\navg_suburban_fare = suburban_data.groupby('city')['Average Fare'].mean()\n\nall_rural_rides = rural_data.groupby('city')['Total Rides'].sum()\navg_rural_fare = rural_data.groupby('city')['Average Fare'].mean()\n", "_____no_output_____" ], [ "## Bubble Plot\n\n# Store driver count as a Numpy Array\nnp_city_driver_count = np.array(city_driver_count)\nnp_city_driver_count = np_city_driver_count * 3\n\n# Add chart note\ntextstr = 'Note: Circle size corresponds to driver count/city'\n\nurban = plt.scatter(all_urban_rides, avg_urban_fare, s=np_city_driver_count, color='lightskyblue', alpha=0.65, edgecolors='none')\nsuburban = plt.scatter(all_suburban_rides, avg_suburban_fare, s=np_city_driver_count, color='gold', alpha=0.65, edgecolors='none')\nrural = plt.scatter(all_rural_rides, avg_rural_fare, s=np_city_driver_count, color='lightcoral', alpha=0.65, edgecolors='none')\n\nplt.grid(linestyle='dotted')\nplt.xlabel('Total Number of Rides (Per City)')\nplt.ylabel('Average Fare ($)')\nplt.title('Pyber Ride Sharing Data (2016)')\nplt.gcf().text(0.95, 0.50, textstr, fontsize=8)\nplt.legend((urban, suburban, rural),('Urban', 'Suburban', 'Rural'),scatterpoints=1,loc='upper right',ncol=1,\\\n fontsize=8, markerscale=0.75,title='City Type', edgecolor='none',framealpha=0.25)\n\nplt.show()\n", "_____no_output_____" ] ], [ [ "## Pie Charts", "_____no_output_____" ], [ "### Total Fares by City Type", "_____no_output_____" ] ], [ [ "## Find Total Fares By City Type\nurban_fare_total = urban_data['City Fare Total'].sum()\nsuburban_fare_total = suburban_data['City Fare Total'].sum()\nrural_fare_total = rural_data['City Fare Total'].sum()\n\n\n# Create a Pie Chart to Express the Above Date\ndriver_type = [\"Urban\", \"Suburban\", \"Rural\"]\ndriver_count = [urban_fare_total, suburban_fare_total, rural_fare_total]\ncolors = [\"lightskyblue\", \"gold\",\"lightcoral\"]\nexplode = (0.1,0,0)\nplt.pie(driver_count, explode=explode, labels=driver_type, colors=colors,\n \tautopct=\"%1.1f%%\", shadow=True, startangle=68)\nplt.title(\"% of Total Fares by City Type\")\nplt.axis(\"equal\")\nplt.show()", "_____no_output_____" ] ], [ [ "### Total Rides by City Type", "_____no_output_____" ] ], [ [ "## Find Total Rides By City Type\nurban_rides_count = urban_data['Total Rides'].sum()\nsuburban_rides_count = suburban_data['Total Rides'].sum()\nrural_rides_count = rural_data['Total Rides'].sum()\n\n\n# Create a Pie Chart to Express the Above Date\nride_type = [\"Urban\", \"Suburban\", \"Rural\"]\nride_count = [urban_rides_count, suburban_rides_count, rural_rides_count]\ncolors = [\"lightskyblue\", \"gold\",\"lightcoral\"]\nexplode = (0.1,0,0)\nplt.pie(ride_count, explode=explode, labels=ride_type, colors=colors,\n \tautopct=\"%1.1f%%\", shadow=True, startangle=60)\nplt.title(\"% of Total Rides by City Type\")\nplt.axis(\"equal\")\nplt.show()\n", "_____no_output_____" ] ], [ [ "### Total Drivers by City Type", "_____no_output_____" ] ], [ [ "## Find Total Drivers By City Type\nurban_driver_count = urban_data['Driver Count'].sum()\nsuburban_driver_count = suburban_data['Driver Count'].sum()\nrural_driver_count = rural_data['Driver Count'].sum()\n\n\n# Create a Pie Chart to Express the Above Date\ndriver_type = [\"Urban\", \"Suburban\", \"Rural\"]\ndriver_count = [urban_driver_count, suburban_driver_count, rural_driver_count]\ncolors = [\"lightskyblue\", \"gold\",\"lightcoral\"]\nexplode = (0.1,0,0)\nplt.pie(driver_count, explode=explode, labels=driver_type, colors=colors,\n \tautopct=\"%1.1f%%\", shadow=True, startangle=40)\nplt.title(\"% of Total Drivers by City Type\")\nplt.axis(\"equal\")\nplt.show()\n", "_____no_output_____" ] ], [ [ "## Average Ride Value Per Driver (by City Type)", "_____no_output_____" ] ], [ [ "# Identify the average fare for drivers in each city type\nurban_avg_driver_pay = urban_fare_total / urban_rides_count\nsuburban_avg_driver_pay = suburban_fare_total / suburban_rides_count\nrural_avg_driver_pay = rural_fare_total / rural_rides_count\n\n# Create a Bar Chart to Express the Above Date\ndriver_type = [\"Urban\", \"Suburban\", \"Rural\"]\navg_driver_pay = [urban_avg_driver_pay, suburban_avg_driver_pay, rural_avg_driver_pay]\nx_axis = np.arange(len(avg_driver_pay))\n\ncolors = [\"lightskyblue\", \"gold\",\"lightcoral\"]\nplt.bar(x_axis, avg_driver_pay, color=colors, align='edge')\ntick_locations = [value+0.4 for value in x_axis]\nplt.xticks(tick_locations, [\"Urban\", \"Suburban\", \"Rural\"])\nplt.ylim(0, max(avg_driver_pay)+1)\nplt.xlim(-0.25, len(driver_type))\n\n\nplt.title(\"Average Per Ride Value for Drivers\")\n\nplt.show()", "_____no_output_____" ] ], [ [ "## Average Fare Distribution Across All Drivers (by City Type)", "_____no_output_____" ] ], [ [ "urban_fare_dist = urban_fare_total / urban_driver_count\nsuburban_fare_dist = suburban_fare_total / suburban_driver_count\nrural_fare_dist = rural_fare_total / rural_driver_count\n\n# Create a Bar Chart to Express the Above Date\ndriver_type = [\"Urban\", \"Suburban\", \"Rural\"]\navg_fare_dist = [urban_fare_dist, suburban_fare_dist, rural_fare_dist]\nx_axis = np.arange(len(avg_fare_dist))\n\ncolors = [\"lightskyblue\", \"gold\",\"lightcoral\"]\nplt.bar(x_axis, avg_fare_dist, color=colors, align='edge')\ntick_locations = [value+0.4 for value in x_axis]\nplt.xticks(tick_locations, [\"Urban\", \"Suburban\", \"Rural\"])\nplt.ylim(0, max(avg_fare_dist)+1)\nplt.xlim(-0.25, len(driver_type))\n\n\nplt.title(\"Average Fare Distribution Across All Drivers\")\n\nplt.show()", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# TODO: Add import statements\n", "_____no_output_____" ], [ "# Assign the data to predictor and outcome variables\n# TODO: Load the data\ntrain_data = None\nX = None\ny = None", "_____no_output_____" ], [ "# TODO: Create the linear regression model with lasso regularization.\nlasso_reg = None", "_____no_output_____" ], [ "# TODO: Fit the model.", "_____no_output_____" ], [ "# TODO: Retrieve and print out the coefficients from the regression model.\nreg_coef = None\nprint(reg_coef)", "_____no_output_____" ] ] ]

[ "code" ]

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# torchserve.ipynb\n\nThis notebook contains code for the portions of the benchmark in [the benchmark notebook](./benchmark.ipynb) that use [TorchServe](https://github.com/pytorch/serve).\n\n", "_____no_output_____" ] ], [ [ "# Imports go here\nimport json\nimport os\nimport requests\n\nimport scipy.special\nimport transformers\n\n# Fix silly warning messages about parallel tokenizers\nos.environ['TOKENIZERS_PARALLELISM'] = 'False'", "_____no_output_____" ], [ "# Constants go here\n\nINTENT_MODEL_NAME = 'mrm8488/t5-base-finetuned-e2m-intent'\nSENTIMENT_MODEL_NAME = 'cardiffnlp/twitter-roberta-base-sentiment'\nQA_MODEL_NAME = 'deepset/roberta-base-squad2'\nGENERATE_MODEL_NAME = 'gpt2'\n\n\nINTENT_INPUT = {\n 'context':\n (\"I came here to eat chips and beat you up, \"\n \"and I'm all out of chips.\")\n}\n\nSENTIMENT_INPUT = {\n 'context': \"We're not happy unless you're not happy.\"\n}\n\nQA_INPUT = {\n 'question': 'What is 1 + 1?',\n 'context': \n \"\"\"Addition (usually signified by the plus symbol +) is one of the four basic operations of \n arithmetic, the other three being subtraction, multiplication and division. The addition of two \n whole numbers results in the total amount or sum of those values combined. The example in the\n adjacent image shows a combination of three apples and two apples, making a total of five apples. \n This observation is equivalent to the mathematical expression \"3 + 2 = 5\" (that is, \"3 plus 2 \n is equal to 5\").\n \"\"\"\n}\n\nGENERATE_INPUT = {\n 'prompt_text': 'All your base are'\n}", "_____no_output_____" ] ], [ [ "## Model Packaging\n\nTorchServe requires models to be packaged up as model archive files. Documentation for this process (such as it is) is [here](https://github.com/pytorch/serve/blob/master/README.md#serve-a-model) and [here](https://github.com/pytorch/serve/blob/master/model-archiver/README.md).\n\n", "_____no_output_____" ], [ "### Intent Model\n\nThe intent model requires the caller to call the pre- and post-processing code manually. Only the model and tokenizer are provided on the model zoo.", "_____no_output_____" ] ], [ [ "# First we need to dump the model into a local directory.\nintent_model = transformers.AutoModelForSeq2SeqLM.from_pretrained(\n INTENT_MODEL_NAME)\nintent_tokenizer = transformers.AutoTokenizer.from_pretrained('t5-base')\n\nintent_model.save_pretrained('torchserve/intent')\nintent_tokenizer.save_pretrained('torchserve/intent')", "_____no_output_____" ] ], [ [ "Next we wrapped the model in a handler class, located at `./torchserve/handler_intent.py`, which \nneeds to be in its own separate Python file in order for the `torch-model-archiver`\nutility to work.\n\nThe following command turns this Python file, plus the data files created by the \nprevious cell, into a model archive (`.mar`) file at `torchserve/model_store/intent.mar`.", "_____no_output_____" ] ], [ [ "%%time\n!mkdir -p torchserve/model_store\n!torch-model-archiver --model-name intent --version 1.0 \\\n --serialized-file torchserve/intent/pytorch_model.bin \\\n --handler torchserve/handler_intent.py \\\n --extra-files \"torchserve/intent/config.json,torchserve/intent/special_tokens_map.json,torchserve/intent/tokenizer_config.json,torchserve/intent/tokenizer.json\" \\\n --export-path torchserve/model_store \\\n --force", "CPU times: user 438 ms, sys: 116 ms, total: 553 ms\nWall time: 54 s\n" ] ], [ [ "### Sentiment Model\n\nThe sentiment model operates similarly to the intent model.", "_____no_output_____" ] ], [ [ "sentiment_tokenizer = transformers.AutoTokenizer.from_pretrained(\n SENTIMENT_MODEL_NAME)\nsentiment_model = (\n transformers.AutoModelForSequenceClassification\n .from_pretrained(SENTIMENT_MODEL_NAME))\n\nsentiment_model.save_pretrained('torchserve/sentiment')\nsentiment_tokenizer.save_pretrained('torchserve/sentiment')", "_____no_output_____" ], [ "contexts = ['hello', 'world']\ninput_batch = sentiment_tokenizer(contexts, padding=True, \n return_tensors='pt')\n\ninference_output = sentiment_model(**input_batch)\n\nscores = inference_output.logits.detach().numpy()\nscores = scipy.special.softmax(scores, axis=1).tolist()\nscores = [{k: v for k, v in zip(['positive', 'neutral', 'negative'], row)}\n for row in scores]\n# return scores\n\nscores", "_____no_output_____" ] ], [ [ "As with the intent model, we created a handler class (located at `torchserve/handler_sentiment.py`), then\npass that class and the serialized model from two cells ago\nthrough the `torch-model-archiver` utility.", "_____no_output_____" ] ], [ [ "%%time\n!torch-model-archiver --model-name sentiment --version 1.0 \\\n --serialized-file torchserve/sentiment/pytorch_model.bin \\\n --handler torchserve/handler_sentiment.py \\\n --extra-files \"torchserve/sentiment/config.json,torchserve/sentiment/special_tokens_map.json,torchserve/sentiment/tokenizer_config.json,torchserve/sentiment/tokenizer.json\" \\\n --export-path torchserve/model_store \\\n --force", "CPU times: user 210 ms, sys: 114 ms, total: 324 ms\nWall time: 24.2 s\n" ] ], [ [ "### Question Answering Model\n\nThe QA model uses a `transformers` pipeline. We squeeze this model into the TorchServe APIs by telling the pipeline to serialize all of its parts to a single directory, then passing the parts that aren't `pytorch_model.bin` in as extra files. At runtime, our custom handler uses the model loading code from `transformers` on the reconstituted model directory.", "_____no_output_____" ] ], [ [ "qa_pipeline = transformers.pipeline('question-answering', model=QA_MODEL_NAME)\nqa_pipeline.save_pretrained('torchserve/qa')", "_____no_output_____" ] ], [ [ "As with the previous models, we wrote a class (located at `torchserve/handler_qa.py`), then\npass that wrapper class and the serialized model through the `torch-model-archiver` utility.", "_____no_output_____" ] ], [ [ "%%time\n!torch-model-archiver --model-name qa --version 1.0 \\\n --serialized-file torchserve/qa/pytorch_model.bin \\\n --handler torchserve/handler_qa.py \\\n --extra-files \"torchserve/qa/config.json,torchserve/qa/merges.txt,torchserve/qa/special_tokens_map.json,torchserve/qa/tokenizer_config.json,torchserve/qa/tokenizer.json,torchserve/qa/vocab.json\" \\\n --export-path torchserve/model_store \\\n --force", "CPU times: user 287 ms, sys: 67.5 ms, total: 354 ms\nWall time: 24.7 s\n" ], [ "data = [QA_INPUT, QA_INPUT]\n\n# Preprocessing\nsamples = [qa_pipeline.create_sample(**r) for r in data]\ngenerators = [qa_pipeline.preprocess(s) for s in samples]\n\n# Inference\ninference_outputs = ((qa_pipeline.forward(example) for example in batch) for batch in generators)\n\npost_results = [qa_pipeline.postprocess(o) for o in inference_outputs]\npost_results", "_____no_output_____" ] ], [ [ "### Natural Language Generation Model\n\nThe text generation model is roughly similar to the QA model, albeit with important differences in how the three stages of the pipeline operate. At least model loading is the same.", "_____no_output_____" ] ], [ [ "generate_pipeline = transformers.pipeline(\n 'text-generation', model=GENERATE_MODEL_NAME)\ngenerate_pipeline.save_pretrained('torchserve/generate')", "_____no_output_____" ], [ "data = [GENERATE_INPUT, GENERATE_INPUT]\n\n\npad_token_id = generate_pipeline.tokenizer.eos_token_id\n\njson_records = data\n\n# preprocess() takes a single input at a time, but we need to do \n# a batch at a time.\ninput_batch = [generate_pipeline.preprocess(**r) for r in json_records]\n\n# forward() takes a single input at a time, but we need to run a\n# batch at a time.\ninference_output = [\n generate_pipeline.forward(r, pad_token_id=pad_token_id)\n for r in input_batch]\n\n# postprocess() takes a single generation result at a time, but we\n# need to run a batch at a time.\ngenerate_result = [generate_pipeline.postprocess(i)\n for i in inference_output]\ngenerate_result", "_____no_output_____" ] ], [ [ "Once again, we wrote a class (located at `torchserve/handler_generate.py`), then\npass that wrapper class and the serialized model through the `torch-model-archiver` utility.", "_____no_output_____" ] ], [ [ "%%time\n!torch-model-archiver --model-name generate --version 1.0 \\\n --serialized-file torchserve/generate/pytorch_model.bin \\\n --handler torchserve/handler_generate.py \\\n --extra-files \"torchserve/generate/config.json,torchserve/generate/merges.txt,torchserve/generate/special_tokens_map.json,torchserve/generate/tokenizer_config.json,torchserve/generate/tokenizer.json,torchserve/generate/vocab.json\" \\\n --export-path torchserve/model_store \\\n --force", "CPU times: user 198 ms, sys: 96 ms, total: 294 ms\nWall time: 24.5 s\n" ] ], [ [ "## Testing\n\nNow we can fire up TorchServe and test our models.\n\nFor some reason, starting TorchServe needs to be done in a proper terminal window. Running the command from this notebook has no effect. The commands to run (from the root of the repository) are:\n\n```\n> conda activate ./env\n> cd notebooks/benchmark/torchserve\n> torchserve --start --ncs --model-store model_store --ts-config torchserve.properties\n```\n\nThen pick up a cup of coffee and a book and wait a while. The startup process is like cold-starting a gas turbine and takes about 10 minutes.\n\nOnce the server has started, we can test our deployed models by making POST requests.", "_____no_output_____" ] ], [ [ "# Probe the management API to verify that TorchServe is running.\nrequests.get('http://127.0.0.1:8081/models').json()", "_____no_output_____" ], [ "port = 8080\n\nintent_result = requests.put(\n f'http://127.0.0.1:{port}/predictions/intent_en', \n json.dumps(INTENT_INPUT)).json()\nprint(f'Intent result: {intent_result}')\n\nsentiment_result = requests.put(\n f'http://127.0.0.1:{port}/predictions/sentiment_en', \n json.dumps(SENTIMENT_INPUT)).json()\nprint(f'Sentiment result: {sentiment_result}')\n\nqa_result = requests.put(\n f'http://127.0.0.1:{port}/predictions/qa_en', \n json.dumps(QA_INPUT)).json()\nprint(f'Question answering result: {qa_result}')\n\ngenerate_result = requests.put(\n f'http://127.0.0.1:{port}/predictions/generate_en', \n json.dumps(GENERATE_INPUT)).json()\nprint(f'Natural language generation result: {generate_result}')", "_____no_output_____" ] ], [ [ "## Cleanup\n\nTorchServe consumes many resources even when it isn't doing anything. When you're done running the baseline portion of the benchmark, be sure to shut down the server by running:\n```\n> torchserve --stop\n```", "_____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" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Lab 1: Markov Decision Processes - Problem 3\n\n\n## Lab Instructions\nAll your answers should be written in this notebook. You shouldn't need to write or modify any other files.\n\n**You should execute every block of code to not miss any dependency.**\n\n*This project was developed by Peter Chen, Rocky Duan, Pieter Abbeel for the Berkeley Deep RL Bootcamp, August 2017. Bootcamp website with slides and lecture videos: https://sites.google.com/view/deep-rl-bootcamp/. It is adapted from CS188 project materials: http://ai.berkeley.edu/project_overview.html.*\n\n--------------------------", "_____no_output_____" ] ], [ [ "import numpy as np, numpy.random as nr, gym\nimport matplotlib.pyplot as plt\n%matplotlib inline\nnp.set_printoptions(precision=3)", "_____no_output_____" ] ], [ [ "### Problem 3: Sampling-based Tabular Q-Learning\n\nSo far we have implemented Value Iteration and Policy Iteration, both of which require access to an MDP's dynamics model. This requirement can sometimes be restrictive - for example, if the environment is given as a blackbox physics simulator, then we won't be able to read off the whole transition model.\n\nWe can however use sampling-based Q-Learning to learn from this type of environments. ", "_____no_output_____" ], [ "For this exercise, we will learn to control a Crawler robot. Let's first try some completely random actions to see how the robot moves and familiarize ourselves with Gym environment interface again.", "_____no_output_____" ] ], [ [ "from crawler_env import CrawlingRobotEnv\n\nenv = CrawlingRobotEnv()\n\nprint(\"We can inspect the observation space and action space of this Gym Environment\")\nprint(\"-----------------------------------------------------------------------------\")\nprint(\"Action space:\", env.action_space)\nprint(\"It's a discrete space with %i actions to take\" % env.action_space.n)\nprint(\"Each action corresponds to increasing/decreasing the angle of one of the joints\")\nprint(\"We can also sample from this action space:\", env.action_space.sample())\nprint(\"Another action sample:\", env.action_space.sample())\nprint(\"Another action sample:\", env.action_space.sample())\nprint(\"Observation space:\", env.observation_space, \", which means it's a 9x13 grid.\")\nprint(\"It's the discretized version of the robot's two joint angles\")", "We can inspect the observation space and action space of this Gym Environment\n-----------------------------------------------------------------------------\nAction space: Discrete(4)\nIt's a discrete space with 4 actions to take\nEach action corresponds to increasing/decreasing the angle of one of the joints\nWe can also sample from this action space: 0\nAnother action sample: 3\nAnother action sample: 1\nObservation space: Tuple(Discrete(9), Discrete(13)) , which means it's a 9x13 grid.\nIt's the discretized version of the robot's two joint angles\n" ], [ "env = CrawlingRobotEnv(\n render=True, # turn render mode on to visualize random motion\n)\n\n# standard procedure for interfacing with a Gym environment\ncur_state = env.reset() # reset environment and get initial state\nret = 0.\ndone = False\ni = 0\nwhile not done:\n action = env.action_space.sample() # sample an action randomly\n next_state, reward, done, info = env.step(action)\n ret += reward\n cur_state = next_state\n i += 1\n if i == 1500:\n break # for the purpose of this visualization, let's only run for 1500 steps\n # also note the GUI won't close automatically", "_____no_output_____" ], [ "# you can close the visualization GUI with the following method \nenv.close_gui()", "_____no_output_____" ] ], [ [ "You will see the random controller can sometimes make progress but it won't get very far. Let's implement Tabular Q-Learning with $\\epsilon$-greedy exploration to find a better policy piece by piece.\n", "_____no_output_____" ] ], [ [ "from collections import defaultdict\nimport random\n\n# dictionary that maps from state, s, to a numpy array of Q values [Q(s, a_1), Q(s, a_2) ... Q(s, a_n)]\n# and everything is initialized to 0.\nq_vals = defaultdict(lambda: np.array([0. for _ in range(env.action_space.n)]))\n\nprint(\"Q-values for state (0, 0): %s\" % q_vals[(0, 0)], \"which is a list of Q values for each action\")\nprint(\"As such, the Q value of taking action 3 in state (1,2), i.e. Q((1,2), 3), can be accessed by q_vals[(1,2)][3]:\", q_vals[(1,2)][3])", "Q-values for state (0, 0): [ 0. 0. 0. 0.] which is a list of Q values for each action\nAs such, the Q value of taking action 3 in state (1,2), i.e. Q((1,2), 3), can be accessed by q_vals[(1,2)][3]: 0.0\n" ], [ "def eps_greedy(q_vals, eps, state):\n \"\"\"\n Inputs:\n q_vals: q value tables\n eps: epsilon\n state: current state\n Outputs:\n random action with probability of eps; argmax Q(s, .) with probability of (1-eps)\n \"\"\"\n # you might want to use random.random() to implement random exploration\n # number of actions can be read off from len(q_vals[state])\n import random\n # YOUR CODE HERE\n \n # MY CODE -------------------------------------------------------------------\n if random.random() <= eps:\n return np.random.randint(0, len(q_vals[state]))\n return np.argmax(q_vals[state])\n #----------------------------------------------------------------------------\n\n# test 1\ndummy_q = defaultdict(lambda: np.array([0. for _ in range(env.action_space.n)]))\ntest_state = (0, 0)\ndummy_q[test_state][0] = 10.\ntrials = 100000\nsampled_actions = [\n int(eps_greedy(dummy_q, 0.3, test_state))\n for _ in range(trials)\n]\nfreq = np.sum(np.array(sampled_actions) == 0) / trials\ntgt_freq = 0.3 / env.action_space.n + 0.7\nif np.isclose(freq, tgt_freq, atol=1e-2):\n print(\"Test1 passed\")\nelse:\n print(\"Test1: Expected to select 0 with frequency %.2f but got %.2f\" % (tgt_freq, freq))\n \n# test 2\ndummy_q = defaultdict(lambda: np.array([0. for _ in range(env.action_space.n)]))\ntest_state = (0, 0)\ndummy_q[test_state][2] = 10.\ntrials = 100000\nsampled_actions = [\n int(eps_greedy(dummy_q, 0.5, test_state))\n for _ in range(trials)\n]\nfreq = np.sum(np.array(sampled_actions) == 2) / trials\ntgt_freq = 0.5 / env.action_space.n + 0.5\nif np.isclose(freq, tgt_freq, atol=1e-2):\n print(\"Test2 passed\")\nelse:\n print(\"Test2: Expected to select 2 with frequency %.2f but got %.2f\" % (tgt_freq, freq))", "Test1 passed\nTest2 passed\n" ] ], [ [ "Next we will implement Q learning update. After we observe a transition $s, a, s', r$,\n\n$$\\textrm{target}(s') = R(s,a,s') + \\gamma \\max_{a'} Q_{\\theta_k}(s',a')$$\n\n\n$$Q_{k+1}(s,a) \\leftarrow (1-\\alpha) Q_k(s,a) + \\alpha \\left[ \\textrm{target}(s') \\right]$$", "_____no_output_____" ] ], [ [ "def q_learning_update(gamma, alpha, q_vals, cur_state, action, next_state, reward):\n \"\"\"\n Inputs:\n gamma: discount factor\n alpha: learning rate\n q_vals: q value table\n cur_state: current state\n action: action taken in current state\n next_state: next state results from taking `action` in `cur_state`\n reward: reward received from this transition\n \n Performs in-place update of q_vals table to implement one step of Q-learning\n \"\"\"\n # YOUR CODE HERE\n \n # MY CODE -------------------------------------------------------------------\n target = reward + gamma*np.max(q_vals[next_state])\n q_vals[cur_state][action] -= alpha*(q_vals[cur_state][action] - target)\n #----------------------------------------------------------------------------\n\n# testing your q_learning_update implementation\ndummy_q = q_vals.copy()\ntest_state = (0, 0)\ntest_next_state = (0, 1)\ndummy_q[test_state][0] = 10.\ndummy_q[test_next_state][1] = 10.\nq_learning_update(0.9, 0.1, dummy_q, test_state, 0, test_next_state, 1.1)\ntgt = 10.01\nif np.isclose(dummy_q[test_state][0], tgt,):\n print(\"Test passed\")\nelse:\n print(\"Q(test_state, 0) is expected to be %.2f but got %.2f\" % (tgt, dummy_q[test_state][0]))", "Test passed\n" ], [ "# now with the main components tested, we can put everything together to create a complete q learning agent\n\nenv = CrawlingRobotEnv() \nq_vals = defaultdict(lambda: np.array([0. for _ in range(env.action_space.n)]))\ngamma = 0.9\nalpha = 0.1\neps = 0.5\ncur_state = env.reset()\n\ndef greedy_eval():\n \"\"\"evaluate greedy policy w.r.t current q_vals\"\"\"\n test_env = CrawlingRobotEnv(horizon=np.inf)\n prev_state = test_env.reset()\n ret = 0.\n done = False\n H = 100\n for i in range(H):\n action = np.argmax(q_vals[prev_state])\n state, reward, done, info = test_env.step(action)\n ret += reward\n prev_state = state\n return ret / H\n\nfor itr in range(300000):\n # YOUR CODE HERE\n # Hint: use eps_greedy & q_learning_update\n \n # MY CODE --------------------------------------------------------------------\n action = eps_greedy(q_vals, eps, cur_state)\n next_state, reward, done, info = env.step(action)\n q_learning_update(gamma, alpha, q_vals, cur_state, action, next_state, reward)\n cur_state = next_state\n #-----------------------------------------------------------------------------\n \n if itr % 50000 == 0: # evaluation\n print(\"Itr %i # Average speed: %.2f\" % (itr, greedy_eval()))\n\n# at the end of learning your crawler should reach a speed of >= 3", "Itr 0 # Average speed: 0.05\nItr 50000 # Average speed: 2.03\nItr 100000 # Average speed: 3.37\nItr 150000 # Average speed: 3.37\nItr 200000 # Average speed: 3.37\nItr 250000 # Average speed: 3.37\n" ] ], [ [ "After the learning is successful, we can visualize the learned robot controller. Remember we learn this just from interacting with the environment instead of peeking into the dynamics model!", "_____no_output_____" ] ], [ [ "env = CrawlingRobotEnv(render=True, horizon=500)\nprev_state = env.reset()\nret = 0.\ndone = False\nwhile not done:\n action = np.argmax(q_vals[prev_state])\n state, reward, done, info = env.step(action)\n ret += reward\n prev_state = state", "_____no_output_____" ], [ "# you can close the visualization GUI with the following method \nenv.close_gui()", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Exercise 6-3\n\n## LSTM\n\nThe following two cells will create a LSTM cell with one neuron.\nWe scale the output of the LSTM linear and add a bias.\nThen the output will be wrapped by a sigmoid activation.\nThe goal is to predict a time series where every $n^{th}$ ($5^{th}$ in the current example) element is 1 and all others are 0.\n\na) Please read and understand the source code below.\n\nb) Consult the output of the predictions. What do you observe? How does the LSTM manage to predict the next element in the sequence?", "_____no_output_____" ] ], [ [ "import tensorflow as tf\nimport numpy as np\nfrom matplotlib import pyplot as plt", "_____no_output_____" ], [ "tf.reset_default_graph()\ntf.set_random_seed(12314)\n\nepochs=50\nzero_steps = 5\nlearning_rate = 0.01\nlstm_neurons = 1\nout_dim = 1\n\nnum_features = 1\nbatch_size = zero_steps\nwindow_size = zero_steps*2\ntime_steps = 5\n\nx = tf.placeholder(tf.float32, [None, window_size, num_features], 'x')\ny = tf.placeholder(tf.float32, [None, out_dim], 'y')\n\nlstm = tf.nn.rnn_cell.LSTMCell(lstm_neurons)\nstate = lstm.zero_state(batch_size, dtype=tf.float32)\n\nregression_w = tf.Variable(tf.random_normal([lstm_neurons]))\nregression_b = tf.Variable(tf.random_normal([out_dim]))\n\noutputs, state = tf.contrib.rnn.static_rnn(lstm, tf.unstack(x, window_size, 1), state)\noutput = outputs[-1]\npredicted = tf.nn.sigmoid(output * regression_w + regression_b)\ncost = tf.reduce_mean(tf.losses.mean_squared_error(y, predicted))\noptimizer = tf.train.RMSPropOptimizer(learning_rate=learning_rate).minimize(cost)\n\n\nforget_gate = output.op.inputs[1].op.inputs[0].op.inputs[0].op.inputs[0]\ninput_gate = output.op.inputs[1].op.inputs[0].op.inputs[1].op.inputs[0]\ncell_candidates = output.op.inputs[1].op.inputs[0].op.inputs[1].op.inputs[1]\noutput_gate_sig = output.op.inputs[0]\noutput_gate_tanh = output.op.inputs[1]\n\n\nX = [\n [[ (shift-n) % zero_steps == 0 ] for n in range(window_size)\n ] for shift in range(batch_size)\n]\nY = [[ shift % zero_steps == 0 ] for shift in range(batch_size) ]\n\n\nwith tf.Session() as sess:\n sess.run(tf.initializers.global_variables())\n\n loss = 1\n epoch = 0\n while loss >= 1e-5:\n epoch += 1\n _, loss = sess.run([optimizer, cost], {x:X, y:Y})\n\n if epoch % (epochs//10) == 0:\n print(\"loss %.5f\" % (loss), end='\\t\\t\\r')\n print()\n \n outs, stat, pred, fg, inpg, cell_cands, outg_sig, outg_tanh = sess.run([outputs, state, predicted, forget_gate, input_gate, cell_candidates, output_gate_sig, output_gate_tanh], {x:X, y:Y})\n outs = np.asarray(outs)\n for batch in reversed(range(batch_size)):\n print(\"input:\")\n print(np.asarray(X)[batch].astype(int).reshape(-1))\n print(\"forget\\t\\t%.4f\\ninput gate\\t%.4f\\ncell cands\\t%.4f\\nout gate sig\\t%.4f\\nout gate tanh\\t%.4f\\nhidden state\\t%.4f\\ncell state\\t%.4f\\npred\\t\\t%.4f\\n\\n\" % (\n fg[batch,0], \n inpg[batch,0], \n cell_cands[batch,0], \n outg_sig[batch,0], \n outg_tanh[batch,0], \n stat.h[batch,0], \n stat.c[batch,0], \n pred[batch,0]))", "loss 0.00001\t\t\ninput:\n[0 0 0 0 1 0 0 0 0 1]\nforget\t\t0.0135\ninput gate\t0.9997\ncell cands\t0.9994\nout gate sig\t1.0000\nout gate tanh\t0.7586\nhidden state\t0.7586\ncell state\t0.9928\npred\t\t0.0000\n\n\ninput:\n[0 0 0 1 0 0 0 0 1 0]\nforget\t\t0.8747\ninput gate\t0.9860\ncell cands\t-0.0476\nout gate sig\t1.0000\nout gate tanh\t0.6759\nhidden state\t0.6759\ncell state\t0.8215\npred\t\t0.0000\n\n\ninput:\n[0 0 1 0 0 0 0 1 0 0]\nforget\t\t0.8504\ninput gate\t0.9877\ncell cands\t-0.1311\nout gate sig\t0.9999\nout gate tanh\t0.5148\nhidden state\t0.5147\ncell state\t0.5692\npred\t\t0.0000\n\n\ninput:\n[0 1 0 0 0 0 1 0 0 0]\nforget\t\t0.7921\ninput gate\t0.9903\ncell cands\t-0.2875\nout gate sig\t0.9999\nout gate tanh\t0.1646\nhidden state\t0.1646\ncell state\t0.1661\npred\t\t0.0052\n\n\ninput:\n[1 0 0 0 0 1 0 0 0 0]\nforget\t\t0.6150\ninput gate\t0.9943\ncell cands\t-0.5732\nout gate sig\t0.9997\nout gate tanh\t-0.4363\nhidden state\t-0.4361\ncell state\t-0.4676\npred\t\t0.9953\n\n\n" ] ], [ [ "LSTM gates:\n\n\n(image source: https://www.stratio.com/wp-content/uploads/2017/10/6-1.jpg)", "_____no_output_____" ], [ "### Answers\n\n* When the current element is 1, then the forget-gate tells \"forget\" (value is close to 0) $\\Rightarrow$ Reset cell state\n* The cell state (long term memory) decreases until it reached some certain point. Then the hidden state is activated and thus the prediction is close to 1.\n* The sigomoid output cell ($o_t$) is always close to 1 $\\Rightarrow$ the hidden layer directly dependent on the cell state (no short term memory is used).\n* The input gate ($i_t$) is always close to 1 thus the cell candidates ($c_t$) will always be accepted\n* The cell candidates ($c_t$) are mainly dependent on $x_t$. It is close to 1 when $x_t$ is one (resetting the counter) and negative if $x_t$ is 0 (decreasing the counter).\n\nNote that with other initial values (different seed) it may result in a different local minimum (the counter could increase, $h_t$ could be negative and be scaled negative, ...)", "_____no_output_____" ] ] ]

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

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# 2020L-WUM Praca domowa 2", "_____no_output_____" ], [ "Kod: **Bartłomiej Eljasiak**", "_____no_output_____" ], [ "## Załadowanie bibliotek", "_____no_output_____" ], [ "Z tych bibliotek będziemy korzystać w wielu miejscach, jednak w niektórych fragmentach kodu znajdą się dodatkowe importowania, lecz w takich sytuacjach użytek załadowanej biblioteki jest ograniczony do 'chunku', w którym została załadowana.", "_____no_output_____" ] ], [ [ "import pandas as pd\nimport seaborn as sns\nimport numpy as np\nimport sklearn", "_____no_output_____" ] ], [ [ "## Wczytanie danych", "_____no_output_____" ] ], [ [ "# local version \n_data=pd.read_csv('allegro-api-transactions.csv')\n\n#online version\n#_data = pd.read_csv('https://www.dropbox.com/s/360xhh2d9lnaek3/allegro-api-transactions.csv?dl=1')\ncdata=_data.copy()", "_____no_output_____" ] ], [ [ "### Przyjrzenie się danym", "_____no_output_____" ] ], [ [ "cdata.head()", "_____no_output_____" ] ], [ [ "# Obróbka danych", "_____no_output_____" ] ], [ [ "len(cdata.it_location.unique())", "_____no_output_____" ], [ "cdata.info()", "<class 'pandas.core.frame.DataFrame'>\nRangeIndex: 420020 entries, 0 to 420019\nData columns (total 14 columns):\nlp 420020 non-null int64\ndate 420020 non-null object\nitem_id 420020 non-null int64\ncategories 420020 non-null object\npay_option_on_delivery 420020 non-null int64\npay_option_transfer 420020 non-null int64\nseller 420020 non-null object\nprice 420020 non-null float64\nit_is_allegro_standard 420020 non-null int64\nit_quantity 420020 non-null int64\nit_is_brand_zone 420020 non-null int64\nit_seller_rating 420020 non-null int64\nit_location 420020 non-null object\nmain_category 420020 non-null object\ndtypes: float64(1), int64(8), object(5)\nmemory usage: 36.9+ MB\n" ] ], [ [ "Na pierwszy rzut oka nie mamy żadnych braków w danych, co znacznie ułatwia nam prace. ", "_____no_output_____" ], [ "# Kodowanie zmiennych kategorycznych", "_____no_output_____" ], [ "Wiemy, że naszym targetem będzie `price`, chcemy więc w tym akapicie zamienić wszystkie zmienne kategoryczne, z których w dalszej części kodu będziemy korzystać, na zmienne liczbowe. Skuteczna zamiana przy użyciu odpowiednich metod takich jak **target encoding** oraz **on-hot encoding** pozwoli nam przekształcić obecne informacje, tak byśmy mogli je wykorzystać przy operacjach matematycznych. Pominiemy jednak w naszych przekształceniach kolumnę `categories`. ", "_____no_output_____" ], [ "## Target encoding dla `it_location`", "_____no_output_____" ] ], [ [ "import category_encoders\ny=cdata.price\nte = category_encoders.target_encoder.TargetEncoder(cdata.it_location, smoothing=100)\nencoded = te.fit_transform(cdata.it_location,y)\nencoded", "_____no_output_____" ] ], [ [ "## Różne rodzaje zakodowania kolumny `main_category`", "_____no_output_____" ], [ "## One-hot Coding", "_____no_output_____" ] ], [ [ "from sklearn.preprocessing import LabelEncoder\nfrom sklearn.preprocessing import OneHotEncoder\n\n# integer encode\nle = LabelEncoder()\ninteger_encoded = le.fit_transform(cdata.main_category)\nprint(integer_encoded)\n\n# binary encode\nonehot_encoder = OneHotEncoder(categories='auto',sparse=False)\ninteger_encoded = integer_encoded.reshape(len(integer_encoded), 1)\nonehot_encoded = onehot_encoder.fit_transform(integer_encoded)\nprint(onehot_encoded)", "[12 18 6 ... 18 5 15]\n[[0. 0. 0. ... 0. 0. 0.]\n [0. 0. 0. ... 0. 0. 0.]\n [0. 0. 0. ... 0. 0. 0.]\n ...\n [0. 0. 0. ... 0. 0. 0.]\n [0. 0. 0. ... 0. 0. 0.]\n [0. 0. 0. ... 0. 0. 0.]]\n" ] ], [ [ "Zamieniliśmy w ten sposób kolumnę kategoryczną na 26 kolumn o wartościach 0 lub 1. Nie jest to złe rozwiązanie, jednak stanowczo zwiększa rozmiar naszej ramki i wydłża czas uczenia się modelu. Prawdopodobnie może istnieć lepsze rozwiązanie.", "_____no_output_____" ], [ "## Helmert Coding\nDocumentation: [scikit-learn](http://contrib.scikit-learn.org/categorical-encoding/helmert.html)", "_____no_output_____" ] ], [ [ "# Pobieramy go z category_encoders\n\nhelmert_encoder = category_encoders.HelmertEncoder()\nhelmert_encoded = helmert_encoder.fit_transform(cdata.main_category)\n\n#showing only first 5 encoded rows\nprint(helmert_encoded.loc[1:5,:].transpose())", " 1 2 3 4 5\nintercept 1.0 1.0 1.0 1.0 1.0\nmain_category_0 1.0 0.0 0.0 1.0 1.0\nmain_category_1 -1.0 2.0 0.0 -1.0 -1.0\nmain_category_2 -1.0 -1.0 3.0 -1.0 -1.0\nmain_category_3 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_4 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_5 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_6 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_7 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_8 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_9 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_10 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_11 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_12 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_13 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_14 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_15 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_16 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_17 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_18 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_19 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_20 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_21 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_22 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_23 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_24 -1.0 -1.0 -1.0 -1.0 -1.0\nmain_category_25 -1.0 -1.0 -1.0 -1.0 -1.0\n" ] ], [ [ "## Backward Difference Coding\nDocumentation: [scikit-learn](http://contrib.scikit-learn.org/categorical-encoding/backward_difference.html#backward-difference-coding)", "_____no_output_____" ] ], [ [ "# Pobieramy go z category_encoders\n\nback_diff_encoder = category_encoders.BackwardDifferenceEncoder()\nback_diff_encoded = back_diff_encoder.fit_transform(cdata.main_category)\n\n#showing only first 5 encoded rows\nprint(back_diff_encoded.loc[1:5,:].transpose())", " 1 2 3 4 5\nintercept 1.000000 1.000000 1.000000 1.000000 1.000000\nmain_category_0 0.037037 0.037037 0.037037 0.037037 0.037037\nmain_category_1 -0.925926 0.074074 0.074074 -0.925926 -0.925926\nmain_category_2 -0.888889 -0.888889 0.111111 -0.888889 -0.888889\nmain_category_3 -0.851852 -0.851852 -0.851852 -0.851852 -0.851852\nmain_category_4 -0.814815 -0.814815 -0.814815 -0.814815 -0.814815\nmain_category_5 -0.777778 -0.777778 -0.777778 -0.777778 -0.777778\nmain_category_6 -0.740741 -0.740741 -0.740741 -0.740741 -0.740741\nmain_category_7 -0.703704 -0.703704 -0.703704 -0.703704 -0.703704\nmain_category_8 -0.666667 -0.666667 -0.666667 -0.666667 -0.666667\nmain_category_9 -0.629630 -0.629630 -0.629630 -0.629630 -0.629630\nmain_category_10 -0.592593 -0.592593 -0.592593 -0.592593 -0.592593\nmain_category_11 -0.555556 -0.555556 -0.555556 -0.555556 -0.555556\nmain_category_12 -0.518519 -0.518519 -0.518519 -0.518519 -0.518519\nmain_category_13 -0.481481 -0.481481 -0.481481 -0.481481 -0.481481\nmain_category_14 -0.444444 -0.444444 -0.444444 -0.444444 -0.444444\nmain_category_15 -0.407407 -0.407407 -0.407407 -0.407407 -0.407407\nmain_category_16 -0.370370 -0.370370 -0.370370 -0.370370 -0.370370\nmain_category_17 -0.333333 -0.333333 -0.333333 -0.333333 -0.333333\nmain_category_18 -0.296296 -0.296296 -0.296296 -0.296296 -0.296296\nmain_category_19 -0.259259 -0.259259 -0.259259 -0.259259 -0.259259\nmain_category_20 -0.222222 -0.222222 -0.222222 -0.222222 -0.222222\nmain_category_21 -0.185185 -0.185185 -0.185185 -0.185185 -0.185185\nmain_category_22 -0.148148 -0.148148 -0.148148 -0.148148 -0.148148\nmain_category_23 -0.111111 -0.111111 -0.111111 -0.111111 -0.111111\nmain_category_24 -0.074074 -0.074074 -0.074074 -0.074074 -0.074074\nmain_category_25 -0.037037 -0.037037 -0.037037 -0.037037 -0.037037\n" ] ], [ [ "# Uzupełnianie braków", "_____no_output_____" ], [ "### Wybranie danych z ramki\nDalej będziemy pracowac tylko na 3 kolumnach danych, ograniczę więc je dla przejżystości poczynań.", "_____no_output_____" ] ], [ [ "data_selected= cdata.loc[:,['price','it_seller_rating','it_quantity']]\ndata_selected.head()", "_____no_output_____" ] ], [ [ "### Usunięcie danych z kolumny", "_____no_output_____" ], [ "Dane z kolumny będziemy usuwać funkcją `df.column.sample(frac)` gdzie `frac` będzie oznaczać procent danych, które chcemy zatrzymać. Gwarantuje nam to w miarę losowe usunięcie danych, które powinno być wystarczające do dalszych działań.", "_____no_output_____" ] ], [ [ "cdata.price.sample(frac=0.9)\n", "_____no_output_____" ] ], [ [ "### Ocena skutecznosci imputacji\nDo oceny skuteczności podanych algorytmów imputacji danych musimy przyjąć jakis sposób liczenia ich. Zgodnie z sugestią prowadzącej skorszystam z [RMSE](https://en.wikipedia.org/wiki/Root_mean_square) czyli root mean square error. Nazwa powinna przybliżyć sposób, którm RMSE jest wyznaczane, jednak ciekawskich zachęcam do wejścia w link.", "_____no_output_____" ], [ "## Imputacja danych\nNapiszmy więc funkcje, która pozwoli nam stestować wybrany sposób imputacji.", "_____no_output_____" ] ], [ [ "from sklearn.experimental import enable_iterative_imputer\nfrom sklearn.impute import IterativeImputer\nfrom sklearn.metrics import mean_squared_error\n\ndef test_imputation(imputer,iterations=10):\n _resoults=[]\n # we always use the same data, so it's taken globally \n for i in range(iterations):\n test_data = data_selected.copy()\n test_data.it_seller_rating = test_data.it_seller_rating.sample(frac = 0.9)\n data_imputed = pd.DataFrame(imputer.fit_transform(test_data))\n _resoults.append(np.sqrt(mean_squared_error(data_selected,data_imputed)))\n return _resoults", "_____no_output_____" ] ], [ [ "I to niech będzie przykład działania takiej funkcji", "_____no_output_____" ] ], [ [ "imputer = IterativeImputer(max_iter=10,random_state=0)\nRMSE_list = test_imputation(imputer,20)\n\n\nprint(\"Średnie RMSE wynosi\", round(np.mean(RMSE_list)))\nprint('Odchylenie standardowe RMSE wynosi: ', round(np.std(RMSE_list))) \nRMSE_list", "Średnie RMSE wynosi 6659.0\nOdchylenie standardowe RMSE wynosi: 64.0\n" ] ], [ [ "Odchylenie standardowe jest dosyć małe, więc metodę imputacji uważam za skuteczną. Chętnym polecam przetestowanie tej funkcji dla innych typów imputacji oraz zmiennej liczbie iteracji. ", "_____no_output_____" ], [ "### Usuwanie danych z wielu kolumn", "_____no_output_____" ], [ "Powtórzmy uprzedni przykład dodając drobną modyfikacje. Tym razem będziemy usuwali dane zarówno z `it_seller_rating` jak i `it_quantity`. Napiszmy do tego odpowiednią funkcję i zobaczmy wyniki.", "_____no_output_____" ] ], [ [ "def test_imputation2(imputer,iterations=10):\n _resoults=[]\n # we always use the same data, so it's taken globally \n for i in range(iterations):\n test_data = data_selected.copy()\n test_data.it_seller_rating = test_data.it_seller_rating.sample(frac = 0.9)\n test_data.it_quantity = test_data.it_quantity.sample(frac = 0.9)\n data_imputed = pd.DataFrame(imputer.fit_transform(test_data))\n _resoults.append(np.sqrt(mean_squared_error(data_selected,data_imputed)))\n return _resoults ", "_____no_output_____" ], [ "imputer = IterativeImputer(max_iter=10,random_state=0)\nRMSE_list = test_imputation2(imputer,20)\n\n \nprint(\"Średnie RMSE wynosi\", round(np.mean(RMSE_list)))\nprint('Odchylenie standardowe RMSE wynosi: ', round(np.std(RMSE_list))) \nRMSE_list", "Średnie RMSE wynosi 7953.0\nOdchylenie standardowe RMSE wynosi: 60.0\n" ] ], [ [ "Tak jak moglibyśmy się spodziewać, średni błąd jest większy w przypadku gdy usuneliśmy więcej danych. Ponownie zachęcam do powtórzenia obliczeń i sprawdzenia wyników. ", "_____no_output_____" ], [ "### Spojrzenie na imputacje typu `IterativeImputer`", "_____no_output_____" ], [ "Wykorzystałem pewien szczególny sposób imputacji danych tzn `IterativeImputer`, o którym nie wspomniałem za dużo podczas korzystania z niego. Chciałbym jednka w tym miejscu go bardziej szczegółowo przedstawić oraz zobaczyć jak liczba iteracji wpływa na jakość imputacji.", "_____no_output_____" ], [ "Nasz imputer będę chiał przetestować dla całego spektrum wartości `max_iter` i dokładnie to zrobię w poniższej pętli.", "_____no_output_____" ], [ "**uwaga poniższy kod wykonuję się dosyć długo**", "_____no_output_____" ] ], [ [ "upper_iter_limit = 30\nlower_iter_limit = 5\nimputation_iterations = 10\n\nmean_RMSE ={\n \"single\": [],\n \"multi\": [],\n}\n\nfor imputer_iterations in range(lower_iter_limit,upper_iter_limit,2):\n _resoults_single = []\n _resoults_multi = []\n imputer = IterativeImputer(max_iter=imputer_iterations,random_state=0)\n\n print(\"max_iter: \", imputer_iterations, \"/\",upper_iter_limit)\n # Data missing from single columns\n _resoults_multi.append(test_imputation(imputer,imputation_iterations))\n\n # Data missing from multiple column\n _resoults_single.append(test_imputation2(imputer,imputation_iterations))\n \n mean_RMSE['single'].append(np.mean(_resoults_single))\n mean_RMSE['multi'].append(np.mean(_resoults_multi))\n ", "max_iter: 5 / 30\nmax_iter: 7 / 30\nmax_iter: 9 / 30\nmax_iter: 11 / 30\nmax_iter: 13 / 30\nmax_iter: 15 / 30\nmax_iter: 17 / 30\nmax_iter: 19 / 30\nmax_iter: 21 / 30\nmax_iter: 23 / 30\nmax_iter: 25 / 30\nmax_iter: 27 / 30\nmax_iter: 29 / 30\n" ] ], [ [ "Przyjrzyjmy się wynikom.", "_____no_output_____" ] ], [ [ "mean_RMSE\n", "_____no_output_____" ] ], [ [ "### Komentarz\nCo ciekawe nie widać dużej różnicy w błędzie imputacji dla różnych współczynników imputacji. Co wiecej nie ma, żadnego typu korelacji, a więc nie opłaca się brać dużego współczynnika iteracji, ponieważ wcale nie daje on lepszych wyników. Pragnę jednak ograniczyć moje wnioski do tego zbioru, ponieważ nie dysponuję w tym momencie wystarczającą liczbą informacji by twierdzić, że jest to zjawisko globalne. Niech ten przykład posłuży jako pretekst do dalszych dyskusji na ten temat.", "_____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" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import os\nimport numpy as np\nimport matplotlib.pyplot as plt\nimport pandas as pd\nfrom keras.applications.resnet50 import ResNet50,preprocess_input,decode_predictions\nfrom keras.layers import Dense,Conv2D,Input,Dropout,MaxPooling2D,Flatten,GlobalAveragePooling2D\nfrom keras.optimizers import Adam\nfrom keras.preprocessing import image\nfrom keras.models import Sequential,Model\nfrom keras.utils import np_utils\nimport random\nimport cv2", "Using TensorFlow backend.\n" ], [ "dirs = os.listdir(\"./men-women-classification/\")\nprint(dirs)", "['men', 'women']\n" ], [ "path = \"./men-women-classification/\"", "_____no_output_____" ], [ "images = []\nlabels = []\n\nlabels_dict = {'men':0,'women':1}\ndict_labels = { 0:\"men\",1:\"women\"}", "_____no_output_____" ], [ "for ix in dirs:\n class_path = path + ix + \"/\"\n img_names = os.listdir(class_path)\n for im in img_names:\n im = image.load_img(class_path + im, target_size=(224,224))\n im_array = image.img_to_array(im)\n images.append(im_array)\n labels.append(labels_dict[ix])\n \nprint (len(images), len(labels))", "1000 1000\n" ], [ "combined = list(zip(images,labels))\nrandom.shuffle(combined)\n\nimages[:],labels[:] = zip(*combined) ", "_____no_output_____" ], [ "X_train = np.array(images)\ny_train = np.array(labels)\ny_train = np_utils.to_categorical(y_train)\nprint (X_train.shape, y_train.shape)", "(1000, 224, 224, 3) (1000, 2)\n" ], [ "res_model = ResNet50(include_top = False , weights = 'imagenet' , input_shape = (224,224,3))", "WARNING: Logging before flag parsing goes to stderr.\nW0728 17:42:03.904074 140521947629376 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:74: The name tf.get_default_graph is deprecated. Please use tf.compat.v1.get_default_graph instead.\n\nW0728 17:42:04.210152 140521947629376 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:517: The name tf.placeholder is deprecated. Please use tf.compat.v1.placeholder instead.\n\nW0728 17:42:04.318412 140521947629376 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:4185: The name tf.truncated_normal is deprecated. Please use tf.random.truncated_normal instead.\n\nW0728 17:42:04.438638 140521947629376 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:174: The name tf.get_default_session is deprecated. Please use tf.compat.v1.get_default_session instead.\n\nW0728 17:42:04.439691 140521947629376 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:181: The name tf.ConfigProto is deprecated. Please use tf.compat.v1.ConfigProto instead.\n\nW0728 17:42:05.331743 140521947629376 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:1834: The name tf.nn.fused_batch_norm is deprecated. Please use tf.compat.v1.nn.fused_batch_norm instead.\n\nW0728 17:42:05.445978 140521947629376 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:3976: The name tf.nn.max_pool is deprecated. Please use tf.nn.max_pool2d instead.\n\n/usr/local/lib/python3.6/dist-packages/keras_applications/resnet50.py:265: UserWarning: The output shape of `ResNet50(include_top=False)` has been changed since Keras 2.2.0.\n warnings.warn('The output shape of `ResNet50(include_top=False)` '\n" ], [ "res_model.summary()", "__________________________________________________________________________________________________\nLayer (type) Output Shape Param # Connected to \n==================================================================================================\ninput_1 (InputLayer) (None, 224, 224, 3) 0 \n__________________________________________________________________________________________________\nconv1_pad (ZeroPadding2D) (None, 230, 230, 3) 0 input_1[0][0] \n__________________________________________________________________________________________________\nconv1 (Conv2D) (None, 112, 112, 64) 9472 conv1_pad[0][0] \n__________________________________________________________________________________________________\nbn_conv1 (BatchNormalization) (None, 112, 112, 64) 256 conv1[0][0] \n__________________________________________________________________________________________________\nactivation_1 (Activation) (None, 112, 112, 64) 0 bn_conv1[0][0] \n__________________________________________________________________________________________________\npool1_pad (ZeroPadding2D) (None, 114, 114, 64) 0 activation_1[0][0] \n__________________________________________________________________________________________________\nmax_pooling2d_1 (MaxPooling2D) (None, 56, 56, 64) 0 pool1_pad[0][0] \n__________________________________________________________________________________________________\nres2a_branch2a (Conv2D) (None, 56, 56, 64) 4160 max_pooling2d_1[0][0] \n__________________________________________________________________________________________________\nbn2a_branch2a (BatchNormalizati (None, 56, 56, 64) 256 res2a_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_2 (Activation) (None, 56, 56, 64) 0 bn2a_branch2a[0][0] \n__________________________________________________________________________________________________\nres2a_branch2b (Conv2D) (None, 56, 56, 64) 36928 activation_2[0][0] \n__________________________________________________________________________________________________\nbn2a_branch2b (BatchNormalizati (None, 56, 56, 64) 256 res2a_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_3 (Activation) (None, 56, 56, 64) 0 bn2a_branch2b[0][0] \n__________________________________________________________________________________________________\nres2a_branch2c (Conv2D) (None, 56, 56, 256) 16640 activation_3[0][0] \n__________________________________________________________________________________________________\nres2a_branch1 (Conv2D) (None, 56, 56, 256) 16640 max_pooling2d_1[0][0] \n__________________________________________________________________________________________________\nbn2a_branch2c (BatchNormalizati (None, 56, 56, 256) 1024 res2a_branch2c[0][0] \n__________________________________________________________________________________________________\nbn2a_branch1 (BatchNormalizatio (None, 56, 56, 256) 1024 res2a_branch1[0][0] \n__________________________________________________________________________________________________\nadd_1 (Add) (None, 56, 56, 256) 0 bn2a_branch2c[0][0] \n bn2a_branch1[0][0] \n__________________________________________________________________________________________________\nactivation_4 (Activation) (None, 56, 56, 256) 0 add_1[0][0] \n__________________________________________________________________________________________________\nres2b_branch2a (Conv2D) (None, 56, 56, 64) 16448 activation_4[0][0] \n__________________________________________________________________________________________________\nbn2b_branch2a (BatchNormalizati (None, 56, 56, 64) 256 res2b_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_5 (Activation) (None, 56, 56, 64) 0 bn2b_branch2a[0][0] \n__________________________________________________________________________________________________\nres2b_branch2b (Conv2D) (None, 56, 56, 64) 36928 activation_5[0][0] \n__________________________________________________________________________________________________\nbn2b_branch2b (BatchNormalizati (None, 56, 56, 64) 256 res2b_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_6 (Activation) (None, 56, 56, 64) 0 bn2b_branch2b[0][0] \n__________________________________________________________________________________________________\nres2b_branch2c (Conv2D) (None, 56, 56, 256) 16640 activation_6[0][0] \n__________________________________________________________________________________________________\nbn2b_branch2c (BatchNormalizati (None, 56, 56, 256) 1024 res2b_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_2 (Add) (None, 56, 56, 256) 0 bn2b_branch2c[0][0] \n activation_4[0][0] \n__________________________________________________________________________________________________\nactivation_7 (Activation) (None, 56, 56, 256) 0 add_2[0][0] \n__________________________________________________________________________________________________\nres2c_branch2a (Conv2D) (None, 56, 56, 64) 16448 activation_7[0][0] \n__________________________________________________________________________________________________\nbn2c_branch2a (BatchNormalizati (None, 56, 56, 64) 256 res2c_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_8 (Activation) (None, 56, 56, 64) 0 bn2c_branch2a[0][0] \n__________________________________________________________________________________________________\nres2c_branch2b (Conv2D) (None, 56, 56, 64) 36928 activation_8[0][0] \n__________________________________________________________________________________________________\nbn2c_branch2b (BatchNormalizati (None, 56, 56, 64) 256 res2c_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_9 (Activation) (None, 56, 56, 64) 0 bn2c_branch2b[0][0] \n__________________________________________________________________________________________________\nres2c_branch2c (Conv2D) (None, 56, 56, 256) 16640 activation_9[0][0] \n__________________________________________________________________________________________________\nbn2c_branch2c (BatchNormalizati (None, 56, 56, 256) 1024 res2c_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_3 (Add) (None, 56, 56, 256) 0 bn2c_branch2c[0][0] \n activation_7[0][0] \n__________________________________________________________________________________________________\nactivation_10 (Activation) (None, 56, 56, 256) 0 add_3[0][0] \n__________________________________________________________________________________________________\nres3a_branch2a (Conv2D) (None, 28, 28, 128) 32896 activation_10[0][0] \n__________________________________________________________________________________________________\nbn3a_branch2a (BatchNormalizati (None, 28, 28, 128) 512 res3a_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_11 (Activation) (None, 28, 28, 128) 0 bn3a_branch2a[0][0] \n__________________________________________________________________________________________________\nres3a_branch2b (Conv2D) (None, 28, 28, 128) 147584 activation_11[0][0] \n__________________________________________________________________________________________________\nbn3a_branch2b (BatchNormalizati (None, 28, 28, 128) 512 res3a_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_12 (Activation) (None, 28, 28, 128) 0 bn3a_branch2b[0][0] \n__________________________________________________________________________________________________\nres3a_branch2c (Conv2D) (None, 28, 28, 512) 66048 activation_12[0][0] \n__________________________________________________________________________________________________\nres3a_branch1 (Conv2D) (None, 28, 28, 512) 131584 activation_10[0][0] \n__________________________________________________________________________________________________\nbn3a_branch2c (BatchNormalizati (None, 28, 28, 512) 2048 res3a_branch2c[0][0] \n__________________________________________________________________________________________________\nbn3a_branch1 (BatchNormalizatio (None, 28, 28, 512) 2048 res3a_branch1[0][0] \n__________________________________________________________________________________________________\nadd_4 (Add) (None, 28, 28, 512) 0 bn3a_branch2c[0][0] \n bn3a_branch1[0][0] \n__________________________________________________________________________________________________\nactivation_13 (Activation) (None, 28, 28, 512) 0 add_4[0][0] \n__________________________________________________________________________________________________\nres3b_branch2a (Conv2D) (None, 28, 28, 128) 65664 activation_13[0][0] \n__________________________________________________________________________________________________\nbn3b_branch2a (BatchNormalizati (None, 28, 28, 128) 512 res3b_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_14 (Activation) (None, 28, 28, 128) 0 bn3b_branch2a[0][0] \n__________________________________________________________________________________________________\nres3b_branch2b (Conv2D) (None, 28, 28, 128) 147584 activation_14[0][0] \n__________________________________________________________________________________________________\nbn3b_branch2b (BatchNormalizati (None, 28, 28, 128) 512 res3b_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_15 (Activation) (None, 28, 28, 128) 0 bn3b_branch2b[0][0] \n__________________________________________________________________________________________________\nres3b_branch2c (Conv2D) (None, 28, 28, 512) 66048 activation_15[0][0] \n__________________________________________________________________________________________________\nbn3b_branch2c (BatchNormalizati (None, 28, 28, 512) 2048 res3b_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_5 (Add) (None, 28, 28, 512) 0 bn3b_branch2c[0][0] \n activation_13[0][0] \n__________________________________________________________________________________________________\nactivation_16 (Activation) (None, 28, 28, 512) 0 add_5[0][0] \n__________________________________________________________________________________________________\nres3c_branch2a (Conv2D) (None, 28, 28, 128) 65664 activation_16[0][0] \n__________________________________________________________________________________________________\nbn3c_branch2a (BatchNormalizati (None, 28, 28, 128) 512 res3c_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_17 (Activation) (None, 28, 28, 128) 0 bn3c_branch2a[0][0] \n__________________________________________________________________________________________________\nres3c_branch2b (Conv2D) (None, 28, 28, 128) 147584 activation_17[0][0] \n__________________________________________________________________________________________________\nbn3c_branch2b (BatchNormalizati (None, 28, 28, 128) 512 res3c_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_18 (Activation) (None, 28, 28, 128) 0 bn3c_branch2b[0][0] \n__________________________________________________________________________________________________\nres3c_branch2c (Conv2D) (None, 28, 28, 512) 66048 activation_18[0][0] \n__________________________________________________________________________________________________\nbn3c_branch2c (BatchNormalizati (None, 28, 28, 512) 2048 res3c_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_6 (Add) (None, 28, 28, 512) 0 bn3c_branch2c[0][0] \n activation_16[0][0] \n__________________________________________________________________________________________________\nactivation_19 (Activation) (None, 28, 28, 512) 0 add_6[0][0] \n__________________________________________________________________________________________________\nres3d_branch2a (Conv2D) (None, 28, 28, 128) 65664 activation_19[0][0] \n__________________________________________________________________________________________________\nbn3d_branch2a (BatchNormalizati (None, 28, 28, 128) 512 res3d_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_20 (Activation) (None, 28, 28, 128) 0 bn3d_branch2a[0][0] \n__________________________________________________________________________________________________\nres3d_branch2b (Conv2D) (None, 28, 28, 128) 147584 activation_20[0][0] \n__________________________________________________________________________________________________\nbn3d_branch2b (BatchNormalizati (None, 28, 28, 128) 512 res3d_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_21 (Activation) (None, 28, 28, 128) 0 bn3d_branch2b[0][0] \n__________________________________________________________________________________________________\nres3d_branch2c (Conv2D) (None, 28, 28, 512) 66048 activation_21[0][0] \n__________________________________________________________________________________________________\nbn3d_branch2c (BatchNormalizati (None, 28, 28, 512) 2048 res3d_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_7 (Add) (None, 28, 28, 512) 0 bn3d_branch2c[0][0] \n activation_19[0][0] \n__________________________________________________________________________________________________\nactivation_22 (Activation) (None, 28, 28, 512) 0 add_7[0][0] \n__________________________________________________________________________________________________\nres4a_branch2a (Conv2D) (None, 14, 14, 256) 131328 activation_22[0][0] \n__________________________________________________________________________________________________\nbn4a_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4a_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_23 (Activation) (None, 14, 14, 256) 0 bn4a_branch2a[0][0] \n__________________________________________________________________________________________________\nres4a_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_23[0][0] \n__________________________________________________________________________________________________\nbn4a_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4a_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_24 (Activation) (None, 14, 14, 256) 0 bn4a_branch2b[0][0] \n__________________________________________________________________________________________________\nres4a_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_24[0][0] \n__________________________________________________________________________________________________\nres4a_branch1 (Conv2D) (None, 14, 14, 1024) 525312 activation_22[0][0] \n__________________________________________________________________________________________________\nbn4a_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4a_branch2c[0][0] \n__________________________________________________________________________________________________\nbn4a_branch1 (BatchNormalizatio (None, 14, 14, 1024) 4096 res4a_branch1[0][0] \n__________________________________________________________________________________________________\nadd_8 (Add) (None, 14, 14, 1024) 0 bn4a_branch2c[0][0] \n bn4a_branch1[0][0] \n__________________________________________________________________________________________________\nactivation_25 (Activation) (None, 14, 14, 1024) 0 add_8[0][0] \n__________________________________________________________________________________________________\nres4b_branch2a (Conv2D) (None, 14, 14, 256) 262400 activation_25[0][0] \n__________________________________________________________________________________________________\nbn4b_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4b_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_26 (Activation) (None, 14, 14, 256) 0 bn4b_branch2a[0][0] \n__________________________________________________________________________________________________\nres4b_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_26[0][0] \n__________________________________________________________________________________________________\nbn4b_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4b_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_27 (Activation) (None, 14, 14, 256) 0 bn4b_branch2b[0][0] \n__________________________________________________________________________________________________\nres4b_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_27[0][0] \n__________________________________________________________________________________________________\nbn4b_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4b_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_9 (Add) (None, 14, 14, 1024) 0 bn4b_branch2c[0][0] \n activation_25[0][0] \n__________________________________________________________________________________________________\nactivation_28 (Activation) (None, 14, 14, 1024) 0 add_9[0][0] \n__________________________________________________________________________________________________\nres4c_branch2a (Conv2D) (None, 14, 14, 256) 262400 activation_28[0][0] \n__________________________________________________________________________________________________\nbn4c_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4c_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_29 (Activation) (None, 14, 14, 256) 0 bn4c_branch2a[0][0] \n__________________________________________________________________________________________________\nres4c_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_29[0][0] \n__________________________________________________________________________________________________\nbn4c_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4c_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_30 (Activation) (None, 14, 14, 256) 0 bn4c_branch2b[0][0] \n__________________________________________________________________________________________________\nres4c_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_30[0][0] \n__________________________________________________________________________________________________\nbn4c_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4c_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_10 (Add) (None, 14, 14, 1024) 0 bn4c_branch2c[0][0] \n activation_28[0][0] \n__________________________________________________________________________________________________\nactivation_31 (Activation) (None, 14, 14, 1024) 0 add_10[0][0] \n__________________________________________________________________________________________________\nres4d_branch2a (Conv2D) (None, 14, 14, 256) 262400 activation_31[0][0] \n__________________________________________________________________________________________________\nbn4d_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4d_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_32 (Activation) (None, 14, 14, 256) 0 bn4d_branch2a[0][0] \n__________________________________________________________________________________________________\nres4d_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_32[0][0] \n__________________________________________________________________________________________________\nbn4d_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4d_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_33 (Activation) (None, 14, 14, 256) 0 bn4d_branch2b[0][0] \n__________________________________________________________________________________________________\nres4d_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_33[0][0] \n__________________________________________________________________________________________________\nbn4d_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4d_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_11 (Add) (None, 14, 14, 1024) 0 bn4d_branch2c[0][0] \n activation_31[0][0] \n__________________________________________________________________________________________________\nactivation_34 (Activation) (None, 14, 14, 1024) 0 add_11[0][0] \n__________________________________________________________________________________________________\nres4e_branch2a (Conv2D) (None, 14, 14, 256) 262400 activation_34[0][0] \n__________________________________________________________________________________________________\nbn4e_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4e_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_35 (Activation) (None, 14, 14, 256) 0 bn4e_branch2a[0][0] \n__________________________________________________________________________________________________\nres4e_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_35[0][0] \n__________________________________________________________________________________________________\nbn4e_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4e_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_36 (Activation) (None, 14, 14, 256) 0 bn4e_branch2b[0][0] \n__________________________________________________________________________________________________\nres4e_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_36[0][0] \n__________________________________________________________________________________________________\nbn4e_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4e_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_12 (Add) (None, 14, 14, 1024) 0 bn4e_branch2c[0][0] \n activation_34[0][0] \n__________________________________________________________________________________________________\nactivation_37 (Activation) (None, 14, 14, 1024) 0 add_12[0][0] \n__________________________________________________________________________________________________\nres4f_branch2a (Conv2D) (None, 14, 14, 256) 262400 activation_37[0][0] \n__________________________________________________________________________________________________\nbn4f_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4f_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_38 (Activation) (None, 14, 14, 256) 0 bn4f_branch2a[0][0] \n__________________________________________________________________________________________________\nres4f_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_38[0][0] \n__________________________________________________________________________________________________\nbn4f_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4f_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_39 (Activation) (None, 14, 14, 256) 0 bn4f_branch2b[0][0] \n__________________________________________________________________________________________________\nres4f_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_39[0][0] \n__________________________________________________________________________________________________\nbn4f_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4f_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_13 (Add) (None, 14, 14, 1024) 0 bn4f_branch2c[0][0] \n activation_37[0][0] \n__________________________________________________________________________________________________\nactivation_40 (Activation) (None, 14, 14, 1024) 0 add_13[0][0] \n__________________________________________________________________________________________________\nres5a_branch2a (Conv2D) (None, 7, 7, 512) 524800 activation_40[0][0] \n__________________________________________________________________________________________________\nbn5a_branch2a (BatchNormalizati (None, 7, 7, 512) 2048 res5a_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_41 (Activation) (None, 7, 7, 512) 0 bn5a_branch2a[0][0] \n__________________________________________________________________________________________________\nres5a_branch2b (Conv2D) (None, 7, 7, 512) 2359808 activation_41[0][0] \n__________________________________________________________________________________________________\nbn5a_branch2b (BatchNormalizati (None, 7, 7, 512) 2048 res5a_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_42 (Activation) (None, 7, 7, 512) 0 bn5a_branch2b[0][0] \n__________________________________________________________________________________________________\nres5a_branch2c (Conv2D) (None, 7, 7, 2048) 1050624 activation_42[0][0] \n__________________________________________________________________________________________________\nres5a_branch1 (Conv2D) (None, 7, 7, 2048) 2099200 activation_40[0][0] \n__________________________________________________________________________________________________\nbn5a_branch2c (BatchNormalizati (None, 7, 7, 2048) 8192 res5a_branch2c[0][0] \n__________________________________________________________________________________________________\nbn5a_branch1 (BatchNormalizatio (None, 7, 7, 2048) 8192 res5a_branch1[0][0] \n__________________________________________________________________________________________________\nadd_14 (Add) (None, 7, 7, 2048) 0 bn5a_branch2c[0][0] \n bn5a_branch1[0][0] \n__________________________________________________________________________________________________\nactivation_43 (Activation) (None, 7, 7, 2048) 0 add_14[0][0] \n__________________________________________________________________________________________________\nres5b_branch2a (Conv2D) (None, 7, 7, 512) 1049088 activation_43[0][0] \n__________________________________________________________________________________________________\nbn5b_branch2a (BatchNormalizati (None, 7, 7, 512) 2048 res5b_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_44 (Activation) (None, 7, 7, 512) 0 bn5b_branch2a[0][0] \n__________________________________________________________________________________________________\nres5b_branch2b (Conv2D) (None, 7, 7, 512) 2359808 activation_44[0][0] \n__________________________________________________________________________________________________\nbn5b_branch2b (BatchNormalizati (None, 7, 7, 512) 2048 res5b_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_45 (Activation) (None, 7, 7, 512) 0 bn5b_branch2b[0][0] \n__________________________________________________________________________________________________\nres5b_branch2c (Conv2D) (None, 7, 7, 2048) 1050624 activation_45[0][0] \n__________________________________________________________________________________________________\nbn5b_branch2c (BatchNormalizati (None, 7, 7, 2048) 8192 res5b_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_15 (Add) (None, 7, 7, 2048) 0 bn5b_branch2c[0][0] \n activation_43[0][0] \n__________________________________________________________________________________________________\nactivation_46 (Activation) (None, 7, 7, 2048) 0 add_15[0][0] \n__________________________________________________________________________________________________\nres5c_branch2a (Conv2D) (None, 7, 7, 512) 1049088 activation_46[0][0] \n__________________________________________________________________________________________________\nbn5c_branch2a (BatchNormalizati (None, 7, 7, 512) 2048 res5c_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_47 (Activation) (None, 7, 7, 512) 0 bn5c_branch2a[0][0] \n__________________________________________________________________________________________________\nres5c_branch2b (Conv2D) (None, 7, 7, 512) 2359808 activation_47[0][0] \n__________________________________________________________________________________________________\nbn5c_branch2b (BatchNormalizati (None, 7, 7, 512) 2048 res5c_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_48 (Activation) (None, 7, 7, 512) 0 bn5c_branch2b[0][0] \n__________________________________________________________________________________________________\nres5c_branch2c (Conv2D) (None, 7, 7, 2048) 1050624 activation_48[0][0] \n__________________________________________________________________________________________________\nbn5c_branch2c (BatchNormalizati (None, 7, 7, 2048) 8192 res5c_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_16 (Add) (None, 7, 7, 2048) 0 bn5c_branch2c[0][0] \n activation_46[0][0] \n__________________________________________________________________________________________________\nactivation_49 (Activation) (None, 7, 7, 2048) 0 add_16[0][0] \n==================================================================================================\nTotal params: 23,587,712\nTrainable params: 23,534,592\nNon-trainable params: 53,120\n__________________________________________________________________________________________________\n" ], [ "avg = GlobalAveragePooling2D()(res_model.output)\nfc1 = Dense(256, activation='relu')(avg)\nfc2 = Dense(2, activation='softmax')(fc1)\n\nmodel = Model(inputs=res_model.inputs, outputs=fc2)\nmodel.summary()\n", "__________________________________________________________________________________________________\nLayer (type) Output Shape Param # Connected to \n==================================================================================================\ninput_1 (InputLayer) (None, 224, 224, 3) 0 \n__________________________________________________________________________________________________\nconv1_pad (ZeroPadding2D) (None, 230, 230, 3) 0 input_1[0][0] \n__________________________________________________________________________________________________\nconv1 (Conv2D) (None, 112, 112, 64) 9472 conv1_pad[0][0] \n__________________________________________________________________________________________________\nbn_conv1 (BatchNormalization) (None, 112, 112, 64) 256 conv1[0][0] \n__________________________________________________________________________________________________\nactivation_1 (Activation) (None, 112, 112, 64) 0 bn_conv1[0][0] \n__________________________________________________________________________________________________\npool1_pad (ZeroPadding2D) (None, 114, 114, 64) 0 activation_1[0][0] \n__________________________________________________________________________________________________\nmax_pooling2d_1 (MaxPooling2D) (None, 56, 56, 64) 0 pool1_pad[0][0] \n__________________________________________________________________________________________________\nres2a_branch2a (Conv2D) (None, 56, 56, 64) 4160 max_pooling2d_1[0][0] \n__________________________________________________________________________________________________\nbn2a_branch2a (BatchNormalizati (None, 56, 56, 64) 256 res2a_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_2 (Activation) (None, 56, 56, 64) 0 bn2a_branch2a[0][0] \n__________________________________________________________________________________________________\nres2a_branch2b (Conv2D) (None, 56, 56, 64) 36928 activation_2[0][0] \n__________________________________________________________________________________________________\nbn2a_branch2b (BatchNormalizati (None, 56, 56, 64) 256 res2a_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_3 (Activation) (None, 56, 56, 64) 0 bn2a_branch2b[0][0] \n__________________________________________________________________________________________________\nres2a_branch2c (Conv2D) (None, 56, 56, 256) 16640 activation_3[0][0] \n__________________________________________________________________________________________________\nres2a_branch1 (Conv2D) (None, 56, 56, 256) 16640 max_pooling2d_1[0][0] \n__________________________________________________________________________________________________\nbn2a_branch2c (BatchNormalizati (None, 56, 56, 256) 1024 res2a_branch2c[0][0] \n__________________________________________________________________________________________________\nbn2a_branch1 (BatchNormalizatio (None, 56, 56, 256) 1024 res2a_branch1[0][0] \n__________________________________________________________________________________________________\nadd_1 (Add) (None, 56, 56, 256) 0 bn2a_branch2c[0][0] \n bn2a_branch1[0][0] \n__________________________________________________________________________________________________\nactivation_4 (Activation) (None, 56, 56, 256) 0 add_1[0][0] \n__________________________________________________________________________________________________\nres2b_branch2a (Conv2D) (None, 56, 56, 64) 16448 activation_4[0][0] \n__________________________________________________________________________________________________\nbn2b_branch2a (BatchNormalizati (None, 56, 56, 64) 256 res2b_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_5 (Activation) (None, 56, 56, 64) 0 bn2b_branch2a[0][0] \n__________________________________________________________________________________________________\nres2b_branch2b (Conv2D) (None, 56, 56, 64) 36928 activation_5[0][0] \n__________________________________________________________________________________________________\nbn2b_branch2b (BatchNormalizati (None, 56, 56, 64) 256 res2b_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_6 (Activation) (None, 56, 56, 64) 0 bn2b_branch2b[0][0] \n__________________________________________________________________________________________________\nres2b_branch2c (Conv2D) (None, 56, 56, 256) 16640 activation_6[0][0] \n__________________________________________________________________________________________________\nbn2b_branch2c (BatchNormalizati (None, 56, 56, 256) 1024 res2b_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_2 (Add) (None, 56, 56, 256) 0 bn2b_branch2c[0][0] \n activation_4[0][0] \n__________________________________________________________________________________________________\nactivation_7 (Activation) (None, 56, 56, 256) 0 add_2[0][0] \n__________________________________________________________________________________________________\nres2c_branch2a (Conv2D) (None, 56, 56, 64) 16448 activation_7[0][0] \n__________________________________________________________________________________________________\nbn2c_branch2a (BatchNormalizati (None, 56, 56, 64) 256 res2c_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_8 (Activation) (None, 56, 56, 64) 0 bn2c_branch2a[0][0] \n__________________________________________________________________________________________________\nres2c_branch2b (Conv2D) (None, 56, 56, 64) 36928 activation_8[0][0] \n__________________________________________________________________________________________________\nbn2c_branch2b (BatchNormalizati (None, 56, 56, 64) 256 res2c_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_9 (Activation) (None, 56, 56, 64) 0 bn2c_branch2b[0][0] \n__________________________________________________________________________________________________\nres2c_branch2c (Conv2D) (None, 56, 56, 256) 16640 activation_9[0][0] \n__________________________________________________________________________________________________\nbn2c_branch2c (BatchNormalizati (None, 56, 56, 256) 1024 res2c_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_3 (Add) (None, 56, 56, 256) 0 bn2c_branch2c[0][0] \n activation_7[0][0] \n__________________________________________________________________________________________________\nactivation_10 (Activation) (None, 56, 56, 256) 0 add_3[0][0] \n__________________________________________________________________________________________________\nres3a_branch2a (Conv2D) (None, 28, 28, 128) 32896 activation_10[0][0] \n__________________________________________________________________________________________________\nbn3a_branch2a (BatchNormalizati (None, 28, 28, 128) 512 res3a_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_11 (Activation) (None, 28, 28, 128) 0 bn3a_branch2a[0][0] \n__________________________________________________________________________________________________\nres3a_branch2b (Conv2D) (None, 28, 28, 128) 147584 activation_11[0][0] \n__________________________________________________________________________________________________\nbn3a_branch2b (BatchNormalizati (None, 28, 28, 128) 512 res3a_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_12 (Activation) (None, 28, 28, 128) 0 bn3a_branch2b[0][0] \n__________________________________________________________________________________________________\nres3a_branch2c (Conv2D) (None, 28, 28, 512) 66048 activation_12[0][0] \n__________________________________________________________________________________________________\nres3a_branch1 (Conv2D) (None, 28, 28, 512) 131584 activation_10[0][0] \n__________________________________________________________________________________________________\nbn3a_branch2c (BatchNormalizati (None, 28, 28, 512) 2048 res3a_branch2c[0][0] \n__________________________________________________________________________________________________\nbn3a_branch1 (BatchNormalizatio (None, 28, 28, 512) 2048 res3a_branch1[0][0] \n__________________________________________________________________________________________________\nadd_4 (Add) (None, 28, 28, 512) 0 bn3a_branch2c[0][0] \n bn3a_branch1[0][0] \n__________________________________________________________________________________________________\nactivation_13 (Activation) (None, 28, 28, 512) 0 add_4[0][0] \n__________________________________________________________________________________________________\nres3b_branch2a (Conv2D) (None, 28, 28, 128) 65664 activation_13[0][0] \n__________________________________________________________________________________________________\nbn3b_branch2a (BatchNormalizati (None, 28, 28, 128) 512 res3b_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_14 (Activation) (None, 28, 28, 128) 0 bn3b_branch2a[0][0] \n__________________________________________________________________________________________________\nres3b_branch2b (Conv2D) (None, 28, 28, 128) 147584 activation_14[0][0] \n__________________________________________________________________________________________________\nbn3b_branch2b (BatchNormalizati (None, 28, 28, 128) 512 res3b_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_15 (Activation) (None, 28, 28, 128) 0 bn3b_branch2b[0][0] \n__________________________________________________________________________________________________\nres3b_branch2c (Conv2D) (None, 28, 28, 512) 66048 activation_15[0][0] \n__________________________________________________________________________________________________\nbn3b_branch2c (BatchNormalizati (None, 28, 28, 512) 2048 res3b_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_5 (Add) (None, 28, 28, 512) 0 bn3b_branch2c[0][0] \n activation_13[0][0] \n__________________________________________________________________________________________________\nactivation_16 (Activation) (None, 28, 28, 512) 0 add_5[0][0] \n__________________________________________________________________________________________________\nres3c_branch2a (Conv2D) (None, 28, 28, 128) 65664 activation_16[0][0] \n__________________________________________________________________________________________________\nbn3c_branch2a (BatchNormalizati (None, 28, 28, 128) 512 res3c_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_17 (Activation) (None, 28, 28, 128) 0 bn3c_branch2a[0][0] \n__________________________________________________________________________________________________\nres3c_branch2b (Conv2D) (None, 28, 28, 128) 147584 activation_17[0][0] \n__________________________________________________________________________________________________\nbn3c_branch2b (BatchNormalizati (None, 28, 28, 128) 512 res3c_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_18 (Activation) (None, 28, 28, 128) 0 bn3c_branch2b[0][0] \n__________________________________________________________________________________________________\nres3c_branch2c (Conv2D) (None, 28, 28, 512) 66048 activation_18[0][0] \n__________________________________________________________________________________________________\nbn3c_branch2c (BatchNormalizati (None, 28, 28, 512) 2048 res3c_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_6 (Add) (None, 28, 28, 512) 0 bn3c_branch2c[0][0] \n activation_16[0][0] \n__________________________________________________________________________________________________\nactivation_19 (Activation) (None, 28, 28, 512) 0 add_6[0][0] \n__________________________________________________________________________________________________\nres3d_branch2a (Conv2D) (None, 28, 28, 128) 65664 activation_19[0][0] \n__________________________________________________________________________________________________\nbn3d_branch2a (BatchNormalizati (None, 28, 28, 128) 512 res3d_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_20 (Activation) (None, 28, 28, 128) 0 bn3d_branch2a[0][0] \n__________________________________________________________________________________________________\nres3d_branch2b (Conv2D) (None, 28, 28, 128) 147584 activation_20[0][0] \n__________________________________________________________________________________________________\nbn3d_branch2b (BatchNormalizati (None, 28, 28, 128) 512 res3d_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_21 (Activation) (None, 28, 28, 128) 0 bn3d_branch2b[0][0] \n__________________________________________________________________________________________________\nres3d_branch2c (Conv2D) (None, 28, 28, 512) 66048 activation_21[0][0] \n__________________________________________________________________________________________________\nbn3d_branch2c (BatchNormalizati (None, 28, 28, 512) 2048 res3d_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_7 (Add) (None, 28, 28, 512) 0 bn3d_branch2c[0][0] \n activation_19[0][0] \n__________________________________________________________________________________________________\nactivation_22 (Activation) (None, 28, 28, 512) 0 add_7[0][0] \n__________________________________________________________________________________________________\nres4a_branch2a (Conv2D) (None, 14, 14, 256) 131328 activation_22[0][0] \n__________________________________________________________________________________________________\nbn4a_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4a_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_23 (Activation) (None, 14, 14, 256) 0 bn4a_branch2a[0][0] \n__________________________________________________________________________________________________\nres4a_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_23[0][0] \n__________________________________________________________________________________________________\nbn4a_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4a_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_24 (Activation) (None, 14, 14, 256) 0 bn4a_branch2b[0][0] \n__________________________________________________________________________________________________\nres4a_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_24[0][0] \n__________________________________________________________________________________________________\nres4a_branch1 (Conv2D) (None, 14, 14, 1024) 525312 activation_22[0][0] \n__________________________________________________________________________________________________\nbn4a_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4a_branch2c[0][0] \n__________________________________________________________________________________________________\nbn4a_branch1 (BatchNormalizatio (None, 14, 14, 1024) 4096 res4a_branch1[0][0] \n__________________________________________________________________________________________________\nadd_8 (Add) (None, 14, 14, 1024) 0 bn4a_branch2c[0][0] \n bn4a_branch1[0][0] \n__________________________________________________________________________________________________\nactivation_25 (Activation) (None, 14, 14, 1024) 0 add_8[0][0] \n__________________________________________________________________________________________________\nres4b_branch2a (Conv2D) (None, 14, 14, 256) 262400 activation_25[0][0] \n__________________________________________________________________________________________________\nbn4b_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4b_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_26 (Activation) (None, 14, 14, 256) 0 bn4b_branch2a[0][0] \n__________________________________________________________________________________________________\nres4b_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_26[0][0] \n__________________________________________________________________________________________________\nbn4b_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4b_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_27 (Activation) (None, 14, 14, 256) 0 bn4b_branch2b[0][0] \n__________________________________________________________________________________________________\nres4b_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_27[0][0] \n__________________________________________________________________________________________________\nbn4b_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4b_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_9 (Add) (None, 14, 14, 1024) 0 bn4b_branch2c[0][0] \n activation_25[0][0] \n__________________________________________________________________________________________________\nactivation_28 (Activation) (None, 14, 14, 1024) 0 add_9[0][0] \n__________________________________________________________________________________________________\nres4c_branch2a (Conv2D) (None, 14, 14, 256) 262400 activation_28[0][0] \n__________________________________________________________________________________________________\nbn4c_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4c_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_29 (Activation) (None, 14, 14, 256) 0 bn4c_branch2a[0][0] \n__________________________________________________________________________________________________\nres4c_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_29[0][0] \n__________________________________________________________________________________________________\nbn4c_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4c_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_30 (Activation) (None, 14, 14, 256) 0 bn4c_branch2b[0][0] \n__________________________________________________________________________________________________\nres4c_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_30[0][0] \n__________________________________________________________________________________________________\nbn4c_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4c_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_10 (Add) (None, 14, 14, 1024) 0 bn4c_branch2c[0][0] \n activation_28[0][0] \n__________________________________________________________________________________________________\nactivation_31 (Activation) (None, 14, 14, 1024) 0 add_10[0][0] \n__________________________________________________________________________________________________\nres4d_branch2a (Conv2D) (None, 14, 14, 256) 262400 activation_31[0][0] \n__________________________________________________________________________________________________\nbn4d_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4d_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_32 (Activation) (None, 14, 14, 256) 0 bn4d_branch2a[0][0] \n__________________________________________________________________________________________________\nres4d_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_32[0][0] \n__________________________________________________________________________________________________\nbn4d_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4d_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_33 (Activation) (None, 14, 14, 256) 0 bn4d_branch2b[0][0] \n__________________________________________________________________________________________________\nres4d_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_33[0][0] \n__________________________________________________________________________________________________\nbn4d_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4d_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_11 (Add) (None, 14, 14, 1024) 0 bn4d_branch2c[0][0] \n activation_31[0][0] \n__________________________________________________________________________________________________\nactivation_34 (Activation) (None, 14, 14, 1024) 0 add_11[0][0] \n__________________________________________________________________________________________________\nres4e_branch2a (Conv2D) (None, 14, 14, 256) 262400 activation_34[0][0] \n__________________________________________________________________________________________________\nbn4e_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4e_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_35 (Activation) (None, 14, 14, 256) 0 bn4e_branch2a[0][0] \n__________________________________________________________________________________________________\nres4e_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_35[0][0] \n__________________________________________________________________________________________________\nbn4e_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4e_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_36 (Activation) (None, 14, 14, 256) 0 bn4e_branch2b[0][0] \n__________________________________________________________________________________________________\nres4e_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_36[0][0] \n__________________________________________________________________________________________________\nbn4e_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4e_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_12 (Add) (None, 14, 14, 1024) 0 bn4e_branch2c[0][0] \n activation_34[0][0] \n__________________________________________________________________________________________________\nactivation_37 (Activation) (None, 14, 14, 1024) 0 add_12[0][0] \n__________________________________________________________________________________________________\nres4f_branch2a (Conv2D) (None, 14, 14, 256) 262400 activation_37[0][0] \n__________________________________________________________________________________________________\nbn4f_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4f_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_38 (Activation) (None, 14, 14, 256) 0 bn4f_branch2a[0][0] \n__________________________________________________________________________________________________\nres4f_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_38[0][0] \n__________________________________________________________________________________________________\nbn4f_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4f_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_39 (Activation) (None, 14, 14, 256) 0 bn4f_branch2b[0][0] \n__________________________________________________________________________________________________\nres4f_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_39[0][0] \n__________________________________________________________________________________________________\nbn4f_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4f_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_13 (Add) (None, 14, 14, 1024) 0 bn4f_branch2c[0][0] \n activation_37[0][0] \n__________________________________________________________________________________________________\nactivation_40 (Activation) (None, 14, 14, 1024) 0 add_13[0][0] \n__________________________________________________________________________________________________\nres5a_branch2a (Conv2D) (None, 7, 7, 512) 524800 activation_40[0][0] \n__________________________________________________________________________________________________\nbn5a_branch2a (BatchNormalizati (None, 7, 7, 512) 2048 res5a_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_41 (Activation) (None, 7, 7, 512) 0 bn5a_branch2a[0][0] \n__________________________________________________________________________________________________\nres5a_branch2b (Conv2D) (None, 7, 7, 512) 2359808 activation_41[0][0] \n__________________________________________________________________________________________________\nbn5a_branch2b (BatchNormalizati (None, 7, 7, 512) 2048 res5a_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_42 (Activation) (None, 7, 7, 512) 0 bn5a_branch2b[0][0] \n__________________________________________________________________________________________________\nres5a_branch2c (Conv2D) (None, 7, 7, 2048) 1050624 activation_42[0][0] \n__________________________________________________________________________________________________\nres5a_branch1 (Conv2D) (None, 7, 7, 2048) 2099200 activation_40[0][0] \n__________________________________________________________________________________________________\nbn5a_branch2c (BatchNormalizati (None, 7, 7, 2048) 8192 res5a_branch2c[0][0] \n__________________________________________________________________________________________________\nbn5a_branch1 (BatchNormalizatio (None, 7, 7, 2048) 8192 res5a_branch1[0][0] \n__________________________________________________________________________________________________\nadd_14 (Add) (None, 7, 7, 2048) 0 bn5a_branch2c[0][0] \n bn5a_branch1[0][0] \n__________________________________________________________________________________________________\nactivation_43 (Activation) (None, 7, 7, 2048) 0 add_14[0][0] \n__________________________________________________________________________________________________\nres5b_branch2a (Conv2D) (None, 7, 7, 512) 1049088 activation_43[0][0] \n__________________________________________________________________________________________________\nbn5b_branch2a (BatchNormalizati (None, 7, 7, 512) 2048 res5b_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_44 (Activation) (None, 7, 7, 512) 0 bn5b_branch2a[0][0] \n__________________________________________________________________________________________________\nres5b_branch2b (Conv2D) (None, 7, 7, 512) 2359808 activation_44[0][0] \n__________________________________________________________________________________________________\nbn5b_branch2b (BatchNormalizati (None, 7, 7, 512) 2048 res5b_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_45 (Activation) (None, 7, 7, 512) 0 bn5b_branch2b[0][0] \n__________________________________________________________________________________________________\nres5b_branch2c (Conv2D) (None, 7, 7, 2048) 1050624 activation_45[0][0] \n__________________________________________________________________________________________________\nbn5b_branch2c (BatchNormalizati (None, 7, 7, 2048) 8192 res5b_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_15 (Add) (None, 7, 7, 2048) 0 bn5b_branch2c[0][0] \n activation_43[0][0] \n__________________________________________________________________________________________________\nactivation_46 (Activation) (None, 7, 7, 2048) 0 add_15[0][0] \n__________________________________________________________________________________________________\nres5c_branch2a (Conv2D) (None, 7, 7, 512) 1049088 activation_46[0][0] \n__________________________________________________________________________________________________\nbn5c_branch2a (BatchNormalizati (None, 7, 7, 512) 2048 res5c_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_47 (Activation) (None, 7, 7, 512) 0 bn5c_branch2a[0][0] \n__________________________________________________________________________________________________\nres5c_branch2b (Conv2D) (None, 7, 7, 512) 2359808 activation_47[0][0] \n__________________________________________________________________________________________________\nbn5c_branch2b (BatchNormalizati (None, 7, 7, 512) 2048 res5c_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_48 (Activation) (None, 7, 7, 512) 0 bn5c_branch2b[0][0] \n__________________________________________________________________________________________________\nres5c_branch2c (Conv2D) (None, 7, 7, 2048) 1050624 activation_48[0][0] \n__________________________________________________________________________________________________\nbn5c_branch2c (BatchNormalizati (None, 7, 7, 2048) 8192 res5c_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_16 (Add) (None, 7, 7, 2048) 0 bn5c_branch2c[0][0] \n activation_46[0][0] \n__________________________________________________________________________________________________\nactivation_49 (Activation) (None, 7, 7, 2048) 0 add_16[0][0] \n__________________________________________________________________________________________________\nglobal_average_pooling2d_1 (Glo (None, 2048) 0 activation_49[0][0] \n__________________________________________________________________________________________________\ndense_1 (Dense) (None, 256) 524544 global_average_pooling2d_1[0][0] \n__________________________________________________________________________________________________\ndense_2 (Dense) (None, 2) 514 dense_1[0][0] \n==================================================================================================\nTotal params: 24,112,770\nTrainable params: 24,059,650\nNon-trainable params: 53,120\n__________________________________________________________________________________________________\n" ], [ "for ix,layers in enumerate(model.layers):\n print (ix, layers)", "0 <keras.engine.input_layer.InputLayer object at 0x7fcd75ec60f0>\n1 <keras.layers.convolutional.ZeroPadding2D object at 0x7fcd75ec66a0>\n2 <keras.layers.convolutional.Conv2D object at 0x7fcd75ec67b8>\n3 <keras.layers.normalization.BatchNormalization object at 0x7fcd75ec6be0>\n4 <keras.layers.core.Activation object at 0x7fcd75ec6dd8>\n5 <keras.layers.convolutional.ZeroPadding2D object at 0x7fcd51f5ee48>\n6 <keras.layers.pooling.MaxPooling2D object at 0x7fcd51efef28>\n7 <keras.layers.convolutional.Conv2D object at 0x7fcd75d99f28>\n8 <keras.layers.normalization.BatchNormalization object at 0x7fcd5062b3c8>\n9 <keras.layers.core.Activation object at 0x7fcd5062bcc0>\n10 <keras.layers.convolutional.Conv2D object at 0x7fcd5061b630>\n11 <keras.layers.normalization.BatchNormalization object at 0x7fcd505f8da0>\n12 <keras.layers.core.Activation object at 0x7fcd505bb748>\n13 <keras.layers.convolutional.Conv2D object at 0x7fcd50512fd0>\n14 <keras.layers.convolutional.Conv2D object at 0x7fcd504b5940>\n15 <keras.layers.normalization.BatchNormalization object at 0x7fcd504eef98>\n16 <keras.layers.normalization.BatchNormalization object at 0x7fcd503efbe0>\n17 <keras.layers.merge.Add object at 0x7fcd503b2eb8>\n18 <keras.layers.core.Activation object at 0x7fcd50331978>\n19 <keras.layers.convolutional.Conv2D object at 0x7fcd75ddd390>\n20 <keras.layers.normalization.BatchNormalization object at 0x7fcd5034d9b0>\n21 <keras.layers.core.Activation object at 0x7fcd75ec90f0>\n22 <keras.layers.convolutional.Conv2D object at 0x7fcd75e18630>\n23 <keras.layers.normalization.BatchNormalization object at 0x7fcd75e16908>\n24 <keras.layers.core.Activation object at 0x7fcd503097f0>\n25 <keras.layers.convolutional.Conv2D object at 0x7fcd50107cc0>\n26 <keras.layers.normalization.BatchNormalization object at 0x7fcd500e77f0>\n27 <keras.layers.merge.Add object at 0x7fcd500aeda0>\n28 <keras.layers.core.Activation object at 0x7fcd50050d30>\n29 <keras.layers.convolutional.Conv2D object at 0x7fcd50050d68>\n30 <keras.layers.normalization.BatchNormalization object at 0x7fcd387a0320>\n31 <keras.layers.core.Activation object at 0x7fcd386ff748>\n32 <keras.layers.convolutional.Conv2D object at 0x7fcd386f3e80>\n33 <keras.layers.normalization.BatchNormalization object at 0x7fcd386d2f28>\n34 <keras.layers.core.Activation object at 0x7fcd38696b00>\n35 <keras.layers.convolutional.Conv2D object at 0x7fcd38634860>\n36 <keras.layers.normalization.BatchNormalization object at 0x7fcd385af1d0>\n37 <keras.layers.merge.Add object at 0x7fcd38540860>\n38 <keras.layers.core.Activation object at 0x7fcd3852ae80>\n39 <keras.layers.convolutional.Conv2D object at 0x7fcd3852aba8>\n40 <keras.layers.normalization.BatchNormalization object at 0x7fcd384a5908>\n41 <keras.layers.core.Activation object at 0x7fcd38478ba8>\n42 <keras.layers.convolutional.Conv2D object at 0x7fcd38427dd8>\n43 <keras.layers.normalization.BatchNormalization object at 0x7fcd383da748>\n44 <keras.layers.core.Activation object at 0x7fcd3839c940>\n45 <keras.layers.convolutional.Conv2D object at 0x7fcd38292630>\n46 <keras.layers.convolutional.Conv2D object at 0x7fcd382b2748>\n47 <keras.layers.normalization.BatchNormalization object at 0x7fcd382eeda0>\n48 <keras.layers.normalization.BatchNormalization object at 0x7fcd381e7860>\n49 <keras.layers.merge.Add object at 0x7fcd381abb00>\n50 <keras.layers.core.Activation object at 0x7fcd3812e6d8>\n51 <keras.layers.convolutional.Conv2D object at 0x7fcd3812e630>\n52 <keras.layers.normalization.BatchNormalization object at 0x7fcd38096f98>\n53 <keras.layers.core.Activation object at 0x7fcd1a7ea208>\n54 <keras.layers.convolutional.Conv2D object at 0x7fcd1a7a1f28>\n55 <keras.layers.normalization.BatchNormalization object at 0x7fcd1a788e10>\n56 <keras.layers.core.Activation object at 0x7fcd1a788be0>\n57 <keras.layers.convolutional.Conv2D object at 0x7fcd1a654e48>\n58 <keras.layers.normalization.BatchNormalization object at 0x7fcd1a6b8908>\n59 <keras.layers.merge.Add object at 0x7fcd1a67cc50>\n60 <keras.layers.core.Activation object at 0x7fcd1a59fe48>\n61 <keras.layers.convolutional.Conv2D object at 0x7fcd1a59fe80>\n62 <keras.layers.normalization.BatchNormalization object at 0x7fcd1a515588>\n63 <keras.layers.core.Activation object at 0x7fcd1a4e7860>\n64 <keras.layers.convolutional.Conv2D object at 0x7fcd1a464f98>\n65 <keras.layers.normalization.BatchNormalization object at 0x7fcd1a446748>\n66 <keras.layers.core.Activation object at 0x7fcd1a38af60>\n67 <keras.layers.convolutional.Conv2D object at 0x7fcd1a3ad978>\n68 <keras.layers.normalization.BatchNormalization object at 0x7fcd1a3052e8>\n69 <keras.layers.merge.Add object at 0x7fcd1a325320>\n70 <keras.layers.core.Activation object at 0x7fcd1a2b9cf8>\n71 <keras.layers.convolutional.Conv2D object at 0x7fcd1a255518>\n72 <keras.layers.normalization.BatchNormalization object at 0x7fcd1a21a9e8>\n73 <keras.layers.core.Activation object at 0x7fcd1a1e8cc0>\n74 <keras.layers.convolutional.Conv2D object at 0x7fcd1a19ad68>\n75 <keras.layers.normalization.BatchNormalization object at 0x7fcd1a151c18>\n76 <keras.layers.core.Activation object at 0x7fcd1a10cf60>\n77 <keras.layers.convolutional.Conv2D object at 0x7fcd1a006748>\n78 <keras.layers.normalization.BatchNormalization object at 0x7fcd1a066e48>\n79 <keras.layers.merge.Add object at 0x7fcd1a028ef0>\n80 <keras.layers.core.Activation object at 0x7fcd19f5d978>\n81 <keras.layers.convolutional.Conv2D object at 0x7fcd19f5d898>\n82 <keras.layers.normalization.BatchNormalization object at 0x7fcd19ef5eb8>\n83 <keras.layers.core.Activation object at 0x7fcd19fa79e8>\n84 <keras.layers.convolutional.Conv2D object at 0x7fcd19e56d30>\n85 <keras.layers.normalization.BatchNormalization object at 0x7fcd19e42b70>\n86 <keras.layers.core.Activation object at 0x7fcd19e34668>\n87 <keras.layers.convolutional.Conv2D object at 0x7fcd19d0aeb8>\n88 <keras.layers.convolutional.Conv2D object at 0x7fcd19d2fc18>\n89 <keras.layers.normalization.BatchNormalization object at 0x7fcd19d67278>\n90 <keras.layers.normalization.BatchNormalization object at 0x7fcd19c6bcf8>\n91 <keras.layers.merge.Add object at 0x7fcd19c25e48>\n92 <keras.layers.core.Activation object at 0x7fcd19b44f60>\n93 <keras.layers.convolutional.Conv2D object at 0x7fcd19b7aba8>\n94 <keras.layers.normalization.BatchNormalization object at 0x7fcd19abe240>\n95 <keras.layers.core.Activation object at 0x7fcd19a3e048>\n96 <keras.layers.convolutional.Conv2D object at 0x7fcd19a75c18>\n97 <keras.layers.normalization.BatchNormalization object at 0x7fcd19a0ef60>\n98 <keras.layers.core.Activation object at 0x7fcd199d59b0>\n99 <keras.layers.convolutional.Conv2D object at 0x7fcd19972710>\n100 <keras.layers.normalization.BatchNormalization object at 0x7fcd1990c7b8>\n101 <keras.layers.merge.Add object at 0x7fcd198cef98>\n102 <keras.layers.core.Activation object at 0x7fcd19868d30>\n103 <keras.layers.convolutional.Conv2D object at 0x7fcd19868a58>\n104 <keras.layers.normalization.BatchNormalization object at 0x7fcd197e0748>\n105 <keras.layers.core.Activation object at 0x7fcd197b2f28>\n106 <keras.layers.convolutional.Conv2D object at 0x7fcd196d84e0>\n107 <keras.layers.normalization.BatchNormalization object at 0x7fcd19732f98>\n108 <keras.layers.core.Activation object at 0x7fcd196f35f8>\n109 <keras.layers.convolutional.Conv2D object at 0x7fcd19611cf8>\n110 <keras.layers.normalization.BatchNormalization object at 0x7fcd19629f60>\n111 <keras.layers.merge.Add object at 0x7fcd195ebba8>\n112 <keras.layers.core.Activation object at 0x7fcd19571e48>\n113 <keras.layers.convolutional.Conv2D object at 0x7fcd19571cf8>\n114 <keras.layers.normalization.BatchNormalization object at 0x7fcd194e4f28>\n115 <keras.layers.core.Activation object at 0x7fcd1945bb38>\n116 <keras.layers.convolutional.Conv2D object at 0x7fcd19405d68>\n117 <keras.layers.normalization.BatchNormalization object at 0x7fcd1937df60>\n118 <keras.layers.core.Activation object at 0x7fcd1937def0>\n119 <keras.layers.convolutional.Conv2D object at 0x7fcd192f05c0>\n120 <keras.layers.normalization.BatchNormalization object at 0x7fcd192cfd30>\n121 <keras.layers.merge.Add object at 0x7fcd192946d8>\n122 <keras.layers.core.Activation object at 0x7fcd19210f98>\n123 <keras.layers.convolutional.Conv2D object at 0x7fcd19210e10>\n124 <keras.layers.normalization.BatchNormalization object at 0x7fcd1918e240>\n125 <keras.layers.core.Activation object at 0x7fcd1910f5f8>\n126 <keras.layers.convolutional.Conv2D object at 0x7fcd190c7c18>\n127 <keras.layers.normalization.BatchNormalization object at 0x7fcd190ddf60>\n128 <keras.layers.core.Activation object at 0x7fcd190a59b0>\n129 <keras.layers.convolutional.Conv2D object at 0x7fcd19025cc0>\n130 <keras.layers.normalization.BatchNormalization object at 0x7fcd18fc2710>\n131 <keras.layers.merge.Add object at 0x7fcd18f9e0f0>\n132 <keras.layers.core.Activation object at 0x7fcd18f37d30>\n133 <keras.layers.convolutional.Conv2D object at 0x7fcd18f37a58>\n134 <keras.layers.normalization.BatchNormalization object at 0x7fcd18eb1780>\n135 <keras.layers.core.Activation object at 0x7fcd18dfeb38>\n136 <keras.layers.convolutional.Conv2D object at 0x7fcd18daa4e0>\n137 <keras.layers.normalization.BatchNormalization object at 0x7fcd18d82f98>\n138 <keras.layers.core.Activation object at 0x7fcd18d445f8>\n139 <keras.layers.convolutional.Conv2D object at 0x7fcd18ce3cf8>\n140 <keras.layers.normalization.BatchNormalization object at 0x7fcd18cfbf60>\n141 <keras.layers.merge.Add object at 0x7fcd18c3f908>\n142 <keras.layers.core.Activation object at 0x7fcd18bc34a8>\n143 <keras.layers.convolutional.Conv2D object at 0x7fcd18bc34e0>\n144 <keras.layers.normalization.BatchNormalization object at 0x7fcd18bb2da0>\n145 <keras.layers.core.Activation object at 0x7fcd18b28b38>\n146 <keras.layers.convolutional.Conv2D object at 0x7fcd18ad2eb8>\n147 <keras.layers.normalization.BatchNormalization object at 0x7fcd18a8b6d8>\n148 <keras.layers.core.Activation object at 0x7fcd18a4dd68>\n149 <keras.layers.convolutional.Conv2D object at 0x7fcd189425c0>\n150 <keras.layers.convolutional.Conv2D object at 0x7fcd189636d8>\n151 <keras.layers.normalization.BatchNormalization object at 0x7fcd1899dd30>\n152 <keras.layers.normalization.BatchNormalization object at 0x7fcd188987f0>\n153 <keras.layers.merge.Add object at 0x7fcd1885ca90>\n154 <keras.layers.core.Activation object at 0x7fcd187f8a90>\n155 <keras.layers.convolutional.Conv2D object at 0x7fcd187f8f60>\n156 <keras.layers.normalization.BatchNormalization object at 0x7fcd18771be0>\n157 <keras.layers.core.Activation object at 0x7fcd186c4e10>\n158 <keras.layers.convolutional.Conv2D object at 0x7fcd18669438>\n159 <keras.layers.normalization.BatchNormalization object at 0x7fcd18645f60>\n160 <keras.layers.core.Activation object at 0x7fcd1860b3c8>\n161 <keras.layers.convolutional.Conv2D object at 0x7fcd185a1c50>\n162 <keras.layers.normalization.BatchNormalization object at 0x7fcd185baf98>\n163 <keras.layers.merge.Add object at 0x7fcd18504860>\n164 <keras.layers.core.Activation object at 0x7fcd18483cf8>\n165 <keras.layers.convolutional.Conv2D object at 0x7fcd18483cc0>\n166 <keras.layers.normalization.BatchNormalization object at 0x7fcd1847ce80>\n167 <keras.layers.core.Activation object at 0x7fcd183eaa90>\n168 <keras.layers.convolutional.Conv2D object at 0x7fcd18367f28>\n169 <keras.layers.normalization.BatchNormalization object at 0x7fcd1834b978>\n170 <keras.layers.core.Activation object at 0x7fcd18311c88>\n171 <keras.layers.convolutional.Conv2D object at 0x7fcd181ff518>\n172 <keras.layers.normalization.BatchNormalization object at 0x7fcd18261fd0>\n173 <keras.layers.merge.Add object at 0x7fcd18222630>\n174 <keras.layers.core.Activation object at 0x7fcd18159748>\n175 <keras.layers.pooling.GlobalAveragePooling2D object at 0x7fcd137181d0>\n176 <keras.layers.core.Dense object at 0x7fcd13718e10>\n177 <keras.layers.core.Dense object at 0x7fcd13705a58>\n" ], [ "for ix in range(171):\n model.layers[ix].trainable = False", "_____no_output_____" ], [ "model.summary()", "__________________________________________________________________________________________________\nLayer (type) Output Shape Param # Connected to \n==================================================================================================\ninput_1 (InputLayer) (None, 224, 224, 3) 0 \n__________________________________________________________________________________________________\nconv1_pad (ZeroPadding2D) (None, 230, 230, 3) 0 input_1[0][0] \n__________________________________________________________________________________________________\nconv1 (Conv2D) (None, 112, 112, 64) 9472 conv1_pad[0][0] \n__________________________________________________________________________________________________\nbn_conv1 (BatchNormalization) (None, 112, 112, 64) 256 conv1[0][0] \n__________________________________________________________________________________________________\nactivation_1 (Activation) (None, 112, 112, 64) 0 bn_conv1[0][0] \n__________________________________________________________________________________________________\npool1_pad (ZeroPadding2D) (None, 114, 114, 64) 0 activation_1[0][0] \n__________________________________________________________________________________________________\nmax_pooling2d_1 (MaxPooling2D) (None, 56, 56, 64) 0 pool1_pad[0][0] \n__________________________________________________________________________________________________\nres2a_branch2a (Conv2D) (None, 56, 56, 64) 4160 max_pooling2d_1[0][0] \n__________________________________________________________________________________________________\nbn2a_branch2a (BatchNormalizati (None, 56, 56, 64) 256 res2a_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_2 (Activation) (None, 56, 56, 64) 0 bn2a_branch2a[0][0] \n__________________________________________________________________________________________________\nres2a_branch2b (Conv2D) (None, 56, 56, 64) 36928 activation_2[0][0] \n__________________________________________________________________________________________________\nbn2a_branch2b (BatchNormalizati (None, 56, 56, 64) 256 res2a_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_3 (Activation) (None, 56, 56, 64) 0 bn2a_branch2b[0][0] \n__________________________________________________________________________________________________\nres2a_branch2c (Conv2D) (None, 56, 56, 256) 16640 activation_3[0][0] \n__________________________________________________________________________________________________\nres2a_branch1 (Conv2D) (None, 56, 56, 256) 16640 max_pooling2d_1[0][0] \n__________________________________________________________________________________________________\nbn2a_branch2c (BatchNormalizati (None, 56, 56, 256) 1024 res2a_branch2c[0][0] \n__________________________________________________________________________________________________\nbn2a_branch1 (BatchNormalizatio (None, 56, 56, 256) 1024 res2a_branch1[0][0] \n__________________________________________________________________________________________________\nadd_1 (Add) (None, 56, 56, 256) 0 bn2a_branch2c[0][0] \n bn2a_branch1[0][0] \n__________________________________________________________________________________________________\nactivation_4 (Activation) (None, 56, 56, 256) 0 add_1[0][0] \n__________________________________________________________________________________________________\nres2b_branch2a (Conv2D) (None, 56, 56, 64) 16448 activation_4[0][0] \n__________________________________________________________________________________________________\nbn2b_branch2a (BatchNormalizati (None, 56, 56, 64) 256 res2b_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_5 (Activation) (None, 56, 56, 64) 0 bn2b_branch2a[0][0] \n__________________________________________________________________________________________________\nres2b_branch2b (Conv2D) (None, 56, 56, 64) 36928 activation_5[0][0] \n__________________________________________________________________________________________________\nbn2b_branch2b (BatchNormalizati (None, 56, 56, 64) 256 res2b_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_6 (Activation) (None, 56, 56, 64) 0 bn2b_branch2b[0][0] \n__________________________________________________________________________________________________\nres2b_branch2c (Conv2D) (None, 56, 56, 256) 16640 activation_6[0][0] \n__________________________________________________________________________________________________\nbn2b_branch2c (BatchNormalizati (None, 56, 56, 256) 1024 res2b_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_2 (Add) (None, 56, 56, 256) 0 bn2b_branch2c[0][0] \n activation_4[0][0] \n__________________________________________________________________________________________________\nactivation_7 (Activation) (None, 56, 56, 256) 0 add_2[0][0] \n__________________________________________________________________________________________________\nres2c_branch2a (Conv2D) (None, 56, 56, 64) 16448 activation_7[0][0] \n__________________________________________________________________________________________________\nbn2c_branch2a (BatchNormalizati (None, 56, 56, 64) 256 res2c_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_8 (Activation) (None, 56, 56, 64) 0 bn2c_branch2a[0][0] \n__________________________________________________________________________________________________\nres2c_branch2b (Conv2D) (None, 56, 56, 64) 36928 activation_8[0][0] \n__________________________________________________________________________________________________\nbn2c_branch2b (BatchNormalizati (None, 56, 56, 64) 256 res2c_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_9 (Activation) (None, 56, 56, 64) 0 bn2c_branch2b[0][0] \n__________________________________________________________________________________________________\nres2c_branch2c (Conv2D) (None, 56, 56, 256) 16640 activation_9[0][0] \n__________________________________________________________________________________________________\nbn2c_branch2c (BatchNormalizati (None, 56, 56, 256) 1024 res2c_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_3 (Add) (None, 56, 56, 256) 0 bn2c_branch2c[0][0] \n activation_7[0][0] \n__________________________________________________________________________________________________\nactivation_10 (Activation) (None, 56, 56, 256) 0 add_3[0][0] \n__________________________________________________________________________________________________\nres3a_branch2a (Conv2D) (None, 28, 28, 128) 32896 activation_10[0][0] \n__________________________________________________________________________________________________\nbn3a_branch2a (BatchNormalizati (None, 28, 28, 128) 512 res3a_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_11 (Activation) (None, 28, 28, 128) 0 bn3a_branch2a[0][0] \n__________________________________________________________________________________________________\nres3a_branch2b (Conv2D) (None, 28, 28, 128) 147584 activation_11[0][0] \n__________________________________________________________________________________________________\nbn3a_branch2b (BatchNormalizati (None, 28, 28, 128) 512 res3a_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_12 (Activation) (None, 28, 28, 128) 0 bn3a_branch2b[0][0] \n__________________________________________________________________________________________________\nres3a_branch2c (Conv2D) (None, 28, 28, 512) 66048 activation_12[0][0] \n__________________________________________________________________________________________________\nres3a_branch1 (Conv2D) (None, 28, 28, 512) 131584 activation_10[0][0] \n__________________________________________________________________________________________________\nbn3a_branch2c (BatchNormalizati (None, 28, 28, 512) 2048 res3a_branch2c[0][0] \n__________________________________________________________________________________________________\nbn3a_branch1 (BatchNormalizatio (None, 28, 28, 512) 2048 res3a_branch1[0][0] \n__________________________________________________________________________________________________\nadd_4 (Add) (None, 28, 28, 512) 0 bn3a_branch2c[0][0] \n bn3a_branch1[0][0] \n__________________________________________________________________________________________________\nactivation_13 (Activation) (None, 28, 28, 512) 0 add_4[0][0] \n__________________________________________________________________________________________________\nres3b_branch2a (Conv2D) (None, 28, 28, 128) 65664 activation_13[0][0] \n__________________________________________________________________________________________________\nbn3b_branch2a (BatchNormalizati (None, 28, 28, 128) 512 res3b_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_14 (Activation) (None, 28, 28, 128) 0 bn3b_branch2a[0][0] \n__________________________________________________________________________________________________\nres3b_branch2b (Conv2D) (None, 28, 28, 128) 147584 activation_14[0][0] \n__________________________________________________________________________________________________\nbn3b_branch2b (BatchNormalizati (None, 28, 28, 128) 512 res3b_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_15 (Activation) (None, 28, 28, 128) 0 bn3b_branch2b[0][0] \n__________________________________________________________________________________________________\nres3b_branch2c (Conv2D) (None, 28, 28, 512) 66048 activation_15[0][0] \n__________________________________________________________________________________________________\nbn3b_branch2c (BatchNormalizati (None, 28, 28, 512) 2048 res3b_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_5 (Add) (None, 28, 28, 512) 0 bn3b_branch2c[0][0] \n activation_13[0][0] \n__________________________________________________________________________________________________\nactivation_16 (Activation) (None, 28, 28, 512) 0 add_5[0][0] \n__________________________________________________________________________________________________\nres3c_branch2a (Conv2D) (None, 28, 28, 128) 65664 activation_16[0][0] \n__________________________________________________________________________________________________\nbn3c_branch2a (BatchNormalizati (None, 28, 28, 128) 512 res3c_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_17 (Activation) (None, 28, 28, 128) 0 bn3c_branch2a[0][0] \n__________________________________________________________________________________________________\nres3c_branch2b (Conv2D) (None, 28, 28, 128) 147584 activation_17[0][0] \n__________________________________________________________________________________________________\nbn3c_branch2b (BatchNormalizati (None, 28, 28, 128) 512 res3c_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_18 (Activation) (None, 28, 28, 128) 0 bn3c_branch2b[0][0] \n__________________________________________________________________________________________________\nres3c_branch2c (Conv2D) (None, 28, 28, 512) 66048 activation_18[0][0] \n__________________________________________________________________________________________________\nbn3c_branch2c (BatchNormalizati (None, 28, 28, 512) 2048 res3c_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_6 (Add) (None, 28, 28, 512) 0 bn3c_branch2c[0][0] \n activation_16[0][0] \n__________________________________________________________________________________________________\nactivation_19 (Activation) (None, 28, 28, 512) 0 add_6[0][0] \n__________________________________________________________________________________________________\nres3d_branch2a (Conv2D) (None, 28, 28, 128) 65664 activation_19[0][0] \n__________________________________________________________________________________________________\nbn3d_branch2a (BatchNormalizati (None, 28, 28, 128) 512 res3d_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_20 (Activation) (None, 28, 28, 128) 0 bn3d_branch2a[0][0] \n__________________________________________________________________________________________________\nres3d_branch2b (Conv2D) (None, 28, 28, 128) 147584 activation_20[0][0] \n__________________________________________________________________________________________________\nbn3d_branch2b (BatchNormalizati (None, 28, 28, 128) 512 res3d_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_21 (Activation) (None, 28, 28, 128) 0 bn3d_branch2b[0][0] \n__________________________________________________________________________________________________\nres3d_branch2c (Conv2D) (None, 28, 28, 512) 66048 activation_21[0][0] \n__________________________________________________________________________________________________\nbn3d_branch2c (BatchNormalizati (None, 28, 28, 512) 2048 res3d_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_7 (Add) (None, 28, 28, 512) 0 bn3d_branch2c[0][0] \n activation_19[0][0] \n__________________________________________________________________________________________________\nactivation_22 (Activation) (None, 28, 28, 512) 0 add_7[0][0] \n__________________________________________________________________________________________________\nres4a_branch2a (Conv2D) (None, 14, 14, 256) 131328 activation_22[0][0] \n__________________________________________________________________________________________________\nbn4a_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4a_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_23 (Activation) (None, 14, 14, 256) 0 bn4a_branch2a[0][0] \n__________________________________________________________________________________________________\nres4a_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_23[0][0] \n__________________________________________________________________________________________________\nbn4a_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4a_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_24 (Activation) (None, 14, 14, 256) 0 bn4a_branch2b[0][0] \n__________________________________________________________________________________________________\nres4a_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_24[0][0] \n__________________________________________________________________________________________________\nres4a_branch1 (Conv2D) (None, 14, 14, 1024) 525312 activation_22[0][0] \n__________________________________________________________________________________________________\nbn4a_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4a_branch2c[0][0] \n__________________________________________________________________________________________________\nbn4a_branch1 (BatchNormalizatio (None, 14, 14, 1024) 4096 res4a_branch1[0][0] \n__________________________________________________________________________________________________\nadd_8 (Add) (None, 14, 14, 1024) 0 bn4a_branch2c[0][0] \n bn4a_branch1[0][0] \n__________________________________________________________________________________________________\nactivation_25 (Activation) (None, 14, 14, 1024) 0 add_8[0][0] \n__________________________________________________________________________________________________\nres4b_branch2a (Conv2D) (None, 14, 14, 256) 262400 activation_25[0][0] \n__________________________________________________________________________________________________\nbn4b_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4b_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_26 (Activation) (None, 14, 14, 256) 0 bn4b_branch2a[0][0] \n__________________________________________________________________________________________________\nres4b_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_26[0][0] \n__________________________________________________________________________________________________\nbn4b_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4b_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_27 (Activation) (None, 14, 14, 256) 0 bn4b_branch2b[0][0] \n__________________________________________________________________________________________________\nres4b_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_27[0][0] \n__________________________________________________________________________________________________\nbn4b_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4b_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_9 (Add) (None, 14, 14, 1024) 0 bn4b_branch2c[0][0] \n activation_25[0][0] \n__________________________________________________________________________________________________\nactivation_28 (Activation) (None, 14, 14, 1024) 0 add_9[0][0] \n__________________________________________________________________________________________________\nres4c_branch2a (Conv2D) (None, 14, 14, 256) 262400 activation_28[0][0] \n__________________________________________________________________________________________________\nbn4c_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4c_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_29 (Activation) (None, 14, 14, 256) 0 bn4c_branch2a[0][0] \n__________________________________________________________________________________________________\nres4c_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_29[0][0] \n__________________________________________________________________________________________________\nbn4c_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4c_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_30 (Activation) (None, 14, 14, 256) 0 bn4c_branch2b[0][0] \n__________________________________________________________________________________________________\nres4c_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_30[0][0] \n__________________________________________________________________________________________________\nbn4c_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4c_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_10 (Add) (None, 14, 14, 1024) 0 bn4c_branch2c[0][0] \n activation_28[0][0] \n__________________________________________________________________________________________________\nactivation_31 (Activation) (None, 14, 14, 1024) 0 add_10[0][0] \n__________________________________________________________________________________________________\nres4d_branch2a (Conv2D) (None, 14, 14, 256) 262400 activation_31[0][0] \n__________________________________________________________________________________________________\nbn4d_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4d_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_32 (Activation) (None, 14, 14, 256) 0 bn4d_branch2a[0][0] \n__________________________________________________________________________________________________\nres4d_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_32[0][0] \n__________________________________________________________________________________________________\nbn4d_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4d_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_33 (Activation) (None, 14, 14, 256) 0 bn4d_branch2b[0][0] \n__________________________________________________________________________________________________\nres4d_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_33[0][0] \n__________________________________________________________________________________________________\nbn4d_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4d_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_11 (Add) (None, 14, 14, 1024) 0 bn4d_branch2c[0][0] \n activation_31[0][0] \n__________________________________________________________________________________________________\nactivation_34 (Activation) (None, 14, 14, 1024) 0 add_11[0][0] \n__________________________________________________________________________________________________\nres4e_branch2a (Conv2D) (None, 14, 14, 256) 262400 activation_34[0][0] \n__________________________________________________________________________________________________\nbn4e_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4e_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_35 (Activation) (None, 14, 14, 256) 0 bn4e_branch2a[0][0] \n__________________________________________________________________________________________________\nres4e_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_35[0][0] \n__________________________________________________________________________________________________\nbn4e_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4e_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_36 (Activation) (None, 14, 14, 256) 0 bn4e_branch2b[0][0] \n__________________________________________________________________________________________________\nres4e_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_36[0][0] \n__________________________________________________________________________________________________\nbn4e_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4e_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_12 (Add) (None, 14, 14, 1024) 0 bn4e_branch2c[0][0] \n activation_34[0][0] \n__________________________________________________________________________________________________\nactivation_37 (Activation) (None, 14, 14, 1024) 0 add_12[0][0] \n__________________________________________________________________________________________________\nres4f_branch2a (Conv2D) (None, 14, 14, 256) 262400 activation_37[0][0] \n__________________________________________________________________________________________________\nbn4f_branch2a (BatchNormalizati (None, 14, 14, 256) 1024 res4f_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_38 (Activation) (None, 14, 14, 256) 0 bn4f_branch2a[0][0] \n__________________________________________________________________________________________________\nres4f_branch2b (Conv2D) (None, 14, 14, 256) 590080 activation_38[0][0] \n__________________________________________________________________________________________________\nbn4f_branch2b (BatchNormalizati (None, 14, 14, 256) 1024 res4f_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_39 (Activation) (None, 14, 14, 256) 0 bn4f_branch2b[0][0] \n__________________________________________________________________________________________________\nres4f_branch2c (Conv2D) (None, 14, 14, 1024) 263168 activation_39[0][0] \n__________________________________________________________________________________________________\nbn4f_branch2c (BatchNormalizati (None, 14, 14, 1024) 4096 res4f_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_13 (Add) (None, 14, 14, 1024) 0 bn4f_branch2c[0][0] \n activation_37[0][0] \n__________________________________________________________________________________________________\nactivation_40 (Activation) (None, 14, 14, 1024) 0 add_13[0][0] \n__________________________________________________________________________________________________\nres5a_branch2a (Conv2D) (None, 7, 7, 512) 524800 activation_40[0][0] \n__________________________________________________________________________________________________\nbn5a_branch2a (BatchNormalizati (None, 7, 7, 512) 2048 res5a_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_41 (Activation) (None, 7, 7, 512) 0 bn5a_branch2a[0][0] \n__________________________________________________________________________________________________\nres5a_branch2b (Conv2D) (None, 7, 7, 512) 2359808 activation_41[0][0] \n__________________________________________________________________________________________________\nbn5a_branch2b (BatchNormalizati (None, 7, 7, 512) 2048 res5a_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_42 (Activation) (None, 7, 7, 512) 0 bn5a_branch2b[0][0] \n__________________________________________________________________________________________________\nres5a_branch2c (Conv2D) (None, 7, 7, 2048) 1050624 activation_42[0][0] \n__________________________________________________________________________________________________\nres5a_branch1 (Conv2D) (None, 7, 7, 2048) 2099200 activation_40[0][0] \n__________________________________________________________________________________________________\nbn5a_branch2c (BatchNormalizati (None, 7, 7, 2048) 8192 res5a_branch2c[0][0] \n__________________________________________________________________________________________________\nbn5a_branch1 (BatchNormalizatio (None, 7, 7, 2048) 8192 res5a_branch1[0][0] \n__________________________________________________________________________________________________\nadd_14 (Add) (None, 7, 7, 2048) 0 bn5a_branch2c[0][0] \n bn5a_branch1[0][0] \n__________________________________________________________________________________________________\nactivation_43 (Activation) (None, 7, 7, 2048) 0 add_14[0][0] \n__________________________________________________________________________________________________\nres5b_branch2a (Conv2D) (None, 7, 7, 512) 1049088 activation_43[0][0] \n__________________________________________________________________________________________________\nbn5b_branch2a (BatchNormalizati (None, 7, 7, 512) 2048 res5b_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_44 (Activation) (None, 7, 7, 512) 0 bn5b_branch2a[0][0] \n__________________________________________________________________________________________________\nres5b_branch2b (Conv2D) (None, 7, 7, 512) 2359808 activation_44[0][0] \n__________________________________________________________________________________________________\nbn5b_branch2b (BatchNormalizati (None, 7, 7, 512) 2048 res5b_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_45 (Activation) (None, 7, 7, 512) 0 bn5b_branch2b[0][0] \n__________________________________________________________________________________________________\nres5b_branch2c (Conv2D) (None, 7, 7, 2048) 1050624 activation_45[0][0] \n__________________________________________________________________________________________________\nbn5b_branch2c (BatchNormalizati (None, 7, 7, 2048) 8192 res5b_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_15 (Add) (None, 7, 7, 2048) 0 bn5b_branch2c[0][0] \n activation_43[0][0] \n__________________________________________________________________________________________________\nactivation_46 (Activation) (None, 7, 7, 2048) 0 add_15[0][0] \n__________________________________________________________________________________________________\nres5c_branch2a (Conv2D) (None, 7, 7, 512) 1049088 activation_46[0][0] \n__________________________________________________________________________________________________\nbn5c_branch2a (BatchNormalizati (None, 7, 7, 512) 2048 res5c_branch2a[0][0] \n__________________________________________________________________________________________________\nactivation_47 (Activation) (None, 7, 7, 512) 0 bn5c_branch2a[0][0] \n__________________________________________________________________________________________________\nres5c_branch2b (Conv2D) (None, 7, 7, 512) 2359808 activation_47[0][0] \n__________________________________________________________________________________________________\nbn5c_branch2b (BatchNormalizati (None, 7, 7, 512) 2048 res5c_branch2b[0][0] \n__________________________________________________________________________________________________\nactivation_48 (Activation) (None, 7, 7, 512) 0 bn5c_branch2b[0][0] \n__________________________________________________________________________________________________\nres5c_branch2c (Conv2D) (None, 7, 7, 2048) 1050624 activation_48[0][0] \n__________________________________________________________________________________________________\nbn5c_branch2c (BatchNormalizati (None, 7, 7, 2048) 8192 res5c_branch2c[0][0] \n__________________________________________________________________________________________________\nadd_16 (Add) (None, 7, 7, 2048) 0 bn5c_branch2c[0][0] \n activation_46[0][0] \n__________________________________________________________________________________________________\nactivation_49 (Activation) (None, 7, 7, 2048) 0 add_16[0][0] \n__________________________________________________________________________________________________\nglobal_average_pooling2d_1 (Glo (None, 2048) 0 activation_49[0][0] \n__________________________________________________________________________________________________\ndense_1 (Dense) (None, 256) 524544 global_average_pooling2d_1[0][0] \n__________________________________________________________________________________________________\ndense_2 (Dense) (None, 2) 514 dense_1[0][0] \n==================================================================================================\nTotal params: 24,112,770\nTrainable params: 1,579,778\nNon-trainable params: 22,532,992\n__________________________________________________________________________________________________\n" ], [ "adam = Adam(lr=0.0003)", "_____no_output_____" ], [ "model.compile(loss='categorical_crossentropy',optimizer=adam,metrics=['accuracy'])", "W0728 17:42:16.628257 140521947629376 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/optimizers.py:790: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead.\n\n" ], [ "model.fit(X_train,y_train,epochs=5,shuffle=True,batch_size=64,validation_split=.20) ", "W0728 17:42:16.857084 140521947629376 deprecation.py:323] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/math_grad.py:1250: add_dispatch_support.<locals>.wrapper (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.\nInstructions for updating:\nUse tf.where in 2.0, which has the same broadcast rule as np.where\n" ], [ "img_path = '/home/vandit/Desktop/My Projects/My_Submissions/Men_Women_Classification/men-women-classification/women/Sophie.jpg'\nimg = image.load_img(img_path, target_size=(224,224))\nx = image.img_to_array(img)\nx = np.expand_dims(x, axis=0)\nx = preprocess_input(x)\nprint(x.shape)\npreds = model.predict(x)\n# decode the results into a list of tuples (class, description, probability)\n# (one such list for each sample in the batch)\n# print('Predicted:', decode_predictions(preds, top=3)[0])\nans = model.predict(x)\nprint(dict_labels[np.argmax(ans)])", "(1, 224, 224, 3)\nwomen\n" ] ] ]

[ "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import requests\nimport pandas as pd\nfrom bs4 import BeautifulSoup\nimport string\nimport re\nimport nltk\nimport json\nimport matplotlib.pyplot as plt\nimport numpy as np", "_____no_output_____" ], [ "def get_text(url):\n response = requests.get(url)\n content = response.content\n parser = BeautifulSoup(content,'html.parser')\n return(parser.text)", "_____no_output_____" ], [ "def clean_text(script):\n script_clean=script.strip()\n script_clean=script_clean.replace(\"\\n\",\"\")\n script_clean=script_clean.replace(\"\\r\",\" \")\n script_clean=script_clean.replace(\"\\r\\n\",\"\")\n script_clean=re.sub(\"([\\(\\[]).*?([\\)\\]])\", \"\", script_clean)\n script_clean=re.sub(r'\\.([a-zA-Z])', r'. \\1', script_clean) #remove missing whitespace between character lines.\n script_clean=re.sub(r'\\!([a-zA-Z])', r'! \\1', script_clean)\n script_clean=re.sub(r'\\?([a-zA-Z])', r'? \\1', script_clean)\n return(script_clean)", "_____no_output_____" ], [ "def get_cast(script_clean):\n tokens=nltk.word_tokenize(script_clean)\n cast=[]\n for word in tokens:\n if re.search(\"\\\\b[A-Z]{3,}\\\\b\", word) is not None:\n cast.append(word)\n return(list(set(cast)))", "_____no_output_____" ], [ "script=get_text('http://www.chakoteya.net/DS9/575.htm')", "_____no_output_____" ], [ "script_clean=clean_text(script)", "_____no_output_____" ], [ "def get_lines(script_clean, cast):\n split_script=script_clean.split(':')\n lines_dict=dict.fromkeys(cast)\n for cast_member in cast:\n lines=[]\n for i in range(len(split_script)-1):\n if cast_member in split_script[i].strip().split(\" \"):\n line=split_script[i+1].strip().split(\" \")\n line=[word for word in line if word != '']\n for member in cast:\n if member in line:\n line.remove(member)\n line=' '.join(line)\n lines.append(line)\n lines_dict[cast_member]=lines\n\n return(lines_dict)", "_____no_output_____" ], [ "def get_page_links():\n top_links=[\"http://www.chakoteya.net/DS9/episodes.htm\", \n \"http://www.chakoteya.net/StarTrek/episodes.htm\", \n \"http://www.chakoteya.net/NextGen/episodes.htm\", \n \"http://www.chakoteya.net/Voyager/episode_listing.htm\", \n \"http://www.chakoteya.net/Enterprise/episodes.htm\"]\n short_links=[\"http://www.chakoteya.net/DS9/\", \n \"http://www.chakoteya.net/StarTrek/\", \n \"http://www.chakoteya.net/NextGen/\", \n \"http://www.chakoteya.net/Voyager/\", \n \"http://www.chakoteya.net/Enterprise/\"]\n links_list=[]\n names_list=[]\n for i, link in enumerate(top_links):\n response = requests.get(link)\n content = response.content\n parser = BeautifulSoup(content,'html.parser')\n urls = parser.find_all('a')\n for page in urls:\n links_list.append(short_links[i]+str(page.get('href')))\n name=page.text\n name=name.replace('\\r\\n',' ')\n names_list.append(name)\n \n \n links_to_remove=['http://www.chakoteya.net/Voyager/fortyseven.htm',\n 'http://www.chakoteya.net/Voyager/LineCountS1-S3.htm',\n 'http://www.chakoteya.net/Voyager/LineCountS4-S7.htm',\n 'http://www.chakoteya.net/Enterprise/fortyseven.htm',\n ]\n links_list=[link for link in links_list if (link.endswith('.htm')) & (link not in links_to_remove)]\n \n return(links_list)", "_____no_output_____" ], [ "# links_list", "_____no_output_____" ], [ "page_links=get_page_links()", "_____no_output_____" ], [ "len(page_links)", "_____no_output_____" ], [ "DS9_links = page_links[0:173]\nTOS_links = page_links[173:253]\nTAS_links = page_links[253:275]\nTNG_links = page_links[275:451]\nVOY_links = page_links[451:611]\nENT_links = page_links[611:708]\n\nlinks=[DS9_links, TOS_links, TAS_links, TNG_links, VOY_links, ENT_links]", "_____no_output_____" ], [ "links_names=['DS9', 'TOS', 'TAS', 'TNG', 'VOY', 'ENT']\nlinks=[DS9_links, TOS_links, TAS_links, TNG_links, VOY_links, ENT_links]\n\nall_series_scripts={}\nfor i,series in enumerate(links):\n series_name=str(links_names[i])\n print(series_name)\n all_series_scripts[series_name]={}\n episode_script={}\n all_cast=[]\n for j,link in enumerate(series):\n episode=\"episode \"+str(j)\n text=get_text(link)\n episode_script[episode]=text\n all_series_scripts[series_name]=episode_script\n\nprint(all_series_scripts)", "DS9\nTOS\nTAS\nTNG\nVOY\nENT\n" ], [ "with open('all_scripts_raw.json', 'w') as data:\n json.dump(all_series_scripts, data)", "_____no_output_____" ], [ "with open('all_scripts_raw.json', 'r') as data:\n all_scripts_raw = json.load(data)", "_____no_output_____" ], [ "\nlinks_names=['DS9', 'TOS', 'TAS', 'TNG', 'VOY', 'ENT']\n\nall_series_lines={}\nfor i,series in enumerate(links_names):\n print(series)\n series_name=str(links_names[i])\n all_series_lines[series_name]={}\n all_lines_dict={}\n all_cast=[]\n #for j,episode in enumerate(all_series_scripts[series]):\n for j,episode in enumerate(all_scripts_raw[series]):\n #script=all_series_scripts[series][episode]\n script=all_scripts_raw[series][episode]\n cleaned_script=clean_text(script)\n cast=get_cast(cleaned_script)\n for member in cast:\n if member not in all_cast:\n all_cast.append(member)\n lines_dict=get_lines(cleaned_script,all_cast)\n all_lines_dict[episode]=lines_dict\n all_series_lines[series]=all_lines_dict\n\nprint(all_series_lines)", "DS9\nTOS\nTAS\nTNG\nVOY\nENT\n" ], [ "with open('all_series_lines.json', 'w') as data:\n json.dump(all_series_lines, data)", "_____no_output_____" ], [ "with open('all_series_lines.json', 'r') as data:\n all_series_lines = json.load(data)", "_____no_output_____" ], [ "#checking against source to make sure the character lines\n#appear in the correct episode\nall_series_lines['TNG']['episode 30']['LAFORGE']", "_____no_output_____" ], [ "#writing the corrected df\n# all_series_lines = pd.DataFrame(data=all_series_lines)\n# all_series_lines.to_csv(r'C:\\Users\\Eric\\Desktop\\Star_Trek_Scripts-master\\Star_Trek_Scripts-master\\data\\all_series_lines.csv')\n#when I wrote it to a df spock ended up getting lines???????????????", "_____no_output_____" ], [ "episodes=all_series_lines['TNG'].keys()", "_____no_output_____" ], [ "total_lines_counts={}\nline_counts_by_episode={}\nfor i,ep in enumerate(episodes):\n if i == 0:\n episode=\"Episode 1 & 2\"\n else:\n episode=\"Episode \"+str(i+2)\n line_counts_by_episode[episode]={}\n if all_series_lines['TNG'][ep] is not np.NaN:\n for member in list(all_series_lines['TNG'][ep].keys()):\n line_counts_by_episode[episode][member]=len(all_series_lines['TNG'][ep][member])\n if member in total_lines_counts.keys():\n total_lines_counts[member]=total_lines_counts[member]+len(all_series_lines['TNG'][ep][member])\n else:\n total_lines_counts[member]=len(all_series_lines['TNG'][ep][member])", "_____no_output_____" ], [ "#checking to make sure Spock doesn't appear, since that was an issue before\nTNG_df_byep = pd.DataFrame(line_counts_by_episode)\n# TNG_df_byep.loc['SPOCK']\n\nTNG_df=pd.DataFrame(list(total_lines_counts.items()), columns=['Character','No. of Lines'])\nTop20=TNG_df.sort_values(by='No. of Lines', ascending=False).head(20)\n\nTop20.plot.bar(x='Character',y='No. of Lines')\nplt.show()\n\nTop20['Character']", "_____no_output_____" ], [ "export_TOP20 = Top20.to_csv(r'C:\\Users\\Eric\\startrek-dash-app\\assets\\top20')", "_____no_output_____" ], [ "export_vis_TNG = TNG_df_byep.to_csv(r'C:\\Users\\Eric\\startrek-dash-app\\assets\\bar_chart_TNG')", "_____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" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Classifying Fashion-MNIST\n\nNow it's your turn to build and train a neural network. You'll be using the [Fashion-MNIST dataset](https://github.com/zalandoresearch/fashion-mnist), a drop-in replacement for the MNIST dataset. MNIST is actually quite trivial with neural networks where you can easily achieve better than 97% accuracy. Fashion-MNIST is a set of 28x28 greyscale images of clothes. It's more complex than MNIST, so it's a better representation of the actual performance of your network, and a better representation of datasets you'll use in the real world.\n\n<img src='assets/fashion-mnist-sprite.png' width=500px>\n\nIn this notebook, you'll build your own neural network. For the most part, you could just copy and paste the code from Part 3, but you wouldn't be learning. It's important for you to write the code yourself and get it to work. Feel free to consult the previous notebooks though as you work through this.\n\nFirst off, let's load the dataset through torchvision.", "_____no_output_____" ] ], [ [ "import torch\nfrom torchvision import datasets, transforms\nimport helper\n\n# Define a transform to normalize the data\ntransform = transforms.Compose([transforms.ToTensor(),\n transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])\n# Download and load the training data\ntrainset = datasets.FashionMNIST('~/.pytorch/F_MNIST_data/', download=True, train=True, transform=transform)\ntrainloader = torch.utils.data.DataLoader(trainset, batch_size=64, shuffle=True)\n\n# Download and load the test data\ntestset = datasets.FashionMNIST('~/.pytorch/F_MNIST_data/', download=True, train=False, transform=transform)\ntestloader = torch.utils.data.DataLoader(testset, batch_size=64, shuffle=True)", "_____no_output_____" ] ], [ [ "Here we can see one of the images.", "_____no_output_____" ] ], [ [ "image, label = next(iter(trainloader))\nhelper.imshow(image[0,:]);", "_____no_output_____" ] ], [ [ "## Building the network\n\nHere you should define your network. As with MNIST, each image is 28x28 which is a total of 784 pixels, and there are 10 classes. You should include at least one hidden layer. We suggest you use ReLU activations for the layers and to return the logits or log-softmax from the forward pass. It's up to you how many layers you add and the size of those layers.", "_____no_output_____" ] ], [ [ "# TODO: Define your network architecture here\nfrom torch import nn\nn_hidden1 = 1024\nn_hidden2 = 128\nn_hidden3 = 64\nmodel = nn.Sequential(nn.Linear(784,n_hidden1),\n nn.ReLU(),\n nn.Linear(n_hidden1,n_hidden2),\n nn.ReLU(),\n nn.Linear(n_hidden2,n_hidden3),\n nn.ReLU(),\n nn.Linear(n_hidden3, 10),\n nn.LogSoftmax(dim=1))", "_____no_output_____" ] ], [ [ "# Train the network\n\nNow you should create your network and train it. First you'll want to define [the criterion](http://pytorch.org/docs/master/nn.html#loss-functions) ( something like `nn.CrossEntropyLoss`) and [the optimizer](http://pytorch.org/docs/master/optim.html) (typically `optim.SGD` or `optim.Adam`).\n\nThen write the training code. Remember the training pass is a fairly straightforward process:\n\n* Make a forward pass through the network to get the logits \n* Use the logits to calculate the loss\n* Perform a backward pass through the network with `loss.backward()` to calculate the gradients\n* Take a step with the optimizer to update the weights\n\nBy adjusting the hyperparameters (hidden units, learning rate, etc), you should be able to get the training loss below 0.4.", "_____no_output_____" ] ], [ [ "# TODO: Create the network, define the criterion and optimizer\nfrom torch import optim\ncriterion = nn.NLLLoss()\noptimizer = optim.SGD(model.parameters(),lr=0.007)", "_____no_output_____" ], [ "# TODO: Train the network here\ndata = iter(trainloader)\nepochs = 10\nfor e in range(epochs):\n running_loss = 0\n for images,labels in trainloader:\n input=images.view(images.shape[0],784)\n optimizer.zero_grad()\n output = model(input)\n loss = criterion(output,labels)\n loss.backward()\n optimizer.step()\n running_loss+=loss.item()\n else:\n print \"Epoch: \", e, \"Training loss: \", running_loss/len(trainloader)", "Epoch: 0 Training loss: 1.4089186108\nEpoch: 1 Training loss: 0.665717209008\nEpoch: 2 Training loss: 0.561825139182\nEpoch: 3 Training loss: 0.505677753738\nEpoch: 4 Training loss: 0.468876876572\nEpoch: 5 Training loss: 0.442242734071\nEpoch: 6 Training loss: 0.423429649252\nEpoch: 7 Training loss: 0.407437733559\nEpoch: 8 Training loss: 0.393918345239\nEpoch: 9 Training loss: 0.382098605384\n" ], [ "%matplotlib inline\n%config InlineBackend.figure_format = 'retina'\n\nimport helper\n\n# Test out your network!\n\ndataiter = iter(testloader)\nimages, labels = dataiter.next()\nimg = images[0]\n# Convert 2D image to 1D vector\nimg = img.resize_(1, 784)\n\n# TODO: Calculate the class probabilities (softmax) for img\nwith torch.no_grad():\n logprobs = model(img)\nps = torch.exp(logprobs)\n\n# Plot the image and probabilities\nhelper.view_classify(img.resize_(1, 28, 28), ps, version='Fashion')", "_____no_output_____" ] ] ]

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

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/reihaneh-torkzadehmahani/MyDPGAN/blob/master/AdvancedDPCGAN.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ], [ "## differential_privacy.analysis.rdp_accountant", "_____no_output_____" ] ], [ [ "# Copyright 2018 The TensorFlow Authors. 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# ==============================================================================\n\"\"\"RDP analysis of the Sampled Gaussian Mechanism.\n\nFunctionality for computing Renyi differential privacy (RDP) of an additive\nSampled Gaussian Mechanism (SGM). Its public interface consists of two methods:\n compute_rdp(q, noise_multiplier, T, orders) computes RDP for SGM iterated\n T times.\n get_privacy_spent(orders, rdp, target_eps, target_delta) computes delta\n (or eps) given RDP at multiple orders and\n a target value for eps (or delta).\n\nExample use:\n\nSuppose that we have run an SGM applied to a function with l2-sensitivity 1.\nIts parameters are given as a list of tuples (q1, sigma1, T1), ...,\n(qk, sigma_k, Tk), and we wish to compute eps for a given delta.\nThe example code would be:\n\n max_order = 32\n orders = range(2, max_order + 1)\n rdp = np.zeros_like(orders, dtype=float)\n for q, sigma, T in parameters:\n rdp += rdp_accountant.compute_rdp(q, sigma, T, orders)\n eps, _, opt_order = rdp_accountant.get_privacy_spent(rdp, target_delta=delta)\n\"\"\"\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport math\nimport sys\n\nimport numpy as np\nfrom scipy import special\nimport six\n\n########################\n# LOG-SPACE ARITHMETIC #\n########################\n\n\ndef _log_add(logx, logy):\n \"\"\"Add two numbers in the log space.\"\"\"\n a, b = min(logx, logy), max(logx, logy)\n if a == -np.inf: # adding 0\n return b\n # Use exp(a) + exp(b) = (exp(a - b) + 1) * exp(b)\n return math.log1p(math.exp(a - b)) + b # log1p(x) = log(x + 1)\n\n\ndef _log_sub(logx, logy):\n \"\"\"Subtract two numbers in the log space. Answer must be non-negative.\"\"\"\n if logx < logy:\n raise ValueError(\"The result of subtraction must be non-negative.\")\n if logy == -np.inf: # subtracting 0\n return logx\n if logx == logy:\n return -np.inf # 0 is represented as -np.inf in the log space.\n\n try:\n # Use exp(x) - exp(y) = (exp(x - y) - 1) * exp(y).\n return math.log(\n math.expm1(logx - logy)) + logy # expm1(x) = exp(x) - 1\n except OverflowError:\n return logx\n\n\ndef _log_print(logx):\n \"\"\"Pretty print.\"\"\"\n if logx < math.log(sys.float_info.max):\n return \"{}\".format(math.exp(logx))\n else:\n return \"exp({})\".format(logx)\n\n\ndef _compute_log_a_int(q, sigma, alpha):\n \"\"\"Compute log(A_alpha) for integer alpha. 0 < q < 1.\"\"\"\n assert isinstance(alpha, six.integer_types)\n\n # Initialize with 0 in the log space.\n log_a = -np.inf\n\n for i in range(alpha + 1):\n log_coef_i = (math.log(special.binom(alpha, i)) + i * math.log(q) +\n (alpha - i) * math.log(1 - q))\n\n s = log_coef_i + (i * i - i) / (2 * (sigma**2))\n log_a = _log_add(log_a, s)\n\n return float(log_a)\n\n\ndef _compute_log_a_frac(q, sigma, alpha):\n \"\"\"Compute log(A_alpha) for fractional alpha. 0 < q < 1.\"\"\"\n # The two parts of A_alpha, integrals over (-inf,z0] and [z0, +inf), are\n # initialized to 0 in the log space:\n log_a0, log_a1 = -np.inf, -np.inf\n i = 0\n\n z0 = sigma**2 * math.log(1 / q - 1) + .5\n\n while True: # do ... until loop\n coef = special.binom(alpha, i)\n log_coef = math.log(abs(coef))\n j = alpha - i\n\n log_t0 = log_coef + i * math.log(q) + j * math.log(1 - q)\n log_t1 = log_coef + j * math.log(q) + i * math.log(1 - q)\n\n log_e0 = math.log(.5) + _log_erfc((i - z0) / (math.sqrt(2) * sigma))\n log_e1 = math.log(.5) + _log_erfc((z0 - j) / (math.sqrt(2) * sigma))\n\n log_s0 = log_t0 + (i * i - i) / (2 * (sigma**2)) + log_e0\n log_s1 = log_t1 + (j * j - j) / (2 * (sigma**2)) + log_e1\n\n if coef > 0:\n log_a0 = _log_add(log_a0, log_s0)\n log_a1 = _log_add(log_a1, log_s1)\n else:\n log_a0 = _log_sub(log_a0, log_s0)\n log_a1 = _log_sub(log_a1, log_s1)\n\n i += 1\n if max(log_s0, log_s1) < -30:\n break\n\n return _log_add(log_a0, log_a1)\n\n\ndef _compute_log_a(q, sigma, alpha):\n \"\"\"Compute log(A_alpha) for any positive finite alpha.\"\"\"\n if float(alpha).is_integer():\n return _compute_log_a_int(q, sigma, int(alpha))\n else:\n return _compute_log_a_frac(q, sigma, alpha)\n\n\ndef _log_erfc(x):\n \"\"\"Compute log(erfc(x)) with high accuracy for large x.\"\"\"\n try:\n return math.log(2) + special.log_ndtr(-x * 2**.5)\n except NameError:\n # If log_ndtr is not available, approximate as follows:\n r = special.erfc(x)\n if r == 0.0:\n # Using the Laurent series at infinity for the tail of the erfc function:\n # erfc(x) ~ exp(-x^2-.5/x^2+.625/x^4)/(x*pi^.5)\n # To verify in Mathematica:\n # Series[Log[Erfc[x]] + Log[x] + Log[Pi]/2 + x^2, {x, Infinity, 6}]\n return (-math.log(math.pi) / 2 - math.log(x) - x**2 - .5 * x**-2 +\n .625 * x**-4 - 37. / 24. * x**-6 + 353. / 64. * x**-8)\n else:\n return math.log(r)\n\n\ndef _compute_delta(orders, rdp, eps):\n \"\"\"Compute delta given a list of RDP values and target epsilon.\n\n Args:\n orders: An array (or a scalar) of orders.\n rdp: A list (or a scalar) of RDP guarantees.\n eps: The target epsilon.\n\n Returns:\n Pair of (delta, optimal_order).\n\n Raises:\n ValueError: If input is malformed.\n\n \"\"\"\n orders_vec = np.atleast_1d(orders)\n rdp_vec = np.atleast_1d(rdp)\n\n if len(orders_vec) != len(rdp_vec):\n raise ValueError(\"Input lists must have the same length.\")\n\n deltas = np.exp((rdp_vec - eps) * (orders_vec - 1))\n idx_opt = np.argmin(deltas)\n return min(deltas[idx_opt], 1.), orders_vec[idx_opt]\n\n\ndef _compute_eps(orders, rdp, delta):\n \"\"\"Compute epsilon given a list of RDP values and target delta.\n\n Args:\n orders: An array (or a scalar) of orders.\n rdp: A list (or a scalar) of RDP guarantees.\n delta: The target delta.\n\n Returns:\n Pair of (eps, optimal_order).\n\n Raises:\n ValueError: If input is malformed.\n\n \"\"\"\n orders_vec = np.atleast_1d(orders)\n rdp_vec = np.atleast_1d(rdp)\n\n if len(orders_vec) != len(rdp_vec):\n raise ValueError(\"Input lists must have the same length.\")\n\n eps = rdp_vec - math.log(delta) / (orders_vec - 1)\n\n idx_opt = np.nanargmin(eps) # Ignore NaNs\n return eps[idx_opt], orders_vec[idx_opt]\n\n\ndef _compute_rdp(q, sigma, alpha):\n \"\"\"Compute RDP of the Sampled Gaussian mechanism at order alpha.\n\n Args:\n q: The sampling rate.\n sigma: The std of the additive Gaussian noise.\n alpha: The order at which RDP is computed.\n\n Returns:\n RDP at alpha, can be np.inf.\n \"\"\"\n if q == 0:\n return 0\n\n if q == 1.:\n return alpha / (2 * sigma**2)\n\n if np.isinf(alpha):\n return np.inf\n\n return _compute_log_a(q, sigma, alpha) / (alpha - 1)\n\n\ndef compute_rdp(q, noise_multiplier, steps, orders):\n \"\"\"Compute RDP of the Sampled Gaussian Mechanism.\n\n Args:\n q: The sampling rate.\n noise_multiplier: The ratio of the standard deviation of the Gaussian noise\n to the l2-sensitivity of the function to which it is added.\n steps: The number of steps.\n orders: An array (or a scalar) of RDP orders.\n\n Returns:\n The RDPs at all orders, can be np.inf.\n \"\"\"\n if np.isscalar(orders):\n rdp = _compute_rdp(q, noise_multiplier, orders)\n else:\n rdp = np.array(\n [_compute_rdp(q, noise_multiplier, order) for order in orders])\n\n return rdp * steps\n\n\ndef get_privacy_spent(orders, rdp, target_eps=None, target_delta=None):\n \"\"\"Compute delta (or eps) for given eps (or delta) from RDP values.\n\n Args:\n orders: An array (or a scalar) of RDP orders.\n rdp: An array of RDP values. Must be of the same length as the orders list.\n target_eps: If not None, the epsilon for which we compute the corresponding\n delta.\n target_delta: If not None, the delta for which we compute the corresponding\n epsilon. Exactly one of target_eps and target_delta must be None.\n\n Returns:\n eps, delta, opt_order.\n\n Raises:\n ValueError: If target_eps and target_delta are messed up.\n \"\"\"\n if target_eps is None and target_delta is None:\n raise ValueError(\n \"Exactly one out of eps and delta must be None. (Both are).\")\n\n if target_eps is not None and target_delta is not None:\n raise ValueError(\n \"Exactly one out of eps and delta must be None. (None is).\")\n\n if target_eps is not None:\n delta, opt_order = _compute_delta(orders, rdp, target_eps)\n return target_eps, delta, opt_order\n else:\n eps, opt_order = _compute_eps(orders, rdp, target_delta)\n return eps, target_delta, opt_order\n", "_____no_output_____" ] ], [ [ "## dp query\n\n", "_____no_output_____" ] ], [ [ "# Copyright 2018, The TensorFlow Authors.\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\n\"\"\"An interface for differentially private query mechanisms.\n\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport abc\n\n\nclass DPQuery(object):\n \"\"\"Interface for differentially private query mechanisms.\"\"\"\n\n __metaclass__ = abc.ABCMeta\n\n @abc.abstractmethod\n def initial_global_state(self):\n \"\"\"Returns the initial global state for the DPQuery.\"\"\"\n pass\n\n @abc.abstractmethod\n def derive_sample_params(self, global_state):\n \"\"\"Given the global state, derives parameters to use for the next sample.\n\n Args:\n global_state: The current global state.\n\n Returns:\n Parameters to use to process records in the next sample.\n \"\"\"\n pass\n\n @abc.abstractmethod\n def initial_sample_state(self, global_state, tensors):\n \"\"\"Returns an initial state to use for the next sample.\n\n Args:\n global_state: The current global state.\n tensors: A structure of tensors used as a template to create the initial\n sample state.\n\n Returns: An initial sample state.\n \"\"\"\n pass\n\n @abc.abstractmethod\n def accumulate_record(self, params, sample_state, record):\n \"\"\"Accumulates a single record into the sample state.\n\n Args:\n params: The parameters for the sample.\n sample_state: The current sample state.\n record: The record to accumulate.\n\n Returns:\n The updated sample state.\n \"\"\"\n pass\n\n @abc.abstractmethod\n def get_noised_result(self, sample_state, global_state):\n \"\"\"Gets query result after all records of sample have been accumulated.\n\n Args:\n sample_state: The sample state after all records have been accumulated.\n global_state: The global state.\n\n Returns:\n A tuple (result, new_global_state) where \"result\" is the result of the\n query and \"new_global_state\" is the updated global state.\n \"\"\"\n pass\n", "_____no_output_____" ] ], [ [ "## gausian query", "_____no_output_____" ] ], [ [ "# Copyright 2018, The TensorFlow Authors.\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\n\"\"\"Implements DPQuery interface for Gaussian average queries.\n\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport collections\n\nimport tensorflow as tf\n\nnest = tf.contrib.framework.nest\n\n\nclass GaussianSumQuery(DPQuery):\n \"\"\"Implements DPQuery interface for Gaussian sum queries.\n\n Accumulates clipped vectors, then adds Gaussian noise to the sum.\n \"\"\"\n\n # pylint: disable=invalid-name\n _GlobalState = collections.namedtuple(\n '_GlobalState', ['l2_norm_clip', 'stddev'])\n\n def __init__(self, l2_norm_clip, stddev):\n \"\"\"Initializes the GaussianSumQuery.\n\n Args:\n l2_norm_clip: The clipping norm to apply to the global norm of each\n record.\n stddev: The stddev of the noise added to the sum.\n \"\"\"\n self._l2_norm_clip = l2_norm_clip\n self._stddev = stddev\n\n def initial_global_state(self):\n \"\"\"Returns the initial global state for the GaussianSumQuery.\"\"\"\n return self._GlobalState(float(self._l2_norm_clip), float(self._stddev))\n\n def derive_sample_params(self, global_state):\n \"\"\"Given the global state, derives parameters to use for the next sample.\n\n Args:\n global_state: The current global state.\n\n Returns:\n Parameters to use to process records in the next sample.\n \"\"\"\n return global_state.l2_norm_clip\n\n def initial_sample_state(self, global_state, tensors):\n \"\"\"Returns an initial state to use for the next sample.\n\n Args:\n global_state: The current global state.\n tensors: A structure of tensors used as a template to create the initial\n sample state.\n\n Returns: An initial sample state.\n \"\"\"\n del global_state # unused.\n return nest.map_structure(tf.zeros_like, tensors)\n\n def accumulate_record(self, params, sample_state, record):\n \"\"\"Accumulates a single record into the sample state.\n\n Args:\n params: The parameters for the sample.\n sample_state: The current sample state.\n record: The record to accumulate.\n\n Returns:\n The updated sample state.\n \"\"\"\n l2_norm_clip = params\n record_as_list = nest.flatten(record)\n clipped_as_list, _ = tf.clip_by_global_norm(record_as_list, l2_norm_clip)\n clipped = nest.pack_sequence_as(record, clipped_as_list)\n\n return nest.map_structure(tf.add, sample_state, clipped)\n\n def get_noised_result(self, sample_state, global_state, add_noise=True):\n \"\"\"Gets noised sum after all records of sample have been accumulated.\n\n Args:\n sample_state: The sample state after all records have been accumulated.\n global_state: The global state.\n\n Returns:\n A tuple (estimate, new_global_state) where \"estimate\" is the estimated\n sum of the records and \"new_global_state\" is the updated global state.\n \"\"\"\n def add_noise(v):\n if add_noise:\n return v + tf.random_normal(tf.shape(v), stddev=global_state.stddev)\n else:\n return v\n\n\n return nest.map_structure(add_noise, sample_state), global_state\n\n\nclass GaussianAverageQuery(DPQuery):\n \"\"\"Implements DPQuery interface for Gaussian average queries.\n\n Accumulates clipped vectors, adds Gaussian noise, and normalizes.\n\n Note that we use \"fixed-denominator\" estimation: the denominator should be\n specified as the expected number of records per sample. Accumulating the\n denominator separately would also be possible but would be produce a higher\n variance estimator.\n \"\"\"\n\n # pylint: disable=invalid-name\n _GlobalState = collections.namedtuple(\n '_GlobalState', ['sum_state', 'denominator'])\n\n def __init__(self, l2_norm_clip, sum_stddev, denominator):\n \"\"\"Initializes the GaussianAverageQuery.\n\n Args:\n l2_norm_clip: The clipping norm to apply to the global norm of each\n record.\n sum_stddev: The stddev of the noise added to the sum (before\n normalization).\n denominator: The normalization constant (applied after noise is added to\n the sum).\n \"\"\"\n self._numerator = GaussianSumQuery(l2_norm_clip, sum_stddev)\n self._denominator = denominator\n\n def initial_global_state(self):\n \"\"\"Returns the initial global state for the GaussianAverageQuery.\"\"\"\n sum_global_state = self._numerator.initial_global_state()\n return self._GlobalState(sum_global_state, float(self._denominator))\n\n def derive_sample_params(self, global_state):\n \"\"\"Given the global state, derives parameters to use for the next sample.\n\n Args:\n global_state: The current global state.\n\n Returns:\n Parameters to use to process records in the next sample.\n \"\"\"\n return self._numerator.derive_sample_params(global_state.sum_state)\n\n def initial_sample_state(self, global_state, tensors):\n \"\"\"Returns an initial state to use for the next sample.\n\n Args:\n global_state: The current global state.\n tensors: A structure of tensors used as a template to create the initial\n sample state.\n\n Returns: An initial sample state.\n \"\"\"\n # GaussianAverageQuery has no state beyond the sum state.\n return self._numerator.initial_sample_state(global_state.sum_state, tensors)\n\n def accumulate_record(self, params, sample_state, record):\n \"\"\"Accumulates a single record into the sample state.\n\n Args:\n params: The parameters for the sample.\n sample_state: The current sample state.\n record: The record to accumulate.\n\n Returns:\n The updated sample state.\n \"\"\"\n \n return self._numerator.accumulate_record(params, sample_state, record)\n\n def get_noised_result(self, sample_state, global_state, add_noise=True):\n \"\"\"Gets noised average after all records of sample have been accumulated.\n\n Args:\n sample_state: The sample state after all records have been accumulated.\n global_state: The global state.\n\n Returns:\n A tuple (estimate, new_global_state) where \"estimate\" is the estimated\n average of the records and \"new_global_state\" is the updated global state.\n \"\"\"\n noised_sum, new_sum_global_state = self._numerator.get_noised_result(\n sample_state, global_state.sum_state, add_noise)\n new_global_state = self._GlobalState(\n new_sum_global_state, global_state.denominator)\n def normalize(v):\n return tf.truediv(v, global_state.denominator)\n\n return nest.map_structure(normalize, noised_sum), new_global_state\n", "\nWARNING: The TensorFlow contrib module will not be included in TensorFlow 2.0.\nFor more information, please see:\n * https://github.com/tensorflow/community/blob/master/rfcs/20180907-contrib-sunset.md\n * https://github.com/tensorflow/addons\nIf you depend on functionality not listed there, please file an issue.\n\n" ] ], [ [ "## our_dp_optimizer", "_____no_output_____" ] ], [ [ "# Copyright 2018, The TensorFlow Authors.\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\"\"\"Differentially private optimizers for TensorFlow.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\n\ndef make_optimizer_class(cls):\n \"\"\"Constructs a DP optimizer class from an existing one.\"\"\"\n if (tf.train.Optimizer.compute_gradients.__code__ is\n not cls.compute_gradients.__code__):\n tf.logging.warning(\n 'WARNING: Calling make_optimizer_class() on class %s that overrides '\n 'method compute_gradients(). Check to ensure that '\n 'make_optimizer_class() does not interfere with overridden version.',\n cls.__name__)\n\n class DPOptimizerClass(cls):\n \"\"\"Differentially private subclass of given class cls.\"\"\"\n\n def __init__(\n self,\n l2_norm_clip,\n noise_multiplier,\n dp_average_query,\n num_microbatches,\n unroll_microbatches=False,\n *args, # pylint: disable=keyword-arg-before-vararg\n **kwargs):\n super(DPOptimizerClass, self).__init__(*args, **kwargs)\n self._dp_average_query = dp_average_query\n self._num_microbatches = num_microbatches\n self._global_state = self._dp_average_query.initial_global_state()\n\n # TODO(b/122613513): Set unroll_microbatches=True to avoid this bug.\n # Beware: When num_microbatches is large (>100), enabling this parameter\n # may cause an OOM error.\n self._unroll_microbatches = unroll_microbatches\n\n def dp_compute_gradients(self,\n loss,\n var_list,\n gate_gradients=tf.train.Optimizer.GATE_OP,\n aggregation_method=None,\n colocate_gradients_with_ops=False,\n grad_loss=None,\n add_noise=True):\n\n # Note: it would be closer to the correct i.i.d. sampling of records if\n # we sampled each microbatch from the appropriate binomial distribution,\n # although that still wouldn't be quite correct because it would be\n # sampling from the dataset without replacement.\n microbatches_losses = tf.reshape(loss,\n [self._num_microbatches, -1])\n sample_params = (self._dp_average_query.derive_sample_params(\n self._global_state))\n\n def process_microbatch(i, sample_state):\n \"\"\"Process one microbatch (record) with privacy helper.\"\"\"\n grads, _ = zip(*super(cls, self).compute_gradients(\n tf.gather(microbatches_losses, [i]), var_list,\n gate_gradients, aggregation_method,\n colocate_gradients_with_ops, grad_loss))\n\n # Converts tensor to list to replace None gradients with zero\n grads1 = list(grads)\n\n for inx in range(0, len(grads)):\n if (grads[inx] == None):\n grads1[inx] = tf.zeros_like(var_list[inx])\n\n grads_list = grads1\n sample_state = self._dp_average_query.accumulate_record(\n sample_params, sample_state, grads_list)\n\n return sample_state\n\n if var_list is None:\n var_list = (tf.trainable_variables() + tf.get_collection(\n tf.GraphKeys.TRAINABLE_RESOURCE_VARIABLES))\n sample_state = self._dp_average_query.initial_sample_state(\n self._global_state, var_list)\n\n if self._unroll_microbatches:\n for idx in range(self._num_microbatches):\n sample_state = process_microbatch(idx, sample_state)\n else:\n # Use of while_loop here requires that sample_state be a nested\n # structure of tensors. In general, we would prefer to allow it to be\n # an arbitrary opaque type.\n\n cond_fn = lambda i, _: tf.less(i, self._num_microbatches)\n body_fn = lambda i, state: [\n tf.add(i, 1), process_microbatch(i, state)\n ]\n\n idx = tf.constant(0)\n _, sample_state = tf.while_loop(cond_fn, body_fn,\n [idx, sample_state])\n\n final_grads, self._global_state = (\n self._dp_average_query.get_noised_result(\n sample_state, self._global_state, add_noise))\n\n return (final_grads)\n\n def minimize(self,\n d_loss_real,\n d_loss_fake,\n global_step=None,\n var_list=None,\n gate_gradients=tf.train.Optimizer.GATE_OP,\n aggregation_method=None,\n colocate_gradients_with_ops=False,\n name=None,\n grad_loss=None):\n \"\"\"Minimize using sanitized gradients\n\n Args:\n d_loss_real: the loss tensor for real data\n d_loss_fake: the loss tensor for fake data\n global_step: the optional global step.\n var_list: the optional variables.\n name: the optional name.\n Returns:\n the operation that runs one step of DP gradient descent.\n \"\"\"\n\n # First validate the var_list\n\n if var_list is None:\n var_list = tf.trainable_variables()\n for var in var_list:\n if not isinstance(var, tf.Variable):\n raise TypeError(\"Argument is not a variable.Variable: %s\" %\n var)\n\n # ------------------ OUR METHOD --------------------------------\n\n r_grads = self.dp_compute_gradients(\n d_loss_real,\n var_list=var_list,\n gate_gradients=gate_gradients,\n aggregation_method=aggregation_method,\n colocate_gradients_with_ops=colocate_gradients_with_ops,\n grad_loss=grad_loss, add_noise = True)\n\n f_grads = self.dp_compute_gradients(\n d_loss_fake,\n var_list=var_list,\n gate_gradients=gate_gradients,\n aggregation_method=aggregation_method,\n colocate_gradients_with_ops=colocate_gradients_with_ops,\n grad_loss=grad_loss,\n add_noise=False)\n\n # Compute the overall gradients \n s_grads = [(r_grads[idx] + f_grads[idx])\n for idx in range(len(r_grads))]\n\n sanitized_grads_and_vars = list(zip(s_grads, var_list))\n self._assert_valid_dtypes(\n [v for g, v in sanitized_grads_and_vars if g is not None])\n\n # Apply the overall gradients\n apply_grads = self.apply_gradients(sanitized_grads_and_vars,\n global_step=global_step,\n name=name)\n\n return apply_grads\n\n # -----------------------------------------------------------------\n\n return DPOptimizerClass\n\n\ndef make_gaussian_optimizer_class(cls):\n \"\"\"Constructs a DP optimizer with Gaussian averaging of updates.\"\"\"\n\n class DPGaussianOptimizerClass(make_optimizer_class(cls)):\n \"\"\"DP subclass of given class cls using Gaussian averaging.\"\"\"\n\n def __init__(\n self,\n l2_norm_clip,\n noise_multiplier,\n num_microbatches,\n unroll_microbatches=False,\n *args, # pylint: disable=keyword-arg-before-vararg\n **kwargs):\n dp_average_query = GaussianAverageQuery(\n l2_norm_clip, l2_norm_clip * noise_multiplier,\n num_microbatches)\n self.l2_norm_clip = l2_norm_clip\n self.noise_multiplier = noise_multiplier\n\n super(DPGaussianOptimizerClass,\n self).__init__(l2_norm_clip, noise_multiplier,\n dp_average_query, num_microbatches,\n unroll_microbatches, *args, **kwargs)\n\n return DPGaussianOptimizerClass\n\n\nDPAdagradOptimizer = make_optimizer_class(tf.train.AdagradOptimizer)\nDPAdamOptimizer = make_optimizer_class(tf.train.AdamOptimizer)\nDPGradientDescentOptimizer = make_optimizer_class(\n tf.train.GradientDescentOptimizer)\n\nDPAdagradGaussianOptimizer = make_gaussian_optimizer_class(\n tf.train.AdagradOptimizer)\nDPAdamGaussianOptimizer = make_gaussian_optimizer_class(tf.train.AdamOptimizer)\nDPGradientDescentGaussianOptimizer = make_gaussian_optimizer_class(\n tf.train.GradientDescentOptimizer)\n", "_____no_output_____" ] ], [ [ "## gan.ops", "_____no_output_____" ] ], [ [ "\"\"\"\nMost codes from https://github.com/carpedm20/DCGAN-tensorflow\n\"\"\"\nimport math\nimport numpy as np\nimport tensorflow as tf\n\n\nif \"concat_v2\" in dir(tf):\n\n def concat(tensors, axis, *args, **kwargs):\n return tf.concat_v2(tensors, axis, *args, **kwargs)\nelse:\n\n def concat(tensors, axis, *args, **kwargs):\n return tf.concat(tensors, axis, *args, **kwargs)\n\n\ndef bn(x, is_training, scope):\n return tf.contrib.layers.batch_norm(x,\n decay=0.9,\n updates_collections=None,\n epsilon=1e-5,\n scale=True,\n is_training=is_training,\n scope=scope)\n\n\ndef conv_out_size_same(size, stride):\n return int(math.ceil(float(size) / float(stride)))\n\n\ndef conv_cond_concat(x, y):\n \"\"\"Concatenate conditioning vector on feature map axis.\"\"\"\n x_shapes = x.get_shape()\n y_shapes = y.get_shape()\n return concat(\n [x, y * tf.ones([x_shapes[0], x_shapes[1], x_shapes[2], y_shapes[3]])],\n 3)\n\n\ndef conv2d(input_,\n output_dim,\n k_h=5,\n k_w=5,\n d_h=2,\n d_w=2,\n stddev=0.02,\n name=\"conv2d\"):\n with tf.variable_scope(name):\n w = tf.get_variable(\n 'w', [k_h, k_w, input_.get_shape()[-1], output_dim],\n initializer=tf.truncated_normal_initializer(stddev=stddev))\n conv = tf.nn.conv2d(input_,\n w,\n strides=[1, d_h, d_w, 1],\n padding='SAME')\n\n biases = tf.get_variable('biases', [output_dim],\n initializer=tf.constant_initializer(0.0))\n conv = tf.reshape(tf.nn.bias_add(conv, biases), conv.get_shape())\n\n return conv\n\n\ndef deconv2d(input_,\n output_shape,\n k_h=5,\n k_w=5,\n d_h=2,\n d_w=2,\n name=\"deconv2d\",\n stddev=0.02,\n with_w=False):\n with tf.variable_scope(name):\n # filter : [height, width, output_channels, in_channels]\n w = tf.get_variable(\n 'w', [k_h, k_w, output_shape[-1],\n input_.get_shape()[-1]],\n initializer=tf.random_normal_initializer(stddev=stddev))\n\n try:\n deconv = tf.nn.conv2d_transpose(input_,\n w,\n output_shape=output_shape,\n strides=[1, d_h, d_w, 1])\n\n # Support for verisons of TensorFlow before 0.7.0\n except AttributeError:\n deconv = tf.nn.deconv2d(input_,\n w,\n output_shape=output_shape,\n strides=[1, d_h, d_w, 1])\n\n biases = tf.get_variable('biases', [output_shape[-1]],\n initializer=tf.constant_initializer(0.0))\n deconv = tf.reshape(tf.nn.bias_add(deconv, biases), deconv.get_shape())\n\n if with_w:\n return deconv, w, biases\n else:\n return deconv\n\n\ndef lrelu(x, leak=0.2, name=\"lrelu\"):\n return tf.maximum(x, leak * x)\n\n\ndef linear(input_,\n output_size,\n scope=None,\n stddev=0.02,\n bias_start=0.0,\n with_w=False):\n shape = input_.get_shape().as_list()\n\n with tf.variable_scope(scope or \"Linear\"):\n matrix = tf.get_variable(\"Matrix\", [shape[1], output_size], tf.float32,\n tf.random_normal_initializer(stddev=stddev))\n bias = tf.get_variable(\"bias\", [output_size],\n initializer=tf.constant_initializer(bias_start))\n if with_w:\n return tf.matmul(input_, matrix) + bias, matrix, bias\n else:\n return tf.matmul(input_, matrix) + bias\n", "_____no_output_____" ] ], [ [ "## OUR DP CGAN", "_____no_output_____" ] ], [ [ "# -*- coding: utf-8 -*-\nfrom __future__ import division\nfrom keras.datasets import cifar10\n\nfrom mlxtend.data import loadlocal_mnist\nfrom sklearn.preprocessing import label_binarize\n\nfrom sklearn.multiclass import OneVsRestClassifier\nfrom sklearn.metrics import roc_curve, auc\nfrom sklearn.linear_model import LogisticRegression\nfrom sklearn.ensemble import RandomForestClassifier\nfrom sklearn.neural_network import MLPClassifier\n\n\nclass OUR_DP_CGAN(object):\n model_name = \"OUR_DP_CGAN\" # name for checkpoint\n\n def __init__(self, sess, epoch, batch_size, z_dim, epsilon, delta, sigma,\n clip_value, lr, dataset_name, base_dir, checkpoint_dir,\n result_dir, log_dir):\n self.sess = sess\n self.dataset_name = dataset_name\n self.base_dir = base_dir\n self.checkpoint_dir = checkpoint_dir\n self.result_dir = result_dir\n self.log_dir = log_dir\n self.epoch = epoch\n self.batch_size = batch_size\n self.epsilon = epsilon\n self.delta = delta\n self.noise_multiplier = sigma\n self.l2_norm_clip = clip_value\n self.lr = lr\n\n if dataset_name == 'mnist' or dataset_name == 'fashion-mnist':\n # parameters\n self.input_height = 28\n self.input_width = 28\n self.output_height = 28\n self.output_width = 28\n\n self.z_dim = z_dim # dimension of noise-vector\n self.y_dim = 10 # dimension of condition-vector (label)\n self.c_dim = 1\n\n # train\n self.learningRateD = self.lr\n self.learningRateG = self.learningRateD * 5\n self.beta1 = 0.5\n self.beta2 = 0.99\n # test\n self.sample_num = 64 # number of generated images to be saved\n\n # load mnist\n self.data_X, self.data_y = load_mnist(train = True)\n\n # get number of batches for a single epoch\n self.num_batches = len(self.data_X) // self.batch_size\n\n elif dataset_name == 'cifar10':\n # parameters\n self.input_height = 32\n self.input_width = 32\n self.output_height = 32\n self.output_width = 32\n\n self.z_dim = 100 # dimension of noise-vector\n self.y_dim = 10 # dimension of condition-vector (label)\n self.c_dim = 3 # color dimension\n\n # train\n # self.learning_rate = 0.0002 # 1e-3, 1e-4\n self.learningRateD = 1e-3\n self.learningRateG = 1e-4\n self.beta1 = 0.5\n self.beta2 = 0.99\n\n # test\n self.sample_num = 64 # number of generated images to be saved\n\n # load cifar10\n self.data_X, self.data_y = load_cifar10(train=True)\n\n self.num_batches = len(self.data_X) // self.batch_size\n\n else:\n raise NotImplementedError\n\n def discriminator(self, x, y, is_training=True, reuse=False):\n # Network Architecture is exactly same as in infoGAN (https://arxiv.org/abs/1606.03657)\n # Architecture : (64)4c2s-(128)4c2s_BL-FC1024_BL-FC1_S\n with tf.variable_scope(\"discriminator\", reuse=reuse):\n\n # merge image and label\n if (self.dataset_name == \"mnist\"):\n y = tf.reshape(y, [self.batch_size, 1, 1, self.y_dim])\n x = conv_cond_concat(x, y)\n\n net = lrelu(conv2d(x, 64, 4, 4, 2, 2, name='d_conv1'))\n net = lrelu(\n bn(conv2d(net, 128, 4, 4, 2, 2, name='d_conv2'),\n is_training=is_training,\n scope='d_bn2'))\n net = tf.reshape(net, [self.batch_size, -1])\n net = lrelu(\n bn(linear(net, 1024, scope='d_fc3'),\n is_training=is_training,\n scope='d_bn3'))\n out_logit = linear(net, 1, scope='d_fc4')\n out = tf.nn.sigmoid(out_logit)\n\n elif (self.dataset_name == \"cifar10\"):\n\n y = tf.reshape(y, [self.batch_size, 1, 1, self.y_dim])\n x = conv_cond_concat(x, y)\n lrelu_slope = 0.2\n kernel_size = 5\n w_init = tf.contrib.layers.xavier_initializer()\n\n net = lrelu(\n conv2d(x,\n 64,\n 5,\n 5,\n 2,\n 2,\n name='d_conv1' + '_' + self.dataset_name))\n \n net = lrelu(\n bn(conv2d(net,\n 128,\n 5,\n 5,\n 2,\n 2,\n name='d_conv2' + '_' + self.dataset_name),\n is_training=is_training,\n scope='d_bn2'))\n \n net = lrelu(\n bn(conv2d(net,\n 256,\n 5,\n 5,\n 2,\n 2,\n name='d_conv3' + '_' + self.dataset_name),\n is_training=is_training,\n scope='d_bn3'))\n \n net = lrelu(\n bn(conv2d(net,\n 512,\n 5,\n 5,\n 2,\n 2,\n name='d_conv4' + '_' + self.dataset_name),\n is_training=is_training,\n scope='d_bn4'))\n \n net = tf.reshape(net, [self.batch_size, -1])\n \n out_logit = linear(net,\n 1,\n scope='d_fc5' + '_' + self.dataset_name)\n \n out = tf.nn.sigmoid(out_logit)\n\n return out, out_logit\n\n def generator(self, z, y, is_training=True, reuse=False):\n # Network Architecture is exactly same as in infoGAN (https://arxiv.org/abs/1606.03657)\n # Architecture : FC1024_BR-FC7x7x128_BR-(64)4dc2s_BR-(1)4dc2s_S\n with tf.variable_scope(\"generator\", reuse=reuse):\n\n if (self.dataset_name == \"mnist\"):\n # merge noise and label\n z = concat([z, y], 1)\n\n net = tf.nn.relu(\n bn(linear(z, 1024, scope='g_fc1'),\n is_training=is_training,\n scope='g_bn1'))\n net = tf.nn.relu(\n bn(linear(net, 128 * 7 * 7, scope='g_fc2'),\n is_training=is_training,\n scope='g_bn2'))\n net = tf.reshape(net, [self.batch_size, 7, 7, 128])\n net = tf.nn.relu(\n bn(deconv2d(net, [self.batch_size, 14, 14, 64],\n 4,\n 4,\n 2,\n 2,\n name='g_dc3'),\n is_training=is_training,\n scope='g_bn3'))\n\n out = tf.nn.sigmoid(\n deconv2d(net, [self.batch_size, 28, 28, 1],\n 4,\n 4,\n 2,\n 2,\n name='g_dc4'))\n\n elif (self.dataset_name == \"cifar10\"):\n h_size = 32\n h_size_2 = 16\n h_size_4 = 8\n h_size_8 = 4\n h_size_16 = 2\n\n z = concat([z, y], 1)\n\n net = linear(z,\n 512 * h_size_16 * h_size_16,\n scope='g_fc1' + '_' + self.dataset_name)\n \n net = tf.nn.relu(\n bn(tf.reshape(\n net, [self.batch_size, h_size_16, h_size_16, 512]),\n is_training=is_training,\n scope='g_bn1'))\n \n net = tf.nn.relu(\n bn(deconv2d(net,\n [self.batch_size, h_size_8, h_size_8, 256],\n 5,\n 5,\n 2,\n 2,\n name='g_dc2' + '_' + self.dataset_name),\n is_training=is_training,\n scope='g_bn2'))\n \n net = tf.nn.relu(\n bn(deconv2d(net,\n [self.batch_size, h_size_4, h_size_4, 128],\n 5,\n 5,\n 2,\n 2,\n name='g_dc3' + '_' + self.dataset_name),\n is_training=is_training,\n scope='g_bn3'))\n \n net = tf.nn.relu(\n bn(deconv2d(net, [self.batch_size, h_size_2, h_size_2, 64],\n 5,\n 5,\n 2,\n 2,\n name='g_dc4' + '_' + self.dataset_name),\n is_training=is_training,\n scope='g_bn4'))\n \n out = tf.nn.tanh(\n deconv2d(net, [\n self.batch_size, self.output_height, self.output_width,\n self.c_dim\n ],\n 5,\n 5,\n 2,\n 2,\n name='g_dc5' + '_' + self.dataset_name))\n\n return out\n\n def build_model(self):\n # some parameters\n image_dims = [self.input_height, self.input_width, self.c_dim]\n bs = self.batch_size\n \n \"\"\" Graph Input \"\"\"\n # images\n self.inputs = tf.placeholder(tf.float32, [bs] + image_dims,\n name='real_images')\n\n # labels\n self.y = tf.placeholder(tf.float32, [bs, self.y_dim], name='y')\n\n # noises\n self.z = tf.placeholder(tf.float32, [bs, self.z_dim], name='z')\n \"\"\" Loss Function \"\"\"\n\n # output of D for real images\n D_real, D_real_logits = self.discriminator(self.inputs,\n self.y,\n is_training=True,\n reuse=False)\n\n # output of D for fake images\n G = self.generator(self.z, self.y, is_training=True, reuse=False)\n D_fake, D_fake_logits = self.discriminator(G,\n self.y,\n is_training=True,\n reuse=True)\n\n # get loss for discriminator\n d_loss_real = tf.reduce_mean(\n tf.nn.sigmoid_cross_entropy_with_logits(\n logits=D_real_logits, labels=tf.ones_like(D_real)))\n d_loss_fake = tf.reduce_mean(\n tf.nn.sigmoid_cross_entropy_with_logits(\n logits=D_fake_logits, labels=tf.zeros_like(D_fake)))\n\n self.d_loss_real_vec = tf.nn.sigmoid_cross_entropy_with_logits(\n logits=D_real_logits, labels=tf.ones_like(D_real))\n self.d_loss_fake_vec = tf.nn.sigmoid_cross_entropy_with_logits(\n logits=D_fake_logits, labels=tf.zeros_like(D_fake))\n\n self.d_loss = d_loss_real + d_loss_fake\n\n # get loss for generator\n self.g_loss = tf.reduce_mean(\n tf.nn.sigmoid_cross_entropy_with_logits(\n logits=D_fake_logits, labels=tf.ones_like(D_fake)))\n \n \"\"\" Training \"\"\"\n # divide trainable variables into a group for D and a group for G\n t_vars = tf.trainable_variables()\n d_vars = [\n var for var in t_vars if var.name.startswith('discriminator')\n ]\n g_vars = [var for var in t_vars if var.name.startswith('generator')]\n\n # optimizers\n with tf.control_dependencies(tf.get_collection(\n tf.GraphKeys.UPDATE_OPS)):\n\n d_optim_init = DPGradientDescentGaussianOptimizer(\n l2_norm_clip=self.l2_norm_clip,\n noise_multiplier=self.noise_multiplier,\n num_microbatches=self.batch_size,\n learning_rate=self.learningRateD)\n\n global_step = tf.train.get_global_step()\n\n self.d_optim = d_optim_init.minimize(\n d_loss_real=self.d_loss_real_vec,\n d_loss_fake=self.d_loss_fake_vec,\n global_step=global_step,\n var_list=d_vars)\n\n optimizer = DPGradientDescentGaussianOptimizer(\n l2_norm_clip=self.l2_norm_clip,\n noise_multiplier=self.noise_multiplier,\n num_microbatches=self.batch_size,\n learning_rate=self.learningRateD)\n\n self.g_optim = tf.train.GradientDescentOptimizer(self.learningRateG) \\\n .minimize(self.g_loss, var_list=g_vars)\n \n \"\"\"\" Testing \"\"\"\n self.fake_images = self.generator(self.z,\n self.y,\n is_training=False,\n reuse=True)\n \"\"\" Summary \"\"\"\n d_loss_real_sum = tf.summary.scalar(\"d_loss_real\", d_loss_real)\n d_loss_fake_sum = tf.summary.scalar(\"d_loss_fake\", d_loss_fake)\n d_loss_sum = tf.summary.scalar(\"d_loss\", self.d_loss)\n g_loss_sum = tf.summary.scalar(\"g_loss\", self.g_loss)\n\n # final summary operations\n self.g_sum = tf.summary.merge([d_loss_fake_sum, g_loss_sum])\n self.d_sum = tf.summary.merge([d_loss_real_sum, d_loss_sum])\n\n def train(self):\n\n # initialize all variables\n tf.global_variables_initializer().run()\n\n # graph inputs for visualize training results\n self.sample_z = np.random.uniform(-1,\n 1,\n size=(self.batch_size, self.z_dim))\n self.test_labels = self.data_y[0:self.batch_size]\n\n # saver to save model\n self.saver = tf.train.Saver()\n\n # summary writer\n self.writer = tf.summary.FileWriter(\n self.log_dir + '/' + self.model_name, self.sess.graph)\n\n # restore check-point if it exits\n could_load, checkpoint_counter = self.load(self.checkpoint_dir)\n if could_load:\n start_epoch = (int)(checkpoint_counter / self.num_batches)\n start_batch_id = checkpoint_counter - start_epoch * self.num_batches\n counter = checkpoint_counter\n print(\" [*] Load SUCCESS\")\n else:\n start_epoch = 0\n start_batch_id = 0\n counter = 1\n print(\" [!] Load failed...\")\n\n # loop for epoch\n epoch = start_epoch\n should_terminate = False\n \n while (epoch < self.epoch and not should_terminate):\n\n # get batch data\n for idx in range(start_batch_id, self.num_batches):\n batch_images = self.data_X[idx * self.batch_size:(idx + 1) *\n self.batch_size]\n batch_labels = self.data_y[idx * self.batch_size:(idx + 1) *\n self.batch_size]\n batch_z = np.random.uniform(\n -1, 1, [self.batch_size, self.z_dim]).astype(np.float32)\n\n # update D network\n\n _, summary_str, d_loss = self.sess.run(\n [self.d_optim, self.d_sum, self.d_loss],\n feed_dict={\n self.inputs: batch_images,\n self.y: batch_labels,\n self.z: batch_z\n })\n self.writer.add_summary(summary_str, counter)\n\n eps = self.compute_epsilon((epoch * self.num_batches) + idx)\n\n if (eps > self.epsilon):\n\n should_terminate = True\n print(\"TERMINATE !! Run out of Privacy Budget.....\")\n epoch = self.epoch\n break\n\n # update G network\n _, summary_str, g_loss = self.sess.run(\n [self.g_optim, self.g_sum, self.g_loss],\n feed_dict={\n self.inputs: batch_images,\n self.y: batch_labels,\n self.z: batch_z\n })\n\n self.writer.add_summary(summary_str, counter)\n\n # display training status\n counter += 1\n _ = self.sess.run(self.fake_images,\n feed_dict={\n self.z: self.sample_z,\n self.y: self.test_labels\n })\n\n # save training results for every 100 steps\n if np.mod(counter, 100) == 0:\n print(\"Iteration : \" + str(idx) + \" Eps: \" + str(eps))\n samples = self.sess.run(self.fake_images,\n feed_dict={\n self.z: self.sample_z,\n self.y: self.test_labels\n })\n tot_num_samples = min(self.sample_num, self.batch_size)\n manifold_h = int(np.floor(np.sqrt(tot_num_samples)))\n manifold_w = int(np.floor(np.sqrt(tot_num_samples)))\n save_images(\n samples[:manifold_h * manifold_w, :, :, :],\n [manifold_h, manifold_w],\n check_folder(self.result_dir + '/' + self.model_dir) +\n '/' + self.model_name +\n '_train_{:02d}_{:04d}.png'.format(epoch, idx))\n epoch = epoch + 1\n\n # After an epoch, start_batch_id is set to zero\n # non-zero value is only for the first epoch after loading pre-trained model\n start_batch_id = 0\n\n # save model\n self.save(self.checkpoint_dir, counter)\n\n # show temporal results\n if (self.dataset_name == 'mnist'):\n self.visualize_results_MNIST(epoch)\n elif (self.dataset_name == 'cifar10'):\n self.visualize_results_CIFAR(epoch)\n\n # save model for final step\n self.save(self.checkpoint_dir, counter)\n\n\n def compute_fpr_tpr_roc(Y_test, Y_score):\n n_classes = Y_score.shape[1]\n false_positive_rate = dict()\n true_positive_rate = dict()\n roc_auc = dict()\n for class_cntr in range(n_classes):\n false_positive_rate[class_cntr], true_positive_rate[\n class_cntr], _ = roc_curve(Y_test[:, class_cntr],\n Y_score[:, class_cntr])\n roc_auc[class_cntr] = auc(false_positive_rate[class_cntr],\n true_positive_rate[class_cntr])\n\n # Compute micro-average ROC curve and ROC area\n false_positive_rate[\"micro\"], true_positive_rate[\n \"micro\"], _ = roc_curve(Y_test.ravel(), Y_score.ravel())\n roc_auc[\"micro\"] = auc(false_positive_rate[\"micro\"],\n true_positive_rate[\"micro\"])\n\n return false_positive_rate, true_positive_rate, roc_auc\n\n def classify(X_train,\n Y_train,\n X_test,\n classiferName,\n random_state_value=0):\n if classiferName == \"lr\":\n classifier = OneVsRestClassifier(\n LogisticRegression(solver='lbfgs',\n multi_class='multinomial',\n random_state=random_state_value))\n elif classiferName == \"mlp\":\n classifier = OneVsRestClassifier(\n MLPClassifier(random_state=random_state_value, alpha=1))\n\n elif classiferName == \"rf\":\n classifier = OneVsRestClassifier(\n RandomForestClassifier(n_estimators=100,\n random_state=random_state_value))\n\n else:\n print(\"Classifier not in the list!\")\n exit()\n Y_score = classifier.fit(X_train, Y_train).predict_proba(X_test)\n return Y_score\n\n batch_size = int(self.batch_size)\n\n if (self.dataset_name == \"mnist\"):\n\n n_class = np.zeros(10)\n n_class[0] = 5923 - batch_size\n n_class[1] = 6742\n n_class[2] = 5958\n n_class[3] = 6131\n n_class[4] = 5842\n n_class[5] = 5421\n n_class[6] = 5918\n n_class[7] = 6265\n n_class[8] = 5851\n n_class[9] = 5949\n\n Z_sample = np.random.uniform(-1, 1, size=(batch_size, self.z_dim))\n y = np.zeros(batch_size, dtype=np.int64) + 0\n y_one_hot = np.zeros((batch_size, self.y_dim))\n y_one_hot[np.arange(batch_size), y] = 1\n images = self.sess.run(self.fake_images,\n feed_dict={\n self.z: Z_sample,\n self.y: y_one_hot\n })\n\n for classLabel in range(0, 10):\n for _ in range(0, int(n_class[classLabel]), batch_size):\n Z_sample = np.random.uniform(-1,\n 1,\n size=(batch_size, self.z_dim))\n y = np.zeros(batch_size, dtype=np.int64) + classLabel\n y_one_hot_init = np.zeros((batch_size, self.y_dim))\n y_one_hot_init[np.arange(batch_size), y] = 1\n\n images = np.append(images,\n self.sess.run(self.fake_images,\n feed_dict={\n self.z: Z_sample,\n self.y: y_one_hot_init\n }),\n axis=0)\n y_one_hot = np.append(y_one_hot, y_one_hot_init, axis=0)\n\n X_test, Y_test = load_mnist(train = False)\n\n Y_test = [int(y) for y in Y_test]\n\n classes = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]\n Y_test = label_binarize(Y_test, classes=classes)\n\n if (self.dataset_name == \"cifar10\"):\n n_class = np.zeros(10)\n for t in range(1, 10):\n n_class[t] = 1000\n \n Z_sample = np.random.uniform(-1, 1, size=(batch_size, self.z_dim))\n y = np.zeros(batch_size, dtype=np.int64) + 0\n y_one_hot = np.zeros((batch_size, self.y_dim))\n y_one_hot[np.arange(batch_size), y] = 1\n images = self.sess.run(self.fake_images,\n feed_dict={\n self.z: Z_sample,\n self.y: y_one_hot\n })\n\n for classLabel in range(0, 10):\n for _ in range(0, int(n_class[classLabel]), batch_size):\n Z_sample = np.random.uniform(-1,\n 1,\n size=(batch_size, self.z_dim))\n y = np.zeros(batch_size, dtype=np.int64) + classLabel\n y_one_hot_init = np.zeros((batch_size, self.y_dim))\n y_one_hot_init[np.arange(batch_size), y] = 1\n\n images = np.append(images,\n self.sess.run(self.fake_images,\n feed_dict={\n self.z: Z_sample,\n self.y: y_one_hot_init\n }),\n axis=0)\n y_one_hot = np.append(y_one_hot, y_one_hot_init, axis=0)\n\n X_test, Y_test = load_cifar10(train=False)\n\n classes = range(0, 10)\n Y_test = label_binarize(Y_test, classes=classes)\n\n print(\" Classifying - Logistic Regression...\")\n\n TwoDim_images = images.reshape(np.shape(images)[0], -2)\n \n X_test = X_test.reshape(np.shape(X_test)[0], -2)\n Y_score = classify(TwoDim_images,\n y_one_hot,\n X_test,\n \"lr\",\n random_state_value=30)\n\n false_positive_rate, true_positive_rate, roc_auc = compute_fpr_tpr_roc(\n Y_test, Y_score)\n\n classification_results_fname = self.base_dir + \"CGAN_AuROC.txt\"\n classification_results = open(classification_results_fname, \"w\")\n\n classification_results.write(\n \"\\nepsilon : {:.2f}, sigma: {:.2f}, clipping value: {:.2f}\".format(\n (self.epsilon), round(self.noise_multiplier, 2),\n round(self.l2_norm_clip, 2)))\n\n classification_results.write(\"\\nAuROC - logistic Regression: \" +\n str(roc_auc[\"micro\"]))\n classification_results.write(\n \"\\n--------------------------------------------------------------------\\n\"\n )\n \n print(\" Classifying - Random Forest...\")\n Y_score = classify(TwoDim_images,\n y_one_hot,\n X_test,\n \"rf\",\n random_state_value=30)\n\n print(\" Computing ROC - Random Forest ...\")\n false_positive_rate, true_positive_rate, roc_auc = compute_fpr_tpr_roc(\n Y_test, Y_score)\n\n classification_results.write(\n \"\\nepsilon : {:.2f}, sigma: {:.2f}, clipping value: {:.2f}\".format(\n (self.epsilon), round(self.noise_multiplier, 2),\n round(self.l2_norm_clip, 2)))\n\n classification_results.write(\"\\nAuROC - random Forest: \" +\n str(roc_auc[\"micro\"]))\n classification_results.write(\n \"\\n--------------------------------------------------------------------\\n\"\n )\n \n print(\" Classifying - multilayer Perceptron ...\")\n Y_score = classify(TwoDim_images,\n y_one_hot,\n X_test,\n \"mlp\",\n random_state_value=30)\n\n print(\" Computing ROC - Multilayer Perceptron ...\")\n false_positive_rate, true_positive_rate, roc_auc = compute_fpr_tpr_roc(\n Y_test, Y_score)\n\n classification_results.write(\n \"\\nepsilon : {:.2f}, sigma: {:.2f}, clipping value: {:.2f}\".format(\n (self.epsilon), round(self.noise_multiplier, 2),\n round(self.l2_norm_clip, 2)))\n\n classification_results.write(\"\\nAuROC - multilayer Perceptron: \" +\n str(roc_auc[\"micro\"]))\n classification_results.write(\n \"\\n--------------------------------------------------------------------\\n\"\n )\n\n # save model for final step\n self.save(self.checkpoint_dir, counter)\n\n def compute_epsilon(self, steps):\n \"\"\"Computes epsilon value for given hyperparameters.\"\"\"\n if self.noise_multiplier == 0.0:\n return float('inf')\n orders = [1 + x / 10. for x in range(1, 100)] + list(range(12, 64))\n sampling_probability = self.batch_size / 60000\n rdp = compute_rdp(q=sampling_probability,\n noise_multiplier=self.noise_multiplier,\n steps=steps,\n orders=orders)\n # Delta is set to 1e-5 because MNIST has 60000 training points.\n return get_privacy_spent(orders, rdp, target_delta=1e-5)[0]\n\n # CIFAR 10\n def visualize_results_CIFAR(self, epoch):\n tot_num_samples = min(self.sample_num, self.batch_size) # 64, 100\n image_frame_dim = int(np.floor(np.sqrt(tot_num_samples))) # 8\n \"\"\" random condition, random noise \"\"\"\n y = np.random.choice(self.y_dim, self.batch_size)\n y_one_hot = np.zeros((self.batch_size, self.y_dim))\n y_one_hot[np.arange(self.batch_size), y] = 1\n\n z_sample = np.random.uniform(-1, 1, size=(self.batch_size,\n self.z_dim)) # 100, 100\n\n samples = self.sess.run(self.fake_images,\n feed_dict={\n self.z: z_sample,\n self.y: y_one_hot\n })\n\n save_matplot_img(\n samples[:image_frame_dim * image_frame_dim, :, :, :],\n [image_frame_dim, image_frame_dim], self.result_dir + '/' +\n self.model_name + '_epoch%03d' % epoch + '_test_all_classes.png')\n\n # MNIST\n def visualize_results_MNIST(self, epoch):\n tot_num_samples = min(self.sample_num, self.batch_size)\n image_frame_dim = int(np.floor(np.sqrt(tot_num_samples)))\n \"\"\" random condition, random noise \"\"\"\n y = np.random.choice(self.y_dim, self.batch_size)\n y_one_hot = np.zeros((self.batch_size, self.y_dim))\n y_one_hot[np.arange(self.batch_size), y] = 1\n\n z_sample = np.random.uniform(-1, 1, size=(self.batch_size, self.z_dim))\n samples = self.sess.run(self.fake_images,\n feed_dict={\n self.z: z_sample,\n self.y: y_one_hot\n })\n\n save_images(\n samples[:image_frame_dim * image_frame_dim, :, :, :],\n [image_frame_dim, image_frame_dim],\n check_folder(self.result_dir + '/' + self.model_dir) + '/' +\n self.model_name + '_epoch%03d' % epoch + '_test_all_classes.png')\n \"\"\" specified condition, random noise \"\"\"\n n_styles = 10 # must be less than or equal to self.batch_size\n\n np.random.seed()\n si = np.random.choice(self.batch_size, n_styles)\n\n for l in range(self.y_dim):\n y = np.zeros(self.batch_size, dtype=np.int64) + l\n y_one_hot = np.zeros((self.batch_size, self.y_dim))\n y_one_hot[np.arange(self.batch_size), y] = 1\n\n samples = self.sess.run(self.fake_images,\n feed_dict={\n self.z: z_sample,\n self.y: y_one_hot\n })\n save_images(\n samples[:image_frame_dim * image_frame_dim, :, :, :],\n [image_frame_dim, image_frame_dim],\n check_folder(self.result_dir + '/' + self.model_dir) + '/' +\n self.model_name + '_epoch%03d' % epoch +\n '_test_class_%d.png' % l)\n\n samples = samples[si, :, :, :]\n\n if l == 0:\n all_samples = samples\n else:\n all_samples = np.concatenate((all_samples, samples), axis=0)\n \"\"\" save merged images to check style-consistency \"\"\"\n canvas = np.zeros_like(all_samples)\n for s in range(n_styles):\n for c in range(self.y_dim):\n canvas[s * self.y_dim +\n c, :, :, :] = all_samples[c * n_styles + s, :, :, :]\n\n save_images(\n canvas, [n_styles, self.y_dim],\n check_folder(self.result_dir + '/' + self.model_dir) + '/' +\n self.model_name + '_epoch%03d' % epoch +\n '_test_all_classes_style_by_style.png')\n\n @property\n def model_dir(self):\n return \"{}_{}_{}_{}\".format(self.model_name, self.dataset_name,\n self.batch_size, self.z_dim)\n\n def save(self, checkpoint_dir, step):\n checkpoint_dir = os.path.join(checkpoint_dir, self.model_dir,\n self.model_name)\n\n if not os.path.exists(checkpoint_dir):\n os.makedirs(checkpoint_dir)\n\n self.saver.save(self.sess,\n os.path.join(checkpoint_dir,\n self.model_name + '.model'),\n global_step=step)\n\n def load(self, checkpoint_dir):\n import re\n print(\" [*] Reading checkpoints...\")\n checkpoint_dir = os.path.join(checkpoint_dir, self.model_dir,\n self.model_name)\n\n ckpt = tf.train.get_checkpoint_state(checkpoint_dir)\n if ckpt and ckpt.model_checkpoint_path:\n ckpt_name = os.path.basename(ckpt.model_checkpoint_path)\n self.saver.restore(self.sess,\n os.path.join(checkpoint_dir, ckpt_name))\n counter = int(\n next(re.finditer(\"(\\d+)(?!.*\\d)\", ckpt_name)).group(0))\n print(\" [*] Success to read {}\".format(ckpt_name))\n return True, counter\n else:\n print(\" [*] Failed to find a checkpoint\")\n return False, 0\n", "Using TensorFlow backend.\n" ] ], [ [ "## gan.utils", "_____no_output_____" ] ], [ [ "\"\"\"\n Most codes from https://github.com/carpedm20/DCGAN-tensorflow\n\"\"\"\nfrom __future__ import division\n\nimport scipy.misc\nimport numpy as np\n\nfrom six.moves import xrange\nimport matplotlib.pyplot as plt\nimport os, gzip\n\nimport tensorflow as tf\nimport tensorflow.contrib.slim as slim\n\nfrom keras.datasets import cifar10\nfrom keras.datasets import mnist\n\n\ndef one_hot(x, n):\n \"\"\"\n convert index representation to one-hot representation\n \"\"\"\n x = np.array(x)\n assert x.ndim == 1\n return np.eye(n)[x]\n\n\ndef prepare_input(data=None, labels=None):\n image_height = 32\n image_width = 32\n image_depth = 3\n assert (data.shape[1] == image_height * image_width * image_depth)\n assert (data.shape[0] == labels.shape[0])\n # do mean normalization across all samples\n mu = np.mean(data, axis=0)\n mu = mu.reshape(1, -1)\n sigma = np.std(data, axis=0)\n sigma = sigma.reshape(1, -1)\n data = data - mu\n data = data / sigma\n is_nan = np.isnan(data)\n is_inf = np.isinf(data)\n if np.any(is_nan) or np.any(is_inf):\n print('data is not well-formed : is_nan {n}, is_inf: {i}'.format(\n n=np.any(is_nan), i=np.any(is_inf)))\n # data is transformed from (no_of_samples, 3072) to (no_of_samples , image_height, image_width, image_depth)\n # make sure the type of the data is no.float32\n data = data.reshape([-1, image_depth, image_height, image_width])\n data = data.transpose([0, 2, 3, 1])\n data = data.astype(np.float32)\n return data, labels\n\n\ndef read_cifar10(filename): # queue one element\n class CIFAR10Record(object):\n pass\n\n result = CIFAR10Record()\n\n label_bytes = 1 # 2 for CIFAR-100\n result.height = 32\n result.width = 32\n result.depth = 3\n\n data = np.load(filename, encoding='latin1')\n\n value = np.asarray(data['data']).astype(np.float32)\n labels = np.asarray(data['labels']).astype(np.int32)\n\n return prepare_input(value, labels)\n\n\ndef load_cifar10(train):\n (x_train, y_train), (x_test, y_test) = cifar10.load_data()\n\n if (train == True):\n\n dataX = x_train.reshape([-1, 32, 32, 3])\n dataY = y_train\n\n else:\n dataX = x_test.reshape([-1, 32, 32, 3])\n dataY = y_test\n\n seed = 547\n np.random.seed(seed)\n np.random.shuffle(dataX)\n np.random.seed(seed)\n np.random.shuffle(dataY)\n\n y_vec = np.zeros((len(dataY), 10), dtype=np.float)\n for i, label in enumerate(dataY):\n y_vec[i, dataY[i]] = 1.0\n\n return dataX / 255., y_vec\n\n\ndef load_mnist(train = True):\n\n def extract_data(filename, num_data, head_size, data_size):\n with gzip.open(filename) as bytestream:\n bytestream.read(head_size)\n buf = bytestream.read(data_size * num_data)\n data = np.frombuffer(buf, dtype=np.uint8).astype(np.float)\n return data\n\n (x_train, y_train), (x_test, y_test) = mnist.load_data()\n \n x_train = x_train.reshape((60000, 28, 28, 1))\n y_train = y_train.reshape((60000))\n\n x_test = x_test.reshape((10000, 28, 28, 1))\n y_test = y_test.reshape((10000))\n\n y_train = np.asarray(y_train)\n y_test = np.asarray(y_test)\n\n if (train == True):\n seed = 547\n np.random.seed(seed)\n np.random.shuffle(x_train)\n np.random.seed(seed)\n np.random.shuffle(y_train)\n\n y_vec = np.zeros((len(y_train), 10), dtype=np.float)\n for i, label in enumerate(y_train):\n y_vec[i, y_train[i]] = 1.0\n return x_train / 255., y_vec\n \n else:\n seed = 547\n np.random.seed(seed)\n np.random.shuffle(x_test)\n np.random.seed(seed)\n np.random.shuffle(y_test)\n\n y_vec = np.zeros((len(y_test), 10), dtype=np.float)\n for i, label in enumerate(y_test):\n y_vec[i, y_test[i]] = 1.0\n return x_test / 255., y_vec\n \n\n\ndef check_folder(log_dir):\n if not os.path.exists(log_dir):\n os.makedirs(log_dir)\n return log_dir\n\n\ndef show_all_variables():\n model_vars = tf.trainable_variables()\n slim.model_analyzer.analyze_vars(model_vars, print_info=True)\n\n\ndef get_image(image_path,\n input_height,\n input_width,\n resize_height=64,\n resize_width=64,\n crop=True,\n grayscale=False):\n image = imread(image_path, grayscale)\n return transform(image, input_height, input_width, resize_height,\n resize_width, crop)\n\n\ndef save_images(images, size, image_path):\n return imsave(inverse_transform(images), size, image_path)\n\n\ndef imread(path, grayscale=False):\n if (grayscale):\n return scipy.misc.imread(path, flatten=True).astype(np.float)\n else:\n return scipy.misc.imread(path).astype(np.float)\n\n\ndef merge_images(images, size):\n return inverse_transform(images)\n\n\ndef merge(images, size):\n h, w = images.shape[1], images.shape[2]\n if (images.shape[3] in (3, 4)):\n c = images.shape[3]\n img = np.zeros((h * size[0], w * size[1], c))\n for idx, image in enumerate(images):\n i = idx % size[1]\n j = idx // size[1]\n img[j * h:j * h + h, i * w:i * w + w, :] = image\n return img\n elif images.shape[3] == 1:\n img = np.zeros((h * size[0], w * size[1]))\n for idx, image in enumerate(images):\n i = idx % size[1]\n j = idx // size[1]\n img[j * h:j * h + h, i * w:i * w + w] = image[:, :, 0]\n return img\n else:\n raise ValueError('in merge(images,size) images parameter '\n 'must have dimensions: HxW or HxWx3 or HxWx4')\n\n\ndef imsave(images, size, path):\n image = np.squeeze(merge(images, size))\n return scipy.misc.imsave(path, image)\n\n\ndef center_crop(x, crop_h, crop_w, resize_h=64, resize_w=64):\n if crop_w is None:\n crop_w = crop_h\n h, w = x.shape[:2]\n j = int(round((h - crop_h) / 2.))\n i = int(round((w - crop_w) / 2.))\n return scipy.misc.imresize(x[j:j + crop_h, i:i + crop_w],\n [resize_h, resize_w])\n\n\ndef transform(image,\n input_height,\n input_width,\n resize_height=64,\n resize_width=64,\n crop=True):\n if crop:\n cropped_image = center_crop(image, input_height, input_width,\n resize_height, resize_width)\n else:\n cropped_image = scipy.misc.imresize(image,\n [resize_height, resize_width])\n return np.array(cropped_image) / 127.5 - 1.\n\n\ndef inverse_transform(images):\n return (images + 1.) / 2.\n\n\n\"\"\" Drawing Tools \"\"\"\n\n\n# borrowed from https://github.com/ykwon0407/variational_autoencoder/blob/master/variational_bayes.ipynb\ndef save_scattered_image(z,\n id,\n z_range_x,\n z_range_y,\n name='scattered_image.jpg'):\n N = 10\n plt.figure(figsize=(8, 6))\n plt.scatter(z[:, 0],\n z[:, 1],\n c=np.argmax(id, 1),\n marker='o',\n edgecolor='none',\n cmap=discrete_cmap(N, 'jet'))\n plt.colorbar(ticks=range(N))\n axes = plt.gca()\n axes.set_xlim([-z_range_x, z_range_x])\n axes.set_ylim([-z_range_y, z_range_y])\n plt.grid(True)\n plt.savefig(name)\n\n\n# borrowed from https://gist.github.com/jakevdp/91077b0cae40f8f8244a\ndef discrete_cmap(N, base_cmap=None):\n \"\"\"Create an N-bin discrete colormap from the specified input map\"\"\"\n\n # Note that if base_cmap is a string or None, you can simply do\n # return plt.cm.get_cmap(base_cmap, N)\n # The following works for string, None, or a colormap instance:\n\n base = plt.cm.get_cmap(base_cmap)\n color_list = base(np.linspace(0, 1, N))\n cmap_name = base.name + str(N)\n return base.from_list(cmap_name, color_list, N)\n\n\ndef save_matplot_img(images, size, image_path):\n # revice image data // M*N*3 // RGB float32 : value must set between 0. with 1.\n for idx in range(64):\n vMin = np.amin(images[idx])\n vMax = np.amax(images[idx])\n img_arr = images[idx].reshape(32 * 32 * 3, 1) # flatten\n for i, v in enumerate(img_arr):\n img_arr[i] = (v - vMin) / (vMax - vMin)\n img_arr = img_arr.reshape(32, 32, 3) # M*N*3\n\n plt.subplot(8, 8, idx + 1), plt.imshow(img_arr,\n interpolation='nearest')\n plt.axis(\"off\")\n\n\n plt.savefig(image_path)\n", "_____no_output_____" ] ], [ [ "## Main", "_____no_output_____" ] ], [ [ "import tensorflow as tf\nimport os\n\nbase_dir = \"./\"\n\nout_dir = base_dir + \"mnist_clip1_sigma0.6_lr0.55\"\n\nif not os.path.exists(out_dir):\n os.mkdir(out_dir)\n\ngpu_options = tf.GPUOptions(visible_device_list=\"0\")\nwith tf.Session(config=tf.ConfigProto(allow_soft_placement=True,\n gpu_options=gpu_options)) as sess:\n\n epoch = 100\n cgan = OUR_DP_CGAN(sess,\n epoch=epoch,\n batch_size=64,\n z_dim=100,\n epsilon=9.6,\n delta=1e-5,\n sigma=0.6,\n clip_value=1,\n lr=0.055,\n dataset_name='mnist',\n checkpoint_dir=out_dir + \"/checkpoint/\",\n result_dir=out_dir + \"/results/\",\n log_dir=out_dir + \"/logs/\",\n base_dir=base_dir)\n\n cgan.build_model()\n print(\" [*] Building model finished!\")\n\n show_all_variables()\n cgan.train()\n print(\" [*] Training finished!\")\n", "Downloading data from https://s3.amazonaws.com/img-datasets/mnist.npz\n11493376/11490434 [==============================] - 1s 0us/step\nWARNING:tensorflow:From /usr/local/lib/python3.6/dist-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 /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/array_grad.py:425: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\nInstructions for updating:\nUse tf.cast instead.\n [*] Building model finished!\n---------\nVariables: name (type shape) [size]\n---------\ndiscriminator/d_conv1/w:0 (float32_ref 4x4x11x64) [11264, bytes: 45056]\ndiscriminator/d_conv1/biases:0 (float32_ref 64) [64, bytes: 256]\ndiscriminator/d_conv2/w:0 (float32_ref 4x4x64x128) [131072, bytes: 524288]\ndiscriminator/d_conv2/biases:0 (float32_ref 128) [128, bytes: 512]\ndiscriminator/d_bn2/beta:0 (float32_ref 128) [128, bytes: 512]\ndiscriminator/d_bn2/gamma:0 (float32_ref 128) [128, bytes: 512]\ndiscriminator/d_fc3/Matrix:0 (float32_ref 6272x1024) [6422528, bytes: 25690112]\ndiscriminator/d_fc3/bias:0 (float32_ref 1024) [1024, bytes: 4096]\ndiscriminator/d_bn3/beta:0 (float32_ref 1024) [1024, bytes: 4096]\ndiscriminator/d_bn3/gamma:0 (float32_ref 1024) [1024, bytes: 4096]\ndiscriminator/d_fc4/Matrix:0 (float32_ref 1024x1) [1024, bytes: 4096]\ndiscriminator/d_fc4/bias:0 (float32_ref 1) [1, bytes: 4]\ngenerator/g_fc1/Matrix:0 (float32_ref 110x1024) [112640, bytes: 450560]\ngenerator/g_fc1/bias:0 (float32_ref 1024) [1024, bytes: 4096]\ngenerator/g_bn1/beta:0 (float32_ref 1024) [1024, bytes: 4096]\ngenerator/g_bn1/gamma:0 (float32_ref 1024) [1024, bytes: 4096]\ngenerator/g_fc2/Matrix:0 (float32_ref 1024x6272) [6422528, bytes: 25690112]\ngenerator/g_fc2/bias:0 (float32_ref 6272) [6272, bytes: 25088]\ngenerator/g_bn2/beta:0 (float32_ref 6272) [6272, bytes: 25088]\ngenerator/g_bn2/gamma:0 (float32_ref 6272) [6272, bytes: 25088]\ngenerator/g_dc3/w:0 (float32_ref 4x4x64x128) [131072, bytes: 524288]\ngenerator/g_dc3/biases:0 (float32_ref 64) [64, bytes: 256]\ngenerator/g_bn3/beta:0 (float32_ref 64) [64, bytes: 256]\ngenerator/g_bn3/gamma:0 (float32_ref 64) [64, bytes: 256]\ngenerator/g_dc4/w:0 (float32_ref 4x4x1x64) [1024, bytes: 4096]\ngenerator/g_dc4/biases:0 (float32_ref 1) [1, bytes: 4]\nTotal size of variables: 13258754\nTotal bytes of variables: 53035016\n [*] Reading checkpoints...\n [*] Failed to find a checkpoint\n [!] Load failed...\nIteration : 98 Eps: 5.750369579644922\n" ] ] ]

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

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

[ "BSD-3-Clause" ]

[ "BSD-3-Clause" ]

[ "BSD-3-Clause" ]

[ [ [ "URL: http://matplotlib.org/examples/pylab_examples/quiver_demo.html\n\nMost examples work across multiple plotting backends, this example is also available for:\n\n* [Matplotlib - quiver_demo](../matplotlib/quiver_demo.ipynb)", "_____no_output_____" ] ], [ [ "import numpy as np\nimport holoviews as hv\nfrom holoviews import opts\nhv.extension('bokeh')", "_____no_output_____" ] ], [ [ "## Define data", "_____no_output_____" ] ], [ [ "xs, ys = np.arange(0, 2 * np.pi, .2), np.arange(0, 2 * np.pi, .2)\nX, Y = np.meshgrid(xs, ys)\nU = np.cos(X)\nV = np.sin(Y)\n\n# Convert to magnitude and angle\nmag = np.sqrt(U**2 + V**2)\nangle = (np.pi/2.) - np.arctan2(U/mag, V/mag)", "_____no_output_____" ] ], [ [ "## Plot", "_____no_output_____" ] ], [ [ "label = 'Arrows scale with plot width, not view'\n\nopts.defaults(opts.VectorField(height=400, width=500))\n\nvectorfield = hv.VectorField((xs, ys, angle, mag))\nvectorfield.relabel(label)", "_____no_output_____" ], [ "label = \"pivot='mid'; every third arrow\"\n\nvf_mid = hv.VectorField((xs[::3], ys[::3], angle[::3, ::3], mag[::3, ::3], ))\npoints = hv.Points((X[::3, ::3].flat, Y[::3, ::3].flat))\n\nopts.defaults(opts.Points(color='red'))\n\n(vf_mid * points).relabel(label)", "_____no_output_____" ], [ "label = \"pivot='tip'; scales with x view\"\n\nvectorfield = hv.VectorField((xs, ys, angle, mag))\npoints = hv.Points((X.flat, Y.flat))\n\n(points * vectorfield).opts(\n opts.VectorField(magnitude='Magnitude', color='Magnitude',\n pivot='tip', line_width=2, title=label))", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "### **PINN eikonal solver for a portion of the Marmousi model**", "_____no_output_____" ] ], [ [ "from google.colab import drive\ndrive.mount('/content/gdrive')", "_____no_output_____" ], [ "cd \"/content/gdrive/My Drive/Colab Notebooks/Codes/PINN_isotropic_eikonal_R1\"", "_____no_output_____" ], [ "!pip install sciann==0.5.4.0\n!pip install tensorflow==2.2.0\n#!pip install keras==2.3.1 ", "_____no_output_____" ], [ "import numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nfrom mpl_toolkits.axes_grid1 import make_axes_locatable\nimport tensorflow as tf\nfrom sciann import Functional, Variable, SciModel, PDE\nfrom sciann.utils import *\nimport scipy.io \nimport time\nimport random\nfrom mpl_toolkits.axes_grid1.inset_locator import zoomed_inset_axes\nfrom mpl_toolkits.axes_grid1.inset_locator import mark_inset\n\ntf.config.threading.set_intra_op_parallelism_threads(1)\ntf.config.threading.set_inter_op_parallelism_threads(1)", "---------------------- SCIANN 0.5.4.0 ---------------------- \nFor details, check out our review paper and the documentation at: \n + \"https://arxiv.org/abs/2005.08803\", \n + \"https://www.sciann.com\". \n\n" ], [ "np.random.seed(123)\ntf.random.set_seed(123)", "_____no_output_____" ], [ "# Loading velocity model\n\nfilename=\"./inputs/marm/model/marm_vz.txt\"\nmarm = pd.read_csv(filename, index_col=None, header=None)\nvelmodel = np.reshape(np.array(marm), (101, 101)).T\n\n# Loading reference solution\n\nfilename=\"./inputs/marm/traveltimes/fmm_or2_marm_s(1,1).txt\"\nT_data = pd.read_csv(filename, index_col=None, header=None)\nT_data = np.reshape(np.array(T_data), (101, 101)).T", "_____no_output_____" ], [ "#Model specifications\n\nzmin = 0.; zmax = 2.; deltaz = 0.02;\nxmin = 0.; xmax = 2.; deltax = 0.02;\n\n\n# Point-source location\nsz = 1.0; sx = 1.0;\n\n# Number of training points\nnum_tr_pts = 3000", "_____no_output_____" ], [ "# Creating grid, calculating refrence traveltimes, and prepare list of grid points for training (X_star)\n\nz = np.arange(zmin,zmax+deltaz,deltaz)\nnz = z.size\n\nx = np.arange(xmin,xmax+deltax,deltax)\nnx = x.size\n\n\nZ,X = np.meshgrid(z,x,indexing='ij')\n\nX_star = [Z.reshape(-1,1), X.reshape(-1,1)]\n\nselected_pts = np.random.choice(np.arange(Z.size),num_tr_pts,replace=False)\nZf = Z.reshape(-1,1)[selected_pts]\nZf = np.append(Zf,sz)\nXf = X.reshape(-1,1)[selected_pts]\nXf = np.append(Xf,sx)\n\n\nX_starf = [Zf.reshape(-1,1), Xf.reshape(-1,1)]", "_____no_output_____" ], [ "# Plot the velocity model with the source location\n\nplt.style.use('default')\n\nplt.figure(figsize=(4,4))\n\nax = plt.gca()\nim = ax.imshow(velmodel, extent=[xmin,xmax,zmax,zmin], aspect=1, cmap=\"jet\")\n\nax.plot(sx,sz,'k*',markersize=8)\n\nplt.xlabel('Offset (km)', fontsize=14)\nplt.xticks(fontsize=10)\n\nplt.ylabel('Depth (km)', fontsize=14)\nplt.yticks(fontsize=10)\n\nax.xaxis.set_major_locator(plt.MultipleLocator(0.5))\nax.yaxis.set_major_locator(plt.MultipleLocator(0.5))\n\ndivider = make_axes_locatable(ax)\ncax = divider.append_axes(\"right\", size=\"6%\", pad=0.15)\n\ncbar = plt.colorbar(im, cax=cax)\n\ncbar.set_label('km/s',size=10)\ncbar.ax.tick_params(labelsize=10)\n\nplt.savefig(\"./figs/marm/velmodel.pdf\", format='pdf', bbox_inches=\"tight\")", "_____no_output_____" ], [ "# Analytical solution for the known traveltime part\nvel = velmodel[int(round(sz/deltaz)),int(round(sx/deltax))] # Velocity at the source location\n\nT0 = np.sqrt((Z-sz)**2 + (X-sx)**2)/vel; \n\npx0 = np.divide(X-sx, T0*vel**2, out=np.zeros_like(T0), where=T0!=0)\npz0 = np.divide(Z-sz, T0*vel**2, out=np.zeros_like(T0), where=T0!=0)", "_____no_output_____" ], [ "# Find source location id in X_star\n\nTOLX = 1e-6\nTOLZ = 1e-6\n\nsids,_ = np.where(np.logical_and(np.abs(X_starf[0]-sz)<TOLZ , np.abs(X_starf[1]-sx)<TOLX))\n\nprint(sids)\nprint(sids.shape)\nprint(X_starf[0][sids,0])\nprint(X_starf[1][sids,0])", "[3000]\n(1,)\n[1.]\n[1.]\n" ], [ "# Preparing the Sciann model object\n\nK.clear_session() \n\nlayers = [20]*10\n\n# Appending source values\nvelmodelf = velmodel.reshape(-1,1)[selected_pts]; velmodelf = np.append(velmodelf,vel)\npx0f = px0.reshape(-1,1)[selected_pts]; px0f = np.append(px0f,0.)\npz0f = pz0.reshape(-1,1)[selected_pts]; pz0f = np.append(pz0f,0.)\nT0f = T0.reshape(-1,1)[selected_pts]; T0f = np.append(T0f,0.)\n\nxt = Variable(\"xt\",dtype='float64')\nzt = Variable(\"zt\",dtype='float64')\nvt = Variable(\"vt\",dtype='float64')\npx0t = Variable(\"px0t\",dtype='float64')\npz0t = Variable(\"pz0t\",dtype='float64')\nT0t = Variable(\"T0t\",dtype='float64')\n\ntau = Functional(\"tau\", [zt, xt], layers, 'l-atan')\n\n# Loss function based on the factored isotropic eikonal equation\nL = (T0t*diff(tau, xt) + tau*px0t)**2 + (T0t*diff(tau, zt) + tau*pz0t)**2 - 1.0/vt**2\n\ntargets = [tau, 20*L, (1-sign(tau*T0t))*abs(tau*T0t)]\ntarget_vals = [(sids, np.ones(sids.shape).reshape(-1,1)), 'zeros', 'zeros']\n\nmodel = SciModel(\n [zt, xt, vt, pz0t, px0t, T0t], \n targets,\n load_weights_from='models/vofz_model-end.hdf5',\n optimizer='scipy-l-BFGS-B'\n)", "_____no_output_____" ], [ "#Model training\n\nstart_time = time.time()\nhist = model.train(\n X_starf + [velmodelf,pz0f,px0f,T0f],\n target_vals,\n batch_size = X_starf[0].size,\n epochs = 12000,\n learning_rate = 0.008,\n verbose=0\n )\nelapsed = time.time() - start_time\nprint('Training time: %.2f seconds' %(elapsed))\n", "\nTotal samples: 3001 \nBatch size: 3001 \nTotal batches: 1 \n\nTraining time: 831.12 seconds\n" ], [ "# Convergence history plot for verification\n\nfig = plt.figure(figsize=(5,3))\nax = plt.axes()\n#ax.semilogy(np.arange(0,300,0.001),hist.history['loss'],LineWidth=2)\nax.semilogy(hist.history['loss'],LineWidth=2)\n\nax.set_xlabel('Epochs (x $10^3$)',fontsize=16)\n\nplt.xticks(fontsize=12)\n#ax.xaxis.set_major_locator(plt.MultipleLocator(50))\n\nax.set_ylabel('Loss',fontsize=16)\nplt.yticks(fontsize=12);\nplt.grid()", "_____no_output_____" ], [ "# Predicting traveltime solution from the trained model\n\nL_pred = L.eval(model, X_star + [velmodel,pz0,px0,T0])\ntau_pred = tau.eval(model, X_star + [velmodel,pz0,px0,T0])\ntau_pred = tau_pred.reshape(Z.shape)\n\nT_pred = tau_pred*T0\n\nprint('Time at source: %.4f'%(tau_pred[int(round(sz/deltaz)),int(round(sx/deltax))]))", "Time at source: 0.9995\n" ], [ "# Plot the PINN solution error\n\nplt.style.use('default')\n\nplt.figure(figsize=(4,4))\n\nax = plt.gca()\nim = ax.imshow(np.abs(T_pred-T_data), extent=[xmin,xmax,zmax,zmin], aspect=1, cmap=\"jet\")\n\n\nplt.xlabel('Offset (km)', fontsize=14)\nplt.xticks(fontsize=10)\n\nplt.ylabel('Depth (km)', fontsize=14)\nplt.yticks(fontsize=10)\n\nax.xaxis.set_major_locator(plt.MultipleLocator(0.5))\nax.yaxis.set_major_locator(plt.MultipleLocator(0.5))\n\ndivider = make_axes_locatable(ax)\ncax = divider.append_axes(\"right\", size=\"6%\", pad=0.15)\n\ncbar = plt.colorbar(im, cax=cax)\n\ncbar.set_label('seconds',size=10)\ncbar.ax.tick_params(labelsize=10)\n\nplt.savefig(\"./figs/marm/pinnerror.pdf\", format='pdf', bbox_inches=\"tight\")", "_____no_output_____" ], [ "# Load fast sweeping traveltims for comparison\n\nT_fsm = np.load('./inputs/marm/traveltimes/Tcomp.npy')", "_____no_output_____" ], [ "# Plot the first order FMM solution error\n\nplt.style.use('default')\n\nplt.figure(figsize=(4,4))\n\nax = plt.gca()\nim = ax.imshow(np.abs(T_fsm-T_data), extent=[xmin,xmax,zmax,zmin], aspect=1, cmap=\"jet\")\n\n\nplt.xlabel('Offset (km)', fontsize=14)\nplt.xticks(fontsize=10)\n\nplt.ylabel('Depth (km)', fontsize=14)\nplt.yticks(fontsize=10)\n\nax.xaxis.set_major_locator(plt.MultipleLocator(0.5))\nax.yaxis.set_major_locator(plt.MultipleLocator(0.5))\n\ndivider = make_axes_locatable(ax)\ncax = divider.append_axes(\"right\", size=\"6%\", pad=0.15)\n\ncbar = plt.colorbar(im, cax=cax)\n\ncbar.set_label('seconds',size=10)\n\ncbar.ax.tick_params(labelsize=10)\n\nplt.savefig(\"./figs/marm/fmm1error.pdf\", format='pdf', bbox_inches=\"tight\")", "_____no_output_____" ], [ "# Traveltime contour plots\n\nfig = plt.figure(figsize=(5,5))\n\nax = plt.gca()\nim1 = ax.contour(T_data, 6, extent=[xmin,xmax,zmin,zmax], colors='r')\nim2 = ax.contour(T_pred, 6, extent=[xmin,xmax,zmin,zmax], colors='k',linestyles = 'dashed')\nim3 = ax.contour(T_fsm, 6, extent=[xmin,xmax,zmin,zmax], colors='b',linestyles = 'dotted')\n\nax.plot(sx,sz,'k*',markersize=8)\n\nplt.xlabel('Offset (km)', fontsize=14)\nplt.ylabel('Depth (km)', fontsize=14)\nax.tick_params(axis='both', which='major', labelsize=8)\nplt.gca().invert_yaxis()\nh1,_ = im1.legend_elements()\nh2,_ = im2.legend_elements()\nh3,_ = im3.legend_elements()\nax.legend([h1[0], h2[0], h3[0]], ['Reference', 'PINN', 'Fast sweeping'],fontsize=12)\n\nax.xaxis.set_major_locator(plt.MultipleLocator(0.5))\nax.yaxis.set_major_locator(plt.MultipleLocator(0.5))\n\nplt.xticks(fontsize=10)\nplt.yticks(fontsize=10)\n\n#ax.arrow(1.9, 1.7, -0.1, -0.1, head_width=0.05, head_length=0.075, fc='red', ec='red',width=0.02)\n\nplt.savefig(\"./figs/marm/contours.pdf\", format='pdf', bbox_inches=\"tight\")", "_____no_output_____" ], [ "print(np.linalg.norm(T_pred-T_data)/np.linalg.norm(T_data))\nprint(np.linalg.norm(T_pred-T_data))", "0.005852164264049101\n0.19346281698941364\n" ] ] ]

[ "markdown", "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# GRIP June'21 - The Sparks Foundation\n\n## Data Science and Business Analytics \n\n## Author: Smriti Gupta\n\n### Task 1: **Prediction using Supervised ML** \n\n* Predict the percentage of an student based on the no. of study hours. \n* What will be predicted score if a student studies for 9.25 hrs/ day? \n* _LANGUAGE:_ Python\n* _DATASET:_ http://bit.ly/w-data\n", "_____no_output_____" ] ], [ [ "# Importing Libraries\n\nimport pandas as pd\nimport numpy as np\nimport matplotlib.pyplot as plt\nimport seaborn as sns\n%matplotlib inline\n%config Completer.use_jedi = False", "_____no_output_____" ], [ "# Reading data from remote link\n\nurl = \"https://raw.githubusercontent.com/AdiPersonalWorks/Random/master/student_scores%20-%20student_scores.csv\"\ndf = pd.read_csv(url)\n\n# Viewing the Data\n\ndf.head(10)", "_____no_output_____" ], [ "# Shape of the Dataset\n\ndf.shape", "_____no_output_____" ], [ "# Checking the information of Data\n\ndf.info()", "<class 'pandas.core.frame.DataFrame'>\nRangeIndex: 25 entries, 0 to 24\nData columns (total 2 columns):\n # Column Non-Null Count Dtype \n--- ------ -------------- ----- \n 0 Hours 25 non-null float64\n 1 Scores 25 non-null int64 \ndtypes: float64(1), int64(1)\nmemory usage: 528.0 bytes\n" ], [ "# Checking the statistical details of Data\n\ndf.describe()", "_____no_output_____" ], [ "# Checking the correlation between Hours and Scores\n\ncorr = df.corr()\ncorr", "_____no_output_____" ], [ "colors = ['#670067','#008080']", "_____no_output_____" ] ], [ [ "# Data Visualization", "_____no_output_____" ] ], [ [ "# 2-D graph to establish relationship between the Data and checking for linearity \n\nsns.set_style('darkgrid')\ndf.plot(x='Hours', y='Scores', style='o') \nplt.title('Hours vs Percentage') \nplt.xlabel('Hours Studied') \nplt.ylabel('Percentage Score') \nplt.show()", "_____no_output_____" ] ], [ [ "# Data Preprocessing", "_____no_output_____" ] ], [ [ "X = df.iloc[:, :-1].values \ny = df.iloc[:, 1].values ", "_____no_output_____" ] ], [ [ "# LINEAR REGRESSION MODEL", "_____no_output_____" ], [ "## Splitting Dataset into training and test sets:", "_____no_output_____" ] ], [ [ "from sklearn.model_selection import train_test_split\nX_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)", "_____no_output_____" ] ], [ [ "## Training the Model", "_____no_output_____" ] ], [ [ "from sklearn.linear_model import LinearRegression\nregressor = LinearRegression() \nregressor.fit(X_train, y_train) \n\nprint('TRAINING COMPLETED.')", "TRAINING COMPLETED.\n" ] ], [ [ "## Predicting the Score", "_____no_output_____" ] ], [ [ "y_predict = regressor.predict(X_test)\nprediction = pd.DataFrame({'Hours': [i[0] for i in X_test], 'Predicted Scores': [k for k in y_predict]})\nprediction", "_____no_output_____" ], [ "print(regressor.intercept_)", "2.813097028256877\n" ], [ "print(regressor.coef_)", "[9.8272078]\n" ], [ "# Plotting the regression line\n\nline = regressor.coef_*X+regressor.intercept_\n\n# Plotting for the test data\n\nplt.scatter(X, y, color = colors[1])\nplt.plot(X, line, color = colors[0]);\nplt.title('Hours vs Percentage') \nplt.xlabel('Hours Studied') \nplt.ylabel('Percentage Score')\nplt.show()", "_____no_output_____" ] ], [ [ "## Checking the Accuracy Scores for training and test set", "_____no_output_____" ] ], [ [ "print('Test Score')\nprint(regressor.score(X_test, y_test))\nprint('Training Score')\nprint(regressor.score(X_train, y_train))", "Test Score\n0.9480612939203932\nTraining Score\n0.953103139564599\n" ] ], [ [ "## Comparing Actual Scores and Predicted Scores", "_____no_output_____" ] ], [ [ "data= pd.DataFrame({'Actual': y_test,'Predicted': y_predict})\ndata", "_____no_output_____" ], [ "# Visualization comparing Actual Scores and Predicted Scores\n\nplt.scatter(X_test, y_test, color = colors[1]) \nplt.plot(X_test, y_predict, color = colors[0]) \nplt.title(\"Hours Studied Vs Percentage (Test Dataset)\") \nplt.xlabel(\"Hour\") \nplt.ylabel(\"Percentage\") \nplt.show()", "_____no_output_____" ] ], [ [ "## Model Evaluation Metrics", "_____no_output_____" ] ], [ [ "#Checking the efficiency of model\n\nfrom sklearn import metrics\nprint('Mean Absolute Error:', metrics.mean_absolute_error(y_test, y_predict))\nprint('Mean Squared Error:', metrics.mean_squared_error(y_test, y_predict))\nprint('Root Mean Squared Error:', np.sqrt(metrics.mean_squared_error(y_test, y_predict)))", "Mean Absolute Error: 4.451593449522768\nMean Squared Error: 25.188194900366135\nRoot Mean Squared Error: 5.018784205399365\n" ] ], [ [ "# What will be predicted score if a student studies for 9.25 hrs/ day?", "_____no_output_____" ] ], [ [ "hours = 9.25\nans = regressor.predict([[hours]])\nprint(\"No of Hours = {}\".format(hours))\nprint(\"Predicted Score = {}\".format(ans[0]))", "No of Hours = 9.25\nPredicted Score = 93.71476919815156\n" ] ] ]

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

[ "BSD-3-Clause" ]

[ "BSD-3-Clause" ]

[ "BSD-3-Clause" ]

[ [ [ "<a href=\"https://www.kaggle.com/egyfirst/detecting-non-groundtruth-datasets?scriptVersionId=86411863\" target=\"_blank\"><img align=\"left\" alt=\"Kaggle\" title=\"Open in Kaggle\" src=\"https://kaggle.com/static/images/open-in-kaggle.svg\"></a>", "_____no_output_____" ] ], [ [ "import pandas as pd\nimport time\nimport os.path\n\nimport warnings\nwarnings.filterwarnings('ignore')", "_____no_output_____" ], [ "# install DenMune clustering algorithm using pip command from the offecial Python repository, PyPi\n# from https://pypi.org/project/denmune/\n!pip install denmune\n\n# now import it\nfrom denmune import DenMune", "Collecting denmune\r\n Downloading denmune-0.0.9.0-py3-none-any.whl (13 kB)\r\nRequirement already satisfied: pandas>=1.0.3 in /opt/conda/lib/python3.7/site-packages (from denmune) (1.3.5)\r\nRequirement already satisfied: seaborn>=0.10.1 in /opt/conda/lib/python3.7/site-packages (from denmune) (0.11.2)\r\nRequirement already satisfied: matplotlib>=3.2.1 in /opt/conda/lib/python3.7/site-packages (from denmune) (3.5.1)\r\nRequirement already satisfied: numpy>=1.18.5 in /opt/conda/lib/python3.7/site-packages (from denmune) (1.20.3)\r\nCollecting anytree>=2.8.0\r\n Downloading anytree-2.8.0-py2.py3-none-any.whl (41 kB)\r\n |████████████████████████████████| 41 kB 98 kB/s \r\n\u001b[?25hRequirement already satisfied: scikit-learn>=0.22.1 in /opt/conda/lib/python3.7/site-packages (from denmune) (0.23.2)\r\nCollecting ngt>=1.11.6\r\n Downloading ngt-1.12.2-cp37-cp37m-manylinux1_x86_64.whl (2.2 MB)\r\n |████████████████████████████████| 2.2 MB 1.5 MB/s \r\n\u001b[?25hCollecting treelib>=1.6.1\r\n Downloading treelib-1.6.1.tar.gz (24 kB)\r\n Preparing metadata (setup.py) ... \u001b[?25l-\b \bdone\r\n\u001b[?25hRequirement already satisfied: six>=1.9.0 in /opt/conda/lib/python3.7/site-packages (from anytree>=2.8.0->denmune) (1.16.0)\r\nRequirement already satisfied: pillow>=6.2.0 in /opt/conda/lib/python3.7/site-packages (from matplotlib>=3.2.1->denmune) (8.2.0)\r\nRequirement already satisfied: packaging>=20.0 in /opt/conda/lib/python3.7/site-packages (from matplotlib>=3.2.1->denmune) (21.3)\r\nRequirement already satisfied: pyparsing>=2.2.1 in /opt/conda/lib/python3.7/site-packages (from matplotlib>=3.2.1->denmune) (3.0.6)\r\nRequirement already satisfied: python-dateutil>=2.7 in /opt/conda/lib/python3.7/site-packages (from matplotlib>=3.2.1->denmune) (2.8.0)\r\nRequirement already satisfied: fonttools>=4.22.0 in /opt/conda/lib/python3.7/site-packages (from matplotlib>=3.2.1->denmune) (4.28.4)\r\nRequirement already satisfied: cycler>=0.10 in /opt/conda/lib/python3.7/site-packages (from matplotlib>=3.2.1->denmune) (0.11.0)\r\nRequirement already satisfied: kiwisolver>=1.0.1 in /opt/conda/lib/python3.7/site-packages (from matplotlib>=3.2.1->denmune) (1.3.2)\r\nRequirement already satisfied: pybind11 in /opt/conda/lib/python3.7/site-packages (from ngt>=1.11.6->denmune) (2.9.0)\r\nRequirement already satisfied: pytz>=2017.3 in /opt/conda/lib/python3.7/site-packages (from pandas>=1.0.3->denmune) (2021.3)\r\nRequirement already satisfied: scipy>=0.19.1 in /opt/conda/lib/python3.7/site-packages (from scikit-learn>=0.22.1->denmune) (1.7.3)\r\nRequirement already satisfied: threadpoolctl>=2.0.0 in /opt/conda/lib/python3.7/site-packages (from scikit-learn>=0.22.1->denmune) (3.0.0)\r\nRequirement already satisfied: joblib>=0.11 in /opt/conda/lib/python3.7/site-packages (from scikit-learn>=0.22.1->denmune) (1.1.0)\r\nRequirement already satisfied: future in /opt/conda/lib/python3.7/site-packages (from treelib>=1.6.1->denmune) (0.18.2)\r\nBuilding wheels for collected packages: treelib\r\n Building wheel for treelib (setup.py) ... \u001b[?25l-\b \b\\\b \bdone\r\n\u001b[?25h Created wheel for treelib: filename=treelib-1.6.1-py3-none-any.whl size=18386 sha256=c9fc88fcc62ccee93209a9c6ee2a4bbe17bb1537e6c3c43bc701bb73b4f3f75f\r\n Stored in directory: /root/.cache/pip/wheels/89/be/94/2c6d949ce599d1443426d83ba4dc93cd35c0f4638260930a53\r\nSuccessfully built treelib\r\nInstalling collected packages: treelib, ngt, anytree, denmune\r\nSuccessfully installed anytree-2.8.0 denmune-0.0.9.0 ngt-1.12.2 treelib-1.6.1\r\n\u001b[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\u001b[0m\r\n" ], [ "# clone datasets from our repository datasets\nif not os.path.exists('datasets'):\n !git clone https://github.com/egy1st/datasets", "Cloning into 'datasets'...\r\nremote: Enumerating objects: 52, done.\u001b[K\r\nremote: Counting objects: 100% (52/52), done.\u001b[K\r\nremote: Compressing objects: 100% (43/43), done.\u001b[K\r\nremote: Total 52 (delta 8), reused 49 (delta 8), pack-reused 0\u001b[K\r\nUnpacking objects: 100% (52/52), 20.40 MiB | 6.04 MiB/s, done.\r\n" ], [ "data_path = 'datasets/denmune/chameleon/' \ndatasets = [\"t7.10k\", \"t4.8k\", \"t5.8k\", \"t8.8k\"]\n\nfor dataset in datasets:\n data_file = data_path + dataset + '.csv'\n X_train = pd.read_csv(data_file, sep=',', header=None)\n\n dm = DenMune(train_data=X_train, k_nearest=39, rgn_tsne=False)\n labels, validity = dm.fit_predict(show_noise=True, show_analyzer=True)\n", "Plotting train data\n" ], [ "dataset = \"clusterable\"\ndata_file = data_path + dataset + '.csv'\nX_train = pd.read_csv(data_file, sep=',', header=None)\n\ndm = DenMune(train_data=X_train, k_nearest=24, rgn_tsne=False)\nlabels, validity = dm.fit_predict(show_noise=True, show_analyzer=True)\n", "Plotting train data\n" ] ] ]

[ "markdown", "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/yohanesnuwara/ccs-gundih/blob/master/main/gundih_historical_production_data.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ] ], [ [ "import numpy as np\nimport matplotlib.pyplot as plt\nimport pandas as pd", "_____no_output_____" ], [ "!git clone https://github.com/yohanesnuwara/ccs-gundih", "Cloning into 'ccs-gundih'...\nremote: Enumerating objects: 94, done.\u001b[K\nremote: Counting objects: 1% (1/94)\u001b[K\rremote: Counting objects: 2% (2/94)\u001b[K\rremote: Counting objects: 3% (3/94)\u001b[K\rremote: Counting objects: 4% (4/94)\u001b[K\rremote: Counting objects: 5% (5/94)\u001b[K\rremote: Counting objects: 6% (6/94)\u001b[K\rremote: Counting objects: 7% (7/94)\u001b[K\rremote: Counting objects: 8% (8/94)\u001b[K\rremote: Counting objects: 9% (9/94)\u001b[K\rremote: Counting objects: 10% (10/94)\u001b[K\rremote: Counting objects: 11% (11/94)\u001b[K\rremote: Counting objects: 12% (12/94)\u001b[K\rremote: Counting objects: 13% (13/94)\u001b[K\rremote: Counting objects: 14% (14/94)\u001b[K\rremote: Counting objects: 15% (15/94)\u001b[K\rremote: Counting objects: 17% (16/94)\u001b[K\rremote: Counting objects: 18% (17/94)\u001b[K\rremote: Counting objects: 19% (18/94)\u001b[K\rremote: Counting objects: 20% (19/94)\u001b[K\rremote: Counting objects: 21% (20/94)\u001b[K\rremote: Counting objects: 22% (21/94)\u001b[K\rremote: Counting objects: 23% (22/94)\u001b[K\rremote: Counting objects: 24% (23/94)\u001b[K\rremote: Counting objects: 25% (24/94)\u001b[K\rremote: Counting objects: 26% (25/94)\u001b[K\rremote: Counting objects: 27% (26/94)\u001b[K\rremote: Counting objects: 28% (27/94)\u001b[K\rremote: Counting objects: 29% (28/94)\u001b[K\rremote: Counting objects: 30% (29/94)\u001b[K\rremote: Counting objects: 31% (30/94)\u001b[K\rremote: Counting objects: 32% (31/94)\u001b[K\rremote: Counting objects: 34% (32/94)\u001b[K\rremote: Counting objects: 35% (33/94)\u001b[K\rremote: Counting objects: 36% (34/94)\u001b[K\rremote: Counting objects: 37% (35/94)\u001b[K\rremote: Counting objects: 38% (36/94)\u001b[K\rremote: Counting objects: 39% (37/94)\u001b[K\rremote: Counting objects: 40% (38/94)\u001b[K\rremote: Counting objects: 41% (39/94)\u001b[K\rremote: Counting objects: 42% (40/94)\u001b[K\rremote: Counting objects: 43% (41/94)\u001b[K\rremote: Counting objects: 44% (42/94)\u001b[K\rremote: Counting objects: 45% (43/94)\u001b[K\rremote: Counting objects: 46% (44/94)\u001b[K\rremote: Counting objects: 47% (45/94)\u001b[K\rremote: Counting objects: 48% (46/94)\u001b[K\rremote: Counting objects: 50% (47/94)\u001b[K\rremote: Counting objects: 51% (48/94)\u001b[K\rremote: Counting objects: 52% (49/94)\u001b[K\rremote: Counting objects: 53% (50/94)\u001b[K\rremote: Counting objects: 54% (51/94)\u001b[K\rremote: Counting objects: 55% (52/94)\u001b[K\rremote: Counting objects: 56% (53/94)\u001b[K\rremote: Counting objects: 57% (54/94)\u001b[K\rremote: Counting objects: 58% (55/94)\u001b[K\rremote: Counting objects: 59% (56/94)\u001b[K\rremote: Counting objects: 60% (57/94)\u001b[K\rremote: Counting objects: 61% (58/94)\u001b[K\rremote: Counting objects: 62% (59/94)\u001b[K\rremote: Counting objects: 63% (60/94)\u001b[K\rremote: Counting objects: 64% (61/94)\u001b[K\rremote: Counting objects: 65% (62/94)\u001b[K\rremote: Counting objects: 67% (63/94)\u001b[K\rremote: Counting objects: 68% (64/94)\u001b[K\rremote: Counting objects: 69% (65/94)\u001b[K\rremote: Counting objects: 70% (66/94)\u001b[K\rremote: Counting objects: 71% (67/94)\u001b[K\rremote: Counting objects: 72% (68/94)\u001b[K\rremote: Counting objects: 73% (69/94)\u001b[K\rremote: Counting objects: 74% (70/94)\u001b[K\rremote: Counting objects: 75% (71/94)\u001b[K\rremote: Counting objects: 76% (72/94)\u001b[K\rremote: Counting objects: 77% (73/94)\u001b[K\rremote: Counting objects: 78% (74/94)\u001b[K\rremote: Counting objects: 79% (75/94)\u001b[K\rremote: Counting objects: 80% (76/94)\u001b[K\rremote: Counting objects: 81% (77/94)\u001b[K\rremote: Counting objects: 82% (78/94)\u001b[K\rremote: Counting objects: 84% (79/94)\u001b[K\rremote: Counting objects: 85% (80/94)\u001b[K\rremote: Counting objects: 86% (81/94)\u001b[K\rremote: Counting objects: 87% (82/94)\u001b[K\rremote: Counting objects: 88% (83/94)\u001b[K\rremote: Counting objects: 89% (84/94)\u001b[K\rremote: Counting objects: 90% (85/94)\u001b[K\rremote: Counting objects: 91% (86/94)\u001b[K\rremote: Counting objects: 92% (87/94)\u001b[K\rremote: Counting objects: 93% (88/94)\u001b[K\rremote: Counting objects: 94% (89/94)\u001b[K\rremote: Counting objects: 95% (90/94)\u001b[K\rremote: Counting objects: 96% (91/94)\u001b[K\rremote: Counting objects: 97% (92/94)\u001b[K\rremote: Counting objects: 98% (93/94)\u001b[K\rremote: Counting objects: 100% (94/94)\u001b[K\rremote: Counting objects: 100% (94/94), done.\u001b[K\nremote: Compressing objects: 100% (94/94), done.\u001b[K\nremote: Total 259 (delta 55), reused 0 (delta 0), pack-reused 165\u001b[K\nReceiving objects: 100% (259/259), 14.37 MiB | 26.09 MiB/s, done.\nResolving deltas: 100% (136/136), done.\n" ] ], [ [ "# Visualize the Historical Production Data", "_____no_output_____" ] ], [ [ "# Read simulation result\n\ncol = np.array(['Date', 'Days', 'WSAT', 'OSAT', 'GSAT', 'GMT', 'OMR', 'GMR', 'GCDI', 'GCDM', 'WCD',\n 'WGR', 'WCT', 'VPR', 'VPT', 'VIR', 'VIT', 'WPR', 'OPR', 'GPR', 'WPT', 'OPT', 'GPT',\n 'PR', 'GIR', 'GIT', 'GOR'])\n\ncase1 = pd.read_excel(r'/content/ccs-gundih/data/CASE_1.xlsx'); case1 = pd.DataFrame(case1, columns=col) #INJ-2, 15 MMSCFD, kv/kh = 0.1\ncase2 = pd.read_excel(r'/content/ccs-gundih/data/CASE_2.xlsx'); case2 = pd.DataFrame(case2, columns=col) #INJ-2, 15 MMSCFD, kv/kh = 0.5", "_____no_output_____" ], [ "case1.head(5)", "_____no_output_____" ], [ "case2.head(5)", "_____no_output_____" ], [ "# convert to Panda datetime\ndate = pd.to_datetime(case1['Date'])\n\n# gas production cumulative \nGPT1 = case1['GPT'] * 35.31 * 1E-06 # convert from m3 to ft3 then to mmscf\nGPT2 = case2['GPT'] * 35.31 * 1E-06\n\n# gas production rate\nGPR1 = case1['GPR'] * 35.31 * 1E-06\nGPR2 = case2['GPR'] * 35.31 * 1E-06\n\n# average pressure\nPR1 = case1['PR'] * 14.5038 # convert from bar to psi\nPR2 = case2['PR'] * 14.5038 ", "_____no_output_____" ], [ "# plot gas production and pressure data from 2014-01-01 to 2016-01-01\n\npd.plotting.register_matplotlib_converters()\n\nplt.figure(figsize=(12, 7))\nplt.plot(date, GPR1, color='blue')\nplt.plot(date, GPR2, color='red')\nplt.xlabel(\"Date\"); plt.ylabel(\"Cumulative Gas Production (MMscf)\")\nplt.title(\"Cumulative Gas Production from 01/01/2014 to 01/01/2019\", pad=20, size=15)\nplt.xlim('2014-01-01', '2019-01-01')\n# plt.ylim(0, 50000)", "_____no_output_____" ], [ "# plot gas production and pressure data from 2014-01-01 to 2016-01-01\n\npd.plotting.register_matplotlib_converters()\n\nfig = plt.figure()\nfig = plt.figure(figsize=(12,7))\nhost = fig.add_subplot(111)\n\npar1 = host.twinx()\npar2 = host.twinx()\n\nhost.set_xlabel(\"Year\")\nhost.set_ylabel(\"Cumulative Gas Production (MMscf)\")\npar1.set_ylabel(\"Gas Production Rate (MMscfd)\")\npar2.set_ylabel(\"Average Reservoir Pressure (psi)\")\n\nhost.set_title(\"Historical Production Data of Gundih Field from 2014 to 2019\", pad=20, size=15)\n\ncolor1 = plt.cm.viridis(0)\ncolor2 = plt.cm.viridis(.5)\ncolor3 = plt.cm.viridis(.8)\n\np1, = host.plot(date, GPR1, color=color1,label=\"Gas production rate (MMscfd)\")\np2, = par1.plot(date, GPT1, color=color2, label=\"Cumulative gas production (MMscf)\")\np3, = par2.plot(date, PR1, color=color3, label=\"Average Pressure (psi)\")\nhost.set_xlim('2014-05-01', '2019-01-01')\nhost.set_ylim(ymin=0)\npar1.set_ylim(0, 40000)\npar2.set_ylim(3400, 4100)\n\nlns = [p1, p2, p3]\nhost.legend(handles=lns, loc='best')\n\n# right, left, top, bottom\npar2.spines['right'].set_position(('outward', 60)) \n\nplt.savefig('/content/ccs-gundih/result/production_curve')", "_____no_output_____" ] ], [ [ "# Dry-Gas Reservoir Analysis", "_____no_output_____" ] ], [ [ "!git clone https://github.com/yohanesnuwara/reservoir-engineering", "Cloning into 'reservoir-engineering'...\nremote: Enumerating objects: 32, done.\u001b[K\nremote: Counting objects: 3% (1/32)\u001b[K\rremote: Counting objects: 6% (2/32)\u001b[K\rremote: Counting objects: 9% (3/32)\u001b[K\rremote: Counting objects: 12% (4/32)\u001b[K\rremote: Counting objects: 15% (5/32)\u001b[K\rremote: Counting objects: 18% (6/32)\u001b[K\rremote: Counting objects: 21% (7/32)\u001b[K\rremote: Counting objects: 25% (8/32)\u001b[K\rremote: Counting objects: 28% (9/32)\u001b[K\rremote: Counting objects: 31% (10/32)\u001b[K\rremote: Counting objects: 34% (11/32)\u001b[K\rremote: Counting objects: 37% (12/32)\u001b[K\rremote: Counting objects: 40% (13/32)\u001b[K\rremote: Counting objects: 43% (14/32)\u001b[K\rremote: Counting objects: 46% (15/32)\u001b[K\rremote: Counting objects: 50% (16/32)\u001b[K\rremote: Counting objects: 53% (17/32)\u001b[K\rremote: Counting objects: 56% (18/32)\u001b[K\rremote: Counting objects: 59% (19/32)\u001b[K\rremote: Counting objects: 62% (20/32)\u001b[K\rremote: Counting objects: 65% (21/32)\u001b[K\rremote: Counting objects: 68% (22/32)\u001b[K\rremote: Counting objects: 71% (23/32)\u001b[K\rremote: Counting objects: 75% (24/32)\u001b[K\rremote: Counting objects: 78% (25/32)\u001b[K\rremote: Counting objects: 81% (26/32)\u001b[K\rremote: Counting objects: 84% (27/32)\u001b[K\rremote: Counting objects: 87% (28/32)\u001b[K\rremote: Counting objects: 90% (29/32)\u001b[K\rremote: Counting objects: 93% (30/32)\u001b[K\rremote: Counting objects: 96% (31/32)\u001b[K\rremote: Counting objects: 100% (32/32)\u001b[K\rremote: Counting objects: 100% (32/32), done.\u001b[K\nremote: Compressing objects: 3% (1/31)\u001b[K\rremote: Compressing objects: 6% (2/31)\u001b[K\rremote: Compressing objects: 9% (3/31)\u001b[K\rremote: Compressing objects: 12% (4/31)\u001b[K\rremote: Compressing objects: 16% (5/31)\u001b[K\rremote: Compressing objects: 19% (6/31)\u001b[K\rremote: Compressing objects: 22% (7/31)\u001b[K\rremote: Compressing objects: 25% (8/31)\u001b[K\rremote: Compressing objects: 29% (9/31)\u001b[K\rremote: Compressing objects: 32% (10/31)\u001b[K\rremote: Compressing objects: 35% (11/31)\u001b[K\rremote: Compressing objects: 38% (12/31)\u001b[K\rremote: Compressing objects: 41% (13/31)\u001b[K\rremote: Compressing objects: 45% (14/31)\u001b[K\rremote: Compressing objects: 48% (15/31)\u001b[K\rremote: Compressing objects: 51% (16/31)\u001b[K\rremote: Compressing objects: 54% (17/31)\u001b[K\rremote: Compressing objects: 58% (18/31)\u001b[K\rremote: Compressing objects: 61% (19/31)\u001b[K\rremote: Compressing objects: 64% (20/31)\u001b[K\rremote: Compressing objects: 67% (21/31)\u001b[K\rremote: Compressing objects: 70% (22/31)\u001b[K\rremote: Compressing objects: 74% (23/31)\u001b[K\rremote: Compressing objects: 77% (24/31)\u001b[K\rremote: Compressing objects: 80% (25/31)\u001b[K\rremote: Compressing objects: 83% (26/31)\u001b[K\rremote: Compressing objects: 87% (27/31)\u001b[K\rremote: Compressing objects: 90% (28/31)\u001b[K\rremote: Compressing objects: 93% (29/31)\u001b[K\rremote: Compressing objects: 96% (30/31)\u001b[K\rremote: Compressing objects: 100% (31/31)\u001b[K\rremote: Compressing objects: 100% (31/31), done.\u001b[K\nReceiving objects: 0% (1/941) \rReceiving objects: 1% (10/941) \rReceiving objects: 2% (19/941) \rReceiving objects: 3% (29/941) \rReceiving objects: 4% (38/941) \rReceiving objects: 5% (48/941) \rReceiving objects: 6% (57/941) \rReceiving objects: 7% (66/941) \rReceiving objects: 8% (76/941) \rReceiving objects: 9% (85/941) \rReceiving objects: 10% (95/941) \rReceiving objects: 11% (104/941) \rReceiving objects: 12% (113/941) \rReceiving objects: 13% (123/941) \rReceiving objects: 14% (132/941) \rReceiving objects: 15% (142/941) \rReceiving objects: 16% (151/941) \rReceiving objects: 17% (160/941) \rReceiving objects: 18% (170/941) \rReceiving objects: 19% (179/941) \rReceiving objects: 20% (189/941) \rReceiving objects: 21% (198/941) \rReceiving objects: 22% (208/941) \rReceiving objects: 23% (217/941) \rReceiving objects: 24% (226/941) \rReceiving objects: 25% (236/941) \rReceiving objects: 26% (245/941) \rReceiving objects: 27% (255/941) \rReceiving objects: 28% (264/941) \rReceiving objects: 29% (273/941) \rReceiving objects: 30% (283/941) \rReceiving objects: 31% (292/941) \rReceiving objects: 32% (302/941) \rReceiving objects: 33% (311/941) \rReceiving objects: 34% (320/941) \rReceiving objects: 35% (330/941) \rReceiving objects: 36% (339/941) \rReceiving objects: 37% (349/941) \rReceiving objects: 38% (358/941) \rReceiving objects: 39% (367/941) \rReceiving objects: 40% (377/941) \rReceiving objects: 41% (386/941) \rReceiving objects: 42% (396/941) \rReceiving objects: 43% (405/941) \rReceiving objects: 44% (415/941) \rReceiving objects: 45% (424/941) \rReceiving objects: 46% (433/941) \rReceiving objects: 47% (443/941) \rReceiving objects: 48% (452/941) \rReceiving objects: 49% (462/941) \rReceiving objects: 50% (471/941) \rReceiving objects: 51% (480/941) \rReceiving objects: 52% (490/941) \rReceiving objects: 53% (499/941) \rReceiving objects: 54% (509/941) \rReceiving objects: 55% (518/941) \rReceiving objects: 56% (527/941) \rReceiving objects: 57% (537/941) \rReceiving objects: 58% (546/941) \rReceiving objects: 59% (556/941) \rReceiving objects: 60% (565/941) \rReceiving objects: 61% (575/941) \rReceiving objects: 62% (584/941) \rReceiving objects: 63% (593/941) \rReceiving objects: 64% (603/941) \rReceiving objects: 65% (612/941) \rReceiving objects: 66% (622/941) \rReceiving objects: 67% (631/941) \rReceiving objects: 68% (640/941) \rReceiving objects: 69% (650/941) \rReceiving objects: 70% (659/941) \rReceiving objects: 71% (669/941) \rReceiving objects: 72% (678/941) \rReceiving objects: 73% (687/941) \rReceiving objects: 74% (697/941) \rReceiving objects: 75% (706/941) \rReceiving objects: 76% (716/941) \rReceiving objects: 77% (725/941) \rReceiving objects: 78% (734/941) \rReceiving objects: 79% (744/941) \rReceiving objects: 80% (753/941) \rReceiving objects: 81% (763/941) \rReceiving objects: 82% (772/941) \rReceiving objects: 83% (782/941) \rReceiving objects: 84% (791/941) \rReceiving objects: 85% (800/941) \rReceiving objects: 86% (810/941) \rReceiving objects: 87% (819/941) \rReceiving objects: 88% (829/941) \rReceiving objects: 89% (838/941) \rReceiving objects: 90% (847/941) \rReceiving objects: 91% (857/941) \rReceiving objects: 92% (866/941) \rReceiving objects: 93% (876/941) \rReceiving objects: 94% (885/941) \rReceiving objects: 95% (894/941) \rReceiving objects: 96% (904/941) \rReceiving objects: 97% (913/941) \rReceiving objects: 98% (923/941) \rremote: Total 941 (delta 15), reused 0 (delta 0), pack-reused 909\u001b[K\nReceiving objects: 99% (932/941) \rReceiving objects: 100% (941/941) \rReceiving objects: 100% (941/941), 12.23 MiB | 45.56 MiB/s, done.\nResolving deltas: 0% (0/410) \rResolving deltas: 1% (5/410) \rResolving deltas: 3% (16/410) \rResolving deltas: 6% (26/410) \rResolving deltas: 16% (69/410) \rResolving deltas: 18% (75/410) \rResolving deltas: 20% (85/410) \rResolving deltas: 27% (113/410) \rResolving deltas: 35% (147/410) \rResolving deltas: 41% (170/410) \rResolving deltas: 61% (252/410) \rResolving deltas: 63% (261/410) \rResolving deltas: 67% (276/410) \rResolving deltas: 68% (280/410) \rResolving deltas: 69% (285/410) \rResolving deltas: 71% (292/410) \rResolving deltas: 72% (296/410) \rResolving deltas: 73% (300/410) \rResolving deltas: 74% (304/410) \rResolving deltas: 75% (311/410) \rResolving deltas: 76% (313/410) \rResolving deltas: 77% (317/410) \rResolving deltas: 78% (320/410) \rResolving deltas: 79% (324/410) \rResolving deltas: 83% (342/410) \rResolving deltas: 84% (345/410) \rResolving deltas: 85% (352/410) \rResolving deltas: 86% (353/410) \rResolving deltas: 87% (357/410) \rResolving deltas: 88% (362/410) \rResolving deltas: 89% (368/410) \rResolving deltas: 90% (371/410) \rResolving deltas: 91% (375/410) \rResolving deltas: 92% (380/410) \rResolving deltas: 93% (382/410) \rResolving deltas: 94% (386/410) \rResolving deltas: 95% (390/410) \rResolving deltas: 96% (397/410) \rResolving deltas: 97% (398/410) \rResolving deltas: 98% (402/410) \rResolving deltas: 100% (410/410) \rResolving deltas: 100% (410/410), done.\n" ], [ "# calculate gas z factor and FVF\n\nimport os, sys\nsys.path.append('/content/reservoir-engineering/Unit 2 Review of Rock and Fluid Properties/functions')\n\nfrom pseudoprops import pseudoprops\nfrom dranchuk_aboukassem import dranchuk\nfrom gasfvf import gasfvf\n\ntemp_f = (temp * 9/5) + 32 # Rankine\npressure = np.array(PR1)\n\nz_arr = []\nBg_arr = []\nfor i in range(len(pressure)):\n P_pr, T_pr = pseudoprops(temp_f, pressure[i], 0.8, 0.00467, 0.23)\n rho_pr, z = dranchuk(T_pr, P_pr)\n temp_r = temp_f + 459.67 \n Bg = 0.0282793 * z * temp_r / pressure[i] # Eq 2.2, temp in Rankine, p in psia, result in res ft3/scf\n z_arr.append(float(z))\n Bg_arr.append(float(Bg))", "_____no_output_____" ], [ "Bg_arr = np.array(Bg_arr)\nF = GPT1 * Bg_arr # MMscf\nEg = Bg_arr - Bg_arr[0]\nF_Eg = F / Eg\n\nplt.figure(figsize=(10,7))\nplt.plot(GPT1, (F_Eg / 1E+03), '.', color='red') # convert F_Eg from MMscf to Bscf\nplt.xlim(xmin=0); plt.ylim(ymin=0)\nplt.title(\"Waterdrive Diagnostic Plot of $F/E_g$ vs. $G_p$\", pad=20, size=15)\nplt.xlabel('Cumulative Gas Production (MMscf)')\nplt.ylabel('$F/E_g$ (Bscf)')\nplt.ylim(ymin=250)", "_____no_output_____" ], [ "date_hist = date.iloc[:1700]\nGPT1_hist = GPT1.iloc[:1700]\nBg_arr = np.array(Bg_arr)\nBg_arr_hist = Bg_arr[:1700]\n\nF_hist = GPT1_hist * Bg_arr_hist # MMscf\nEg_hist = Bg_arr_hist - Bg_arr_hist[0]\nF_Eg_hist = F_hist / Eg_hist\n\nplt.figure(figsize=(10,7))\nplt.plot(GPT1_hist, (F_Eg_hist / 1E+03), '.')\nplt.title(\"Waterdrive Diagnostic Plot of $F/E_g$ vs. $G_p$\", pad=20, size=15)\nplt.xlabel('Cumulative Gas Production (MMscf)')\nplt.ylabel('$F/E_g$ (Bscf)')\nplt.xlim(xmin=0); plt.ylim(300, 350)", "_____no_output_____" ], [ "p_z = PR1 / z_arr\nplt.plot(GPT1, p_z, '.')", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# 1A.1 - Deviner un nombre aléatoire (correction)\n\nOn reprend la fonction introduite dans l'énoncé et qui permet de saisir un nombre.", "_____no_output_____" ] ], [ [ "import random", "_____no_output_____" ], [ "nombre = input(\"Entrez un nombre\")", "Entrez un nombre7\n" ], [ "nombre", "_____no_output_____" ] ], [ [ "**Q1 :** Ecrire une jeu dans lequel python choisi aléatoirement un nombre entre 0 et 100, et essayer de trouver ce nombre en 10 étapes.", "_____no_output_____" ] ], [ [ "n = random.randint(0,100)\nappreciation = \"?\"\nwhile True:\n var = input(\"Entrez un nombre\")\n var = int(var)\n if var < n : \n appreciation = \"trop bas\"\n print(var, appreciation)\n \n else : \n appreciation = \"trop haut\"\n print(var, appreciation)\n if var == n:\n appreciation = \"bravo !\"\n print(var, appreciation)\n break", "Entrez un nombre7\n7 trop bas\nEntrez un nombre10\n10 trop bas\nEntrez un nombre100\n100 trop haut\nEntrez un nombre50\n50 trop bas\nEntrez un nombre75\n75 trop haut\nEntrez un nombre60\n60 trop haut\nEntrez un nombre55\n55 trop haut\nEntrez un nombre53\n53 trop haut\nEntrez un nombre52\n52 trop haut\nEntrez un nombre51\n51 trop haut\n51 bravo !\n" ] ], [ [ "**Q2 :** Transformer ce jeu en une fonction ``jeu(nVies)`` où ``nVies`` est le nombre d'itérations maximum.", "_____no_output_____" ] ], [ [ "import random\nn = random.randint(0,100)\nvies = 10\nappreciation = \"?\"\nwhile vies > 0:\n var = input(\"Entrez un nombre\")\n var = int(var)\n if var < n : \n appreciation = \"trop bas\"\n print(vies, var, appreciation)\n else : \n appreciation = \"trop haut\"\n print(vies, var, appreciation)\n if var == n:\n appreciation = \"bravo !\"\n print(vies, var, appreciation)\n break\n\n vies -= 1", "Entrez un nombre50\n10 50 trop bas\nEntrez un nombre75\n9 75 trop haut\nEntrez un nombre60\n8 60 trop bas\nEntrez un nombre65\n7 65 trop bas\nEntrez un nombre70\n6 70 trop bas\nEntrez un nombre73\n5 73 trop haut\nEntrez un nombre72\n4 72 trop haut\nEntrez un nombre71\n3 71 trop haut\n3 71 bravo !\n" ] ], [ [ "**Q3 :** Adapter le code pour faire une classe joueur avec une méthode jouer, où un joueur est défini par un pseudo et son nombre de vies. Faire jouer deux joueurs et déterminer le vainqueur.", "_____no_output_____" ] ], [ [ "class joueur:\n def __init__(self, vies, pseudo):\n self.vies = vies\n self.pseudo = pseudo\n \n def jouer(self):\n appreciation = \"?\"\n n = random.randint(0,100)\n while self.vies > 0:\n message = appreciation + \" -- \" + self.pseudo + \" : \" + str(self.vies) + \" vies restantes. Nombre choisi : \"\n var = input(message)\n var = int(var)\n if var < n : \n appreciation = \"trop bas\"\n print(vies, var, appreciation)\n else : \n appreciation = \"trop haut\"\n print(vies, var, appreciation)\n if var == n:\n appreciation = \"bravo !\"\n print(vies, var, appreciation)\n break\n\n self.vies -= 1\n\n# Initialisation des deux joueurs\nj1 = joueur(10, \"joueur 1\")\nj2 = joueur(10, \"joueur 2\")\n\n# j1 et j2 jouent\nj1.jouer()\nj2.jouer()\n\n# Nombre de vies restantes à chaque joueur\nprint(\"Nombre de vies restantes à chaque joueur\")\nprint(j1.pseudo + \" : \" + str(j1.vies) + \" restantes\")\nprint(j2.pseudo + \" : \" + str(j2.vies) + \" restantes\")\n\n# Résultat de la partie\nprint(\"Résultat de la partie\")\nif j1.vies < j2.vies:\n print(j1.pseudo + \"a gagné la partie\")\nelif j1.vies == j2.vies:\n print(\"match nul\")\nelse: print(j2.pseudo + \" a gagné la partie\")", "? -- joueur 1 : 10 vies restantes. Nombre choisi : 50\n3 50 trop haut\ntrop haut -- joueur 1 : 9 vies restantes. Nombre choisi : 25\n3 25 trop haut\ntrop haut -- joueur 1 : 8 vies restantes. Nombre choisi : 10\n3 10 trop haut\ntrop haut -- joueur 1 : 7 vies restantes. Nombre choisi : 5\n3 5 trop bas\ntrop bas -- joueur 1 : 6 vies restantes. Nombre choisi : 7\n3 7 trop haut\n3 7 bravo !\n? -- joueur 2 : 10 vies restantes. Nombre choisi : 50\n3 50 trop haut\ntrop haut -- joueur 2 : 9 vies restantes. Nombre choisi : 25\n3 25 trop bas\ntrop bas -- joueur 2 : 8 vies restantes. Nombre choisi : 30\n3 30 trop haut\n3 30 bravo !\nNombre de vies restantes à chaque joueur\njoueur 1 : 6 restantes\njoueur 2 : 8 restantes\nRésultat de la partie\njoueur 1a gagné la partie\n" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import sys\nimport os\nimport numpy as np\nmodule_path = os.path.abspath(os.path.join('..'))\nif module_path not in sys.path:\n sys.path.append(module_path + \"/src/simulations_v2\")\n\nfrom analysis_helpers import poisson_waiting_function, \\\n run_multiple_trajectories, \\\n plot_aip_vs_t, \\\n plot_cip_vs_t, \\\n run_sensitivity_sims, \\\n extract_cips", "_____no_output_____" ], [ "# what percent of self-reports are from severe symptoms? \n# in reality I think this value will vary a lot in the first few days, \n# and then reach some kind of steady-state, and I'm not sure what makes the most\n# sense to use here. I am setting it to the very pessimistic value of 100% of\n# self-reporters are severe, which yields the smallest infectious window size\npct_self_reports_severe = 0.6\n\ndaily_self_report_severe = 0.85\ndaily_self_report_mild = 0.1\n\n# avg_infectious_window = (avg time in ID state) + (avg time in Sy state prior to self-reporting)\navg_infectious_window = 4 + pct_self_reports_severe * (1 / daily_self_report_severe) + \\\n (1-pct_self_reports_severe) * (1 / daily_self_report_mild)\nprint(avg_infectious_window)\npre_reopen_population = 1500\npre_reopen_daily_contacts = 7\n\nreopen_population = 2500\nreopen_daily_contacts = 10\n\npre_reopen_params = {\n 'max_time_exposed': 4,\n 'exposed_time_function': poisson_waiting_function(max_time=4, mean_time=1),\n \n 'max_time_pre_ID': 4,\n 'pre_ID_time_function': poisson_waiting_function(max_time=4, mean_time=1),\n \n 'max_time_ID': 8,\n 'ID_time_function': poisson_waiting_function(max_time=8, mean_time=4),\n \n 'max_time_SyID_mild': 14,\n 'SyID_mild_time_function': poisson_waiting_function(max_time=14, mean_time=10),\n \n 'max_time_SyID_severe': 14,\n 'SyID_severe_time_function': poisson_waiting_function(max_time=14, mean_time=10),\n \n 'sample_QI_exit_function': (lambda n: np.random.binomial(n, 0.05)),\n 'sample_QS_exit_function': (lambda n: np.random.binomial(n, 0.3)),\n \n 'exposed_infection_p': 0.026,\n 'expected_contacts_per_day': pre_reopen_daily_contacts,\n \n 'mild_symptoms_p': 0.4,\n 'mild_symptoms_daily_self_report_p': daily_self_report_mild,\n 'severe_symptoms_daily_self_report_p': daily_self_report_severe,\n \n 'days_between_tests': 300,\n 'test_population_fraction': 0,\n \n 'test_protocol_QFNR': 0.1,\n 'test_protocol_QFPR': 0.005,\n \n 'perform_contact_tracing': True,\n 'contact_tracing_constant': 0.5,\n 'contact_tracing_delay': 1,\n 'contact_trace_infectious_window': avg_infectious_window,\n \n 'pre_ID_state': 'detectable',\n \n 'population_size': pre_reopen_population,\n 'initial_E_count': 0,\n 'initial_pre_ID_count': 2,\n 'initial_ID_count': 0,\n 'initial_ID_prevalence': 0.001,\n 'initial_SyID_mild_count': 0,\n 'initial_SyID_severe_count': 0\n}\n\nreopen_params = pre_reopen_params.copy()\nreopen_params['population_size'] = reopen_population\nreopen_params['expected_contacts_per_day'] = reopen_daily_contacts", "8.705882352941178\n" ] ], [ [ "# Run sims to understand sensitivity of 'contact_tracing_constant'", "_____no_output_____" ] ], [ [ "ctc_range = [0.1 * x for x in range(11)]\ndfs_ctc_pre_reopen = run_sensitivity_sims(pre_reopen_params, param_to_vary='contact_tracing_constant',\n param_values = ctc_range, trajectories_per_config=250, time_horizon=100)\ndfs_ctc_post_reopen = run_sensitivity_sims(reopen_params, param_to_vary='contact_tracing_constant',\n param_values = ctc_range, trajectories_per_config=250, time_horizon=100)", "Done simulating contact_tracing_constant equal to 0.0\nDone simulating contact_tracing_constant equal to 0.1\nDone simulating contact_tracing_constant equal to 0.2\nDone simulating contact_tracing_constant equal to 0.30000000000000004\nDone simulating contact_tracing_constant equal to 0.4\nDone simulating contact_tracing_constant equal to 0.5\nDone simulating contact_tracing_constant equal to 0.6000000000000001\nDone simulating contact_tracing_constant equal to 0.7000000000000001\nDone simulating contact_tracing_constant equal to 0.8\nDone simulating contact_tracing_constant equal to 0.9\nDone simulating contact_tracing_constant equal to 1.0\nDone simulating contact_tracing_constant equal to 0.0\nDone simulating contact_tracing_constant equal to 0.1\nDone simulating contact_tracing_constant equal to 0.2\nDone simulating contact_tracing_constant equal to 0.30000000000000004\nDone simulating contact_tracing_constant equal to 0.4\nDone simulating contact_tracing_constant equal to 0.5\nDone simulating contact_tracing_constant equal to 0.6000000000000001\nDone simulating contact_tracing_constant equal to 0.7000000000000001\nDone simulating contact_tracing_constant equal to 0.8\nDone simulating contact_tracing_constant equal to 0.9\nDone simulating contact_tracing_constant equal to 1.0\n" ], [ "import matplotlib.pyplot as plt\ndef plot_many_dfs_threshold(dfs_dict, threshold=0.1, xlabel=\"\", title=\"\", figsize=(10,6)):\n plt.figure(figsize=figsize)\n for df_label, dfs_varied in dfs_dict.items():\n p_thresholds = []\n xs = sorted(list(dfs_varied.keys()))\n for x in xs:\n cips = extract_cips(dfs_varied[x])\n cip_exceed_thresh = [cip for cip in cips if cip >= threshold]\n p_thresholds.append(len(cip_exceed_thresh) / len(cips) * 100)\n plt.plot([x * 100 for x in xs], p_thresholds, marker='o', label=df_label)\n plt.xlabel(xlabel)\n plt.ylabel(\"Probability at least {:.0f}% infected (%)\".format(threshold * 100))\n plt.title(title)\n plt.legend(loc='best')\n plt.show()\n\ntitle = \"\"\"Outbreak Likelihood vs. Contact Tracing Effectiveness\"\"\"\nplot_many_dfs_threshold({'Post-Reopen (Population-size 2500, Contacts/person/day 10)': dfs_ctc_post_reopen,\n 'Pre-Reopen (Population-size 1500, Contacts/person/day 7)': dfs_ctc_pre_reopen, \n }, \n xlabel=\"Percentage of contacts recalled in contact tracing (%)\",\n title=title)", "_____no_output_____" ] ] ]

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

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "empty" ] ] ]

[ "empty" ]

[ [ "empty" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Artificial Intelligence Nanodegree\n## Machine Translation Project\nIn this notebook, sections that end with **'(IMPLEMENTATION)'** in the header indicate that the following blocks of code will require additional functionality which you must provide. Please be sure to read the instructions carefully!\n\n## Introduction\nIn this notebook, you will build a deep neural network that functions as part of an end-to-end machine translation pipeline. Your completed pipeline will accept English text as input and return the French translation.\n\n- **Preprocess** - You'll convert text to sequence of integers.\n- **Models** Create models which accepts a sequence of integers as input and returns a probability distribution over possible translations. After learning about the basic types of neural networks that are often used for machine translation, you will engage in your own investigations, to design your own model!\n- **Prediction** Run the model on English text.", "_____no_output_____" ] ], [ [ "%load_ext autoreload\n%aimport helper, tests\n%autoreload 1", "_____no_output_____" ], [ "import collections\n\nimport helper\nimport numpy as np\nimport project_tests as tests\n\nfrom keras.preprocessing.text import Tokenizer\nfrom keras.preprocessing.sequence import pad_sequences\nfrom keras.models import Model\nfrom keras.layers import GRU, Input, Dense, TimeDistributed, Activation, RepeatVector, Bidirectional\nfrom keras.layers.embeddings import Embedding\nfrom keras.optimizers import Adam\nfrom keras.losses import sparse_categorical_crossentropy", "Using TensorFlow backend.\n" ] ], [ [ "### Verify access to the GPU\nThe following test applies only if you expect to be using a GPU, e.g., while running in a Udacity Workspace or using an AWS instance with GPU support. Run the next cell, and verify that the device_type is \"GPU\".\n- If the device is not GPU & you are running from a Udacity Workspace, then save your workspace with the icon at the top, then click \"enable\" at the bottom of the workspace.\n- If the device is not GPU & you are running from an AWS instance, then refer to the cloud computing instructions in the classroom to verify your setup steps.", "_____no_output_____" ] ], [ [ "from tensorflow.python.client import device_lib\nprint(device_lib.list_local_devices())", "[name: \"/cpu:0\"\ndevice_type: \"CPU\"\nmemory_limit: 268435456\nlocality {\n}\nincarnation: 15440441665772238793\n, name: \"/gpu:0\"\ndevice_type: \"GPU\"\nmemory_limit: 357433344\nlocality {\n bus_id: 1\n}\nincarnation: 3061658136091710780\nphysical_device_desc: \"device: 0, name: Tesla K80, pci bus id: 0000:00:04.0\"\n]\n" ] ], [ [ "## Dataset\nWe begin by investigating the dataset that will be used to train and evaluate your pipeline. The most common datasets used for machine translation are from [WMT](http://www.statmt.org/). However, that will take a long time to train a neural network on. We'll be using a dataset we created for this project that contains a small vocabulary. You'll be able to train your model in a reasonable time with this dataset.\n### Load Data\nThe data is located in `data/small_vocab_en` and `data/small_vocab_fr`. The `small_vocab_en` file contains English sentences with their French translations in the `small_vocab_fr` file. Load the English and French data from these files from running the cell below.", "_____no_output_____" ] ], [ [ "# Load English data\nenglish_sentences = helper.load_data('data/small_vocab_en')\n# Load French data\nfrench_sentences = helper.load_data('data/small_vocab_fr')\n\nprint('Dataset Loaded')", "Dataset Loaded\n" ] ], [ [ "### Files\nEach line in `small_vocab_en` contains an English sentence with the respective translation in each line of `small_vocab_fr`. View the first two lines from each file.", "_____no_output_____" ] ], [ [ "for sample_i in range(2):\n print('small_vocab_en Line {}: {}'.format(sample_i + 1, english_sentences[sample_i]))\n print('small_vocab_fr Line {}: {}'.format(sample_i + 1, french_sentences[sample_i]))", "small_vocab_en Line 1: new jersey is sometimes quiet during autumn , and it is snowy in april .\nsmall_vocab_fr Line 1: new jersey est parfois calme pendant l' automne , et il est neigeux en avril .\nsmall_vocab_en Line 2: the united states is usually chilly during july , and it is usually freezing in november .\nsmall_vocab_fr Line 2: les états-unis est généralement froid en juillet , et il gèle habituellement en novembre .\n" ] ], [ [ "From looking at the sentences, you can see they have been preprocessed already. The puncuations have been delimited using spaces. All the text have been converted to lowercase. This should save you some time, but the text requires more preprocessing.\n### Vocabulary\nThe complexity of the problem is determined by the complexity of the vocabulary. A more complex vocabulary is a more complex problem. Let's look at the complexity of the dataset we'll be working with.", "_____no_output_____" ] ], [ [ "english_words_counter = collections.Counter([word for sentence in english_sentences for word in sentence.split()])\nfrench_words_counter = collections.Counter([word for sentence in french_sentences for word in sentence.split()])\n\nprint('{} English words.'.format(len([word for sentence in english_sentences for word in sentence.split()])))\nprint('{} unique English words.'.format(len(english_words_counter)))\nprint('10 Most common words in the English dataset:')\nprint('\"' + '\" \"'.join(list(zip(*english_words_counter.most_common(10)))[0]) + '\"')\nprint()\nprint('{} French words.'.format(len([word for sentence in french_sentences for word in sentence.split()])))\nprint('{} unique French words.'.format(len(french_words_counter)))\nprint('10 Most common words in the French dataset:')\nprint('\"' + '\" \"'.join(list(zip(*french_words_counter.most_common(10)))[0]) + '\"')", "1823250 English words.\n227 unique English words.\n10 Most common words in the English dataset:\n\"is\" \",\" \".\" \"in\" \"it\" \"during\" \"the\" \"but\" \"and\" \"sometimes\"\n\n1961295 French words.\n355 unique French words.\n10 Most common words in the French dataset:\n\"est\" \".\" \",\" \"en\" \"il\" \"les\" \"mais\" \"et\" \"la\" \"parfois\"\n" ] ], [ [ "For comparison, _Alice's Adventures in Wonderland_ contains 2,766 unique words of a total of 15,500 words.\n## Preprocess\nFor this project, you won't use text data as input to your model. Instead, you'll convert the text into sequences of integers using the following preprocess methods:\n1. Tokenize the words into ids\n2. Add padding to make all the sequences the same length.\n\nTime to start preprocessing the data...\n### Tokenize (IMPLEMENTATION)\nFor a neural network to predict on text data, it first has to be turned into data it can understand. Text data like \"dog\" is a sequence of ASCII character encodings. Since a neural network is a series of multiplication and addition operations, the input data needs to be number(s).\n\nWe can turn each character into a number or each word into a number. These are called character and word ids, respectively. Character ids are used for character level models that generate text predictions for each character. A word level model uses word ids that generate text predictions for each word. Word level models tend to learn better, since they are lower in complexity, so we'll use those.\n\nTurn each sentence into a sequence of words ids using Keras's [`Tokenizer`](https://keras.io/preprocessing/text/#tokenizer) function. Use this function to tokenize `english_sentences` and `french_sentences` in the cell below.\n\nRunning the cell will run `tokenize` on sample data and show output for debugging.", "_____no_output_____" ] ], [ [ "def tokenize(x):\n \"\"\"\n Tokenize x\n :param x: List of sentences/strings to be tokenized\n :return: Tuple of (tokenized x data, tokenizer used to tokenize x)\n \"\"\"\n # TODO: Implement\n tokenizer = Tokenizer()\n tokenizer.fit_on_texts(x)\n \n return tokenizer.texts_to_sequences(x), tokenizer\ntests.test_tokenize(tokenize)\n\n# Tokenize Example output\ntext_sentences = [\n 'The quick brown fox jumps over the lazy dog .',\n 'By Jove , my quick study of lexicography won a prize .',\n 'This is a short sentence .']\ntext_tokenized, text_tokenizer = tokenize(text_sentences)\nprint(text_tokenizer.word_index)\nprint()\nfor sample_i, (sent, token_sent) in enumerate(zip(text_sentences, text_tokenized)):\n print('Sequence {} in x'.format(sample_i + 1))\n print(' Input: {}'.format(sent))\n print(' Output: {}'.format(token_sent))", "{'the': 1, 'quick': 2, 'a': 3, 'brown': 4, 'fox': 5, 'jumps': 6, 'over': 7, 'lazy': 8, 'dog': 9, 'by': 10, 'jove': 11, 'my': 12, 'study': 13, 'of': 14, 'lexicography': 15, 'won': 16, 'prize': 17, 'this': 18, 'is': 19, 'short': 20, 'sentence': 21}\n\nSequence 1 in x\n Input: The quick brown fox jumps over the lazy dog .\n Output: [1, 2, 4, 5, 6, 7, 1, 8, 9]\nSequence 2 in x\n Input: By Jove , my quick study of lexicography won a prize .\n Output: [10, 11, 12, 2, 13, 14, 15, 16, 3, 17]\nSequence 3 in x\n Input: This is a short sentence .\n Output: [18, 19, 3, 20, 21]\n" ] ], [ [ "### Padding (IMPLEMENTATION)\nWhen batching the sequence of word ids together, each sequence needs to be the same length. Since sentences are dynamic in length, we can add padding to the end of the sequences to make them the same length.\n\nMake sure all the English sequences have the same length and all the French sequences have the same length by adding padding to the **end** of each sequence using Keras's [`pad_sequences`](https://keras.io/preprocessing/sequence/#pad_sequences) function.", "_____no_output_____" ] ], [ [ "def pad(x, length=None):\n \"\"\"\n Pad x\n :param x: List of sequences.\n :param length: Length to pad the sequence to. If None, use length of longest sequence in x.\n :return: Padded numpy array of sequences\n \"\"\"\n # TODO: Implement\n if length is None:\n length = max([len(sentence) for sentence in x])\n \n return pad_sequences(x, maxlen=length, padding='post')\ntests.test_pad(pad)\n\n# Pad Tokenized output\ntest_pad = pad(text_tokenized)\nfor sample_i, (token_sent, pad_sent) in enumerate(zip(text_tokenized, test_pad)):\n print('Sequence {} in x'.format(sample_i + 1))\n print(' Input: {}'.format(np.array(token_sent)))\n print(' Output: {}'.format(pad_sent))", "Sequence 1 in x\n Input: [1 2 4 5 6 7 1 8 9]\n Output: [1 2 4 5 6 7 1 8 9 0]\nSequence 2 in x\n Input: [10 11 12 2 13 14 15 16 3 17]\n Output: [10 11 12 2 13 14 15 16 3 17]\nSequence 3 in x\n Input: [18 19 3 20 21]\n Output: [18 19 3 20 21 0 0 0 0 0]\n" ] ], [ [ "### Preprocess Pipeline\nYour focus for this project is to build neural network architecture, so we won't ask you to create a preprocess pipeline. Instead, we've provided you with the implementation of the `preprocess` function.", "_____no_output_____" ] ], [ [ "def preprocess(x, y):\n \"\"\"\n Preprocess x and y\n :param x: Feature List of sentences\n :param y: Label List of sentences\n :return: Tuple of (Preprocessed x, Preprocessed y, x tokenizer, y tokenizer)\n \"\"\"\n preprocess_x, x_tk = tokenize(x)\n preprocess_y, y_tk = tokenize(y)\n\n preprocess_x = pad(preprocess_x)\n preprocess_y = pad(preprocess_y)\n\n # Keras's sparse_categorical_crossentropy function requires the labels to be in 3 dimensions\n preprocess_y = preprocess_y.reshape(*preprocess_y.shape, 1)\n\n return preprocess_x, preprocess_y, x_tk, y_tk\n\npreproc_english_sentences, preproc_french_sentences, english_tokenizer, french_tokenizer =\\\n preprocess(english_sentences, french_sentences)\n \nmax_english_sequence_length = preproc_english_sentences.shape[1]\nmax_french_sequence_length = preproc_french_sentences.shape[1]\nenglish_vocab_size = len(english_tokenizer.word_index)\nfrench_vocab_size = len(french_tokenizer.word_index)\n\nprint('Data Preprocessed')\nprint(\"Max English sentence length:\", max_english_sequence_length)\nprint(\"Max French sentence length:\", max_french_sequence_length)\nprint(\"English vocabulary size:\", english_vocab_size)\nprint(\"French vocabulary size:\", french_vocab_size)", "Data Preprocessed\nMax English sentence length: 15\nMax French sentence length: 21\nEnglish vocabulary size: 199\nFrench vocabulary size: 344\n" ] ], [ [ "## Models\nIn this section, you will experiment with various neural network architectures.\nYou will begin by training four relatively simple architectures.\n- Model 1 is a simple RNN\n- Model 2 is a RNN with Embedding\n- Model 3 is a Bidirectional RNN\n- Model 4 is an optional Encoder-Decoder RNN\n\nAfter experimenting with the four simple architectures, you will construct a deeper architecture that is designed to outperform all four models.\n### Ids Back to Text\nThe neural network will be translating the input to words ids, which isn't the final form we want. We want the French translation. The function `logits_to_text` will bridge the gab between the logits from the neural network to the French translation. You'll be using this function to better understand the output of the neural network.", "_____no_output_____" ] ], [ [ "def logits_to_text(logits, tokenizer):\n \"\"\"\n Turn logits from a neural network into text using the tokenizer\n :param logits: Logits from a neural network\n :param tokenizer: Keras Tokenizer fit on the labels\n :return: String that represents the text of the logits\n \"\"\"\n index_to_words = {id: word for word, id in tokenizer.word_index.items()}\n index_to_words[0] = '<PAD>'\n\n return ' '.join([index_to_words[prediction] for prediction in np.argmax(logits, 1)])\n\nprint('`logits_to_text` function loaded.')", "`logits_to_text` function loaded.\n" ] ], [ [ "### Model 1: RNN (IMPLEMENTATION)\n\nA basic RNN model is a good baseline for sequence data. In this model, you'll build a RNN that translates English to French.", "_____no_output_____" ] ], [ [ "def simple_model(input_shape, output_sequence_length, english_vocab_size, french_vocab_size):\n \"\"\"\n Build and train a basic RNN on x and y\n :param input_shape: Tuple of input shape\n :param output_sequence_length: Length of output sequence\n :param english_vocab_size: Number of unique English words in the dataset\n :param french_vocab_size: Number of unique French words in the dataset\n :return: Keras model built, but not trained\n \"\"\"\n # TODO: Build the layers\n inputs = Input(input_shape[1:])\n outputs = GRU(256, return_sequences=True)(inputs) \n outputs = TimeDistributed(Dense(french_vocab_size, activation=\"softmax\"))(outputs)\n \n model = Model(inputs, outputs)\n \n learning_rate = 0.001\n model.compile(loss=sparse_categorical_crossentropy,\n optimizer=Adam(learning_rate),\n metrics=['accuracy'])\n return model\ntests.test_simple_model(simple_model)\n\n# Reshaping the input to work with a basic RNN\ntmp_x = pad(preproc_english_sentences, max_french_sequence_length)\ntmp_x = tmp_x.reshape((-1, preproc_french_sentences.shape[-2], 1))\n\n# Train the neural network\nsimple_rnn_model = simple_model(\n tmp_x.shape,\n max_french_sequence_length,\n english_vocab_size,\n french_vocab_size)\nsimple_rnn_model.fit(tmp_x, preproc_french_sentences, batch_size=1024, epochs=10, validation_split=0.2)\n\n# Print prediction(s)\nprint(logits_to_text(simple_rnn_model.predict(tmp_x[:1])[0], french_tokenizer))", "Train on 110288 samples, validate on 27573 samples\nEpoch 1/10\n110288/110288 [==============================] - 12s 109us/step - loss: 2.5907 - acc: 0.4812 - val_loss: nan - val_acc: 0.5582\nEpoch 2/10\n110288/110288 [==============================] - 10s 91us/step - loss: 1.6567 - acc: 0.5862 - val_loss: nan - val_acc: 0.6038\nEpoch 3/10\n110288/110288 [==============================] - 10s 91us/step - loss: 1.4347 - acc: 0.6124 - val_loss: nan - val_acc: 0.6220\nEpoch 4/10\n110288/110288 [==============================] - 10s 91us/step - loss: 1.3146 - acc: 0.6324 - val_loss: nan - val_acc: 0.6414\nEpoch 5/10\n110288/110288 [==============================] - 10s 91us/step - loss: 1.2245 - acc: 0.6483 - val_loss: nan - val_acc: 0.6516\nEpoch 6/10\n110288/110288 [==============================] - 10s 92us/step - loss: 1.1588 - acc: 0.6597 - val_loss: nan - val_acc: 0.6692\nEpoch 7/10\n110288/110288 [==============================] - 10s 92us/step - loss: 1.1090 - acc: 0.6685 - val_loss: nan - val_acc: 0.6745\nEpoch 8/10\n110288/110288 [==============================] - 10s 91us/step - loss: 1.0710 - acc: 0.6729 - val_loss: nan - val_acc: 0.6732\nEpoch 9/10\n110288/110288 [==============================] - 10s 91us/step - loss: 1.0381 - acc: 0.6766 - val_loss: nan - val_acc: 0.6822\nEpoch 10/10\n110288/110288 [==============================] - 10s 91us/step - loss: 1.0106 - acc: 0.6810 - val_loss: nan - val_acc: 0.6816\nnew jersey est parfois calme en mois et il est il est en en <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD>\n" ] ], [ [ "### Model 2: Embedding (IMPLEMENTATION)\n\nYou've turned the words into ids, but there's a better representation of a word. This is called word embeddings. An embedding is a vector representation of the word that is close to similar words in n-dimensional space, where the n represents the size of the embedding vectors.\n\nIn this model, you'll create a RNN model using embedding.", "_____no_output_____" ] ], [ [ "def embed_model(input_shape, output_sequence_length, english_vocab_size, french_vocab_size):\n \"\"\"\n Build and train a RNN model using word embedding on x and y\n :param input_shape: Tuple of input shape\n :param output_sequence_length: Length of output sequence\n :param english_vocab_size: Number of unique English words in the dataset\n :param french_vocab_size: Number of unique French words in the dataset\n :return: Keras model built, but not trained\n \"\"\"\n # TODO: Implement\n inputs = Input(input_shape[1:])\n \n outputs = Embedding(english_vocab_size, output_sequence_length)(inputs)\n outputs = GRU(256, return_sequences=True)(outputs) \n outputs = TimeDistributed(Dense(french_vocab_size, activation=\"softmax\"))(outputs)\n \n model = Model(inputs, outputs)\n \n learning_rate = 0.001\n model.compile(loss=sparse_categorical_crossentropy,\n optimizer=Adam(learning_rate),\n metrics=['accuracy'])\n return model\ntests.test_embed_model(embed_model)\n\n\n# TODO: Reshape the input\ntmp_x = pad(preproc_english_sentences, preproc_french_sentences.shape[1])\ntmp_x = tmp_x.reshape((-1, preproc_french_sentences.shape[-2]))\n\n# TODO: Train the neural network\nembed_rnn_model = embed_model(\n tmp_x.shape,\n preproc_french_sentences.shape[1],\n english_vocab_size,\n french_vocab_size)\n\nembed_rnn_model.fit(tmp_x, preproc_french_sentences, batch_size=1024, epochs=10, validation_split=0.2)\n\n# TODO: Print prediction(s)\nprint(logits_to_text(embed_rnn_model.predict(tmp_x[:1])[0], french_tokenizer))", "Train on 110288 samples, validate on 27573 samples\nEpoch 1/10\n110288/110288 [==============================] - 12s 104us/step - loss: 3.3620 - acc: 0.4088 - val_loss: nan - val_acc: 0.4603\nEpoch 2/10\n110288/110288 [==============================] - 11s 102us/step - loss: 2.5504 - acc: 0.4842 - val_loss: nan - val_acc: 0.5128\nEpoch 3/10\n110288/110288 [==============================] - 11s 102us/step - loss: 2.0617 - acc: 0.5411 - val_loss: nan - val_acc: 0.5635\nEpoch 4/10\n110288/110288 [==============================] - 11s 101us/step - loss: 1.6021 - acc: 0.6006 - val_loss: nan - val_acc: 0.6337\nEpoch 5/10\n110288/110288 [==============================] - 11s 102us/step - loss: 1.3343 - acc: 0.6570 - val_loss: nan - val_acc: 0.6817\nEpoch 6/10\n110288/110288 [==============================] - 11s 102us/step - loss: 1.1285 - acc: 0.7028 - val_loss: nan - val_acc: 0.7282\nEpoch 7/10\n110288/110288 [==============================] - 11s 102us/step - loss: 0.9586 - acc: 0.7497 - val_loss: nan - val_acc: 0.7712\nEpoch 8/10\n110288/110288 [==============================] - 11s 101us/step - loss: 0.8094 - acc: 0.7879 - val_loss: nan - val_acc: 0.8058\nEpoch 9/10\n110288/110288 [==============================] - 11s 101us/step - loss: 0.6796 - acc: 0.8175 - val_loss: nan - val_acc: 0.8296\nEpoch 10/10\n110288/110288 [==============================] - 11s 101us/step - loss: 0.5832 - acc: 0.8390 - val_loss: nan - val_acc: 0.8477\nnew jersey est parfois calme en l' et il est il est en <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD>\n" ] ], [ [ "### Model 3: Bidirectional RNNs (IMPLEMENTATION)\n\nOne restriction of a RNN is that it can't see the future input, only the past. This is where bidirectional recurrent neural networks come in. They are able to see the future data.", "_____no_output_____" ] ], [ [ "def bd_model(input_shape, output_sequence_length, english_vocab_size, french_vocab_size):\n \"\"\"\n Build and train a bidirectional RNN model on x and y\n :param input_shape: Tuple of input shape\n :param output_sequence_length: Length of output sequence\n :param english_vocab_size: Number of unique English words in the dataset\n :param french_vocab_size: Number of unique French words in the dataset\n :return: Keras model built, but not trained\n \"\"\"\n # TODO: Implement\n inputs = Input(input_shape[1:])\n outputs = Bidirectional(GRU(256, return_sequences=True))(inputs) \n outputs = TimeDistributed(Dense(french_vocab_size, activation=\"softmax\"))(outputs)\n \n model = Model(inputs, outputs)\n \n learning_rate = 0.001\n model.compile(loss=sparse_categorical_crossentropy,\n optimizer=Adam(learning_rate),\n metrics=['accuracy'])\n return model\ntests.test_bd_model(bd_model)\n\n\n# TODO: Train and Print prediction(s)\ntmp_x = pad(preproc_english_sentences, preproc_french_sentences.shape[1])\ntmp_x = tmp_x.reshape((-1, preproc_french_sentences.shape[-2]))\n\nbd_rnn_model = embed_model(\n tmp_x.shape,\n preproc_french_sentences.shape[1],\n english_vocab_size,\n french_vocab_size)\n\nbd_rnn_model.fit(tmp_x, preproc_french_sentences, batch_size=1024, epochs=10, validation_split=0.2)\n\nprint(logits_to_text(bd_rnn_model.predict(tmp_x[:1])[0], french_tokenizer))", "Train on 110288 samples, validate on 27573 samples\nEpoch 1/10\n110288/110288 [==============================] - 12s 106us/step - loss: 3.3402 - acc: 0.4192 - val_loss: nan - val_acc: 0.4769\nEpoch 2/10\n110288/110288 [==============================] - 11s 101us/step - loss: 2.4795 - acc: 0.4908 - val_loss: nan - val_acc: 0.5142\nEpoch 3/10\n110288/110288 [==============================] - 11s 101us/step - loss: 1.9030 - acc: 0.5528 - val_loss: nan - val_acc: 0.5912\nEpoch 4/10\n110288/110288 [==============================] - 11s 101us/step - loss: 1.4347 - acc: 0.6258 - val_loss: nan - val_acc: 0.6628\nEpoch 5/10\n110288/110288 [==============================] - 11s 102us/step - loss: 1.1867 - acc: 0.6922 - val_loss: nan - val_acc: 0.7193\nEpoch 6/10\n110288/110288 [==============================] - 11s 102us/step - loss: 0.9938 - acc: 0.7419 - val_loss: nan - val_acc: 0.7685\nEpoch 7/10\n110288/110288 [==============================] - 11s 101us/step - loss: 0.8202 - acc: 0.7881 - val_loss: nan - val_acc: 0.8065\nEpoch 8/10\n110288/110288 [==============================] - 11s 101us/step - loss: 0.6827 - acc: 0.8198 - val_loss: nan - val_acc: 0.8325\nEpoch 9/10\n110288/110288 [==============================] - 11s 101us/step - loss: 0.5852 - acc: 0.8410 - val_loss: nan - val_acc: 0.8509\nEpoch 10/10\n110288/110288 [==============================] - 11s 101us/step - loss: 0.5119 - acc: 0.8583 - val_loss: nan - val_acc: 0.8654\nnew jersey est parfois calme en l' et et il et il avril <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD>\n" ] ], [ [ "### Model 4: Encoder-Decoder (OPTIONAL)\nTime to look at encoder-decoder models. This model is made up of an encoder and decoder. The encoder creates a matrix representation of the sentence. The decoder takes this matrix as input and predicts the translation as output.\n\nCreate an encoder-decoder model in the cell below.", "_____no_output_____" ] ], [ [ "def encdec_model(input_shape, output_sequence_length, english_vocab_size, french_vocab_size):\n \"\"\"\n Build and train an encoder-decoder model on x and y\n :param input_shape: Tuple of input shape\n :param output_sequence_length: Length of output sequence\n :param english_vocab_size: Number of unique English words in the dataset\n :param french_vocab_size: Number of unique French words in the dataset\n :return: Keras model built, but not trained\n \"\"\"\n # OPTIONAL: Implement\n inputs = Input(input_shape[1:])\n encoded = GRU(256, return_sequences=False)(inputs)\n \n decoded = RepeatVector(output_sequence_length)(encoded)\n \n outputs = GRU(256, return_sequences=True)(decoded)\n outputs = Dense(french_vocab_size, activation=\"softmax\")(outputs)\n \n model = Model(inputs, outputs)\n \n learning_rate = 0.001\n model.compile(loss=sparse_categorical_crossentropy,\n optimizer=Adam(learning_rate),\n metrics=['accuracy'])\n return model\ntests.test_encdec_model(encdec_model)\n\n\n# OPTIONAL: Train and Print prediction(s)\ntmp_x = pad(preproc_english_sentences, preproc_french_sentences.shape[1])\ntmp_x = tmp_x.reshape((-1, preproc_french_sentences.shape[-2]))\n\nencdec_rnn_model = embed_model(\n tmp_x.shape,\n preproc_french_sentences.shape[1],\n english_vocab_size,\n french_vocab_size)\n\nencdec_rnn_model.fit(tmp_x, preproc_french_sentences, batch_size=1024, epochs=10, validation_split=0.2)\n\nprint(logits_to_text(encdec_rnn_model.predict(tmp_x[:1])[0], french_tokenizer))", "Train on 110288 samples, validate on 27573 samples\nEpoch 1/10\n110288/110288 [==============================] - 12s 106us/step - loss: 3.3761 - acc: 0.4076 - val_loss: nan - val_acc: 0.4515\nEpoch 2/10\n110288/110288 [==============================] - 11s 100us/step - loss: 2.5353 - acc: 0.4798 - val_loss: nan - val_acc: 0.5145\nEpoch 3/10\n110288/110288 [==============================] - 11s 101us/step - loss: 2.0491 - acc: 0.5406 - val_loss: nan - val_acc: 0.5683\nEpoch 4/10\n110288/110288 [==============================] - 11s 101us/step - loss: 1.5999 - acc: 0.5997 - val_loss: nan - val_acc: 0.6365\nEpoch 5/10\n110288/110288 [==============================] - 11s 101us/step - loss: 1.3258 - acc: 0.6664 - val_loss: nan - val_acc: 0.6911\nEpoch 6/10\n110288/110288 [==============================] - 11s 101us/step - loss: 1.1065 - acc: 0.7156 - val_loss: nan - val_acc: 0.7447\nEpoch 7/10\n110288/110288 [==============================] - 11s 101us/step - loss: 0.9247 - acc: 0.7625 - val_loss: nan - val_acc: 0.7824\nEpoch 8/10\n110288/110288 [==============================] - 11s 101us/step - loss: 0.7729 - acc: 0.7968 - val_loss: nan - val_acc: 0.8140\nEpoch 9/10\n110288/110288 [==============================] - 11s 101us/step - loss: 0.6530 - acc: 0.8240 - val_loss: nan - val_acc: 0.8355\nEpoch 10/10\n110288/110288 [==============================] - 11s 101us/step - loss: 0.5646 - acc: 0.8437 - val_loss: nan - val_acc: 0.8511\nnew jersey est parfois calme en l' et il est neigeux en avril <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD>\n" ] ], [ [ "### Model 5: Custom (IMPLEMENTATION)\nUse everything you learned from the previous models to create a model that incorporates embedding and a bidirectional rnn into one model.", "_____no_output_____" ] ], [ [ "def model_final(input_shape, output_sequence_length, english_vocab_size, french_vocab_size):\n \"\"\"\n Build and train a model that incorporates embedding, encoder-decoder, and bidirectional RNN on x and y\n :param input_shape: Tuple of input shape\n :param output_sequence_length: Length of output sequence\n :param english_vocab_size: Number of unique English words in the dataset\n :param french_vocab_size: Number of unique French words in the dataset\n :return: Keras model built, but not trained\n \"\"\"\n # TODO: Implement\n learning_rate = 0.001\n \n inputs = Input(shape=input_shape[1:])\n \n encoded = Embedding(english_vocab_size, 300)(inputs)\n encoded = Bidirectional(GRU(512, dropout=0.2))(encoded)\n encoded = Dense(512, activation='relu')(encoded)\n \n decoded = RepeatVector(output_sequence_length)(encoded)\n decoded = Bidirectional(GRU(512, dropout=0.2, return_sequences=True))(decoded)\n decoded = TimeDistributed(Dense(french_vocab_size))(decoded)\n \n predictions = Activation('softmax')(decoded)\n model = Model(inputs=inputs, outputs=predictions)\n \n model.compile(loss=sparse_categorical_crossentropy,\n optimizer=Adam(learning_rate),\n metrics=['accuracy'])\n return model\ntests.test_model_final(model_final)\n\n\nprint('Final Model Loaded')\n# TODO: Train the final model\ntmp_x = pad(preproc_english_sentences, preproc_french_sentences.shape[1])\ntmp_x = tmp_x.reshape((-1, preproc_french_sentences.shape[-2]))\n\nfinal_model = model_final(\n tmp_x.shape,\n preproc_french_sentences.shape[1],\n english_vocab_size,\n french_vocab_size)\n\nfinal_model.fit(tmp_x, preproc_french_sentences, batch_size=1024, epochs=10, validation_split=0.2)\nprint(logits_to_text(encdec_rnn_model.predict(tmp_x[:1])[0], french_tokenizer))", "Final Model Loaded\nTrain on 110288 samples, validate on 27573 samples\nEpoch 1/10\n110288/110288 [==============================] - 91s 824us/step - loss: 2.4410 - acc: 0.4865 - val_loss: nan - val_acc: 0.5841\nEpoch 2/10\n110288/110288 [==============================] - 89s 807us/step - loss: 1.4055 - acc: 0.6253 - val_loss: nan - val_acc: 0.6683\nEpoch 3/10\n110288/110288 [==============================] - 89s 806us/step - loss: 1.0897 - acc: 0.6953 - val_loss: nan - val_acc: 0.7250\nEpoch 4/10\n110288/110288 [==============================] - 89s 806us/step - loss: 0.9146 - acc: 0.7333 - val_loss: nan - val_acc: 0.7512\nEpoch 5/10\n110288/110288 [==============================] - 89s 806us/step - loss: 0.7888 - acc: 0.7662 - val_loss: nan - val_acc: 0.7899\nEpoch 6/10\n110288/110288 [==============================] - 89s 806us/step - loss: 0.6854 - acc: 0.7937 - val_loss: nan - val_acc: 0.8163\nEpoch 7/10\n110288/110288 [==============================] - 89s 806us/step - loss: 0.5770 - acc: 0.8241 - val_loss: nan - val_acc: 0.8468\nEpoch 8/10\n110288/110288 [==============================] - 89s 805us/step - loss: 0.4818 - acc: 0.8538 - val_loss: nan - val_acc: 0.8753\nEpoch 9/10\n110288/110288 [==============================] - 89s 805us/step - loss: 0.3899 - acc: 0.8836 - val_loss: nan - val_acc: 0.9093\nEpoch 10/10\n110288/110288 [==============================] - 89s 805us/step - loss: 0.3099 - acc: 0.9085 - val_loss: nan - val_acc: 0.9293\nnew jersey est parfois calme en l' et il est neigeux en avril <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD>\n" ] ], [ [ "## Prediction (IMPLEMENTATION)", "_____no_output_____" ] ], [ [ "def final_predictions(x, y, x_tk, y_tk):\n \"\"\"\n Gets predictions using the final model\n :param x: Preprocessed English data\n :param y: Preprocessed French data\n :param x_tk: English tokenizer\n :param y_tk: French tokenizer\n \"\"\"\n # TODO: Train neural network using model_final\n model = model_final(x.shape, \n y.shape[1], \n len(x_tk.word_index), \n len(y_tk.word_index))\n \n model.fit(x, y, batch_size=1024, epochs=12, validation_split=0.2)\n\n \n ## DON'T EDIT ANYTHING BELOW THIS LINE\n y_id_to_word = {value: key for key, value in y_tk.word_index.items()}\n y_id_to_word[0] = '<PAD>'\n\n sentence = 'he saw a old yellow truck'\n sentence = [x_tk.word_index[word] for word in sentence.split()]\n sentence = pad_sequences([sentence], maxlen=x.shape[-1], padding='post')\n sentences = np.array([sentence[0], x[0]])\n predictions = model.predict(sentences, len(sentences))\n\n print('Sample 1:')\n print(' '.join([y_id_to_word[np.argmax(x)] for x in predictions[0]]))\n print('Il a vu un vieux camion jaune')\n print('Sample 2:')\n print(' '.join([y_id_to_word[np.argmax(x)] for x in predictions[1]]))\n print(' '.join([y_id_to_word[np.max(x)] for x in y[0]]))\n\n\nfinal_predictions(preproc_english_sentences, preproc_french_sentences, english_tokenizer, french_tokenizer)", "Train on 110288 samples, validate on 27573 samples\nEpoch 1/12\n110288/110288 [==============================] - 81s 733us/step - loss: 2.3220 - acc: 0.5011 - val_loss: nan - val_acc: 0.5999\nEpoch 2/12\n110288/110288 [==============================] - 79s 713us/step - loss: 1.3477 - acc: 0.6375 - val_loss: nan - val_acc: 0.6822\nEpoch 3/12\n110288/110288 [==============================] - 79s 713us/step - loss: 1.0208 - acc: 0.7100 - val_loss: nan - val_acc: 0.7388\nEpoch 4/12\n110288/110288 [==============================] - 79s 713us/step - loss: 0.8375 - acc: 0.7518 - val_loss: nan - val_acc: 0.7783\nEpoch 5/12\n110288/110288 [==============================] - 79s 713us/step - loss: 0.7114 - acc: 0.7849 - val_loss: nan - val_acc: 0.8114\nEpoch 6/12\n110288/110288 [==============================] - 79s 714us/step - loss: 0.5968 - acc: 0.8179 - val_loss: nan - val_acc: 0.8416\nEpoch 7/12\n110288/110288 [==============================] - 79s 713us/step - loss: 0.4868 - acc: 0.8513 - val_loss: nan - val_acc: 0.8810\nEpoch 8/12\n110288/110288 [==============================] - 79s 713us/step - loss: 0.3818 - acc: 0.8846 - val_loss: nan - val_acc: 0.9105\nEpoch 9/12\n110288/110288 [==============================] - 79s 713us/step - loss: 0.2995 - acc: 0.9113 - val_loss: nan - val_acc: 0.9290\nEpoch 10/12\n110288/110288 [==============================] - 79s 713us/step - loss: 0.2387 - acc: 0.9299 - val_loss: nan - val_acc: 0.9451\nEpoch 11/12\n110288/110288 [==============================] - 79s 713us/step - loss: 0.1893 - acc: 0.9444 - val_loss: nan - val_acc: 0.9505\nEpoch 12/12\n110288/110288 [==============================] - 79s 713us/step - loss: 0.1621 - acc: 0.9521 - val_loss: nan - val_acc: 0.9566\nSample 1:\nil a vu un vieux camion jaune <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD>\nIl a vu un vieux camion jaune\nSample 2:\nnew jersey est parfois calme pendant l' automne et il est neigeux en avril <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD>\nnew jersey est parfois calme pendant l' automne et il est neigeux en avril <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD>\n" ] ], [ [ "## Submission\nWhen you're ready to submit, complete the following steps:\n1. Review the [rubric](https://review.udacity.com/#!/rubrics/1004/view) to ensure your submission meets all requirements to pass\n2. Generate an HTML version of this notebook\n\n - Run the next cell to attempt automatic generation (this is the recommended method in Workspaces)\n - Navigate to **FILE -> Download as -> HTML (.html)**\n - Manually generate a copy using `nbconvert` from your shell terminal\n```\n$ pip install nbconvert\n$ python -m nbconvert machine_translation.ipynb\n```\n \n3. Submit the project\n\n - If you are in a Workspace, simply click the \"Submit Project\" button (bottom towards the right)\n \n - Otherwise, add the following files into a zip archive and submit them \n - `helper.py`\n - `machine_translation.ipynb`\n - `machine_translation.html`\n - You can export the notebook by navigating to **File -> Download as -> HTML (.html)**.", "_____no_output_____" ], [ "### Generate the html\n\n**Save your notebook before running the next cell to generate the HTML output.** Then submit your project.", "_____no_output_____" ] ], [ [ "# Save before you run this cell!\n!!jupyter nbconvert *.ipynb", "_____no_output_____" ] ], [ [ "## Optional Enhancements\n\nThis project focuses on learning various network architectures for machine translation, but we don't evaluate the models according to best practices by splitting the data into separate test & training sets -- so the model accuracy is overstated. Use the [`sklearn.model_selection.train_test_split()`](http://scikit-learn.org/stable/modules/generated/sklearn.model_selection.train_test_split.html) function to create separate training & test datasets, then retrain each of the models using only the training set and evaluate the prediction accuracy using the hold out test set. Does the \"best\" model change?", "_____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" ]

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# Unit 5 - Financial Planning\n", "_____no_output_____" ] ], [ [ "# Initial imports\nimport os\nimport requests\nimport pandas as pd\nfrom dotenv import load_dotenv\nimport alpaca_trade_api as tradeapi\nfrom MCForecastTools import MCSimulation\n\n# date here\nfrom datetime import date\n\n\n%matplotlib inline", "_____no_output_____" ], [ "# Load .env enviroment variables\nload_dotenv()", "_____no_output_____" ] ], [ [ "## Part 1 - Personal Finance Planner", "_____no_output_____" ], [ "## Collect Crypto Prices Using the `requests` Library", "_____no_output_____" ] ], [ [ "# Set current amount of crypto assets\n# YOUR CODE HERE!\nmy_btc = 1.2\nmy_eth = 5.3", "_____no_output_____" ], [ "# Crypto API URLs\nbtc_url = \"https://api.alternative.me/v2/ticker/Bitcoin/?convert=CAD\"\neth_url = \"https://api.alternative.me/v2/ticker/Ethereum/?convert=CAD\"", "_____no_output_____" ], [ "# response_data = requests.get(create_deck_url).json()\n# response_data\nbtc_resp = requests.get(btc_url).json()\nbtc_price = btc_resp['data']['1']['quotes']['USD']['price']\nmy_btc_value =my_btc * btc_price\n\neth_resp = requests.get(eth_url).json()\neth_price = eth_resp['data']['1027']['quotes']['USD']['price']\nmy_eth_value = my_eth * eth_price\n\nprint(f\"The current value of your {my_btc} BTC is ${my_btc_value:0.2f}\")\nprint(f\"The current value of your {my_eth} ETH is ${my_eth_value:0.2f}\")", "The current value of your 1.2 BTC is $60097.20\nThe current value of your 5.3 ETH is $11897.07\n" ] ], [ [ "### Collect Investments Data Using Alpaca: `SPY` (stocks) and `AGG` (bonds)", "_____no_output_____" ] ], [ [ "# Current amount of shares\n# Create two variables named my_agg and my_spy and set them equal to 200 and 50, respectively.\n# YOUR CODE HERE!\nmy_agg = 200\nmy_spy = 50", "_____no_output_____" ], [ "# Set Alpaca API key and secret\n# YOUR CODE HERE!\nalpaca_api_key = os.getenv(\"ALPACA_API_KEY\")\nalpaca_secret_key = os.getenv(\"ALPACA_SECRET_KEY\")\n\n# Create the Alpaca API object\n# YOUR CODE HERE!\napi = tradeapi.REST(\n alpaca_api_key,\n alpaca_secret_key,\n api_version = \"v2\"\n)", "_____no_output_____" ], [ "# Format current date as ISO format\n# YOUR CODE HERE!\n# use \"2021-04-16\" so weekend gives actual data\nstart_date = pd.Timestamp(\"2021-04-16\", tz=\"America/New_York\").isoformat()\ntoday_date = pd.Timestamp(date.today(), tz=\"America/New_York\").isoformat()\n\n# Set the tickers\ntickers = [\"AGG\", \"SPY\"]\n\n# Set timeframe to '1D' for Alpaca API\ntimeframe = \"1D\"\n\n# Get current closing prices for SPY and AGG\n# YOUR CODE HERE!\nticker_data = api.get_barset(\n tickers,\n timeframe,\n start=start_date,\n end=start_date,\n).df\n# Preview DataFrame\n# YOUR CODE HERE!\nticker_data", "_____no_output_____" ], [ "# Pick AGG and SPY close prices\n# YOUR CODE HERE!\nagg_close_price = ticker_data['AGG']['close'][0]\nspy_close_price = ticker_data['SPY']['close'][0]\n\n# Print AGG and SPY close prices\nprint(f\"Current AGG closing price: ${agg_close_price}\")\nprint(f\"Current SPY closing price: ${spy_close_price}\")\n", "Current AGG closing price: $114.54\nCurrent SPY closing price: $417.31\n" ], [ "# Compute the current value of shares\n# YOUR CODE HERE!\nmy_spy_value = spy_close_price * my_spy\nmy_agg_value = agg_close_price * my_agg\n\n# Print current value of share\nprint(f\"The current value of your {my_spy} SPY shares is ${my_spy_value:0.2f}\")\nprint(f\"The current value of your {my_agg} AGG shares is ${my_agg_value:0.2f}\")", "The current value of your 50 SPY shares is $20865.50\nThe current value of your 200 AGG shares is $22908.00\n" ] ], [ [ "### Savings Health Analysis", "_____no_output_____" ] ], [ [ "# Set monthly household income\n# YOUR CODE HERE!\nmonthly_income = 12000\n\n# Create savings DataFrame\n# YOUR CODE HERE!\ndf_savings = pd.DataFrame([my_btc_value+ my_eth_value, my_spy_value + my_agg_value], columns = ['amount'], index = ['crypto', 'shares'])\n\n# Display savings DataFrame\ndisplay(df_savings)", "_____no_output_____" ], [ "# Plot savings pie chart\n# YOUR CODE HERE!", "_____no_output_____" ], [ "df_savings['amount'].plot.pie(y= ['crypto', 'shares'])\n# how to put label", "_____no_output_____" ], [ "# Set ideal emergency fund\nemergency_fund = monthly_income * 3\n\n# Calculate total amount of savings\n# YOUR CODE HERE!\ntotal_saving = df_savings['amount'].sum()\n\n# Validate saving health\n# YOUR CODE HERE!\n\n# If total savings are greater than the emergency fund, display a message congratulating the person for having enough money in this fund.\n# If total savings are equal to the emergency fund, display a message congratulating the person on reaching this financial goal.\n# If total savings are less than the emergency fund, display a message showing how many dollars away the person is from reaching the goal.\n\nif total_saving > emergency_fund:\n print ('Congratulations! You have enough money in your emergency fund.')\nelif total_saving == emergency_fund:\n print ('Contratulations you reached your financial goals ')\nelse:\n print ('You are making great progress. You need to save $ {round((emergency_fund - total_saving), 2)}')", "Congratulations! You have enough money in your emergency fund.\n" ] ], [ [ "## Part 2 - Retirement Planning\n\n### Monte Carlo Simulation\n\n#### Hassan's Note for some reason AlPaca would not let me get more than 1000 records. To get 5 years data (252 * 5), I had to break it up into two reads and concat the data. ", "_____no_output_____" ] ], [ [ "# Set start and end dates of five years back from today.\n# Sample results may vary from the solution based on the time frame chosen\nstart_date1 = pd.Timestamp('2015-08-07', tz='America/New_York').isoformat()\nend_date1 = pd.Timestamp('2017-08-07', tz='America/New_York').isoformat()\nstart_date2 = pd.Timestamp('2017-08-08', tz='America/New_York').isoformat()\nend_date2 = pd.Timestamp('2020-08-07', tz='America/New_York').isoformat()\n# end_date1 = pd.Timestamp('2019-08-07', tz='America/New_York').isoformat()\n# hits 1000 item limit, have to do in two batches and pringt", "_____no_output_____" ], [ "# Get 5 years' worth of historical data for SPY and AGG\n# YOUR CODE HERE!\n\n# Display sample data\ndf_stock_data.head()", "_____no_output_____" ], [ "# Get 5 years' worth of historical data for SPY and AGG\n# create two dataframes and concaternate them \n# first period data frame\ndf_stock_data1 = api.get_barset(\n tickers,\n timeframe,\n start = start_date1,\n end = end_date1,\n limit = 1000\n).df\n\n# second period dataframe \ndf_stock_data2 = api.get_barset(\n tickers,\n timeframe,\n start = start_date2,\n end = end_date2,\n limit = 1000\n).df\n\ndf_stock_data = pd.concat ([df_stock_data1, df_stock_data2], axis = 0, join = 'inner')\nprint (f'stock data head: ')\nprint (df_stock_data.head(5))\nprint (f'\\nstock data 1 tail: ')\nprint (df_stock_data.tail(5))\n# print (f'stock data 1 head: {start_date1}')\n# print (df_stock_data1.head(5))\n# print (f'\\nstock data 1 tail: {end_date1}')\n# print (df_stock_data1.tail(5))\n# print (f'\\nstock data 2 head: {start_date2}')\n# print (df_stock_data2.head(5))\n# print (f'\\nstock data 2 tail: {end_date2}')\n# print (df_stock_data2.tail(5))\n", "stock data head: \n AGG \\\n open high low close volume \ntime \n2015-08-07 00:00:00-04:00 109.14 109.2750 109.035 109.21 2041167.0 \n2015-08-10 00:00:00-04:00 109.15 109.1700 108.920 109.06 1149778.0 \n2015-08-11 00:00:00-04:00 109.42 109.5765 109.284 109.42 1420907.0 \n2015-08-12 00:00:00-04:00 109.55 109.7100 109.350 109.36 1468979.0 \n2015-08-13 00:00:00-04:00 109.36 109.3651 109.110 109.15 1465173.0 \n\n SPY \n open high low close volume \ntime \n2015-08-07 00:00:00-04:00 208.16 208.34 206.87 207.93 87669782 \n2015-08-10 00:00:00-04:00 209.28 210.67 209.28 210.58 66755890 \n2015-08-11 00:00:00-04:00 208.98 209.47 207.76 208.63 88424557 \n2015-08-12 00:00:00-04:00 207.11 209.14 205.36 208.89 136171450 \n2015-08-13 00:00:00-04:00 208.73 209.55 208.01 208.63 77197796 \n\nstock data 1 tail: \n AGG \\\n open high low close volume \ntime \n2020-08-03 00:00:00-04:00 119.37 119.40 119.1903 119.400 17837420.0 \n2020-08-04 00:00:00-04:00 119.42 119.63 119.4200 119.630 21512268.0 \n2020-08-05 00:00:00-04:00 119.39 119.49 119.3100 119.400 34175883.0 \n2020-08-06 00:00:00-04:00 119.62 119.73 119.5300 119.580 9009216.0 \n2020-08-07 00:00:00-04:00 119.66 119.73 119.3950 119.445 8830420.0 \n\n SPY \n open high low close volume \ntime \n2020-08-03 00:00:00-04:00 328.3200 329.62 327.73 328.76 71741125 \n2020-08-04 00:00:00-04:00 327.8600 330.06 327.86 330.03 73684427 \n2020-08-05 00:00:00-04:00 331.4700 332.39 331.18 332.06 72846458 \n2020-08-06 00:00:00-04:00 331.4799 334.46 331.13 334.31 76900649 \n2020-08-07 00:00:00-04:00 333.2800 334.88 332.30 334.55 98710236 \n" ], [ "# delete\n# # Get 5 years' worth of historical data for SPY and AGG\n# # YOUR CODE HERE!\n# df_stock_data = api.get_barset(\n# tickers,\n# timeframe,\n# start = start_date,\n# end = end_date,\n# limit = 1000\n# ).df\n\n# # Display sample data\n\n# # to fix.\n# df_stock_data.head()", "_____no_output_____" ], [ "# Configuring a Monte Carlo simulation to forecast 30 years cumulative returns\n# YOUR CODE HERE!\nMC_even_dist = MCSimulation(\n portfolio_data = df_stock_data,\n weights = [.4, .6],\n num_simulation = 500,\n num_trading_days = 252*30\n)", "_____no_output_____" ], [ "# Printing the simulation input data\n# YOUR CODE HERE!", "_____no_output_____" ], [ "# Printing the simulation input data\n# YOUR CODE HERE!\nMC_even_dist.portfolio_data.head()", "_____no_output_____" ], [ "# Running a Monte Carlo simulation to forecast 30 years cumulative returns\n# YOUR CODE HERE!", "_____no_output_____" ], [ "# Running a Monte Carlo simulation to forecast 30 years cumulative returns\n# YOUR CODE HERE!\nMC_even_dist.calc_cumulative_return()", "Running Monte Carlo simulation number 0.\nRunning Monte Carlo simulation number 10.\nRunning Monte Carlo simulation number 20.\nRunning Monte Carlo simulation number 30.\nRunning Monte Carlo simulation number 40.\nRunning Monte Carlo simulation number 50.\nRunning Monte Carlo simulation number 60.\nRunning Monte Carlo simulation number 70.\nRunning Monte Carlo simulation number 80.\nRunning Monte Carlo simulation number 90.\nRunning Monte Carlo simulation number 100.\nRunning Monte Carlo simulation number 110.\nRunning Monte Carlo simulation number 120.\nRunning Monte Carlo simulation number 130.\nRunning Monte Carlo simulation number 140.\nRunning Monte Carlo simulation number 150.\nRunning Monte Carlo simulation number 160.\nRunning Monte Carlo simulation number 170.\nRunning Monte Carlo simulation number 180.\nRunning Monte Carlo simulation number 190.\nRunning Monte Carlo simulation number 200.\nRunning Monte Carlo simulation number 210.\nRunning Monte Carlo simulation number 220.\nRunning Monte Carlo simulation number 230.\nRunning Monte Carlo simulation number 240.\nRunning Monte Carlo simulation number 250.\nRunning Monte Carlo simulation number 260.\nRunning Monte Carlo simulation number 270.\nRunning Monte Carlo simulation number 280.\nRunning Monte Carlo simulation number 290.\nRunning Monte Carlo simulation number 300.\nRunning Monte Carlo simulation number 310.\nRunning Monte Carlo simulation number 320.\nRunning Monte Carlo simulation number 330.\nRunning Monte Carlo simulation number 340.\nRunning Monte Carlo simulation number 350.\nRunning Monte Carlo simulation number 360.\nRunning Monte Carlo simulation number 370.\nRunning Monte Carlo simulation number 380.\nRunning Monte Carlo simulation number 390.\nRunning Monte Carlo simulation number 400.\nRunning Monte Carlo simulation number 410.\nRunning Monte Carlo simulation number 420.\nRunning Monte Carlo simulation number 430.\nRunning Monte Carlo simulation number 440.\nRunning Monte Carlo simulation number 450.\nRunning Monte Carlo simulation number 460.\nRunning Monte Carlo simulation number 470.\nRunning Monte Carlo simulation number 480.\nRunning Monte Carlo simulation number 490.\n" ], [ "# Plot simulation outcomes\n# YOUR CODE HERE!", "_____no_output_____" ], [ "# Plot simulation outcomes\nline_plot = MC_even_dist.plot_simulation()", "_____no_output_____" ], [ "# delete\n# Plot probability distribution and confidence intervals\n# YOUR CODE HERE!", "_____no_output_____" ], [ "# Plot probability distribution and confidence intervals\n# YOUR CODE HERE!\ndist_plot = MC_even_dist.plot_distribution()", "_____no_output_____" ] ], [ [ "### Retirement Analysis", "_____no_output_____" ] ], [ [ "# delete\n# Fetch summary statistics from the Monte Carlo simulation results\n# YOUR CODE HERE!\n\n# Print summary statistics\n# YOUR CODE HERE!", "_____no_output_____" ], [ "# Fetch summary statistics from the Monte Carlo simulation results\n# YOUR CODE HERE!\ntbl = MC_even_dist.summarize_cumulative_return()\n\n# Print summary statistics\nprint (tbl)", "count 500.000000\nmean 9.683146\nstd 6.514698\nmin 1.320640\n25% 5.124632\n50% 8.047245\n75% 12.513634\nmax 51.017155\n95% CI Lower 2.421887\n95% CI Upper 25.440141\nName: 7560, dtype: float64\n" ] ], [ [ "### Calculate the expected portfolio return at the 95% lower and upper confidence intervals based on a `$20,000` initial investment.", "_____no_output_____" ] ], [ [ "# delete\n# # Set initial investment\n# initial_investment = 20000\n\n# # Use the lower and upper `95%` confidence intervals to calculate the range of the possible outcomes of our $20,000\n# # YOUR CODE HERE!\n\n# # Print results\n# print(f\"There is a 95% chance that an initial investment of ${initial_investment} in the portfolio\"\n# f\" over the next 30 years will end within in the range of\"\n# f\" ${ci_lower} and ${ci_upper}\")", "_____no_output_____" ], [ "# Set initial investment\ninitial_investment = 20000\n\n# Use the lower and upper `95%` confidence intervals to calculate the range of the possible outcomes of our $20,000\n# YOUR CODE HERE!\nci_lower = round(tbl[8]*initial_investment,2)\nci_upper = round(tbl[9]*initial_investment,2)\n\n\n# Print results\nprint(f\"There is a 95% chance that an initial investment of ${initial_investment} in the portfolio\"\n f\" over the next 30 years will end within in the range of\"\n f\" ${ci_lower} and ${ci_upper}\")", "There is a 95% chance that an initial investment of $20000 in the portfolio over the next 30 years will end within in the range of $48437.75 and $508802.81\n" ] ], [ [ "### Calculate the expected portfolio return at the `95%` lower and upper confidence intervals based on a `50%` increase in the initial investment.", "_____no_output_____" ] ], [ [ "# delete\n# # Set initial investment\n# initial_investment = 20000 * 1.5\n\n# # Use the lower and upper `95%` confidence intervals to calculate the range of the possible outcomes of our $30,000\n# # YOUR CODE HERE!\n\n# # Print results\n# print(f\"There is a 95% chance that an initial investment of ${initial_investment} in the portfolio\"\n# f\" over the next 30 years will end within in the range of\"\n# f\" ${ci_lower} and ${ci_upper}\")", "_____no_output_____" ], [ "# Set initial investment\ninitial_investment = 20000 * 1.5\n\n# Use the lower and upper `95%` confidence intervals to calculate the range of the possible outcomes of our $30,000\n# YOUR CODE HERE!\nci_lower = round(tbl[8]*initial_investment,2)\nci_upper = round(tbl[9]*initial_investment,2)\n\n# Print results\nprint(f\"There is a 95% chance that an initial investment of ${initial_investment} in the portfolio\"\n f\" over the next 30 years will end within in the range of\"\n f\" ${ci_lower} and ${ci_upper}\")", "There is a 95% chance that an initial investment of $30000.0 in the portfolio over the next 30 years will end within in the range of $72656.62 and $763204.22\n" ] ], [ [ "## Optional Challenge - Early Retirement\n\n\n### Five Years Retirement Option", "_____no_output_____" ] ], [ [ "# Configuring a Monte Carlo simulation to forecast 5 years cumulative returns\n# YOUR CODE HERE!\nMC_even_dist = MCSimulation(\n portfolio_data = df_stock_data,\n weights = [.4, .6],\n num_simulation = 500,\n num_trading_days = 252*5\n)", "_____no_output_____" ], [ "# Running a Monte Carlo simulation to forecast 5 years cumulative returns\n# YOUR CODE HERE!", "_____no_output_____" ], [ "# Running a Monte Carlo simulation to forecast 5 years cumulative returns\nMC_even_dist.calc_cumulative_return()", "Running Monte Carlo simulation number 0.\nRunning Monte Carlo simulation number 10.\nRunning Monte Carlo simulation number 20.\nRunning Monte Carlo simulation number 30.\nRunning Monte Carlo simulation number 40.\nRunning Monte Carlo simulation number 50.\nRunning Monte Carlo simulation number 60.\nRunning Monte Carlo simulation number 70.\nRunning Monte Carlo simulation number 80.\nRunning Monte Carlo simulation number 90.\nRunning Monte Carlo simulation number 100.\nRunning Monte Carlo simulation number 110.\nRunning Monte Carlo simulation number 120.\nRunning Monte Carlo simulation number 130.\nRunning Monte Carlo simulation number 140.\nRunning Monte Carlo simulation number 150.\nRunning Monte Carlo simulation number 160.\nRunning Monte Carlo simulation number 170.\nRunning Monte Carlo simulation number 180.\nRunning Monte Carlo simulation number 190.\nRunning Monte Carlo simulation number 200.\nRunning Monte Carlo simulation number 210.\nRunning Monte Carlo simulation number 220.\nRunning Monte Carlo simulation number 230.\nRunning Monte Carlo simulation number 240.\nRunning Monte Carlo simulation number 250.\nRunning Monte Carlo simulation number 260.\nRunning Monte Carlo simulation number 270.\nRunning Monte Carlo simulation number 280.\nRunning Monte Carlo simulation number 290.\nRunning Monte Carlo simulation number 300.\nRunning Monte Carlo simulation number 310.\nRunning Monte Carlo simulation number 320.\nRunning Monte Carlo simulation number 330.\nRunning Monte Carlo simulation number 340.\nRunning Monte Carlo simulation number 350.\nRunning Monte Carlo simulation number 360.\nRunning Monte Carlo simulation number 370.\nRunning Monte Carlo simulation number 380.\nRunning Monte Carlo simulation number 390.\nRunning Monte Carlo simulation number 400.\nRunning Monte Carlo simulation number 410.\nRunning Monte Carlo simulation number 420.\nRunning Monte Carlo simulation number 430.\nRunning Monte Carlo simulation number 440.\nRunning Monte Carlo simulation number 450.\nRunning Monte Carlo simulation number 460.\nRunning Monte Carlo simulation number 470.\nRunning Monte Carlo simulation number 480.\nRunning Monte Carlo simulation number 490.\n" ], [ "# Plot simulation outcomes\n# YOUR CODE HERE!", "_____no_output_____" ], [ "# Plot simulation outcomes\nline_plot = MC_even_dist.plot_simulation()", "_____no_output_____" ], [ "# Plot probability distribution and confidence intervals\n# YOUR CODE HERE!", "_____no_output_____" ], [ "# Plot probability distribution and confidence intervals\n# YOUR CODE HERE!\ndist_plot = MC_even_dist.plot_distribution()", "_____no_output_____" ], [ "# Fetch summary statistics from the Monte Carlo simulation results\n# YOUR CODE HERE!\n\n# Print summary statistics\n# YOUR CODE HERE!", "_____no_output_____" ], [ "# Fetch summary statistics from the Monte Carlo simulation results\n# YOUR CODE HERE!\ntbl = MC_even_dist.summarize_cumulative_return()\n\n# Print summary statistics\nprint (tbl)", "count 500.000000\nmean 1.469826\nstd 0.393313\nmin 0.630926\n25% 1.172070\n50% 1.420760\n75% 1.721414\nmax 2.939185\n95% CI Lower 0.854639\n95% CI Upper 2.365985\nName: 1260, dtype: float64\n" ], [ "# Set initial investment\n# YOUR CODE HERE!\n\n# Use the lower and upper `95%` confidence intervals to calculate the range of the possible outcomes of our $60,000\n# YOUR CODE HERE!\n\n# Print results\nprint(f\"There is a 95% chance that an initial investment of ${initial_investment} in the portfolio\"\n f\" over the next 5 years will end within in the range of\"\n f\" ${ci_lower_five} and ${ci_upper_five}\")", "There is a 95% chance that an initial investment of $30000.0 in the portfolio over the next 5 years will end within in the range of $50062.36 and $148745.63\n" ], [ "# Set initial investment\n# YOUR CODE HERE!\ninitial_investment = 60000\n# Use the lower and upper `95%` confidence intervals to calculate the range of the possible outcomes of our $60,000\n# YOUR CODE HERE!\nci_lower_five = round(tbl[8]*initial_investment,2)\nci_upper_five = round(tbl[9]*initial_investment,2)\n# Print results\nprint(f\"There is a 95% chance that an initial investment of ${initial_investment} in the portfolio\"\n f\" over the next 5 years will end within in the range of\"\n f\" ${ci_lower_five} and ${ci_upper_five}\")", "There is a 95% chance that an initial investment of $60000 in the portfolio over the next 5 years will end within in the range of $51278.36 and $141959.07\n" ] ], [ [ "### Ten Years Retirement Option", "_____no_output_____" ] ], [ [ "# Configuring a Monte Carlo simulation to forecast 10 years cumulative returns\n# YOUR CODE HERE!\nMC_even_dist = MCSimulation(\n portfolio_data = df_stock_data,\n weights = [.4, .6],\n num_simulation = 500,\n num_trading_days = 252*10\n)", "_____no_output_____" ], [ "# Running a Monte Carlo simulation to forecast 10 years cumulative returns\n# YOUR CODE HERE!", "_____no_output_____" ], [ "MC_even_dist.calc_cumulative_return()", "Running Monte Carlo simulation number 0.\nRunning Monte Carlo simulation number 10.\nRunning Monte Carlo simulation number 20.\nRunning Monte Carlo simulation number 30.\nRunning Monte Carlo simulation number 40.\nRunning Monte Carlo simulation number 50.\nRunning Monte Carlo simulation number 60.\nRunning Monte Carlo simulation number 70.\nRunning Monte Carlo simulation number 80.\nRunning Monte Carlo simulation number 90.\nRunning Monte Carlo simulation number 100.\nRunning Monte Carlo simulation number 110.\nRunning Monte Carlo simulation number 120.\nRunning Monte Carlo simulation number 130.\nRunning Monte Carlo simulation number 140.\nRunning Monte Carlo simulation number 150.\nRunning Monte Carlo simulation number 160.\nRunning Monte Carlo simulation number 170.\nRunning Monte Carlo simulation number 180.\nRunning Monte Carlo simulation number 190.\nRunning Monte Carlo simulation number 200.\nRunning Monte Carlo simulation number 210.\nRunning Monte Carlo simulation number 220.\nRunning Monte Carlo simulation number 230.\nRunning Monte Carlo simulation number 240.\nRunning Monte Carlo simulation number 250.\nRunning Monte Carlo simulation number 260.\nRunning Monte Carlo simulation number 270.\nRunning Monte Carlo simulation number 280.\nRunning Monte Carlo simulation number 290.\nRunning Monte Carlo simulation number 300.\nRunning Monte Carlo simulation number 310.\nRunning Monte Carlo simulation number 320.\nRunning Monte Carlo simulation number 330.\nRunning Monte Carlo simulation number 340.\nRunning Monte Carlo simulation number 350.\nRunning Monte Carlo simulation number 360.\nRunning Monte Carlo simulation number 370.\nRunning Monte Carlo simulation number 380.\nRunning Monte Carlo simulation number 390.\nRunning Monte Carlo simulation number 400.\nRunning Monte Carlo simulation number 410.\nRunning Monte Carlo simulation number 420.\nRunning Monte Carlo simulation number 430.\nRunning Monte Carlo simulation number 440.\nRunning Monte Carlo simulation number 450.\nRunning Monte Carlo simulation number 460.\nRunning Monte Carlo simulation number 470.\nRunning Monte Carlo simulation number 480.\nRunning Monte Carlo simulation number 490.\n" ], [ "# Plot simulation outcomes\n# YOUR CODE HERE!", "_____no_output_____" ], [ "# Plot simulation outcomes\nline_plot = MC_even_dist.plot_simulation()", "_____no_output_____" ], [ "# Plot probability distribution and confidence intervals\n# YOUR CODE HERE!", "_____no_output_____" ], [ "# Plot probability distribution and confidence intervals\ndist_plot = MC_even_dist.plot_distribution()", "_____no_output_____" ], [ "# Fetch summary statistics from the Monte Carlo simulation results\n# YOUR CODE HERE!\n\n# Print summary statistics\n# YOUR CODE HERE!", "_____no_output_____" ], [ "# Fetch summary statistics from the Monte Carlo simulation results\n# YOUR CODE HERE!\ntbl = MC_even_dist.summarize_cumulative_return()\n\n# Print summary statistics\nprint (tbl)", "count 500.000000\nmean 2.134640\nstd 0.847886\nmin 0.587560\n25% 1.573795\n50% 1.988818\n75% 2.559716\nmax 8.737952\n95% CI Lower 0.919582\n95% CI Upper 4.016863\nName: 2520, dtype: float64\n" ], [ "# Set initial investment\n# YOUR CODE HERE!\n\n# Use the lower and upper `95%` confidence intervals to calculate the range of the possible outcomes of our $60,000\n# YOUR CODE HERE!\n\n# Print results\nprint(f\"There is a 95% chance that an initial investment of ${initial_investment} in the portfolio\"\n f\" over the next 10 years will end within in the range of\"\n f\" ${ci_lower_ten} and ${ci_upper_ten}\")", "There is a 95% chance that an initial investment of $60000 in the portfolio over the next 10 years will end within in the range of $53013.83 and $246126.55\n" ], [ "# Set initial investment\n# YOUR CODE HERE!\ninitial_investment = 60000\n\n# Use the lower and upper `95%` confidence intervals to calculate the range of the possible outcomes of our $60,000\n# YOUR CODE HERE!\nci_lower_ten = round(tbl[8]*initial_investment,2)\nci_upper_ten = round(tbl[9]*initial_investment,2)\n\n# Print results\nprint(f\"There is a 95% chance that an initial investment of ${initial_investment} in the portfolio\"\n f\" over the next 10 years will end within in the range of\"\n f\" ${ci_lower_ten} and ${ci_upper_ten}\")", "There is a 95% chance that an initial investment of $60000 in the portfolio over the next 10 years will end within in the range of $55174.9 and $241011.77\n" ] ] ]

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

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/rjthompson22/bMSUS_XRay_Model/blob/master/mBSUS_XRay_Test_Program.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ] ], [ [ "!pip install h5py pyyaml\nimport tensorflow as tf\nfrom tensorflow import keras\n", "Requirement already satisfied: h5py in /usr/local/lib/python3.6/dist-packages (2.10.0)\nRequirement already satisfied: pyyaml in /usr/local/lib/python3.6/dist-packages (3.13)\nRequirement already satisfied: six in /usr/local/lib/python3.6/dist-packages (from h5py) (1.15.0)\nRequirement already satisfied: numpy>=1.7 in /usr/local/lib/python3.6/dist-packages (from h5py) (1.18.5)\n" ], [ "import urllib.request\n\nprint('Beginning file download with urllib2...')\n\nurl = 'https://drive.google.com/file/d/1Z0dlwhtS03lYhHv-f3ZT7nf_PgbIXBMB/view?usp=sharing'\nurllib.request.urlretrieve(url, 'basic_CNN.h5')\nbasic_CNN = keras.models.load_model('basic_CNN.h5')", "Beginning file download with urllib2...\n" ], [ "from google.colab import drive\ndrive.mount('/content/gdrive')\n\n%cd /content/gdrive/My Drive/Colab Notebooks/mBSUS", "/root/.keras/datasets/basic_CNN.h5\n" ], [ "#basic_CNN = keras.models.load_model(\"basic_CNN.h5\")", "_____no_output_____" ], [ "import numpy as np\nfrom google.colab import files\nfrom tensorflow.keras.preprocessing import image\n\nos.chdir('/content/')\n\nuploaded = files.upload()\n\nfor fn in uploaded.keys():\n path = '/content/' + fn\n img = image.load_img(path, target_size = (300,300))\n x = image.img_to_array(img)\n x = np.expand_dims(x, axis=0)\n\n images = np.vstack([x])\n \n classes = basic_CNN.predict(images, batch_size = 10)\n\n print(classes[0])\n\n if classes[0] > 0.5:\n print (fn + \" is pneumonia: \" + str(classes[0]))\n else:\n print(fn + \" is not pneumonia: \" + str(classes[0]))\n", "_____no_output_____" ] ] ]

[ "markdown", "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "### Abstract Factory Design Pattern", "_____no_output_____" ], [ ">An abstract factory is a generative design pattern that allows you to create families of related objects without getting attached to specific classes of created objects. The pattern is being implemented by creating an abstract class (for example - Factory), which is represented as an interface for creating system components. Then the classes that implement this interface are being written.", "_____no_output_____" ], [ "https://py.checkio.org/blog/design-patterns-part-1/", "_____no_output_____" ] ], [ [ "class AbstractFactory:\n def create_chair(self):\n raise NotImplementedError()\n \n def create_sofa(self):\n raise NotImplementedError()\n \n def create_table(self):\n raise NotImplementedError()", "_____no_output_____" ], [ "class Chair:\n def __init__(self, name):\n self._name = name\n \n def __str__(self):\n return self._name\n \nclass Sofa:\n def __init__(self, name):\n self._name = name\n \n def __str__(self):\n return self._name\n \nclass Table:\n def __init__(self, name):\n self._name = name\n \n def __str__(self):\n return self._name", "_____no_output_____" ], [ "class VictorianFactory(AbstractFactory):\n def create_chair(self):\n return Chair('victorian chair')\n \n def create_sofa(self):\n return Sofa('victorian sofa')\n \n def create_table(self):\n return Table('victorian table')", "_____no_output_____" ], [ "class ModernFactory(AbstractFactory):\n def create_chair(self):\n return Chair('modern chair')\n\n def create_sofa(self):\n return Sofa('modern sofa')\n\n def create_table(self):\n return Table('modern table')", "_____no_output_____" ], [ "class FuturisticFactory(AbstractFactory):\n def create_chair(self):\n return Chair('futuristic chair')\n\n def create_sofa(self):\n return Sofa('futuristic sofa')\n\n def create_table(self):\n return Table('futuristic table')", "_____no_output_____" ], [ "factory_1 = VictorianFactory()\nfactory_2 = ModernFactory()\nfactory_3 = FuturisticFactory()\nprint(factory_1.create_chair())\nprint(factory_1.create_sofa())\nprint(factory_1.create_table())\nprint(factory_2.create_chair())\nprint(factory_2.create_sofa())\nprint(factory_2.create_table())\nprint(factory_3.create_chair())\nprint(factory_3.create_sofa())\nprint(factory_3.create_table())", "victorian chair\nvictorian sofa\nvictorian table\nmodern chair\nmodern sofa\nmodern table\nfuturistic chair\nfuturistic sofa\nfuturistic table\n" ] ], [ [ "Example - https://py.checkio.org/mission/army-units/solve/", "_____no_output_____" ] ], [ [ "class Army:\n def train_swordsman(self, name):\n return NotImplementedError()\n \n def train_lancer(self, name):\n return NotImplementedError()\n \n def train_archer(self, name):\n return NotImplementedError()\n\nclass Swordsman:\n def __init__(self, soldier_type, name, army_type):\n self.army_type = army_type + ' swordsman'\n self.name = name\n self.soldier_type = soldier_type\n \n def introduce(self):\n return '{} {}, {}'.format(self.soldier_type, self.name, self.army_type)\n\nclass Lancer:\n def __init__(self, soldier_type, name, army_type):\n self.army_type = army_type + ' lancer'\n self.name = name\n self.soldier_type = soldier_type\n \n def introduce(self):\n return '{} {}, {}'.format(self.soldier_type, self.name, self.army_type)\n\nclass Archer:\n def __init__(self, soldier_type, name, army_type):\n self.army_type = army_type + ' archer'\n self.name = name\n self.soldier_type = soldier_type\n \n def introduce(self):\n return '{} {}, {}'.format(self.soldier_type, self.name, self.army_type)\n\nclass AsianArmy(Army):\n def __init__(self):\n self.army_type = 'Asian'\n \n def train_swordsman(self, name):\n return Swordsman('Samurai', name, self.army_type)\n \n def train_lancer(self, name):\n return Lancer('Ronin', name, self.army_type)\n \n def train_archer(self, name):\n return Archer('Shinobi', name, self.army_type)\n\nclass EuropeanArmy(Army):\n def __init__(self):\n self.army_type = 'European'\n \n def train_swordsman(self, name):\n return Swordsman('Knight', name, self.army_type)\n \n def train_lancer(self, name):\n return Lancer('Raubritter', name, self.army_type)\n \n def train_archer(self, name):\n return Archer('Ranger', name, self.army_type)\n\n\nif __name__ == '__main__':\n #These \"asserts\" using only for self-checking and not necessary for auto-testing\n\n my_army = EuropeanArmy()\n enemy_army = AsianArmy()\n\n soldier_1 = my_army.train_swordsman(\"Jaks\")\n soldier_2 = my_army.train_lancer(\"Harold\")\n soldier_3 = my_army.train_archer(\"Robin\")\n\n soldier_4 = enemy_army.train_swordsman(\"Kishimoto\")\n soldier_5 = enemy_army.train_lancer(\"Ayabusa\")\n soldier_6 = enemy_army.train_archer(\"Kirigae\")\n\n assert soldier_1.introduce() == \"Knight Jaks, European swordsman\"\n assert soldier_2.introduce() == \"Raubritter Harold, European lancer\"\n assert soldier_3.introduce() == \"Ranger Robin, European archer\"\n \n assert soldier_4.introduce() == \"Samurai Kishimoto, Asian swordsman\"\n assert soldier_5.introduce() == \"Ronin Ayabusa, Asian lancer\"\n assert soldier_6.introduce() == \"Shinobi Kirigae, Asian archer\"\n\n print(\"Coding complete? Let's try tests!\")\n", "Coding complete? Let's try tests!\n" ], [ "class Army:\n def train_swordsman(self, name):\n return Swordsman(self, name)\n\n def train_lancer(self, name):\n return Lancer(self, name)\n\n def train_archer(self, name):\n return Archer(self, name)\n\n def introduce(self, name, army_type):\n return f'{self.title[army_type]} {name}, {self.region} {army_type}'\n\n\nclass Fighter:\n def __init__(self, army, name):\n self.army = army\n self.name = name\n\n def introduce(self):\n return self.army.introduce(self.name, self.army_type)\n\n\nclass Swordsman(Fighter):\n army_type = 'swordsman'\n\nclass Lancer(Fighter):\n army_type = 'lancer'\n\nclass Archer(Fighter):\n army_type = 'archer'\n\nclass AsianArmy(Army):\n title = {'swordsman': 'Samurai', 'lancer': 'Ronin', 'archer': 'Shinobi'}\n region = 'Asian'\n\nclass EuropeanArmy(Army):\n title = {'swordsman': 'Knight', 'lancer': 'Raubritter', 'archer': 'Ranger'}\n region = 'European'\n\n\n\nif __name__ == '__main__':\n #These \"asserts\" using only for self-checking and not necessary for auto-testing\n\n my_army = EuropeanArmy()\n enemy_army = AsianArmy()\n\n soldier_1 = my_army.train_swordsman(\"Jaks\")\n soldier_2 = my_army.train_lancer(\"Harold\")\n soldier_3 = my_army.train_archer(\"Robin\")\n\n soldier_4 = enemy_army.train_swordsman(\"Kishimoto\")\n soldier_5 = enemy_army.train_lancer(\"Ayabusa\")\n soldier_6 = enemy_army.train_archer(\"Kirigae\")\n\n print(soldier_1.introduce())\n print(\"Knight Jaks, European swordsman\")\n print(soldier_2.introduce())\n print(soldier_3.introduce())\n assert soldier_1.introduce() == \"Knight Jaks, European swordsman\"\n assert soldier_2.introduce() == \"Raubritter Harold, European lancer\"\n assert soldier_3.introduce() == \"Ranger Robin, European archer\"\n \n assert soldier_4.introduce() == \"Samurai Kishimoto, Asian swordsman\"\n assert soldier_5.introduce() == \"Ronin Ayabusa, Asian lancer\"\n assert soldier_6.introduce() == \"Shinobi Kirigae, Asian archer\"\n\n print(\"Coding complete? Let's try tests!\")", "Knight Jaks, European swordsman\nKnight Jaks, European swordsman\nRaubritter Harold, European lancer\nRanger Robin, European archer\nCoding complete? Let's try tests!\n" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "## Cosine prediction\n\n### Data preparation\nIn this notebook, we will use fost to predict a cosine curve. Let's import `pandas` and `numpy` first.", "_____no_output_____" ] ], [ [ "import pandas as pd\nimport numpy as np", "_____no_output_____" ] ], [ [ "Next, generate cosine data.", "_____no_output_____" ] ], [ [ "train_df = pd.DataFrame()\n#we don't have actual timestamp in this case, use a default one\ntrain_df['Date'] = [pd.datetime(year=2000,month=1,day=1)+pd.Timedelta(days=i) for i in range(2000)] \ntrain_df['TARGET'] = [np.cos(i/2) for i in range(2000)]\n#there is not a 'Node' concept in this dataset, thus all the Node are 0\ntrain_df.loc[:, 'Node'] = 0", "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:4: FutureWarning: The pandas.datetime class is deprecated and will be removed from pandas in a future version. Import from datetime instead.\n after removing the cwd from sys.path.\n" ], [ "#show first 40 data\ntrain_df.iloc[:40].set_index('Date')['TARGET'].plot()", "_____no_output_____" ] ], [ [ "FOST only supports file path as input, save the data to `train.csv`", "_____no_output_____" ] ], [ [ "train_df.to_csv('train.csv',index=False)", "_____no_output_____" ] ], [ [ "### Load fost to predict future", "_____no_output_____" ] ], [ [ "import fostool\nfrom fostool.pipeline import Pipeline", "_____no_output_____" ], [ "#predict future 10 steps\nlookahead = 10\nfost = Pipeline(lookahead=lookahead, train_path='train.csv')", "2021-11-08 09:46:43 fostool/task/config_handler.py 26 \\ - INFO - yaml handler load path: /root/HierST/src/config/default.yaml\n2021-11-08 09:46:43 fostool/dataset/data_utils.py 402 \\ - INFO - Detected Sample Frequency: <Day>.\n2021-11-08 09:46:43 fostool/dataset/data_utils.py 426 \\ - INFO - 2000 Rows Before Time Reindex.\n2021-11-08 09:46:43 fostool/dataset/data_utils.py 428 \\ - INFO - 2000 Rows After Time Reindex.\n2021-11-08 09:46:43 fostool/dataset/data_utils.py 429 \\ - INFO - --------------------\n2021-11-08 09:46:43 fostool/dataset/data_utils.py 457 \\ - INFO - 2000 Rows Before Fill Missing.\n2021-11-08 09:46:43 fostool/dataset/data_utils.py 461 \\ - INFO - 2000 Rows After Fill Missing.\n2021-11-08 09:46:43 fostool/dataset/data_utils.py 462 \\ - INFO - --------------------\n" ], [ "#fit in one line\nfost.fit()", "2021-11-08 09:46:59 fostool/tools/trainer.py 129 \\ - INFO - On epoch 0, train loss 0.9837532142798106, val loss 1.1207878589630127\n2021-11-08 09:46:59 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:46:59 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 1.1207878589630127\n2021-11-08 09:46:59 fostool/tools/trainer.py 129 \\ - INFO - On epoch 1, train loss 0.8563873469829559, val loss 1.0255659818649292\n2021-11-08 09:46:59 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:46:59 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 1.0255659818649292\n2021-11-08 09:46:59 fostool/tools/trainer.py 129 \\ - INFO - On epoch 2, train loss 0.7682058115800222, val loss 0.9656972289085388\n2021-11-08 09:46:59 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:46:59 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.9656972289085388\n2021-11-08 09:46:59 fostool/tools/trainer.py 129 \\ - INFO - On epoch 3, train loss 0.6545088986555735, val loss 0.8323457837104797\n2021-11-08 09:46:59 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:46:59 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.8323457837104797\n2021-11-08 09:46:59 fostool/tools/trainer.py 129 \\ - INFO - On epoch 4, train loss 0.5395551969607671, val loss 0.7392954230308533\n2021-11-08 09:46:59 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:46:59 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.7392954230308533\n2021-11-08 09:46:59 fostool/tools/trainer.py 129 \\ - INFO - On epoch 5, train loss 0.41803252696990967, val loss 0.6145674586296082\n2021-11-08 09:46:59 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:46:59 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.6145674586296082\n2021-11-08 09:47:00 fostool/tools/trainer.py 129 \\ - INFO - On epoch 6, train loss 0.32809186975161236, val loss 0.4214208722114563\n2021-11-08 09:47:00 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:00 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.4214208722114563\n2021-11-08 09:47:00 fostool/tools/trainer.py 129 \\ - INFO - On epoch 7, train loss 0.20533942927916846, val loss 0.18617308139801025\n2021-11-08 09:47:00 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:00 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.18617308139801025\n2021-11-08 09:47:00 fostool/tools/trainer.py 129 \\ - INFO - On epoch 8, train loss 0.12697207058469454, val loss 0.09616108983755112\n2021-11-08 09:47:00 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:00 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.09616108983755112\n2021-11-08 09:47:00 fostool/tools/trainer.py 129 \\ - INFO - On epoch 9, train loss 0.05325042083859444, val loss 0.05969296023249626\n2021-11-08 09:47:00 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:00 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.05969296023249626\n2021-11-08 09:47:00 fostool/tools/trainer.py 129 \\ - INFO - On epoch 10, train loss 0.0261272427936395, val loss 0.011045633815228939\n2021-11-08 09:47:00 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:00 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.011045633815228939\n2021-11-08 09:47:00 fostool/tools/trainer.py 129 \\ - INFO - On epoch 11, train loss 0.012711387127637863, val loss 0.003730224911123514\n2021-11-08 09:47:00 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:00 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.003730224911123514\n2021-11-08 09:47:00 fostool/tools/trainer.py 129 \\ - INFO - On epoch 12, train loss 0.008007198184107741, val loss 0.004111938178539276\n2021-11-08 09:47:00 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:00 fostool/tools/trainer.py 129 \\ - INFO - On epoch 13, train loss 0.006259295002867778, val loss 0.0038592375349253416\n2021-11-08 09:47:00 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:01 fostool/tools/trainer.py 129 \\ - INFO - On epoch 14, train loss 0.005420396104454994, val loss 0.0015936356503516436\n2021-11-08 09:47:01 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:01 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.0015936356503516436\n2021-11-08 09:47:01 fostool/tools/trainer.py 129 \\ - INFO - On epoch 15, train loss 0.0032053233978028097, val loss 0.0009960811585187912\n2021-11-08 09:47:01 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:01 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.0009960811585187912\n2021-11-08 09:47:01 fostool/tools/trainer.py 129 \\ - INFO - On epoch 16, train loss 0.0030816590103010335, val loss 0.000641592894680798\n2021-11-08 09:47:01 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:01 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.000641592894680798\n2021-11-08 09:47:01 fostool/tools/trainer.py 129 \\ - INFO - On epoch 17, train loss 0.0024648708446572223, val loss 0.0006953171687200665\n2021-11-08 09:47:01 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:01 fostool/tools/trainer.py 129 \\ - INFO - On epoch 18, train loss 0.0018245108852473397, val loss 0.0006199749186635017\n2021-11-08 09:47:01 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:01 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.0006199749186635017\n2021-11-08 09:47:01 fostool/tools/trainer.py 129 \\ - INFO - On epoch 19, train loss 0.0013166117617705215, val loss 0.0006043225876055658\n2021-11-08 09:47:01 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:01 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.0006043225876055658\n2021-11-08 09:47:01 fostool/tools/trainer.py 129 \\ - INFO - On epoch 20, train loss 0.0011962133382136624, val loss 0.00043523628846742213\n2021-11-08 09:47:01 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:01 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.00043523628846742213\n2021-11-08 09:47:01 fostool/tools/trainer.py 129 \\ - INFO - On epoch 21, train loss 0.0010729129620206852, val loss 0.00041648070327937603\n2021-11-08 09:47:01 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:01 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.00041648070327937603\n2021-11-08 09:47:01 fostool/tools/trainer.py 129 \\ - INFO - On epoch 22, train loss 0.0009700010802286366, val loss 0.00017646198102738708\n2021-11-08 09:47:01 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:01 fostool/tools/trainer.py 137 \\ - INFO - For model KRNNModel_290a_10, current best val loss 0.00017646198102738708\n2021-11-08 09:47:02 fostool/tools/trainer.py 129 \\ - INFO - On epoch 23, train loss 0.0008185463472424696, val loss 0.0001829482353059575\n2021-11-08 09:47:02 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:02 fostool/tools/trainer.py 129 \\ - INFO - On epoch 24, train loss 0.0008309520975065728, val loss 0.00022877224546391517\n2021-11-08 09:47:02 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:02 fostool/tools/trainer.py 129 \\ - INFO - On epoch 25, train loss 0.0007969176222104579, val loss 0.00020417450286913663\n2021-11-08 09:47:02 fostool/tools/trainer.py 130 \\ - INFO - ------------\n2021-11-08 09:47:02 fostool/tools/trainer.py 129 \\ - INFO - On epoch 26, train loss 0.0007901331215786437, val loss 0.0003080877650063485\n2021-11-08 09:47:02 fostool/tools/trainer.py 130 \\ - INFO - ------------\n" ], [ "#get predict result\nres = fost.predict()", "2021-11-08 09:48:53 fostool/task/fusion.py 67 \\ - INFO - val_loss model_name\n0 0.000035 KRNNModel_290a_10\n1 0.000038 KRNNModel_3bf8_20\n2 0.000005 KRNNModel_d82f_30\n7 0.000031 SandwichModel_3f7a_30\n" ], [ "#plot prediction\nfost.plot(res, lookback_size=lookahead)", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Pytorch Basic", "_____no_output_____" ] ], [ [ "import torch\nimport torch.nn as nn\nimport torchvision\nimport torchvision.transforms as transforms\nimport matplotlib.pyplot as plt\nfrom IPython.display import clear_output", "_____no_output_____" ], [ "torch.cuda.is_available()", "_____no_output_____" ] ], [ [ "## Device", "_____no_output_____" ] ], [ [ "device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')", "_____no_output_____" ] ], [ [ "## Hyper Parameter", "_____no_output_____" ] ], [ [ "input_size = 784\nhidden_size = 500\nnum_class = 10\nepochs = 5\nbatch_size = 100\nlr = 0.001", "_____no_output_____" ] ], [ [ "## Load MNIST Dataset", "_____no_output_____" ] ], [ [ "train_dataset = torchvision.datasets.MNIST(root='../data', \n train=True, \n transform=transforms.ToTensor(), \n download=True)\ntest_dataset = torchvision.datasets.MNIST(root='../data', \n train=False, \n transform=transforms.ToTensor())", "_____no_output_____" ], [ "print('train dataset shape : ',train_dataset.data.shape)\nprint('test dataset shape : ',test_dataset.data.shape)\nplt.imshow(train_dataset.data[0])", "train dataset shape : torch.Size([60000, 28, 28])\ntest dataset shape : torch.Size([10000, 28, 28])\n" ], [ "train_loader = torch.utils.data.DataLoader(dataset=train_dataset, \n batch_size=batch_size, \n shuffle=True)\ntest_loader = torch.utils.data.DataLoader(dataset=test_dataset, \n batch_size=batch_size, \n shuffle=False)", "_____no_output_____" ] ], [ [ "## Simple Model", "_____no_output_____" ] ], [ [ "class NeuralNet(nn.Module):\n def __init__(self, input_size, hidden_size, num_class):\n super(NeuralNet, self).__init__()\n self.fc1 = nn.Linear(input_size,hidden_size)\n self.relu = nn.ReLU()\n self.fc2 = nn.Linear(hidden_size, num_class)\n \n def forward(self, x):\n out = self.fc1(x)\n out = self.relu(out)\n out = self.fc2(out)\n return out", "_____no_output_____" ], [ "model = NeuralNet(input_size,hidden_size,num_class).to(device)", "_____no_output_____" ] ], [ [ "## Loss and Optimizer", "_____no_output_____" ] ], [ [ "criterion = nn.CrossEntropyLoss()\noptimizer = torch.optim.Adam(model.parameters(), lr=lr) ", "_____no_output_____" ] ], [ [ "## Train", "_____no_output_____" ] ], [ [ "total_step = len(train_loader)\n\nfor epoch in range(epochs):\n for i, (images, labels) in enumerate(train_loader):\n images = images.reshape(-1,28*28).to(device)\n labels = labels.to(device)\n \n outputs = model(images)\n loss = criterion(outputs, labels)\n \n optimizer.zero_grad()\n loss.backward()\n optimizer.step()\n \n if (i+1) % 100 == 0:\n clear_output()\n print('EPOCH [{}/{}] STEP [{}/{}] Loss {: .4f})'\n .format(epoch+1, epochs, i+1, total_step, loss.item()))", "EPOCH [5/5] STEP [600/600] Loss 0.0343)\n" ] ], [ [ "## Test", "_____no_output_____" ] ], [ [ "with torch.no_grad():\n correct = 0\n total = 0\n for images, labels in test_loader:\n images = images.reshape(-1, 28*28).to(device)\n labels = labels.to(device)\n outputs = model(images)\n _, predicted = torch.max(outputs.data, 1)\n total += labels.size(0)\n correct += (predicted == labels).sum().item()\n\n print('Accuracy of the network on the 10000 test images: {} %'.format(100 * correct / total))", "Accuracy of the network on the 10000 test images: 97.85 %\n" ] ], [ [ "## save", "_____no_output_____" ] ], [ [ "torch.save(model.state_dict(), 'model.ckpt')", "_____no_output_____" ] ], [ [ "---", "_____no_output_____" ] ] ]

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

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

[ "ADSL" ]

[ "ADSL" ]

[ "ADSL" ]

[ [ [ "# import dependencies \nimport os\nimport pandas as pd", "_____no_output_____" ], [ "# generate path to csv\npath = '../web_design_challenge/Resources/cities.csv'\ncities= pd.read_csv(path)", "_____no_output_____" ], [ "# create html table and save to file \ncities_csv.to_html('table.html', index=False, classes=['table', 'table-striped', 'table-hover'])", "_____no_output_____" ] ] ]

[ "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "### Importar librerías y series de datos", "_____no_output_____" ] ], [ [ "import time\nstart = time.time()\n\n#importar datos y librerias\nimport numpy as np\nimport pandas as pd\nfrom datetime import datetime\nimport matplotlib.pyplot as plt\nfrom scipy import signal\nfrom sklearn.linear_model import LinearRegression\nfrom statsmodels.tsa.seasonal import seasonal_decompose\nfrom scipy.stats import boxcox\nfrom scipy import special\n#leer excel de datos y de dias especiales\ngeneral = pd.read_excel (r'C:\\Users\\Diana\\PAP\\Data\\Data1.xlsx')\nspecial_days= pd.read_excel (r'C:\\Users\\Diana\\PAP\\Data\\Christmas.xlsx')\n#convertir dias especiales a fechas en python\nfor column in special_days.columns:\n special_days[column] = pd.to_datetime(special_days[column])\ngeneral = general.set_index('fecha')", "_____no_output_____" ] ], [ [ "### Establecer las funciones a utilizar", "_____no_output_____" ] ], [ [ "def kronecker(data1:'Dataframe 1',data2:'Dataframe 2'):\n x=0\n data1_kron=data1[x:x+1]\n data2_kron=data2[x:x+1]\n Combinacion=np.kron(data1_kron,data2_kron)\n Combinacion=pd.DataFrame(Combinacion)\n for x in range(1,len(data1)):\n data1_kron=data1[x:x+1]\n data2_kron=data2[x:x+1]\n kron=np.kron(data1_kron,data2_kron)\n Kron=pd.DataFrame(kron)\n Combinacion=Combinacion.append(Kron)\n return Combinacion", "_____no_output_____" ], [ "def regresion_linear(X:'variables para regresion',y:'datos'):\n global model\n model.fit(X, y)\n coefficients=model.coef_\n return model.predict(X)", "_____no_output_____" ], [ "def comparacion(real,pred):\n comparacion=pd.DataFrame(columns=['real','prediccion','error'])\n comparacion.real=real\n comparacion.prediccion=pred\n comparacion.error=np.abs((comparacion.real.values-comparacion.prediccion)/comparacion.real)*100\n return comparacion", "_____no_output_____" ] ], [ [ "### Hacer variables dummies", "_____no_output_____" ] ], [ [ "n=-10\nfinal=general.MWh.tail(-n)\nonlyMWh=pd.DataFrame(general.MWh)\ngeneral['Month'] = general.index.month\ngeneral['Weekday_Name'] = general.index.weekday_name\ndates=general.index\ndummies = pd.get_dummies(general['Weekday_Name']).astype(int)\ndummies2 = pd.get_dummies(general['Month']).astype(int)\nDum=pd.DataFrame(dummies.join(dummies2))\nt=np.arange(0,len(onlyMWh))\nDum[\"t\"]= np.arange(0,len(onlyMWh))\nDum[\"tiempo\"]= np.arange(1,len(onlyMWh)+1)\nDum[\"ones\"]=np.ones(len(t))\nDum= Dum.set_index('t')", "_____no_output_____" ], [ "Dum[\"Dom santo\"]=0\nDum[\"NewYear\"]=0\nDum[\"Constitucion\"]=0\nDum[\"Benito\"]=0\nDum[\"Jue santo\"]=0\nDum[\"Vie santo\"]=0\nDum[\"Trabajo\"]=0\nDum[\"Madre\"]=0\nDum[\"Grito\"]=0\nDum[\"virgen\"]=0\nDum[\"muertos\"]=0\nDum[\"Virgen2\"]=0\nDum[\"Navidad\"]=0\nDum[\"elecciones\"]=0\nDum[\"toma\"]=0\nDum[\"sab santo\"]=0\nDum[\"rev\"]=0\n\nind=0 \nfor date in general.index:\n for date2 in special_days[\"Dom santo\"]:\n if date ==date2:\n Dum.iloc[ind,21]=1\n for date2 in special_days[\"NewYear\"]:\n if date ==date2:\n Dum.iloc[ind,22]=1\n for date2 in special_days[\"Constitucion\"]:\n if date ==date2:\n Dum.iloc[ind,23]=1\n for date2 in special_days[\"Benito\"]:\n if date ==date2:\n Dum.iloc[ind,24]=1\n for date2 in special_days[\"Jue santo\"]:\n if date ==date2:\n Dum.iloc[ind,25]=1\n for date2 in special_days[\"Vie santo\"]:\n if date ==date2:\n Dum.iloc[ind,26]=1\n for date2 in special_days[\"Trabajo\"]:\n if date ==date2:\n Dum.iloc[ind,27]=1\n for date2 in special_days[\"Madre\"]:\n if date ==date2:\n Dum.iloc[ind,28]=1\n for date2 in special_days[\"Grito\"]:\n if date ==date2:\n Dum.iloc[ind,29]=1\n for date2 in special_days[\"virgen\"]:\n if date ==date2:\n Dum.iloc[ind,30]=1\n for date2 in special_days[\"muertos\"]:\n if date ==date2:\n Dum.iloc[ind,31]=1\n for date2 in special_days[\"Virgen2\"]:\n if date ==date2:\n Dum.iloc[ind,32]=1\n for date2 in special_days[\"Navidad\"]:\n if date ==date2:\n Dum.iloc[ind,33]=1\n for date2 in special_days[\"elecciones\"]:\n if date ==date2:\n Dum.iloc[ind,34]=1\n for date2 in special_days[\"toma\"]:\n if date ==date2:\n Dum.iloc[ind,35]=1\n for date2 in special_days[\"sab santo\"]:\n if date ==date2:\n Dum.iloc[ind,36]=1\n for date2 in special_days[\"rev\"]:\n if date ==date2:\n Dum.iloc[ind,37]=1\n ind+=1\ndel Dum[\"Friday\"]\nDum.drop(Dum.columns[[15]], axis=1,inplace=True)", "_____no_output_____" ] ], [ [ "### Observar descomposición", "_____no_output_____" ] ], [ [ "part=general.MWh.tail(100)\nresult=seasonal_decompose(part, model='multiplicative')\nfig = result.seasonal.plot(figsize=(20,5))", "_____no_output_____" ] ], [ [ "Al ver la decomposición, se puede ver por la forma que fourier debe estblecerse en senos y cosenos absolutos, para que se parezca a la estacionalidad de la serie. Se agrega a las variables dummies esta estacionalidad semanal, que parece ser fundamental en los datos", "_____no_output_____" ], [ "### Detectar efecto de las variables dummies", "_____no_output_____" ] ], [ [ "t=np.arange(1,len(onlyMWh)+1)\nTiempo=pd.DataFrame(t)\nTiempo[\"one\"]=np.ones(len(onlyMWh))\nTiempo['sen']=np.abs(np.sin(((2*np.pi)/14)*t))\nTiempo['cos']=np.abs(np.cos(((2*np.pi)/14)*t))\nCombinacion=kronecker(Dum,Tiempo)", "_____no_output_____" ], [ "model = LinearRegression()\nprediction=regresion_linear(Combinacion[:n],general.MWh.values[:n])\nplt.figure(figsize=(10,5))\nplt.plot(onlyMWh.MWh.values[:n],label =\"Datos\")\nplt.plot(prediction,label=\"Predicción\")\nplt.ylabel(\"demanda en MWh\")\nplt.xlabel(\"días\")\nplt.legend()\n#plt.axis([1630,1650,120000,160000])\nplt.show()\ncomp=comparacion(onlyMWh.MWh.values[:n],prediction)\nMAPE=comp.error.mean()\nprint(\"MAPE = \",round(MAPE,4),\"%\")", "_____no_output_____" ] ], [ [ "### Obtener error de datos con variables dummies vs datos reales", "_____no_output_____" ] ], [ [ "Tabla=pd.DataFrame(columns=['regresion','datos','resta'])\nTabla[\"regresion\"]=prediction\nTabla[\"datos\"]=onlyMWh.MWh.values[:n]\nTabla[\"resta\"]=Tabla.datos-Tabla.regresion\nplt.plot(Tabla.resta)\nplt.show()", "_____no_output_____" ] ], [ [ "### Establecer las frecuencias que se debe considerar en la serie de fourier", "_____no_output_____" ] ], [ [ "f, Pxx_den = signal.periodogram(Tabla.resta, 1)\nplt.plot(1/f, Pxx_den)\nplt.xlabel('periodo')\nplt.ylabel('PSD')\nplt.show()", "C:\\Users\\Diana\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:2: RuntimeWarning: divide by zero encountered in true_divide\n \n" ], [ "top_50_periods = {}\n# get indices for 3 highest Pxx values\ntop50_freq_indices = np.flip(np.argsort(Pxx_den), 0)[2:12]\n\nfreqs = f[top50_freq_indices]\npower = Pxx_den[top50_freq_indices]\nperiods = 1 / np.array(freqs)\nmatrix=pd.DataFrame(columns=[\"power\",\"periods\"])\nmatrix.power=power\nmatrix.periods=periods\nprint(matrix)", " power periods\n0 8.749328e+09 1911.333333\n1 5.225123e+09 45.507937\n2 4.883263e+09 819.142857\n3 4.626260e+09 1146.800000\n4 4.413181e+09 36.522293\n5 4.324008e+09 955.666667\n6 4.069662e+09 573.400000\n7 4.064267e+09 521.272727\n8 3.968931e+09 61.000000\n9 3.914966e+09 91.015873\n" ] ], [ [ "### Hacer la regresión del efecto cruzado de variables dummies y senos/cosenos absolutos de frecuencia de error", "_____no_output_____" ] ], [ [ "sencos = pd.DataFrame()\nsencos[\"t\"]=np.arange(1,len(onlyMWh)+1)\nfor i in matrix.periods:\n sencos[\"{}_sen\".format(i)] = np.abs(np.sin(((2*np.pi)/i)*t))\n sencos[\"{}_cos\".format(i)] = np.abs(np.cos(((2*np.pi)/i)*t))\nsencos[\"unos\"] = 1\nsencos['sen']=np.abs(np.sin(((2*np.pi)/14)*t))\nsencos['cos']=np.abs(np.cos(((2*np.pi)/14)*t))\nsencos['sen1']=np.abs(np.sin(((2*np.pi)/365)*t))\nsencos['cos1']=np.abs(np.cos(((2*np.pi)/365)*t))\nsencos['sen2']=np.abs(np.sin(((2*np.pi)/28)*t))\nsencos['cos2']=np.abs(np.cos(((2*np.pi)/28)*t))", "_____no_output_____" ], [ "sencos_test=sencos[n:]\nsencos_train=sencos[0:n]\nDum_test=Dum[n:]\nDum_train=Dum[0:n]\nCombinacion=kronecker(Dum_train,sencos_train)", "_____no_output_____" ], [ "model = LinearRegression()\nprediction=regresion_linear(Combinacion,general.MWh.values[0:n])", "_____no_output_____" ] ], [ [ "### MAPE de la regresion", "_____no_output_____" ] ], [ [ "plt.figure(figsize=(10,5))\nplt.plot(onlyMWh.MWh[0:n].values,label =\"Datos\")\nplt.plot(prediction,label=\"Predicción\")\nplt.ylabel(\"demanda en MWh\")\nplt.xlabel(\"días\")\nplt.legend()\nplt.show()\n#%%obtener mape de regresión\ncomp=comparacion(onlyMWh.MWh.values[:n],prediction)\nMAPE=comp.error.mean()\nprint(\"MAPE = \",round(MAPE,4),\"%\")", "_____no_output_____" ] ], [ [ "### Graficar residuales de la regresión", "_____no_output_____" ] ], [ [ "Tabla=pd.DataFrame(columns=['regresion','datos','resta'])\nTabla[\"regresion\"]=prediction\nTabla[\"datos\"]=onlyMWh.MWh[0:n].values\nTabla[\"resta\"]=Tabla.datos-Tabla.regresion\nplt.plot(Tabla.resta)\nplt.show()", "_____no_output_____" ], [ "plt.hist(Tabla[\"resta\"],bins=50)\nplt.show()", "_____no_output_____" ], [ "resta=pd.DataFrame(Tabla[\"resta\"])", "_____no_output_____" ], [ "from statsmodels.tsa.arima_model import ARIMA\nmod = ARIMA(resta, order=(1,0,4))\nresults = mod.fit()\nplt.plot(resta)\nplt.plot(results.fittedvalues, color='red')", "_____no_output_____" ], [ "T=pd.DataFrame(columns=['regresion','datos','nuevo'])\nT[\"regresion\"]=results.fittedvalues\nT[\"datos\"]=resta\nT[\"nuevo\"]=T.datos-T.regresion\nplt.plot(T.nuevo)\nplt.show()\n", "_____no_output_____" ], [ "plt.figure(figsize=(10,5))\nplt.plot(onlyMWh.MWh[0:n].values,label=\"Reales\")\nplt.plot(prediction+results.fittedvalues,label=\"Predicción\")\n#plt.axis([1630,1650,120000,160000])\nplt.ylabel(\"demanda en MWh\")\nplt.xlabel(\"días\")\nplt.legend()\nplt.show()\n#%%obtener mape de regresión\ncomp=comparacion(onlyMWh.MWh[0:n].values,prediction+results.fittedvalues)\nMAPE=comp.error.mean()\nprint(\"MAPE = \",round(MAPE,4),\"%\")", "_____no_output_____" ] ], [ [ "### Gráfica de manera dinámica", "_____no_output_____" ] ], [ [ "extra=results.predict(len(onlyMWh.MWh[0:n]),len(onlyMWh.MWh[0:n])-n)\nextra=extra.iloc[1:]", "_____no_output_____" ], [ "from sklearn.linear_model import Lasso\nCombinaciontest=kronecker(Dum_test,sencos_test)\n#Initializing the Lasso Regressor with Normalization Factor as True\nlasso_reg = Lasso(normalize=True)\n#Fitting the Training data to the Lasso regressor\nlasso_reg.fit(Combinacion,onlyMWh.MWh[0:n])\ncoeff = lasso_reg.coef_\n#coeff\n#Predicting for X_test\ny_pred_lass =lasso_reg.predict(Combinaciontest)", "_____no_output_____" ], [ "coeff = np.sum(abs(lasso_reg.coef_)==0)\ncoeff", "_____no_output_____" ], [ "len(lasso_reg.coef_)", "_____no_output_____" ], [ "#comb=Combinacion\n#comb2=Combinaciontest\n#x=np.where(lasso_reg.coef_==0)", "_____no_output_____" ], [ "#comb=comb.drop(comb.columns[x], axis=1)\n#comb2=comb2.drop(comb2.columns[x], axis=1)", "_____no_output_____" ], [ "#from sklearn.linear_model import HuberRegressor\n#huber = HuberRegressor().fit(comb,onlyMWh.MWh[0:n])\n#hubpredict=huber.predict(comb2)", "_____no_output_____" ] ], [ [ "### todo para pronóstico", "_____no_output_____" ] ], [ [ "comp_pronostico=comparacion(final,y_pred_lass+extra.values)\n#comp_pronostico=comparacion(final,hubpredict+extra.values)\nMAPE=comp_pronostico.error.mean()\nplt.figure(figsize=(10,5))\nplt.plot(final,label=\"Real\")\nplt.plot(comp_pronostico.prediccion,label=\"Pronóstico\")\nplt.ylabel(\"demanda en MWh\")\nplt.xlabel(\"días\")\nplt.legend()\nplt.show()\nprint(\"MAPE = \",round(MAPE,4),\"%\")", "_____no_output_____" ], [ "comp_pronostico", "_____no_output_____" ], [ "end = time.time()\nprint((end - start)/60)", "6.8972692886988325\n" ], [ "model =LinearRegression()\nmodel.fit(comb,onlyMWh.MWh[0:n])\nprediction=model.predict(comb2)\ncomp_pronostico=comparacion(final,prediction+extra.values)\nMAPE=comp_pronostico.error.mean()\nplt.figure(figsize=(10,5))\nplt.plot(final,label=\"Real\")\nplt.plot(comp_pronostico.prediccion,label=\"Pronóstico\")\nplt.ylabel(\"demanda en MWh\")\nplt.xlabel(\"días\")\nplt.legend()\nplt.show()\nprint(\"MAPE = \",round(MAPE,4),\"%\")", "_____no_output_____" ], [ "comp_pronostico 799,39.39 58.39 13.01", "_____no_output_____" ], [ "lasso_reg = Lasso(normalize=True)\n#Fitting the Training data to the Lasso regressor\nlasso_reg.fit(comb,onlyMWh.MWh[0:n])\ncoeff = lasso_reg.coef_\n#coeff\n#Predicting for X_test\ny_pred_lass =lasso_reg.predict(comb2)\ncomp_pronostico=comparacion(final,y_pred_lass+extra.values)\nMAPE=comp_pronostico.error.mean()\nplt.figure(figsize=(10,5))\nplt.plot(final,label=\"Real\")\nplt.plot(comp_pronostico.prediccion,label=\"Pronóstico\")\nplt.ylabel(\"demanda en MWh\")\nplt.xlabel(\"días\")\nplt.legend()\nplt.show()\nprint(\"MAPE = \",round(MAPE,4),\"%\")", "_____no_output_____" ], [ "#coeff = lasso_reg.coef_\n#coeff", "_____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", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# МАДМО\n\n<a href=\"https://mipt.ru/science/labs/laboratoriya-neyronnykh-sistem-i-glubokogo-obucheniya/\"><img align=\"right\" src=\"https://avatars1.githubusercontent.com/u/29918795?v=4&s=200\" alt=\"DeepHackLab\" style=\"position:relative;top:-40px;right:10px;height:100px;\" /></a>\n\n\n\n### Физтех-Школа Прикладной математики и информатики МФТИ \n### Лаборатория нейронных сетей и глубокого обучения (DeepHackLab) \nДомашнее задание необходимо загрузить в общий репозиторий с именной папкой \n", "_____no_output_____" ], [ "## Домашнее задание 1\n### Основы Python и пакет NumPy\n---\n", "_____no_output_____" ] ], [ [ "import numpy as np\nimport random\nimport scipy.stats as sps", "_____no_output_____" ] ], [ [ "### Задача 1\nВ первой задаче вам предлагается перемножить две квадратные матрицы двумя способами -- без использования пакета ***numpy*** и с ним.", "_____no_output_____" ] ], [ [ "# Для генерации матриц используем фукнцию random -- она используется для генерации случайных объектов \n# функция sample создает случайную выборку. В качестве аргумента ей передается кортеж (i,j), здесь i -- число строк,\n# j -- число столбцов.\na = np.random.sample((1000,1000))\nb = np.random.sample((1000,1000))\n# выведите ранг каждой матрицы с помощью функции np.linalg.rank.\n# Используйте функцию shape, что она вывела?\n# ========\nrank_a = np.linalg.matrix_rank(a)\nprint(rank_a)\nprint(a.shape)\nprint(np.linalg.matrix_rank(b))\nprint(b.shape)\n# ========\n#print(a)\n#print(b)", "1000\n(1000, 1000)\n1000\n(1000, 1000)\n" ], [ " # здесь напишите перемножение матриц без \n # использования NumPy и выведите результат \n\ndef mult(a, b):\n rows_a = len(a)\n cols_a = len(a[0])\n rows_b = len(b)\n cols_b = len(b[0])\n if cols_a != rows_b:\n return 'Error'\n c = [[0 for row in range(cols_b)] for col in range(rows_a)] \n for i in range(rows_a):\n for j in range(cols_b):\n for k in range(cols_a):\n c[i][j] += a[i][k] * b[k][j]\n #return c \n\n", "_____no_output_____" ], [ "def np_mult(a, b):\n # здесь напишите перемножение матриц с\n # использованием NumPy и выведите результат\n return np.dot(a,b)", "_____no_output_____" ], [ "%%time\n# засечем время работы функции без NumPy\nmult(a,b)\n", "Wall time: 18min 9s\n" ], [ "%%time\n# засечем время работы функции с NumPy\nnp_mult(a,b)", "Wall time: 46.8 ms\n" ] ], [ [ "### Задача 2\nНапишите функцию, которая по данной последовательности $\\{A_i\\}_{i=1}^n$ строит последовательность $S_n$, где $S_k = \\frac{A_1 + ... + A_k}{k}$. \nАналогично -- с помощью библиотеки **NumPy** и без нее. Сравните скорость, объясните результат.", "_____no_output_____" ] ], [ [ "\n# функция, решающая задачу с помощью NumPy\ndef sec_av(A):\n return np.cumsum(A)/list(range(1,len(A)+1))\n pass", "_____no_output_____" ], [ "# функция без NumPy\ndef stupid_sec_av(A):\n S = [0 for i in range(len(A))]\n S[0] = A[0]\n \n for i in range(len(A)-1):\n S[i+1] = A[i+1] + S[i]\n \n numb = list(range(1,len(A)+1))\n \n for i in range(len(A)):\n S[i] = S[i] / numb[i]\n \n return S\n\n# зададим некоторую последовательность и проверим ее на ваших функциях. \n# Первая функция должна работать ~ в 50 раз быстрее\nA = sps.uniform.rvs(size=10 ** 7) \n\n%time S1 = sec_av(A)\n%time S2 = stupid_sec_av(A)\n#проверим корректность:\nnp.abs(S1 - S2).sum()", "Wall time: 1.39 s\nWall time: 10.7 s\n" ] ], [ [ "### Задача 3\n\nПусть задан некоторый массив $X$. Надо построить новый массив, где все элементы с нечетными индексами требуется заменить на число $a$ (если оно не указано, то на 1). Все элементы с четными индексами исходного массива нужно возвести в куб и записать в обратном порядке относительно позиций этих элементов. Массив $X$ при этом должен остаться без изменений. В конце требуется слить массив X с преобразованным X и вывести в обратном порядке. ", "_____no_output_____" ] ], [ [ "# функция, решающая задачу с помощью NumPy\n\ndef transformation(X, a = 1):\n \n X[1::2] = a\n \n X[::2] **= 3\n \n X[::2] = X[::2][::-1]\n \n \n return X", "_____no_output_____" ], [ "# функция, решающая задачу без NumPy\n\ndef stupid_transformation(X):\n t_odd = []\n t_ev = []\n t_ev_inv = []\n Y = []\n\n t_odd = int(round(len(X)/2)) * [1] \n\n for i in range(0,len(X),2):\n t_ev = t_ev + [round(X[i]**3,8)] \n \n for i in range(len(t_ev),0,-1):\n t_ev_inv = t_ev_inv + [temp_ev[i-1]]\n \n for i in range(min(len(t_ev_inv), len(t_odd))):\n Y = Y + [t_ev_inv[i]] + [t_odd[i]]\n\n if len(t_ev_inv) > len(t_odd):\n Y = Y + [t_ev_inv[-1]]\n \n if len(t_ev_inv) < len(t_odd):\n Y = Y + [t_odd[-1]]\n \n \n return Y", "_____no_output_____" ], [ "X = sps.uniform.rvs(size=10 ** 7) \n# здесь код эффективнее примерно в 20 раз. \n# если Вы вдруг соберетесь печатать массив без np -- лучше сначала посмотрите на его размер\n%time S1 = transformation(X)\n%time S2 = stupid_transformation(X)\n# проверим корректность:\nnp.abs(S1 - S2).sum()", "Wall time: 172 ms\n" ] ], [ [ "Почему методы ***numpy*** оказываются эффективнее?", "_____no_output_____" ] ], [ [ "# Написаны на C", "_____no_output_____" ] ], [ [ "## Дополнительные задачи", "_____no_output_____" ], [ "Дополнительные задачи подразумевают, что Вы самостоятельно разберётесь в некоторых функциях ***numpy***, чтобы их сделать. \n\nЭти задачи не являются обязательными, но могут повлиять на Ваш рейтинг в лучшую сторону (точные правила учёта доп. задач будут оглашены позже).", "_____no_output_____" ], [ "### Задача 4*", "_____no_output_____" ], [ "Дана функция двух переменных: $f(x, y) = sin(x)cos(y)$ (это просто такой красивый 3D-график), а также дана функция для отрисовки $f(x, y)$ (`draw_f()`), которая принимает на вход двумерную сетку, на которой будет вычисляться функция. \n\nВам нужно разобраться в том, как строить такие сетки (подсказка - это одна конкретная функция ***numpy***), и подать такую сетку на вход функции отрисовки.", "_____no_output_____" ] ], [ [ "from matplotlib import pyplot as plt\nfrom mpl_toolkits.mplot3d import Axes3D\n%matplotlib inline\n\ndef f(x, y):\n '''Функция двух переменных'''\n return np.sin(x) * np.cos(y)\n\ndef draw_f(grid_x, grid_y):\n '''Функция отрисовки функции f(x, y)'''\n fig = plt.figure(figsize=(10, 8))\n ax = Axes3D(fig)\n ax.plot_surface(grid_x, grid_y, f(grid_x, grid_y), cmap='inferno')\n plt.show()", "_____no_output_____" ], [ "\n\ni = np.arange(-1, 1, 0.02)\n\ngrid_x, grid_y = np.meshgrid(i, i)\n\ndraw_f(grid_x, grid_y)", "_____no_output_____" ] ], [ [ "### Задача 5*", "_____no_output_____" ], [ "Выберите любую картинку и загрузите ее в папку с кодом. При загрузке её размерность равна 3: **(w, h, num_channels)**, где **w** - ширина картинки в пикселях, **h** - высота картинки в пикселях, **num_channels** - количество каналов *(R, G, B, alpha)*.\n\nВам нужно \"развернуть\" картинку в одномерный массив размера w \\* h \\* num_channels, написав **одну строку кода**.", "_____no_output_____" ] ], [ [ "from matplotlib import pyplot as plt\n%matplotlib inline", "_____no_output_____" ], [ "path_to_image = './image.png'\nimage_array = plt.imread(path_to_image)\nplt.imshow(image_array);\n", "_____no_output_____" ], [ "flat_image_array = = image_array.flatten()", "_____no_output_____" ], [ "print(len(flat_image_array))", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Colab-pytorch-image-classification\r\nOriginal repo: [bentrevett/pytorch-image-classification](https://github.com/bentrevett/pytorch-image-classification)\r\n\r\n[SqueezeNet code](https://github.com/pytorch/vision/blob/master/torchvision/models/squeezenet.py): [pytorch/vision](https://github.com/pytorch/vision)\r\n\r\nMy fork: [styler00dollar/Colab-image-classification](https://github.com/styler00dollar/Colab-image-classification)\r\n\r\nThis colab is a combination of [this Colab](https://colab.research.google.com/github/bentrevett/pytorch-image-classification/blob/master/5_resnet.ipynb) and [my other Colab](https://colab.research.google.com/github/styler00dollar/Colab-image-classification/blob/master/5_(small)_ResNet.ipynb) to do SqueezeNet training. ", "_____no_output_____" ] ], [ [ "!nvidia-smi", "_____no_output_____" ] ], [ [ "# DATASET CREATION", "_____no_output_____" ] ], [ [ "#@title Mount Google Drive\r\nfrom google.colab import drive\r\ndrive.mount('/content/drive')\r\nprint('Google Drive connected.')", "_____no_output_____" ], [ "# copy data somehow\r\n!mkdir '/content/classification'\r\n!mkdir '/content/classification/images'\r\n!cp \"/content/drive/MyDrive/classification_v2.7z\" \"/content/classification/images/classification.7z\"\r\n%cd /content/classification/images\r\n!7z x \"classification.7z\"\r\n!rm -rf /content/classification/images/classification.7z", "_____no_output_____" ], [ "#@title dataset creation\nTRAIN_RATIO = 0.90 #@param {type:\"number\"}\nimport os\nimport shutil\nfrom tqdm import tqdm\n#data_dir = os.path.join(ROOT, 'CUB_200_2011')\ndata_dir = '/content/classification' #@param {type:\"string\"}\nimages_dir = os.path.join(data_dir, 'images')\ntrain_dir = os.path.join(data_dir, 'train')\ntest_dir = os.path.join(data_dir, 'test')\n\nif os.path.exists(train_dir):\n shutil.rmtree(train_dir) \nif os.path.exists(test_dir):\n shutil.rmtree(test_dir)\n \nos.makedirs(train_dir)\nos.makedirs(test_dir)\n\nclasses = os.listdir(images_dir)\n\nfor c in classes:\n \n class_dir = os.path.join(images_dir, c)\n \n images = os.listdir(class_dir)\n \n n_train = int(len(images) * TRAIN_RATIO)\n \n train_images = images[:n_train]\n test_images = images[n_train:]\n \n os.makedirs(os.path.join(train_dir, c), exist_ok = True)\n os.makedirs(os.path.join(test_dir, c), exist_ok = True)\n \n for image in tqdm(train_images):\n image_src = os.path.join(class_dir, image)\n image_dst = os.path.join(train_dir, c, image) \n shutil.copyfile(image_src, image_dst)\n \n for image in tqdm(test_images):\n image_src = os.path.join(class_dir, image)\n image_dst = os.path.join(test_dir, c, image) \n shutil.copyfile(image_src, image_dst)", "_____no_output_____" ] ], [ [ "# CALC MEANS & STDS", "_____no_output_____" ] ], [ [ "#@title print means and stds\nimport torch\nimport torchvision.transforms as transforms\nimport torchvision.datasets as datasets\nfrom tqdm import tqdm\ntrain_data = datasets.ImageFolder(root = train_dir, \n transform = transforms.ToTensor())\n\nmeans = torch.zeros(3)\nstds = torch.zeros(3)\n\nfor img, label in tqdm(train_data):\n means += torch.mean(img, dim = (1,2))\n stds += torch.std(img, dim = (1,2))\n\nmeans /= len(train_data)\nstds /= len(train_data)\n\nprint(\"\\n\")\nprint(f'Calculated means: {means}')\nprint(f'Calculated stds: {stds}')", "_____no_output_____" ] ], [ [ "# TRAIN", "_____no_output_____" ] ], [ [ "#@title import, seed, transforms, dataloader, functions, plot, model, parameter\n%cd /content/\nfrom tqdm import tqdm\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torch.optim as optim\nimport torch.optim.lr_scheduler as lr_scheduler\nfrom torch.optim.lr_scheduler import _LRScheduler\nimport torch.utils.data as data\n\nimport torchvision.transforms as transforms\nimport torchvision.datasets as datasets\nimport torchvision.models as models\n\nfrom sklearn import decomposition\nfrom sklearn import manifold\nfrom sklearn.metrics import confusion_matrix\nfrom sklearn.metrics import ConfusionMatrixDisplay\nimport matplotlib.pyplot as plt\nimport numpy as np\n\nimport copy\nfrom collections import namedtuple\nimport os\nimport random\nimport shutil\n\nSEED = 1234 #@param {type:\"number\"}\n\nrandom.seed(SEED)\nnp.random.seed(SEED)\ntorch.manual_seed(SEED)\ntorch.cuda.manual_seed(SEED)\ntorch.backends.cudnn.deterministic = True\n\ntrain_dir = '/content/classification/train' #@param {type:\"string\"}\ntest_dir = '/content/classification/test' #@param {type:\"string\"}\npretrained_size = 256 #@param {type:\"number\"}\npretrained_means = [0.6838, 0.6086, 0.6063] #@param {type:\"raw\"}\npretrained_stds= [0.2411, 0.2403, 0.2306] #@param {type:\"raw\"}\n\n\n#https://github.com/mit-han-lab/data-efficient-gans/blob/master/DiffAugment_pytorch.py\nimport torch\nimport torch.nn.functional as F\n\n\ndef DiffAugment(x, policy='', channels_first=True):\n if policy:\n if not channels_first:\n x = x.permute(0, 3, 1, 2)\n for p in policy.split(','):\n for f in AUGMENT_FNS[p]:\n x = f(x)\n if not channels_first:\n x = x.permute(0, 2, 3, 1)\n x = x.contiguous()\n return x\n\n\ndef rand_brightness(x):\n x = x + (torch.rand(x.size(0), 1, 1, 1, dtype=x.dtype, device=x.device) - 0.5)\n return x\n\n\ndef rand_saturation(x):\n x_mean = x.mean(dim=1, keepdim=True)\n x = (x - x_mean) * (torch.rand(x.size(0), 1, 1, 1, dtype=x.dtype, device=x.device) * 2) + x_mean\n return x\n\n\ndef rand_contrast(x):\n x_mean = x.mean(dim=[1, 2, 3], keepdim=True)\n x = (x - x_mean) * (torch.rand(x.size(0), 1, 1, 1, dtype=x.dtype, device=x.device) + 0.5) + x_mean\n return x\n\n\ndef rand_translation(x, ratio=0.125):\n shift_x, shift_y = int(x.size(2) * ratio + 0.5), int(x.size(3) * ratio + 0.5)\n translation_x = torch.randint(-shift_x, shift_x + 1, size=[x.size(0), 1, 1], device=x.device)\n translation_y = torch.randint(-shift_y, shift_y + 1, size=[x.size(0), 1, 1], device=x.device)\n grid_batch, grid_x, grid_y = torch.meshgrid(\n torch.arange(x.size(0), dtype=torch.long, device=x.device),\n torch.arange(x.size(2), dtype=torch.long, device=x.device),\n torch.arange(x.size(3), dtype=torch.long, device=x.device),\n )\n grid_x = torch.clamp(grid_x + translation_x + 1, 0, x.size(2) + 1)\n grid_y = torch.clamp(grid_y + translation_y + 1, 0, x.size(3) + 1)\n x_pad = F.pad(x, [1, 1, 1, 1, 0, 0, 0, 0])\n x = x_pad.permute(0, 2, 3, 1).contiguous()[grid_batch, grid_x, grid_y].permute(0, 3, 1, 2)\n return x\n\n\ndef rand_cutout(x, ratio=0.5):\n cutout_size = int(x.size(2) * ratio + 0.5), int(x.size(3) * ratio + 0.5)\n offset_x = torch.randint(0, x.size(2) + (1 - cutout_size[0] % 2), size=[x.size(0), 1, 1], device=x.device)\n offset_y = torch.randint(0, x.size(3) + (1 - cutout_size[1] % 2), size=[x.size(0), 1, 1], device=x.device)\n grid_batch, grid_x, grid_y = torch.meshgrid(\n torch.arange(x.size(0), dtype=torch.long, device=x.device),\n torch.arange(cutout_size[0], dtype=torch.long, device=x.device),\n torch.arange(cutout_size[1], dtype=torch.long, device=x.device),\n )\n grid_x = torch.clamp(grid_x + offset_x - cutout_size[0] // 2, min=0, max=x.size(2) - 1)\n grid_y = torch.clamp(grid_y + offset_y - cutout_size[1] // 2, min=0, max=x.size(3) - 1)\n mask = torch.ones(x.size(0), x.size(2), x.size(3), dtype=x.dtype, device=x.device)\n mask[grid_batch, grid_x, grid_y] = 0\n x = x * mask.unsqueeze(1)\n return x\n\n\nAUGMENT_FNS = {\n 'color': [rand_brightness, rand_saturation, rand_contrast],\n 'translation': [rand_translation],\n 'cutout': [rand_cutout],\n}\n\n\ntrain_transforms = transforms.Compose([\n transforms.Resize(pretrained_size),\n transforms.RandomRotation(5),\n transforms.RandomHorizontalFlip(0.5),\n transforms.RandomCrop(pretrained_size, padding = 10),\n transforms.ToTensor(),\n transforms.Normalize(mean = pretrained_means, \n std = pretrained_stds)\n ])\n\ntest_transforms = transforms.Compose([\n transforms.Resize(pretrained_size),\n transforms.CenterCrop(pretrained_size),\n transforms.ToTensor(),\n transforms.Normalize(mean = pretrained_means, \n std = pretrained_stds)\n ])\n\ntrain_data = datasets.ImageFolder(root = train_dir, \n transform = train_transforms)\n\ntest_data = datasets.ImageFolder(root = test_dir, \n transform = test_transforms) \n\nVALID_RATIO = 0.90 #@param {type:\"number\"}\n\nn_train_examples = int(len(train_data) * VALID_RATIO)\nn_valid_examples = len(train_data) - n_train_examples\n\ntrain_data, valid_data = data.random_split(train_data, \n [n_train_examples, n_valid_examples])\n\nvalid_data = copy.deepcopy(valid_data)\nvalid_data.dataset.transform = test_transforms\n\nprint(f'Number of training examples: {len(train_data)}')\nprint(f'Number of validation examples: {len(valid_data)}')\nprint(f'Number of testing examples: {len(test_data)}')\n\nBATCH_SIZE = 32 #@param {type:\"number\"}\n\ntrain_iterator = data.DataLoader(train_data, \n shuffle = True, \n batch_size = BATCH_SIZE)\n\nvalid_iterator = data.DataLoader(valid_data, \n batch_size = BATCH_SIZE)\n\ntest_iterator = data.DataLoader(test_data, \n batch_size = BATCH_SIZE)\n\ndef normalize_image(image):\n image_min = image.min()\n image_max = image.max()\n image.clamp_(min = image_min, max = image_max)\n image.add_(-image_min).div_(image_max - image_min + 1e-5)\n return image \ndef plot_images(images, labels, classes, normalize = True):\n\n n_images = len(images)\n\n rows = int(np.sqrt(n_images))\n cols = int(np.sqrt(n_images))\n\n fig = plt.figure(figsize = (15, 15))\n\n for i in range(rows*cols):\n\n ax = fig.add_subplot(rows, cols, i+1)\n \n image = images[i]\n\n if normalize:\n image = normalize_image(image)\n\n ax.imshow(image.permute(1, 2, 0).cpu().numpy())\n label = classes[labels[i]]\n ax.set_title(label)\n ax.axis('off')\n\nN_IMAGES = 25 #@param {type:\"number\"}\n\nimages, labels = zip(*[(image, label) for image, label in \n [train_data[i] for i in range(N_IMAGES)]])\n\nclasses = test_data.classes\n\nplot_images(images, labels, classes)\n\ndef format_label(label):\n label = label.split('.')[-1]\n label = label.replace('_', ' ')\n label = label.title()\n label = label.replace(' ', '')\n return label\n\ntest_data.classes = [format_label(c) for c in test_data.classes]\n\nclasses = test_data.classes\n\nplot_images(images, labels, classes)\n\n\n#https://github.com/pytorch/vision/blob/master/torchvision/models/squeezenet.py\nimport torch\nimport torch.nn as nn\nimport torch.nn.init as init\n#from .utils import load_state_dict_from_url\nfrom typing import Any\n\n#__all__ = ['SqueezeNet', 'squeezenet1_0', 'squeezenet1_1']\n\nmodel_urls = {\n '1_0': 'https://download.pytorch.org/models/squeezenet1_0-a815701f.pth',\n '1_1': 'https://download.pytorch.org/models/squeezenet1_1-f364aa15.pth',\n}\n\n\nclass Fire(nn.Module):\n\n def __init__(\n self,\n inplanes: int,\n squeeze_planes: int,\n expand1x1_planes: int,\n expand3x3_planes: int\n ) -> None:\n super(Fire, self).__init__()\n self.inplanes = inplanes\n self.squeeze = nn.Conv2d(inplanes, squeeze_planes, kernel_size=1)\n self.squeeze_activation = nn.ReLU(inplace=True)\n self.expand1x1 = nn.Conv2d(squeeze_planes, expand1x1_planes,\n kernel_size=1)\n self.expand1x1_activation = nn.ReLU(inplace=True)\n self.expand3x3 = nn.Conv2d(squeeze_planes, expand3x3_planes,\n kernel_size=3, padding=1)\n self.expand3x3_activation = nn.ReLU(inplace=True)\n\n def forward(self, x: torch.Tensor) -> torch.Tensor:\n x = self.squeeze_activation(self.squeeze(x))\n return torch.cat([\n self.expand1x1_activation(self.expand1x1(x)),\n self.expand3x3_activation(self.expand3x3(x))\n ], 1)\n\n\nclass SqueezeNet(nn.Module):\n\n def __init__(\n self,\n version: str = '1_0',\n num_classes: int = 1000\n ) -> None:\n super(SqueezeNet, self).__init__()\n self.num_classes = num_classes\n if version == '1_0':\n self.features = nn.Sequential(\n nn.Conv2d(3, 96, kernel_size=7, stride=2),\n nn.ReLU(inplace=True),\n nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True),\n Fire(96, 16, 64, 64),\n Fire(128, 16, 64, 64),\n Fire(128, 32, 128, 128),\n nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True),\n Fire(256, 32, 128, 128),\n Fire(256, 48, 192, 192),\n Fire(384, 48, 192, 192),\n Fire(384, 64, 256, 256),\n nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True),\n Fire(512, 64, 256, 256),\n )\n elif version == '1_1':\n self.features = nn.Sequential(\n nn.Conv2d(3, 64, kernel_size=3, stride=2),\n nn.ReLU(inplace=True),\n nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True),\n Fire(64, 16, 64, 64),\n Fire(128, 16, 64, 64),\n nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True),\n Fire(128, 32, 128, 128),\n Fire(256, 32, 128, 128),\n nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True),\n Fire(256, 48, 192, 192),\n Fire(384, 48, 192, 192),\n Fire(384, 64, 256, 256),\n Fire(512, 64, 256, 256),\n )\n else:\n # FIXME: Is this needed? SqueezeNet should only be called from the\n # FIXME: squeezenet1_x() functions\n # FIXME: This checking is not done for the other models\n raise ValueError(\"Unsupported SqueezeNet version {version}:\"\n \"1_0 or 1_1 expected\".format(version=version))\n\n # Final convolution is initialized differently from the rest\n final_conv = nn.Conv2d(512, self.num_classes, kernel_size=1)\n self.classifier = nn.Sequential(\n nn.Dropout(p=0.5),\n final_conv,\n nn.ReLU(inplace=True),\n nn.AdaptiveAvgPool2d((1, 1))\n )\n\n for m in self.modules():\n if isinstance(m, nn.Conv2d):\n if m is final_conv:\n init.normal_(m.weight, mean=0.0, std=0.01)\n else:\n init.kaiming_uniform_(m.weight)\n if m.bias is not None:\n init.constant_(m.bias, 0)\n\n def forward(self, x: torch.Tensor) -> torch.Tensor:\n x = self.features(x)\n x = self.classifier(x)\n return torch.flatten(x, 1)\n\n\ndef _squeezenet(version: str, pretrained: bool, progress: bool, **kwargs: Any) -> SqueezeNet:\n model = SqueezeNet(version, **kwargs)\n if pretrained:\n arch = 'squeezenet' + version\n state_dict = load_state_dict_from_url(model_urls[arch],\n progress=progress)\n model.load_state_dict(state_dict)\n return model\n\n\ndef squeezenet1_0(pretrained: bool = False, progress: bool = True, **kwargs: Any) -> SqueezeNet:\n r\"\"\"SqueezeNet model architecture from the `\"SqueezeNet: AlexNet-level\n accuracy with 50x fewer parameters and <0.5MB model size\"\n <https://arxiv.org/abs/1602.07360>`_ paper.\n\n Args:\n pretrained (bool): If True, returns a model pre-trained on ImageNet\n progress (bool): If True, displays a progress bar of the download to stderr\n \"\"\"\n return _squeezenet('1_0', pretrained, progress, **kwargs)\n\n\ndef squeezenet1_1(pretrained: bool = False, progress: bool = True, **kwargs: Any) -> SqueezeNet:\n r\"\"\"SqueezeNet 1.1 model from the `official SqueezeNet repo\n <https://github.com/DeepScale/SqueezeNet/tree/master/SqueezeNet_v1.1>`_.\n SqueezeNet 1.1 has 2.4x less computation and slightly fewer parameters\n than SqueezeNet 1.0, without sacrificing accuracy.\n\n Args:\n pretrained (bool): If True, returns a model pre-trained on ImageNet\n progress (bool): If True, displays a progress bar of the download to stderr\n \"\"\"\n return _squeezenet('1_1', pretrained, progress, **kwargs)\n\n\"\"\"\n#https://github.com/pytorch/vision/blob/master/torchvision/models/utils.py\ntry:\n from torch.hub import load_state_dict_from_url\nexcept ImportError:\n from torch.utils.model_zoo import load_url as load_state_dict_from_url\n\"\"\"\n\nmodel_train = '1_1' #@param [\"1_0\", \"1_1\"] {type:\"string\"}\nif model_train == '1_0':\n model = SqueezeNet(num_classes=len(test_data.classes), version='1_0')\n #state_dict = load_state_dict_from_url(model_urls[model_train],\n # progress=True)\n #model.load_state_dict(state_dict)\nelif model_train == '1_1':\n model = SqueezeNet(num_classes=len(test_data.classes), version='1_1')\n #state_dict = load_state_dict_from_url(model_urls[model_train],\n # progress=True)\n #model.load_state_dict(state_dict)\n\n\ndef count_parameters(model):\n return sum(p.numel() for p in model.parameters() if p.requires_grad)\n\nprint(f'The model has {count_parameters(model):,} trainable parameters')\n\nSTART_LR = 1e-7 #@param {type:\"number\"}\n\noptimizer = optim.Adam(model.parameters(), lr=START_LR)\n\ndevice = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n\ncriterion = nn.CrossEntropyLoss()\n\nmodel = model.to(device)\ncriterion = criterion.to(device)\n\nclass LRFinder:\n def __init__(self, model, optimizer, criterion, device):\n \n self.optimizer = optimizer\n self.model = model\n self.criterion = criterion\n self.device = device\n \n torch.save(model.state_dict(), 'init_params.pt')\n\n def range_test(self, iterator, end_lr = 10, num_iter = 100, \n smooth_f = 0.05, diverge_th = 5):\n \n lrs = []\n losses = []\n best_loss = float('inf')\n\n lr_scheduler = ExponentialLR(self.optimizer, end_lr, num_iter)\n \n iterator = IteratorWrapper(iterator)\n \n for iteration in tqdm(range(num_iter)):\n\n loss = self._train_batch(iterator)\n\n #update lr\n lr_scheduler.step()\n \n lrs.append(lr_scheduler.get_lr()[0])\n\n if iteration > 0:\n loss = smooth_f * loss + (1 - smooth_f) * losses[-1]\n \n if loss < best_loss:\n best_loss = loss\n\n losses.append(loss)\n \n if loss > diverge_th * best_loss:\n print(\"Stopping early, the loss has diverged\")\n break\n \n #reset model to initial parameters\n model.load_state_dict(torch.load('init_params.pt'))\n \n return lrs, losses\n\n def _train_batch(self, iterator):\n \n self.model.train()\n \n self.optimizer.zero_grad()\n \n x, y = iterator.get_batch()\n \n x = x.to(self.device)\n y = y.to(self.device)\n \n y_pred, _ = self.model(x)\n \n loss = self.criterion(y_pred, y)\n \n loss.backward()\n \n self.optimizer.step()\n \n return loss.item()\n\nclass ExponentialLR(_LRScheduler):\n def __init__(self, optimizer, end_lr, num_iter, last_epoch=-1):\n self.end_lr = end_lr\n self.num_iter = num_iter\n super(ExponentialLR, self).__init__(optimizer, last_epoch)\n\n def get_lr(self):\n curr_iter = self.last_epoch + 1\n r = curr_iter / self.num_iter\n return [base_lr * (self.end_lr / base_lr) ** r for base_lr in self.base_lrs]\n\nclass IteratorWrapper:\n def __init__(self, iterator):\n self.iterator = iterator\n self._iterator = iter(iterator)\n\n def __next__(self):\n try:\n inputs, labels = next(self._iterator)\n except StopIteration:\n self._iterator = iter(self.iterator)\n inputs, labels, *_ = next(self._iterator)\n\n return inputs, labels\n\n def get_batch(self):\n return next(self)\n\n\ndef calculate_topk_accuracy(y_pred, y, k = 5):\n with torch.no_grad():\n batch_size = y.shape[0]\n\n _, top_pred = y_pred.topk(k=1)\n \n top_pred = top_pred.t()\n correct = top_pred.eq(y.view(1, -1).expand_as(top_pred))\n correct_1 = correct[:1].view(-1).float().sum(0, keepdim = True)\n #correct_k = correct[:k].view(-1).float().sum(0, keepdim = True)\n acc_1 = correct_1 / batch_size\n #acc_k = correct_k / batch_size\n acc_k = 0\n return acc_1, acc_k\n\n\ndef train(model, iterator, optimizer, criterion, scheduler, device, current_epoch):\n \n epoch_loss = 0\n epoch_acc_1 = 0\n epoch_acc_5 = 0\n \n model.train()\n policy = 'color,translation,cutout' #@param {type:\"string\"}\n diffaug_activate = True #@param [\"False\", \"True\"] {type:\"raw\"}\n #https://stackoverflow.com/questions/45465031/printing-text-below-tqdm-progress-bar\n with tqdm(iterator, position=1, bar_format='{desc}') as desc:\n\n for (x, y) in tqdm(iterator, position=0):\n x = x.to(device)\n y = y.to(device)\n \n optimizer.zero_grad()\n \n if diffaug_activate == False:\n y_pred = model(x)\n else:\n y_pred = model(DiffAugment(x, policy=policy))\n\n loss = criterion(y_pred, y)\n \n acc_1, acc_5 = calculate_topk_accuracy(y_pred, y)\n\n \n loss.backward()\n \n optimizer.step()\n \n scheduler.step()\n \n epoch_loss += loss.item()\n epoch_acc_1 += acc_1.item()\n #epoch_acc_5 += acc_5.item()\n \n epoch_loss /= len(iterator)\n epoch_acc_1 /= len(iterator)\n desc.set_description(f'Epoch: {current_epoch+1}')\n desc.set_description(f'\\tTrain Loss: {epoch_loss:.3f} | Train Acc @1: {epoch_acc_1*100:6.2f}% | ' \\\n f'Train Acc @5: {epoch_acc_5*100:6.2f}%')\n\n\n return epoch_loss, epoch_acc_1, epoch_acc_5\n\ndef evaluate(model, iterator, criterion, device):\n \n epoch_loss = 0\n epoch_acc_1 = 0\n epoch_acc_5 = 0\n \n model.eval()\n \n with torch.no_grad():\n with tqdm(iterator, position=0, bar_format='{desc}', leave=True) as desc:\n for (x, y) in iterator:\n\n x = x.to(device)\n y = y.to(device)\n\n y_pred = model(x)\n\n loss = criterion(y_pred, y)\n\n acc_1, acc_5 = calculate_topk_accuracy(y_pred, y)\n\n epoch_loss += loss.item()\n epoch_acc_1 += acc_1.item()\n #epoch_acc_5 += acc_5.item()\n \n epoch_loss /= len(iterator)\n epoch_acc_1 /= len(iterator)\n #epoch_acc_5 /= len(iterator)\n \n desc.set_description(f'\\tValid Loss: {epoch_loss:.3f} | Valid Acc @1: {epoch_acc_1*100:6.2f}% | ' \\\n f'Valid Acc @5: {epoch_acc_5*100:6.2f}%')\n return epoch_loss, epoch_acc_1, epoch_acc_5\n\ndef epoch_time(start_time, end_time):\n elapsed_time = end_time - start_time\n elapsed_mins = int(elapsed_time / 60)\n elapsed_secs = int(elapsed_time - (elapsed_mins * 60))\n return elapsed_mins, elapsed_secs", "_____no_output_____" ], [ "#@title lr_finder\nEND_LR = 10 #@param {type:\"number\"}\nNUM_ITER = 100#@param {type:\"number\"} #100\n\nlr_finder = LRFinder(model, optimizer, criterion, device)\nlrs, losses = lr_finder.range_test(train_iterator, END_LR, NUM_ITER)", "_____no_output_____" ], [ "#@title plot_lr_finder\ndef plot_lr_finder(lrs, losses, skip_start = 5, skip_end = 5):\n \n if skip_end == 0:\n lrs = lrs[skip_start:]\n losses = losses[skip_start:]\n else:\n lrs = lrs[skip_start:-skip_end]\n losses = losses[skip_start:-skip_end]\n \n fig = plt.figure(figsize = (16,8))\n ax = fig.add_subplot(1,1,1)\n ax.plot(lrs, losses)\n ax.set_xscale('log')\n ax.set_xlabel('Learning rate')\n ax.set_ylabel('Loss')\n ax.grid(True, 'both', 'x')\n plt.show()\n\nplot_lr_finder(lrs, losses, skip_start = 30, skip_end = 30)", "_____no_output_____" ], [ "#@title config\nFOUND_LR = 2e-4 #@param {type:\"number\"}\n\"\"\"\nparams = [\n {'params': model.conv1.parameters(), 'lr': FOUND_LR / 10},\n {'params': model.bn1.parameters(), 'lr': FOUND_LR / 10},\n {'params': model.layer1.parameters(), 'lr': FOUND_LR / 8},\n {'params': model.layer2.parameters(), 'lr': FOUND_LR / 6},\n {'params': model.layer3.parameters(), 'lr': FOUND_LR / 4},\n {'params': model.layer4.parameters(), 'lr': FOUND_LR / 2},\n {'params': model.fc.parameters()}\n ]\n\n\"\"\"\n#optimizer = optim.Adam(params, lr = FOUND_LR)\noptimizer = optim.Adam(model.parameters(), lr = FOUND_LR)\n\nEPOCHS = 100 #@param {type:\"number\"}\nSTEPS_PER_EPOCH = len(train_iterator)\nTOTAL_STEPS = EPOCHS * STEPS_PER_EPOCH\n\nMAX_LRS = [p['lr'] for p in optimizer.param_groups]\n\nscheduler = lr_scheduler.OneCycleLR(optimizer,\n max_lr = MAX_LRS,\n total_steps = TOTAL_STEPS)", "_____no_output_____" ], [ "#@title training without topk\nimport time\nbest_valid_loss = float('inf')\nbest_valid_accuracy = 0\n\nfor epoch in range(EPOCHS):\n start_time = time.monotonic()\n \n train_loss, train_acc_1, train_acc_5 = train(model, train_iterator, optimizer, criterion, scheduler, device, epoch)\n valid_loss, valid_acc_1, valid_acc_5 = evaluate(model, valid_iterator, criterion, device)\n \n if valid_loss < best_valid_loss:\n best_valid_loss = valid_loss\n torch.save(model.state_dict(), 'best-validation-loss.pt')\n if best_valid_accuracy < valid_acc_1:\n best_valid_accuracy = valid_acc_1\n torch.save(model.state_dict(), 'best-validation-accuracy.pt')\n\n end_time = time.monotonic()\n\n epoch_mins, epoch_secs = epoch_time(start_time, end_time)", "_____no_output_____" ] ], [ [ "#####################################################################################################", "_____no_output_____" ], [ "# TESTING", "_____no_output_____" ] ], [ [ "#@title Calc test loss\nmodel.load_state_dict(torch.load('best-validation-accuracy.pt'))\nprint(\"best-validation-accuracy.pt\")\ntest_loss, test_acc_1, test_acc_5 = evaluate(model, test_iterator, criterion, device)\nprint(\"-----------------------------\")\nmodel.load_state_dict(torch.load('best-validation-loss.pt'))\nprint(\"best-validation-loss.pt\")\ntest_loss, test_acc_1, test_acc_5 = evaluate(model, test_iterator, criterion, device)", "_____no_output_____" ], [ "#@title plot_confusion_matrix\ndef get_predictions(model, iterator):\n\n model.eval()\n\n images = []\n labels = []\n probs = []\n\n with torch.no_grad():\n\n for (x, y) in iterator:\n\n x = x.to(device)\n\n y_pred = model(x)\n\n y_prob = F.softmax(y_pred, dim = -1)\n top_pred = y_prob.argmax(1, keepdim = True)\n\n images.append(x.cpu())\n labels.append(y.cpu())\n probs.append(y_prob.cpu())\n\n images = torch.cat(images, dim = 0)\n labels = torch.cat(labels, dim = 0)\n probs = torch.cat(probs, dim = 0)\n\n return images, labels, probs\n\nimages, labels, probs = get_predictions(model, test_iterator)\npred_labels = torch.argmax(probs, 1)\n\ndef plot_confusion_matrix(labels, pred_labels, classes):\n \n fig = plt.figure(figsize = (50, 50));\n ax = fig.add_subplot(1, 1, 1);\n cm = confusion_matrix(labels, pred_labels);\n cm = ConfusionMatrixDisplay(cm, display_labels = classes);\n cm.plot(values_format = 'd', cmap = 'Blues', ax = ax)\n fig.delaxes(fig.axes[1]) #delete colorbar\n plt.xticks(rotation = 90)\n plt.xlabel('Predicted Label', fontsize = 50)\n plt.ylabel('True Label', fontsize = 50)\n\nplot_confusion_matrix(labels, pred_labels, classes)", "_____no_output_____" ], [ "#@title plot\r\ncorrects = torch.eq(labels, pred_labels)\r\n\r\nincorrect_examples = []\r\n\r\nfor image, label, prob, correct in zip(images, labels, probs, corrects):\r\n if not correct:\r\n incorrect_examples.append((image, label, prob))\r\n\r\nincorrect_examples.sort(reverse = True, key = lambda x: torch.max(x[2], dim = 0).values)\r\n\r\ndef plot_most_incorrect(incorrect, classes, n_images, normalize = True):\r\n\r\n rows = int(np.sqrt(n_images))\r\n cols = int(np.sqrt(n_images))\r\n\r\n fig = plt.figure(figsize = (25, 20))\r\n\r\n for i in range(rows*cols):\r\n\r\n ax = fig.add_subplot(rows, cols, i+1)\r\n \r\n image, true_label, probs = incorrect[i]\r\n image = image.permute(1, 2, 0)\r\n true_prob = probs[true_label]\r\n incorrect_prob, incorrect_label = torch.max(probs, dim = 0)\r\n true_class = classes[true_label]\r\n incorrect_class = classes[incorrect_label]\r\n\r\n if normalize:\r\n image = normalize_image(image)\r\n\r\n ax.imshow(image.cpu().numpy())\r\n ax.set_title(f'true label: {true_class} ({true_prob:.3f})\\n' \\\r\n f'pred label: {incorrect_class} ({incorrect_prob:.3f})')\r\n ax.axis('off')\r\n \r\n fig.subplots_adjust(hspace=0.4)\r\n\r\nN_IMAGES = 36\r\n\r\nplot_most_incorrect(incorrect_examples, classes, N_IMAGES)", "_____no_output_____" ], [ "#@title plot_representations\ndef get_representations(model, iterator):\n\n model.eval()\n\n outputs = []\n intermediates = []\n labels = []\n\n with torch.no_grad():\n \n for (x, y) in iterator:\n\n x = x.to(device)\n\n y_pred, _ = model(x)\n\n outputs.append(y_pred.cpu())\n labels.append(y)\n \n outputs = torch.cat(outputs, dim = 0)\n labels = torch.cat(labels, dim = 0)\n\n return outputs, labels\n\noutputs, labels = get_representations(model, train_iterator)\n\ndef get_pca(data, n_components = 2):\n pca = decomposition.PCA()\n pca.n_components = n_components\n pca_data = pca.fit_transform(data)\n return pca_data\n \ndef plot_representations(data, labels, classes, n_images = None):\n \n if n_images is not None:\n data = data[:n_images]\n labels = labels[:n_images]\n \n fig = plt.figure(figsize = (15, 15))\n ax = fig.add_subplot(111)\n scatter = ax.scatter(data[:, 0], data[:, 1], c = labels, cmap = 'hsv')\n #handles, _ = scatter.legend_elements(num = None)\n #legend = plt.legend(handles = handles, labels = classes)\n\noutput_pca_data = get_pca(outputs)\nplot_representations(output_pca_data, labels, classes)", "_____no_output_____" ], [ "#@title get_tsne\ndef get_tsne(data, n_components = 2, n_images = None):\n \n if n_images is not None:\n data = data[:n_images]\n \n tsne = manifold.TSNE(n_components = n_components, random_state = 0)\n tsne_data = tsne.fit_transform(data)\n return tsne_data\n\noutput_tsne_data = get_tsne(outputs)\nplot_representations(output_tsne_data, labels, classes)", "_____no_output_____" ], [ "#@title plot_filtered_images\ndef plot_filtered_images(images, filters, n_filters = None, normalize = True):\n\n images = torch.cat([i.unsqueeze(0) for i in images], dim = 0).cpu()\n filters = filters.cpu()\n\n if n_filters is not None:\n filters = filters[:n_filters]\n\n n_images = images.shape[0]\n n_filters = filters.shape[0]\n\n filtered_images = F.conv2d(images, filters)\n\n fig = plt.figure(figsize = (30, 30))\n\n for i in range(n_images):\n\n image = images[i]\n\n if normalize:\n image = normalize_image(image)\n\n ax = fig.add_subplot(n_images, n_filters+1, i+1+(i*n_filters))\n ax.imshow(image.permute(1,2,0).numpy())\n ax.set_title('Original')\n ax.axis('off')\n\n for j in range(n_filters):\n image = filtered_images[i][j]\n\n if normalize:\n image = normalize_image(image)\n\n ax = fig.add_subplot(n_images, n_filters+1, i+1+(i*n_filters)+j+1)\n ax.imshow(image.numpy(), cmap = 'bone')\n ax.set_title(f'Filter {j+1}')\n ax.axis('off');\n\n fig.subplots_adjust(hspace = -0.7)\n\nN_IMAGES = 5\nN_FILTERS = 7\n\nimages = [image for image, label in [train_data[i] for i in range(N_IMAGES)]]\nfilters = model.conv1.weight.data\n\nplot_filtered_images(images, filters, N_FILTERS)", "_____no_output_____" ], [ "#@title plot_filters\n#filters = model.conv1.weight.data\n\ndef plot_filters(filters, normalize = True):\n\n filters = filters.cpu()\n\n n_filters = filters.shape[0]\n\n rows = int(np.sqrt(n_filters))\n cols = int(np.sqrt(n_filters))\n\n fig = plt.figure(figsize = (30, 15))\n\n for i in range(rows*cols):\n\n image = filters[i]\n\n if normalize:\n image = normalize_image(image)\n\n ax = fig.add_subplot(rows, cols, i+1)\n ax.imshow(image.permute(1, 2, 0))\n ax.axis('off')\n \n fig.subplots_adjust(wspace = -0.9)\n\nplot_filters(filters)", "_____no_output_____" ] ] ]

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

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "class Solution:\n def reverseWords(self, s: str) -> str:\n s_list = s.split(' ')\n res = [w[::-1] for w in s_list]\n return ' '.join(res)", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# From Variables to Classes\n## A short Introduction \n\nPython - as any programming language - has many extensions and libraries at its disposal. Basically, there are libraries for everything. \n\n<center>But what are **libraries**? </center> \n\nBasically, **libraries** are a collection of methods (_small pieces of code where you put sth in and get sth else out_) which you can use to analyse your data, visualise your data, run models ... do anything you like. \n\nAs said, methods usually take _something_ as input. That _something_ is usually a **variable**. \n\nIn the following, we will work our way from **variables** to **libraries**.", "_____no_output_____" ], [ "## Variables \n\nVariables are one of the simplest types of objects in a programming language. An [object](https://en.wikipedia.org/wiki/Object_(computer_science) is a value stored in the memory of your computer, marked by a specific identifyer. Variables can have different types, such as [strings, numbers, and booleans](https://www.learnpython.org/en/Variables_and_Types). Differently to other programming languages, you do not need to declare the type of a variable, as variables are handled as objects in Python. \n\n```python\nx = 4.2 # floating point number\ny = 'Hello World!' # string\nz = True # boolean\n```", "_____no_output_____" ] ], [ [ "x = 4.2 \nprint(type(x))\n\ny = 'Hello World!'\nprint(type(y))\n\nz = True\nprint(type(z))", "<class 'float'>\n<class 'str'>\n<class 'bool'>\n" ] ], [ [ "We can use operations (normal arithmetic operations) to use variables for getting results we want. With numbers, you can add, substract, multiply, divide, basically taking the values from the memory assigned to the variable name and performing calculations. \n\nLet's have a look at operations with numbers and strings. We leave booleans to the side for the moment. We will simply add the variables below. \n\n```python \nn1 = 7 \nn2 = 42\n\ns1 = 'Looking good, '\ns2 = 'you are.'\n```", "_____no_output_____" ] ], [ [ "n1 = 7\nn2 = 42\n\ns1 = 'Looking good, ' \ns2 = 'you are.'\n\nfirst_sum = n1 + n2\nprint(first_sum)\nfirst_conc = s1 + s2\nprint(first_conc)", "49\nLooking good, you are.\n" ] ], [ [ "Variables can be more than just a number. If you think of an Excel-Spreadsheet, a variable can be the content of a single cell, or multiple cells can be combined in one variable (e.g. one column of an Excel table). \nSo let's create a list -_a collection of variables_ - from `x`, `n1`, and `n2`. Lists in python are created using [ ]. \nNow, if you want to calculate the sum of this list, it is really exhausting to sum up every item of this list manually. \n\n```python\nfirst_list = [x, n1, n2] \n# a sum of a list could look like\nsecond_sum = some_list[0] + some_list[1] + ... + some_list[n] # where n is the last item of the list, e.g. 2 for first_list. \n```", "_____no_output_____" ], [ "Actually, writing the second sum like this is the same as before. It would be great, if this step of calculating the sum could be used many times without writing it out. And this is, what functions are for. For example, there already exists a sum function: \n\n```python \nsum(first_list)```", "_____no_output_____" ] ], [ [ "first_list = [x, n1, n2]\nsecond_sum = first_list[0] + first_list[1] + first_list[2]\nprint('manual sum {}'.format(second_sum))\n\n# This can also be done with a function \nprint('sum function {}'.format(sum(first_list)))", "manual sum 53.2\nsum function 53.2\n" ] ], [ [ "## Functions \nThe `sum()` method we used above is a **function**. \nFunctions (later we will call them methods) are pieces of code, which take an input, perform some kind of operation, and (_optionally_) return an output. ", "_____no_output_____" ], [ "In Python, functions are written like: \n\n```python\ndef func(input):\n \"\"\"\n Description of the functions content # called the function header\n \"\"\"\n some kind of operation on input # called the function body\n \n return output\n```\nAs an example, we write a `sumup` function which sums up a list.", "_____no_output_____" ] ], [ [ "def sumup(inp):\n \"\"\"\n input: inp - list/array with floating point or integer numbers\n return: sumd - scalar value of the summed up list\n \"\"\"\n val = 0\n for i in inp:\n val = val + i\n return val\n\n# let's compare the implemented standard sum function with the new sumup function\nsum1 = sum(first_list)\nsum2 = sumup(first_list)\nprint(\"The python sum function yields {}, \\nand our sumup function yields {}.\".format(*(sum1,sum2)))", "The python sum function yields 53.2, \nand our sumup function yields 53.2.\n" ], [ "# summing up the numbers from 1 to 100\nimport numpy as np\nar_2_sum = np.linspace(1,100,100, dtype='i')\n\nprint(\"the sum of the array is: {}\".format(sumup(ar_2_sum)))", "the sum of the array is: 5050\n" ] ], [ [ "As we see above, functions are quite practical and save a lot of time. Further, they help structuring your code. Some functions are directly available in python without any libraries or other external software. In the example above however, you might have noticed, that we `import`ed a library called `numpy`. \nIn those libraries, functions are merged to one package, having the advantage that you don't need to import each single function at a time. \nImagine you move and have to pack all your belongings. You can think of libraries as packing things with similar purpose in the same box (= library).", "_____no_output_____" ], [ "## Functions to Methods as part of classes\nWhen we talk about functions in the environment of classes, we usually call them methods. But what are **classes**? \n[Classes](https://docs.python.org/3/tutorial/classes.html) are ways to bundle functionality together. Logically, functionality with similar purpose (or different kind of similarity). \nOne example could be: think of **apples**. \n\nApples are now a class. You can apply methods to this class, such as `eat()` or `cut()`. Or more sophisticated methods including various recipes using apples comprised in a cookbook. \nThe `eat()` method is straight forward. But the `cut()` method may be more interesting, since there are various ways to cut an apple.\n", "_____no_output_____" ], [ "Let's assume there are two apples to be cut differently. In python, once you have assigned a class to a variable, you have created an **instance** of that class. Then, methods of are applied to that instance by using a . notation.\n```python\nGolden_Delicious = apple()\nYoya = apple()\n\nGolden_Delicious.cut(4)\nYoya.cut(8)\n```\nThe two apples Golden Delicious and Yoya are _instances_ of the class apple. Real _incarnations_ of the abstract concept _apple_. The Golden Delicious is cut into 4 pieces, while the Yoya is cut into 8 pieces. ", "_____no_output_____" ], [ "This is similar to more complex libraries, such as the `scikit-learn`. In one exercise, you used the command: \n```python\nfrom sklearn.cluster import KMeans\n```\nwhich simply imports the **class** `KMeans` from the library part `sklearn.cluster`. `KMeans` comprises several methods for clustering, which you can use by calling them similar to the apple example before.", "_____no_output_____" ], [ " For this, you need to create an _instance_ of the `KMeans` class. \n```python\n...\nkmeans_inst = KMeans(n_clusters=n_clusters) # first we create the instance of the KMeans class called kmeans_inst\nkmeans_inst.fit(data) # then we apply a method to the instance kmeans_inst\n...\n```\nAn example:", "_____no_output_____" ] ], [ [ "# here we just create the data for clustering\nfrom sklearn.datasets.samples_generator import make_blobs\nimport matplotlib.pyplot as plt\n%matplotlib inline\nX, y = make_blobs(n_samples=100, centers=3, cluster_std= 0.5,\n random_state=0)\n\nplt.scatter(X[:,0], X[:,1], s=70)", "_____no_output_____" ], [ "# now we create an instance of the KMeans class\nfrom sklearn.cluster import KMeans\nnr_of_clusters = 3 # because we see 3 clusters in the plot above\nkmeans_inst = KMeans(n_clusters= nr_of_clusters) # create the instance kmeans_inst\nkmeans_inst.fit(X) # apply a method to the instance\ny_predict = kmeans_inst.predict(X) # apply another method to the instance and save it in another variable", "_____no_output_____" ], [ "# lets plot the predicted cluster centers colored in the cluster color\nplt.scatter(X[:, 0], X[:, 1], c=y_predict, s=50, cmap='Accent')\ncenters = kmeans_inst.cluster_centers_ # apply the method to find the new centers of the determined clusters\nplt.scatter(centers[:, 0], centers[:, 1], c='red', s=200, alpha=0.6); # plot the cluster centers", "_____no_output_____" ] ], [ [ "## Summary \nThis short presentation is meant to make you familiar with the concept of variables, functions, methods and classes. All of which are objects! \n* Variables are normally declared by the user and link a value stored in the memory of your pc to a variable name. They are usually the input of functions \n* Functions are pieces of code taking an input and performing some operation on said input. Optionally, they return directly an output value \n* To facilitate the use of functions, they are sometimes bundled as methods within classes. Classes in turn can build up whole libraries in python. ", "_____no_output_____" ], [ "* Similar to real book libraries, python libraries contain a collection of _recipes_ which can be applied to your data. \n* In terms of apples: You own different kinds of apples. A book about apple dishes (_class_) from the library contains different recipes (_methods_) which can be used for your different apples (_instances of the class_).", "_____no_output_____" ], [ "## Further links \n* [Data Science Handbook](https://jakevdp.github.io/PythonDataScienceHandbook/)\n* [Python for Geosciences](https://github.com/koldunovn/python_for_geosciences) \n* [Introduction to Python for Geoscientists](http://ggorman.github.io/Introduction-to-programming-for-geoscientists/) \n* [Full Video course on Object Oriented Programming](https://www.youtube.com/watch?v=ZDa-Z5JzLYM&list=PL-osiE80TeTsqhIuOqKhwlXsIBIdSeYtc)", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import pandas as pd\nimport numpy as np\n\npd.set_option('display.max_columns', None) \npd.set_option('display.max_rows', None)\npd.set_option('display.max_colwidth', -1)\n\n%matplotlib inline\n\nfrom sidecar import Sidecar\nfrom ipywidgets import IntSlider\n\nsc = Sidecar(title='Sidecar Output')\nsl = IntSlider(description='Some slider')", "_____no_output_____" ], [ "# Remove unrelated columns form data and get their name\n\nfolder_path = '../../../datalcdem/data/optima/dementia_18July/'\n\npatient_df = pd.read_csv(folder_path + 'optima_patients.csv')\ndisplay(patient_df.head(5))\n\n#patient_df[['MMS1', 'MMS2']].hist()\n\npatient_com_df = pd.read_csv(folder_path + 'optima_patients_comorbidities.csv').groupby(by=['patient_id', 'EPISODE_DATE'], as_index=False).agg(lambda x: x.tolist())[['patient_id', 'EPISODE_DATE', 'Comorbidity_cui']]\ndisplay(patient_com_df.head(10))\n\npatient_filt_df = pd.read_csv(folder_path + 'optima_patients_filtered.csv')\n#display(patient_filt_df.head(5))\n\n\npatient_treat_df = pd.read_csv(folder_path + 'optima_patients_treatments.csv').groupby(by=['patient_id', 'EPISODE_DATE'], as_index=False).agg(lambda x: x.tolist())[['patient_id', 'EPISODE_DATE', 'Medication_cui']]\ndisplay(patient_treat_df.head(5))", "_____no_output_____" ], [ "len(set(patient_com_df['patient_id'].tolist())), len(set(patient_treat_df['patient_id'].tolist()))", "_____no_output_____" ], [ "patient_treat_df['EPISODE_DATE'] = pd.to_datetime(patient_treat_df['EPISODE_DATE'])\npatient_com_df['EPISODE_DATE'] = pd.to_datetime(patient_com_df['EPISODE_DATE'])\n\npatient_com_treat_df = pd.merge(patient_com_df, patient_treat_df,on=['patient_id', 'EPISODE_DATE'], how='outer')\n#pd.concat([patient_com_df, patient_treat_df], keys=['patient_id', 'EPISODE_DATE'], ignore_index=True, sort=False) #patient_com_df.append(patient_treat_df, sort=False) #pd.concat([patient_com_df, patient_treat_df], axis=0, sort=False)\n#patient_treat_com_df = patient_treat_df.join(patient_com_df, on=['patient_id', 'EPISODE_DATE'], how='outer')\nprint (patient_com_treat_df.shape)\nprint (len(set(patient_com_treat_df['patient_id'].tolist())))\npatient_com_treat_df.sort_values(by=['patient_id', 'EPISODE_DATE'],axis=0, inplace=True, ascending=True)\npatient_com_treat_df.reset_index(drop=True, inplace=True)\npatient_com_treat_df.head(10)", "(3690, 4)\n820\n" ], [ "patient_com_treat_df.to_csv('../../../datalcdem/data/optima/optima_ahmad/patient_com_treat_df.csv')", "_____no_output_____" ], [ "folder_path = '../../../datalcdem/data/optima/dementia_18July/'", "_____no_output_____" ], [ "df_datarequest = pd.read_excel(folder_path+'Data_Request_Jan_2019_final.xlsx')\ndf_datarequest.head(5)\ndf_datarequest_mmse = df_datarequest[['GLOBAL_PATIENT_DB_ID', 'Age At Episode', 'EPISODE_DATE', 'CAMDEX SCORES: MINI MENTAL SCORE']]\ndf_datarequest_mmse_1 = df_datarequest_mmse.rename(columns={'GLOBAL_PATIENT_DB_ID':'patient_id'})\ndf_datarequest_mmse_1.head(10)", "_____no_output_____" ], [ "#patient_com_treat_df.astype('datetime')\npatient_com_treat_df['EPISODE_DATE'] = pd.to_datetime(patient_com_treat_df['EPISODE_DATE'])\nprint (df_datarequest_mmse_1.dtypes, patient_com_treat_df.dtypes)\npatient_com_treat_df = pd.merge(patient_com_treat_df,df_datarequest_mmse_1,on=['patient_id', 'EPISODE_DATE'], how='left')\npatient_com_treat_df.shape, patient_com_treat_df.head(10)", "patient_id int64 \nAge At Episode int64 \nEPISODE_DATE datetime64[ns]\nCAMDEX SCORES: MINI MENTAL SCORE float64 \ndtype: object patient_id int64 \nEPISODE_DATE datetime64[ns]\nComorbidity_cui object \nMedication_cui object \ndtype: object\n" ], [ "len(set (patient_com_treat_df['patient_id'].tolist()))\npatient_com_treat_df.sort_values(by=['patient_id', 'EPISODE_DATE'],axis=0, inplace=True, ascending=True)\npatient_com_treat_df.head(20)\npatient_com_treat_df.reset_index(inplace=True, drop=True)\npatient_com_treat_df.head(5)", "_____no_output_____" ], [ "def setLineNumber(lst):\n lst_dict = {ide:0 for ide in lst}\n lineNumber_list = []\n \n for idx in lst:\n if idx in lst_dict:\n lst_dict[idx] = lst_dict[idx] + 1 \n lineNumber_list.append(lst_dict[idx])\n \n return lineNumber_list\n \n\npatient_com_treat_df['lineNumber'] = setLineNumber(patient_com_treat_df['patient_id'].tolist())\npatient_com_treat_df.tail(20)", "_____no_output_____" ], [ "df = patient_com_treat_df\n\nid_dict = {i:0 for i in df['patient_id'].tolist()}\nfor x in df['patient_id'].tolist():\n if x in id_dict:\n id_dict[x]=id_dict[x]+1\n\nline_updated = [int(j) for i in id_dict.values() for j in range(1,i+1)]\nprint (line_updated[0:10])\ndf.update(pd.Series(line_updated, name='lineNumber'),errors='ignore')\ndisplay(df.head(20))\n\n\n#patients merging based on id and creating new columns\nr = df['lineNumber'].max()\nprint ('Max line:',r)\nl = [df[df['lineNumber']==i] for i in range(1, int(r+1))]\nprint('Number of Dfs to merge: ',len(l))\ndf_new = pd.DataFrame()\ntmp_id = []\nfor i, df_l in enumerate(l):\n df_l = df_l[~df_l['patient_id'].isin(tmp_id)]\n for j, df_ll in enumerate(l[i+1:]):\n #df_l = df_l.merge(df_ll, on='id', how='left', suffix=(str(j), str(j+1))) #suffixe is not working\n #print (j)\n df_l = df_l.join(df_ll.set_index('patient_id'), on='patient_id', rsuffix='_'+str(j+1))\n tmp_id = tmp_id + df_l['patient_id'].tolist()\n #display(df_l)\n df_new = df_new.append(df_l, ignore_index=True, sort=False)\ndisplay(df_new.head(20))\ndisplay(df_new[['patient_id']+[col for col in df_new.columns if 'line' in col or 'DATE' in col]].head(10))", "[1, 2, 3, 1, 2, 3, 4, 5, 6, 7]\n" ], [ "fltr_linnum = ['_'+str(i) for i in range(10, 27)]\nprint (fltr_linnum)\ndf_new.drop(columns=[col for col in df_new.columns for i in fltr_linnum if i in col],inplace=True)", "['_10', '_11', '_12', '_13', '_14', '_15', '_16', '_17', '_18', '_19', '_20', '_21', '_22', '_23', '_24', '_25', '_26']\n" ], [ "df_new.to_csv(folder_path+'dementialTreatmentLine_preData_line_episode.csv', index=False)\ndf_new = df_new.drop([col for col in df_new.columns if 'lineNumber' in col or 'EPISODE_DATE' in col], axis=1).reset_index(drop=True)\ndf_new.to_csv(folder_path+'dementialTreatmentLine_preData.csv', index=False)", "_____no_output_____" ], [ "# Calculate matching intial episode in the data\n\n''' df_episode= pd.read_csv('../../../datalcdem/data/optima/dementialTreatmentLine_preData_line_episode.csv')\ndf_patients = pd.read_csv('../../../datalcdem/data/optima/patients.csv')\ndisplay(df_episode.columns, df_patients.columns)\n\ndf_pat_ep = pd.merge(df_episode[['patient_id', 'EPISODE_DATE']], df_patients[['patient_id', 'epDateInicial', 'mmseInicial']])\n\ndf_episode.shape, df_patients.shape, df_pat_ep.shape\n\ndf_pat_ep['dateEqual']=df_pat_ep['EPISODE_DATE']==df_pat_ep['epDateInicial']\ndisplay(sum(df_pat_ep['dateEqual'].tolist()))\ndf_pat_ep.head(10)\ndisplay(sum(df_pat_ep['mmseInicial']<24))'''", "_____no_output_____" ], [ "df_new.head(10)", "_____no_output_____" ], [ "# Take Some other features from API\ndf_patient_api = pd.read_csv(folder_path+'patients.csv')\ndisplay(df_patient_api.head(10))\ndf_patient_api = df_patient_api[['patient_id', 'gender', 'dementia', 'smoker', 'alcohol', 'education',\n 'bmi', 'weight', 'apoe']]\ndisplay(df_patient_api.head(10))\ndisplay(df_new.head(10))\ndf_patient_new = df_patient_api.merge(df_new, on=['patient_id'], how='inner')\ndf_patient_new.head(10)", "_____no_output_____" ], [ "df_patient_new.to_csv(folder_path+'patients_new.csv', index=False)", "_____no_output_____" ], [ "def removeNANvalues(lst):\n return lst[~numpy.isnan(lst)]\n\ncomorbidity_cui_lst = df_patient_new[[col for col in df_patient_new.columns if 'Comorbidity_cui' in col]].values.flatten()\nmedication_cui_lst = df_patient_new[[col for col in df_patient_new.columns if 'Medication_cui' in col]].values.flatten()\nx = x[~numpy.isnan(x)]\n\nmedication_cui_lst", "_____no_output_____" ], [ "[val for lst_1 in medication_cui.flatten() for val in lst_1]", "_____no_output_____" ] ] ]

[ "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import numpy as np\nimport matplotlib.pyplot as plt\nfrom sklearn import datasets\nfrom tqdm import tqdm\n\nfrom collections import defaultdict\nfrom itertools import combinations\nimport re\nimport time", "_____no_output_____" ], [ "### You may import any Python's standard library here (Do not import other external libraries) ###\n", "_____no_output_____" ], [ "pass_test1_1_1 = False\npass_test1_1_2 = False\npass_test1_2 = False\npass_test2_1 = False\npass_test2_2 = False", "_____no_output_____" ] ], [ [ "## Implementing LSH algorithm", "_____no_output_____" ], [ "### 0. Dataset", "_____no_output_____" ], [ "#### 0.1 Import 20-news dataset", "_____no_output_____" ] ], [ [ "newsgroup_dataset = datasets.fetch_20newsgroups(data_home='./dataset/', subset='train', \n remove=('headers', 'footers', 'quotes'), download_if_missing=True)", "_____no_output_____" ], [ "raw_documents = newsgroup_dataset['data'][:]\nlen(raw_documents)", "_____no_output_____" ], [ "raw_documents[0]", "_____no_output_____" ] ], [ [ "#### 0.2 Preprocess the documents", "_____no_output_____" ] ], [ [ "K = 5 # number of word tokens to shinlge", "_____no_output_____" ], [ "def preprocess(documents):\n processed_words = defaultdict(list)\n cnt = 0\n \n for doc in documents:\n # first, filter out some uncesseary symbols like punctuations\n doc = re.sub('\\/|\\-|\\'|\\@|\\%|\\$|\\#|\\,|\\(|\\)|\\}|\\\"|\\{|\\?|\\.|\\!|\\;|\\:', '', doc)\n \n # second, split the document into the words\n for word in doc.split():\n \n # third, let word to be the lower-case\n if word.isalpha():\n processed_words[cnt].append(word.lower())\n \n # fourth, filter out the articles that has less than k shingles\n if len(processed_words[cnt]) < K:\n continue\n else:\n processed_words[cnt] = ' '.join(processed_words[cnt])\n cnt += 1\n\n return list(processed_words.values())", "_____no_output_____" ], [ "documents = preprocess(raw_documents)\ndel raw_documents\nlen(documents)", "_____no_output_____" ], [ "documents[0]", "_____no_output_____" ] ], [ [ "### 1. Shingling", "_____no_output_____" ] ], [ [ "########################################################################################################################\n# Programming 1 [15pt] #\n# In this section, you will implement the shingling algorithm to convert the document into the characteristic matrix. #\n# However, since storing the whole characteristic matrix in the form of a dense matrix is expensivein terms of space, #\n# your implementation should store the characteristic matrix in the form of a dictionary. #\n# #\n# i) get the all unique shingles from the documents [10pt] #\n# ii) create the dictionary that maps each document to the list of shingles [5pt] #\n# #\n# Note that, shingling is divided into 2-steps just for the readability of the algorithm #\n# #\n########################################################################################################################", "_____no_output_____" ] ], [ [ "#### 1.1 Get Shingles from the documents", "_____no_output_____" ] ], [ [ "def get_shingles(documents):\n ######################################################################################\n # Programming 1.1 [10pt] #\n # Implement 'get_shingles' function to get 1-singles from the preprocessed documents #\n # You should especially be take care of your algorithm's computational efficiency #\n # #\n # Parameters: #\n # documents (dict) #\n # #\n # Returns: #\n # shingles (set) set of tuples where each element is a k-shingle #\n # ex) shingles = {('its', 'hard', 'to', 'say', 'whether'), #\n # ('known', 'bugs', 'in', 'the', 'warning') ...} #\n ######################################################################################\n shingles = set()\n for doc in documents:\n doc_split = doc.split()\n for i in range(len(doc_split) - (K-1)):\n shingles.add(tuple(doc_split[i:i+K]))\n \n return shingles", "_____no_output_____" ], [ "start = time.time()\nshingles = get_shingles(documents)\nend = time.time()\n\n# Check whether your implementation is correct [5pt]\nif len(shingles) == 1766049:\n pass_test1_1_1 = True\n print('Test1 passed')\n \n # Check whether your implementation is efficient enough [5pt]\n # With 4-lines of my implementations, it took 4.8 seconds with i7-8700 cpu\n if (end - start) < 20:\n pass_test1_1_2 = True\n print('Test2 passed')\n ", "Test1 passed\nTest2 passed\n" ], [ "print(end - start)", "1.0299334526062012\n" ] ], [ [ "#### 1.2 Build document to shingles dictionary", "_____no_output_____" ] ], [ [ "def build_doc_to_shingle_dictionary(documents, shingles):\n ################################################################################################################################\n # Programming 1.2 [5pt] #\n # Implement 'build_doc_to_shingle_dictionary' function to convert documents into shingle dictionary with documents & shingles #\n # You need to construct and utilize a shingle2idx dictionary that maps each shingle into the uniuqe integer index. #\n # #\n # Parameters: #\n # documents (dict) # \n # shingles (set) # \n # #\n # Returns: #\n # doc_to_shingles (dict) #\n # key: index of the documents #\n # value: list of the shingle indexes #\n # ex) doc_to_shingles = {0: [1705196, 422880, 491967, ...], #\n # 1: [863922, 1381606, 1524066, ...], #\n # ... } #\n ################################################################################################################################\n doc_to_shingles = {}\n shingle2idx = {}\n for idx, shingle in enumerate(shingles):\n shingle2idx[shingle] = idx\n for idx, doc in enumerate(documents):\n shingle_list = [shingle2idx[s] for s in get_shingles([doc])]\n doc_to_shingles[idx] = shingle_list\n \n return doc_to_shingles", "_____no_output_____" ], [ "doc_to_shingles = build_doc_to_shingle_dictionary(documents, shingles)\n\n# Check whether your implementation is correct [5pt]\nif len(doc_to_shingles) == 10882 and len(doc_to_shingles[0]) == 84:\n pass_test1_2 = True\n print('Test passed')", "Test passed\n" ] ], [ [ "### 2. Min-Hashing", "_____no_output_____" ] ], [ [ "############################################################################################################################\n# Programming 2 [25pt] #\n# In this section, you will implement the min-hashing algorithm to convert the characteristic matrix into the signatures. #\n# #\n# i) implement the jaccard-similarity algorithm [5pt] #\n# ii) implement the min-hash algorithm to create the signatures for the documents [20pt] #\n# #\n############################################################################################################################", "_____no_output_____" ] ], [ [ "#### 2.1 Generate Prime numbers for Universal Hashing", "_____no_output_____" ] ], [ [ "def is_prime(n):\n for i in range(2,int(np.sqrt(n))+1):\n if not n % i:\n return False\n return True\n\ndef generate_prime_numbers(M, N):\n # this function generate the M prime numbers where each prime number is greater than N\n primes = []\n cnt = 0\n n = N + 1\n \n while cnt < M:\n if is_prime(n):\n primes.append(n)\n cnt += 1\n n += 1\n return primes", "_____no_output_____" ], [ "# Test prime number generation\ngenerate_prime_numbers(M = 3, N = 3)", "_____no_output_____" ] ], [ [ "#### 2.2 Jaccard Similarity", "_____no_output_____" ] ], [ [ "def jaccard_similarity(s1, s2):\n ##################################################################################\n # Programming 2.2 [5pt] # \n # Implement the jaccard similarity algorithm to get the similarity of two sets #\n # #\n # Parameters #\n # s1 (set) #\n # s2 (set) #\n # Returns #\n # similarity (float) #\n ##################################################################################\n similarity = len(s1&s2) / len(s1|s2)\n return similarity", "_____no_output_____" ], [ "s1 = {1, 3, 4}\ns2 = {3, 4, 6}\n\nif (jaccard_similarity(s1, s2) - 0.5) < 1e-3:\n pass_test2_1 = True\n print('Test passed')", "Test passed\n" ] ], [ [ "#### 2.3 Min Hash", "_____no_output_____" ] ], [ [ "M = 100 # Number of Hash functions to use\nN = len(shingles)\n\n# First we will create M universal hashing functions\n# You can also modify or implement your own hash functions for implementing min_hash function\n\nclass Hash():\n def __init__(self, M, N):\n self.M = M\n self.N = N\n self.p = generate_prime_numbers(M, N)\n \n self.a = np.random.choice(9999, M)\n self.b = np.random.choice(9999, M)\n \n def __call__(self, x):\n return np.mod(np.mod((self.a * x + self.b), self.p), self.N)\n \n def __len__(self):\n return M\n \n#primes = generate_prime_numbers(M, N)\nhash_functions = Hash(M, N)", "_____no_output_____" ], [ "def min_hash(doc_to_shingles, hash_functions):\n ###########################################################################################\n # Programming 2.3 [20pt] # \n # Implement the min-hash algorithm to create the signatures for the documents #\n # It would take about ~10 minutes to finish computation, #\n # while would take ~20 seconds if you parallelize your hash functions # \n # #\n # Parameters #\n # doc_to_shingles: (dict) dictionary that maps each document to the list of shingles #\n # hash_functions: [list] list of hash functions #\n # Returns #\n # signatures (np.array) numpy array of size (M, C) where C is the number of documents #\n # #\n ###########################################################################################\n \n C = len(doc_to_shingles)\n M = len(hash_functions)\n signatures = np.array(np.ones((M, C)) * 999999999999, dtype = np.int)\n \n for doc_id in range(C):\n shingles = doc_to_shingles[doc_id]\n for shingle in shingles:\n hash_shingle = hash_functions(shingle)\n signatures[:,doc_id] = np.where(hash_shingle < signatures[:,doc_id], hash_shingle, signatures[:,doc_id])\n \n return signatures", "_____no_output_____" ], [ "def compare(signatures, doc_to_shingles, trials = 10000):\n M, C = signatures.shape\n diff_list = []\n \n for t in tqdm(range(trials)):\n doc1, doc2 = np.random.choice(C, 2, replace = False)\n \n shingle1, shingle2 = set(doc_to_shingles[doc1]), set(doc_to_shingles[doc2])\n sig1, sig2 = signatures[:,doc1], signatures[:,doc2]\n \n true_sim = jaccard_similarity(shingle1, shingle2)\n approx_sim = sum(np.equal(sig1, sig2)) / M\n \n diff_list.append(abs(true_sim - approx_sim))\n \n return diff_list", "_____no_output_____" ], [ "start = time.time()\nsignatures = min_hash(doc_to_shingles, hash_functions)\nend = time.time()\n\ndiff_list = compare(signatures, doc_to_shingles)\n\n# Check whether your implementation is correct [20pt]\n# Average difference of document's jaccard similarity between characteristic matrix and signatures should be at most 1%\n# With 10 random seeds, difference was around 1e-5 ~ 1e-6%\nif np.mean(diff_list) < 0.01:\n pass_test2_2 = True\n print('Test passed')", "100%|██████████| 10000/10000 [00:04<00:00, 2437.12it/s]" ] ], [ [ "#### 2.4 Qualitive Analysis", "_____no_output_____" ] ], [ [ "print('Document 3542')\nprint(documents[3542])\nprint('-------------')\nprint('Document 8033')\nprint(documents[8033])\nprint('-------------')\n\nprint('true jaccard similarity:' ,jaccard_similarity(set(doc_to_shingles[3542]), set(doc_to_shingles[8033])))\nprint('approx jaccard similarity:',sum(np.equal(signatures[:,3542], signatures[:,8033])) / M)\nprint('Do you think signature well reflects the characteristic matrix?')", "Document 3542\ni have one complaint for the cameramen doing the jerseypitt series show the shots not the hits on more than one occassion the camera zoomed in on a check along the boards while the puck was in the slot they panned back to show the rebound maybe moms camera people were a little more experienced\n-------------\nDocument 8033\ni have one complaint for the cameramen doing the jerseypitt series show the shots not the hits on more than one occassion the camera zoomed in on a check along the boards while the puck was in the slot they panned back to show the rebound maybe moms camera people were a little more experienced joseph stiehm exactly that is my biggest complaint about the coverage so far follow that damn puck ravi shah\n-------------\ntrue jaccard similarity: 0.7285714285714285\napprox jaccard similarity: 0.74\nDo you think signature well reflects the characteristic matrix?\n" ] ], [ [ "### 3. Locality Sensitive Hashing", "_____no_output_____" ] ], [ [ "########################################################################################################################\n# Programming 3 [35pt] #\n# In this section, you will implement the Min-Hash based Locality Sensitive Hashing algorithm to convert signatures #\n# into the similar document pair candidates #\n# Finally, we will test our results based on the precision, recall and F1 score #\n# #\n# 1) get the similar document pair candidates [20pt] #\n# 2) calculate precision, recall, and f1 score [10pt] #\n# #\n########################################################################################################################", "_____no_output_____" ] ], [ [ "#### 3.1 Min-Hash based LSH", "_____no_output_____" ] ], [ [ "def lsh(signatures, b, r):\n #########################################################################################################\n # Programming 3.1 [20pt] #\n # Implement the min-hash based LSH algorithm to find the candidate pairs of the similar documents. #\n # In the implementation, use python's dictionary to make your hash table, #\n # where each column is hashed into a bucket. #\n # Convert each column vector (within a band) into the tuple and use it as a key of the dictionary. #\n # #\n # Parameters #\n # signatures: (np.array) numpy array of size (M, C) where #\n # M is the number of min-hash functions, C is the number of documents #\n # b: (int) the number of bands #\n # r: (int) the number of rows per each band #\n # #\n # Requirements #\n # 1) M should be equivalent to b * r #\n # #\n # Returns #\n # candidatePairs (Set[Tuple[int, int]]) set of the pairs of indexes of candidate document pairs #\n # #\n #########################################################################################################\n M = signatures.shape[0] # The number of min-hash functions\n C = signatures.shape[1] # The number of documents\n\n assert M == b * r\n\n candidatePairs = set()\n\n # TODO: Write down your code here \n for num_b in range(b):\n bucket = {}\n bands = signatures[num_b*r:(num_b+1)*r]\n for col in range(C):\n if tuple(bands[:,col]) in bucket.keys():\n bucket[tuple(bands[:,col])].append(col)\n else:\n bucket[tuple(bands[:,col])] = [col]\n\n for value in bucket.values():\n if len(value) >= 2:\n combi = combinations(value, 2)\n candidatePairs.update(list(combi))\n #import ipdb; ipdb.set_trace()\n ### Implementation End ###\n\n return candidatePairs", "_____no_output_____" ], [ "# You can test your implementation here\nb = 10\nn = 0\ntmpPairs = list(lsh(signatures, b, M // b))\nprint(f\"b={b}\")\nprint(f\"# of candidate pairs = {len(tmpPairs)}\")\nsamplePair = tmpPairs[n]\nshingle1, shingle2 = set(doc_to_shingles[samplePair[0]]), set(doc_to_shingles[samplePair[1]])\nprint(f\"{n}th sample pair: {samplePair}\")\nprint(f\"Jaccard similarity: {jaccard_similarity(shingle1, shingle2)}\")\nprint('-------------')\nprint(documents[samplePair[0]])\nprint('-------------')\nprint(documents[samplePair[1]])\nprint('-------------')", "b=10\n# of candidate pairs = 162\n0th sample pair: (1658, 5780)\nJaccard similarity: 0.8108108108108109\n-------------\nfrom paynecrldeccom andrew payne messageid organization dec cambridge research lab date tue apr gmt does anyone know if a source for the modem chips as used in the baycom and my pmp modems ideally something that is geared toward hobbyists small quantity mail order etc for years weve been buying them from a distributor marshall by the hundreds for pmp kits but orders have dropped to the point where we can no longer afford to offer this service and all of the distributors ive checked have some crazy minimum order or so id like to find a source for those still interested in building pmp kits any suggestions andrew c payne dec cambridge research lab\n-------------\ndoes anyone know if a source for the modem chips as used in the baycom and my pmp modems ideally something that is geared toward hobbyists small quantity mail order etc for years weve been buying them from a distributor marshall by the hundreds for pmp kits but orders have dropped to the point where we can no longer afford to offer this service and all of the distributors ive checked have some crazy minimum order or so id like to find a source for those still interested in building pmp kits any suggestions\n-------------\n" ] ], [ [ "#### 3.2 Compute the precision, recall, and F1-score", "_____no_output_____" ] ], [ [ "# Compute the number of condition positives, which is the number of every document pair whose Jaccard similarity is greater than or equal to the threshold\n\ns = 0.8 # similarity threshold for checking condition positives\nnumConditionPositives = 151 # This is the computed result when s=0.8, but I gave it to you to save your time.\n\ncomputeConditionPositives = False # If you want to calculate it, then change it to True. It will take 30 minutes to compute.\nif computeConditionPositives:\n numConditionPositives = 0\n\n numDocs = len(documents)\n for i in tqdm(range(numDocs)):\n shingle1 = set(doc_to_shingles[i])\n\n for j in range(i+1, numDocs):\n shingle2 = set(doc_to_shingles[j])\n true_sim = jaccard_similarity(shingle1, shingle2)\n if true_sim >= s:\n numConditionPositives += 1\n\nprint(f\"The number of condition positives: {numConditionPositives} when s={s}\")", "The number of condition positives: 151 when s=0.8\n" ], [ "def query_analysis(signatures, b, s, numConditionPositives):\n ###########################################################################################################\n # Programming 3.2 [10pt] #\n # Calculate the query time, precision, recall, and F1 score for the given configuration #\n # #\n # Parameters #\n # signatures: (np.array) numpy array of size (M, C) where #\n # M is the number of min-hash functions, C is the number of documents #\n # b: (int) the number of bands #\n # s: (float) similarity threshold for checking condition positives #\n # numConditionPositives: (int) the number of condition positives #\n # #\n # Requirements #\n # 1) b should be the divisor of M #\n # 2) 0 <= s <= 1 #\n # #\n # Returns #\n # query time: (float) the execution time of the codes which find the similar document candidate pairs #\n # precision: (float) #\n # recall: (float) #\n # f1: (float) F1-Score #\n # #\n ###########################################################################################################\n M = signatures.shape[0] # The number of min-hash functions\n assert M % b == 0\n\n # TODO: Write down your code here\n TP = 0\n t = time.time()\n candidatePairs = lsh(signatures, b, M // b)\n query_time = time.time() - t\n\n for pair in candidatePairs:\n shingle1, shingle2 = set(doc_to_shingles[pair[0]]), set(doc_to_shingles[pair[1]])\n if jaccard_similarity(shingle1, shingle2) >= s:\n TP += 1\n \n precision = TP / len(candidatePairs)\n recall = TP / numConditionPositives\n f1 = 2 * precision * recall / (precision + recall)\n ### Implementation End ###\n\n return query_time, precision, recall, f1", "_____no_output_____" ], [ "# Return the list of every divisor of given integer\ndef find_divisors(x):\n divisors = list()\n for i in range(1, x + 1):\n if x % i == 0:\n divisors.append(i)\n return divisors", "_____no_output_____" ], [ "b_list = find_divisors(M)\n\nquery_time_list = list()\nprecision_list = list()\nrecall_list = list()\nf1_list = list()\n\nfor b in tqdm(b_list):\n query_time, precision, recall, f1 = query_analysis(signatures, b, s, numConditionPositives)\n \n query_time_list.append(query_time)\n precision_list.append(precision)\n recall_list.append(recall)\n f1_list.append(f1)", "100%|██████████| 9/9 [00:24<00:00, 2.67s/it]\n" ], [ "print(\"b: \", b_list)\nprint(\"Query times: \", query_time_list)\nprint(\"Precisions: \", precision_list)\nprint(\"Recalls: \", recall_list)\nprint(\"F1 scores: \", f1_list)", "b: [1, 2, 4, 5, 10, 20, 25, 50, 100]\nQuery times: [0.4562802314758301, 0.284761905670166, 0.5345544815063477, 0.5704622268676758, 0.9000422954559326, 1.581861972808838, 1.9276137351989746, 3.863903522491455, 7.186660051345825]\nPrecisions: [1.0, 1.0, 1.0, 1.0, 0.9197530864197531, 0.5571955719557196, 0.3511627906976744, 0.050333333333333334, 0.002708908901725808]\nRecalls: [0.6556291390728477, 0.6688741721854304, 0.7549668874172185, 0.8145695364238411, 0.9867549668874173, 1.0, 1.0, 1.0, 1.0]\nF1 scores: [0.792, 0.8015873015873015, 0.8603773584905661, 0.8978102189781023, 0.952076677316294, 0.7156398104265402, 0.5197934595524957, 0.09584258965407808, 0.005403181078131429]\n" ], [ "plt.title(f\"Query time (s={s})\")\nplt.xlabel(\"b\")\nplt.ylabel(\"Query time [sec]\")\nplt.plot(b_list, query_time_list)\nplt.show()", "_____no_output_____" ], [ "plt.title(f\"Precision (s={s})\")\nplt.xlabel(\"b\")\nplt.ylabel(\"Precision\")\nplt.plot(b_list, precision_list)\nplt.show()", "_____no_output_____" ], [ "plt.title(f\"Recall (s={s})\")\nplt.xlabel(\"b\")\nplt.ylabel(\"Recall\")\nplt.plot(b_list, recall_list)\nplt.show()", "_____no_output_____" ], [ "plt.title(f\"F1 Score (s={s})\")\nplt.xlabel(\"b\")\nplt.ylabel(\"F1 Score\")\nplt.plot(b_list, f1_list)\nplt.show()", "_____no_output_____" ], [ "# Check whether the test passed\ntest_msg = {True: \"Passed\", False: \"Failed\"}\n\nprint(\"-----Test results-----\")\nprint(f\"[Test 1.1 (1)]: {test_msg[pass_test1_1_1]}\")\nprint(f\"[Test 1.1 (2)]: {test_msg[pass_test1_1_2]}\")\nprint(f\"[Test 1.2]: {test_msg[pass_test1_2]}\")\nprint(f\"[Test 2.1]: {test_msg[pass_test2_1]}\")\nprint(f\"[Test 2.2]: {test_msg[pass_test2_2]}\")\nprint(\"----------------------\")", "-----Test results-----\n[Test 1.1 (1)]: Passed\n[Test 1.1 (2)]: Passed\n[Test 1.2]: Passed\n[Test 2.1]: Passed\n[Test 2.2]: Passed\n----------------------\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" ]

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

[ "BSD-3-Clause" ]

[ "BSD-3-Clause" ]

[ "BSD-3-Clause" ]

[ [ [ "import sys\nsys.path.append(\"..\")", "_____no_output_____" ], [ "import numpy as np\nnp.seterr(divide=\"ignore\")\nimport logging\nimport pickle\nimport glob\n\nfrom sklearn.metrics import roc_curve\nfrom sklearn.metrics import roc_auc_score\nfrom sklearn.preprocessing import RobustScaler\nfrom sklearn.utils import check_random_state\n\nfrom scipy import interp\n\nfrom recnn.preprocessing import rewrite_content\nfrom recnn.preprocessing import permute_by_pt\nfrom recnn.preprocessing import extract\nfrom recnn.preprocessing import sequentialize_by_pt\nfrom recnn.preprocessing import randomize\n\n%matplotlib inline\nimport matplotlib.pyplot as plt\nplt.rcParams[\"figure.figsize\"] = (6, 6)", "_____no_output_____" ] ], [ [ "# Plotting functions", "_____no_output_____" ] ], [ [ "from recnn.preprocessing import sequentialize_by_pt\n\ndef load_tf(filename_train, preprocess=None, n_events_train=-1):\n # Make training data\n print(\"Loading training data...\")\n\n fd = open(filename_train, \"rb\")\n X, y = pickle.load(fd)\n fd.close()\n \n y = np.array(y)\n \n if n_events_train > 0:\n indices = check_random_state(123).permutation(len(X))[:n_events_train]\n X = [X[i] for i in indices]\n y = y[indices]\n\n print(\"\\tfilename = %s\" % filename_train)\n print(\"\\tX size = %d\" % len(X))\n print(\"\\ty size = %d\" % len(y))\n\n # Preprocessing \n print(\"Preprocessing...\")\n X = [rewrite_content(jet) for jet in X]\n \n if preprocess:\n X = [preprocess(jet) for jet in X]\n \n X = [extract(permute_by_pt(jet)) for jet in X]\n tf = RobustScaler().fit(np.vstack([jet[\"content\"] for jet in X]))\n \n return tf\n\ndef load_test(tf, filename_test, preprocess=None, cropping=True):\n # Make test data \n print(\"Loading test data...\")\n\n fd = open(filename_test, \"rb\")\n X, y = pickle.load(fd)\n fd.close()\n y = np.array(y)\n\n print(\"\\tfilename = %s\" % filename_test)\n print(\"\\tX size = %d\" % len(X))\n print(\"\\ty size = %d\" % len(y))\n\n # Preprocessing \n print(\"Preprocessing...\")\n X = [rewrite_content(jet) for jet in X]\n \n if preprocess:\n X = [preprocess(jet) for jet in X]\n \n X = [extract(permute_by_pt(jet)) for jet in X]\n\n for jet in X:\n jet[\"content\"] = tf.transform(jet[\"content\"])\n \n if not cropping:\n return X, y\n \n # Cropping\n X_ = [j for j in X if 250 < j[\"pt\"] < 300 and 50 < j[\"mass\"] < 110]\n y_ = [y[i] for i, j in enumerate(X) if 250 < j[\"pt\"] < 300 and 50 < j[\"mass\"] < 110]\n\n X = X_\n y = y_\n y = np.array(y)\n \n print(\"\\tX size = %d\" % len(X))\n print(\"\\ty size = %d\" % len(y))\n \n # Weights for flatness in pt\n w = np.zeros(len(y)) \n \n X0 = [X[i] for i in range(len(y)) if y[i] == 0]\n pdf, edges = np.histogram([j[\"pt\"] for j in X0], density=True, range=[250, 300], bins=50)\n pts = [j[\"pt\"] for j in X0]\n indices = np.searchsorted(edges, pts) - 1\n inv_w = 1. / pdf[indices]\n inv_w /= inv_w.sum()\n w[y==0] = inv_w\n \n X1 = [X[i] for i in range(len(y)) if y[i] == 1]\n pdf, edges = np.histogram([j[\"pt\"] for j in X1], density=True, range=[250, 300], bins=50)\n pts = [j[\"pt\"] for j in X1]\n indices = np.searchsorted(edges, pts) - 1\n inv_w = 1. / pdf[indices]\n inv_w /= inv_w.sum()\n w[y==1] = inv_w\n \n return X, y, w", "_____no_output_____" ], [ "from recnn.recnn import grnn_transform_simple\nfrom recnn.recnn import grnn_predict_simple\nfrom recnn.recnn import grnn_predict_gated\nfrom recnn.recnn import grnn_predict_simple_join\n\n\ndef predict(X, filename, func=grnn_predict_simple):\n fd = open(filename, \"rb\")\n params = pickle.load(fd)\n fd.close()\n y_pred = func(params, X)\n return y_pred\n\n\ndef evaluate_models(X, y, w, pattern, func=grnn_predict_simple):\n rocs = []\n fprs = []\n tprs = []\n \n for filename in glob.glob(pattern):\n print(\"Loading %s\" % filename),\n \n y_pred = predict(X, filename, func=func)\n \n # Roc\n rocs.append(roc_auc_score(y, y_pred, sample_weight=w))\n fpr, tpr, _ = roc_curve(y, y_pred, sample_weight=w)\n \n fprs.append(fpr)\n tprs.append(tpr)\n \n print(\"ROC AUC = %.4f\" % rocs[-1])\n \n print(\"Mean ROC AUC = %.4f\" % np.mean(rocs))\n \n return rocs, fprs, tprs\n\ndef build_rocs(prefix_train, prefix_test, model_pattern, preprocess=None, gated=False):\n tf = load_tf(\"../data/w-vs-qcd/final/%s-train.pickle\" % prefix_train, preprocess=preprocess)\n X, y, w = load_test(tf, \"../data/w-vs-qcd/final/%s-test.pickle\" % prefix_test, preprocess=preprocess) \n \n if not gated:\n rocs, fprs, tprs = evaluate_models(X, y, w, \n \"../models/jet-study-2/model-w-s-%s-[0-9]*.pickle\" % model_pattern)\n else:\n rocs, fprs, tprs = evaluate_models(X, y, w, \n \"../models/jet-study-2/model-w-g-%s-[0-9]*.pickle\" % model_pattern, func=grnn_predict_gated)\n \n return rocs, fprs, tprs", "_____no_output_____" ], [ "def remove_outliers(rocs, fprs, tprs):\n inv_fprs = []\n base_tpr = np.linspace(0.05, 1, 476)\n\n for fpr, tpr in zip(fprs, tprs):\n inv_fpr = interp(base_tpr, tpr, 1. / fpr)\n inv_fprs.append(inv_fpr)\n\n inv_fprs = np.array(inv_fprs)\n scores = inv_fprs[:, 225]\n \n p25 = np.percentile(scores, 1 / 6. * 100.)\n p75 = np.percentile(scores, 5 / 6. * 100)\n \n robust_mean = np.mean([scores[i] for i in range(len(scores)) if p25 <= scores[i] <= p75])\n robust_std = np.std([scores[i] for i in range(len(scores)) if p25 <= scores[i] <= p75])\n \n indices = [i for i in range(len(scores)) if robust_mean - 3*robust_std <= scores[i] <= robust_mean + 3*robust_std]\n \n new_r, new_f, new_t = [], [], []\n \n for i in indices:\n new_r.append(rocs[i])\n new_f.append(fprs[i])\n new_t.append(tprs[i])\n \n return new_r, new_f, new_t\n\n\ndef report_score(rocs, fprs, tprs, label, latex=False, input=\"particles\", short=False): \n inv_fprs = []\n base_tpr = np.linspace(0.05, 1, 476)\n \n for fpr, tpr in zip(fprs, tprs):\n inv_fpr = interp(base_tpr, tpr, 1. / fpr)\n inv_fprs.append(inv_fpr)\n \n inv_fprs = np.array(inv_fprs)\n mean_inv_fprs = inv_fprs.mean(axis=0)\n \n if not latex:\n print(\"%32s\\tROC AUC=%.4f+-%.2f\\t1/FPR@TPR=0.5=%.2f+-%.2f\" % (label, \n np.mean(rocs), \n np.std(rocs),\n np.mean(inv_fprs[:, 225]),\n np.std(inv_fprs[:, 225])))\n else:\n if not short:\n print(\"%10s \\t& %30s \\t& %.4f $\\pm$ %.4f \\t& %.1f $\\pm$ %.1f \\\\\\\\\" % \n (input,\n label,\n np.mean(rocs), \n np.std(rocs),\n np.mean(inv_fprs[:, 225]),\n np.std(inv_fprs[:, 225])))\n else:\n print(\"%30s \\t& %.4f $\\pm$ %.4f \\t& %.1f $\\pm$ %.1f \\\\\\\\\" % \n (label,\n np.mean(rocs), \n np.std(rocs),\n np.mean(inv_fprs[:, 225]),\n np.std(inv_fprs[:, 225])))\n \ndef plot_rocs(rocs, fprs, tprs, label=\"\", color=\"r\", show_all=False):\n inv_fprs = []\n base_tpr = np.linspace(0.05, 1, 476)\n \n for fpr, tpr in zip(fprs, tprs):\n inv_fpr = interp(base_tpr, tpr, 1. / fpr)\n inv_fprs.append(inv_fpr)\n if show_all:\n plt.plot(base_tpr, inv_fpr, alpha=0.1, color=color)\n \n inv_fprs = np.array(inv_fprs)\n mean_inv_fprs = inv_fprs.mean(axis=0)\n\n\n plt.plot(base_tpr, mean_inv_fprs, color, \n label=\"%s\" % label)\n \ndef plot_show(filename=None):\n plt.xlabel(\"Signal efficiency\")\n plt.ylabel(\"1 / Background efficiency\")\n plt.xlim([0.1, 1.0])\n plt.ylim(1, 500)\n plt.yscale(\"log\")\n plt.legend(loc=\"best\")\n plt.grid()\n \n if filename:\n plt.savefig(filename)\n \n plt.show()", "_____no_output_____" ] ], [ [ "# Count parameters", "_____no_output_____" ] ], [ [ "def count(params):\n def _count(thing):\n if isinstance(thing, list):\n c = 0\n for stuff in thing:\n c += _count(stuff)\n return c \n\n elif isinstance(thing, np.ndarray):\n return np.prod(thing.shape)\n \n c = 0\n for k, v in params.items():\n c += _count(v)\n return c\n \n# Simple vs gated\nfd = open(\"../models/jet-study-2/model-w-s-antikt-kt-1.pickle\", \"rb\")\nparams = pickle.load(fd)\nfd.close()\nprint(\"Simple =\", count(params)) \n\nfd = open(\"../models/jet-study-2/model-w-g-antikt-kt-1.pickle\", \"rb\")\nparams = pickle.load(fd)\nfd.close()\nprint(\"Gated =\", count(params))", "('Simple =', 8481)\n('Gated =', 48761)\n" ], [ "# double\n# Simple vs gated\nfd = open(\"../models/jet-study-2/model-w-sd-antikt-kt-1.pickle\", \"rb\")\nparams = pickle.load(fd)\nfd.close()\nprint(\"Simple =\", count(params)) \n\nfd = open(\"../models/jet-study-2/model-w-gd-antikt-kt-1.pickle\", \"rb\")\nparams = pickle.load(fd)\nfd.close()\nprint(\"Gated =\", count(params))", "('Simple =', 10081)\n('Gated =', 50361)\n" ] ], [ [ "# Embedding visualization", "_____no_output_____" ] ], [ [ "prefix_train = \"antikt-kt\"\nprefix_test = prefix_train\ntf = load_tf(\"../data/w-vs-qcd/final/%s-train.pickle\" % prefix_train)\nX, y, w = load_test(tf, \"../data/w-vs-qcd/final/%s-test.pickle\" % prefix_test) ", "_____no_output_____" ], [ "fd = open(\"../models/jet-study-2/model-w-s-antikt-kt-1.pickle\", \"rb\")\nparams = pickle.load(fd)\nfd.close()", "_____no_output_____" ], [ "Xt = grnn_transform_simple(params, X[:5000])", "_____no_output_____" ], [ "from sklearn.manifold import TSNE\nXtt = TSNE(n_components=2).fit_transform(Xt)", "_____no_output_____" ], [ "for i in range(5000):\n plt.scatter(Xtt[i, 0], Xtt[i, 1], color=\"b\" if y[i] == 1 else \"r\", alpha=0.5)\n\nplt.show()", "_____no_output_____" ], [ "from sklearn.decomposition import PCA\nXtt = PCA(n_components=2).fit_transform(Xt)\n\nfor i in range(5000):\n plt.scatter(Xtt[i, 0], Xtt[i, 1], color=\"b\" if y[i] == 1 else \"r\", alpha=0.5)\n\nplt.show()", "_____no_output_____" ] ], [ [ "# Generate all ROCs", "_____no_output_____" ] ], [ [ "for pattern, gated in [\n # Simple\n ## Particles\n (\"antikt-kt\", False),\n (\"antikt-cambridge\", False),\n (\"antikt-antikt\", False),\n (\"antikt-random\", False),\n (\"antikt-seqpt\", False),\n (\"antikt-seqpt-reversed\", False),\n ## Towers\n (\"antikt-kt-delphes\", False),\n (\"antikt-cambridge-delphes\", False),\n (\"antikt-antikt-delphes\", False),\n (\"antikt-random-delphes\", False),\n (\"antikt-seqpt-delphes\", False),\n (\"antikt-seqpt-reversed-delphes\", False),\n ## Images\n (\"antikt-kt-images\", False),\n\n # Gated\n ## Particles\n (\"antikt-kt\", True),\n (\"antikt-antikt\", True),\n (\"antikt-seqpt\", True),\n (\"antikt-seqpt-reversed\", True),\n (\"antikt-cambridge\", True),\n (\"antikt-random\", True),\n ## Towers\n (\"antikt-kt-delphes\", True),\n (\"antikt-antikt-delphes\", True),\n (\"antikt-seqpt-delphes\", True),\n (\"antikt-seqpt-reversed-delphes\", True),\n (\"antikt-cambridge-delphes\", True),\n (\"antikt-random-delphes\", True),\n ## Images\n (\"antikt-kt-images\", True)\n ]:\n r, f, t = build_rocs(pattern, pattern, pattern, gated=gated)\n \n # Save\n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s.pickle\" % (\"s\" if not gated else \"g\", pattern), \"wb\")\n pickle.dump((r, f, t), fd)\n fd.close()", "_____no_output_____" ], [ "# sd/gd == contatenate embeddings of h1_L + h1_R\nfor pattern, gated in [\n # Simple\n ## Particles\n (\"antikt-kt\", False),\n ## Towers\n (\"antikt-kt-delphes\", False),\n ## Images\n (\"antikt-kt-images\", False),\n\n # Gated\n ## Particles\n (\"antikt-kt\", True),\n ## Towers\n (\"antikt-kt-delphes\", True),\n ## Images\n (\"antikt-kt-images\", True)\n ]:\n r, f, t = build_rocs(pattern, pattern, pattern, gated=gated)\n \n # Save\n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s.pickle\" % (\"sd\" if not gated else \"gd\", pattern), \"wb\")\n pickle.dump((r, f, t), fd)\n fd.close()", "_____no_output_____" ] ], [ [ "# Table", "_____no_output_____" ] ], [ [ "for pattern, gated, label in [\n # Simple\n ## Particles\n (\"antikt-kt\", False, \"RNN $k_t$\"),\n (\"antikt-cambridge\", False, \"RNN C/A\"),\n (\"antikt-antikt\", False, \"RNN anti-$k_t$\"),\n (\"antikt-random\", False, \"RNN random\"),\n (\"antikt-seqpt\", False, \"RNN asc-$p_T$\"),\n (\"antikt-seqpt-reversed\", False, \"RNN desc-$p_T$\"),\n ## Towers\n (\"antikt-kt-delphes\", False, \"RNN $k_t$\"),\n (\"antikt-cambridge-delphes\", False, \"RNN C/A\"),\n (\"antikt-antikt-delphes\", False, \"RNN anti-$k_t$\"),\n (\"antikt-random-delphes\", False, \"RNN random\"),\n (\"antikt-seqpt-delphes\", False, \"RNN asc-$p_T$\"),\n (\"antikt-seqpt-reversed-delphes\", False, \"RNN desc-$p_T$\"),\n ## Images\n (\"antikt-kt-images\", False, \"RNN $k_t$\"),\n\n # Gated\n ## Particles\n (\"antikt-kt\", True, \"RNN $k_t$ (gated)\"),\n (\"antikt-cambridge\", True, \"RNN C/A (gated)\"),\n (\"antikt-antikt\", True, \"RNN anti-$k_t$ (gated)\"),\n (\"antikt-random\", True, \"RNN random (gated)\"),\n (\"antikt-seqpt\", True, \"RNN asc-$p_T$ (gated)\"),\n (\"antikt-seqpt-reversed\", True, \"RNN desc-$p_T$ (gated)\"),\n ## Towers\n (\"antikt-kt-delphes\", True, \"RNN $k_t$ (gated)\"),\n (\"antikt-cambridge-delphes\", True, \"RNN C/A (gated)\"),\n (\"antikt-antikt-delphes\", True, \"RNN anti-$k_t$ (gated)\"),\n (\"antikt-random-delphes\", True, \"RNN random (gated)\"),\n (\"antikt-seqpt-delphes\", True, \"RNN asc-$p_T$ (gated)\"),\n (\"antikt-seqpt-reversed-delphes\", True, \"RNN desc-$p_T$ (gated)\"),\n \n # Images\n (\"antikt-kt-images\", False, \"RNN $k_t$\"),\n (\"antikt-kt-images\", True, \"RNN $k_t$ (gated)\")\n ]:\n \n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s.pickle\" % (\"s\" if not gated else \"g\", pattern), \"rb\")\n r, f, t = pickle.load(fd)\n fd.close()\n \n r, f, t = remove_outliers(r, f, t)\n \n report_score(r, f, t, label=label, \n latex=True, \n input=\"particles\" if \"delphes\" not in pattern and \"images\" not in pattern else \"towers\")", " particles \t& RNN $k_t$ \t& 0.9185 $\\pm$ 0.0006 \t& 68.3 $\\pm$ 1.8 \\\\\n particles \t& RNN C/A \t& 0.9192 $\\pm$ 0.0008 \t& 68.3 $\\pm$ 3.6 \\\\\n particles \t& RNN anti-$k_t$ \t& 0.9096 $\\pm$ 0.0013 \t& 51.7 $\\pm$ 3.5 \\\\\n particles \t& RNN random \t& 0.9121 $\\pm$ 0.0008 \t& 51.1 $\\pm$ 2.0 \\\\\n particles \t& RNN asc-$p_T$ \t& 0.9130 $\\pm$ 0.0031 \t& 52.5 $\\pm$ 7.3 \\\\\n particles \t& RNN desc-$p_T$ \t& 0.9189 $\\pm$ 0.0009 \t& 70.4 $\\pm$ 3.6 \\\\\n towers \t& RNN $k_t$ \t& 0.8807 $\\pm$ 0.0010 \t& 24.1 $\\pm$ 0.6 \\\\\n towers \t& RNN C/A \t& 0.8831 $\\pm$ 0.0010 \t& 24.2 $\\pm$ 0.7 \\\\\n towers \t& RNN anti-$k_t$ \t& 0.8737 $\\pm$ 0.0017 \t& 22.3 $\\pm$ 0.8 \\\\\n towers \t& RNN random \t& 0.8704 $\\pm$ 0.0011 \t& 20.4 $\\pm$ 0.3 \\\\\n towers \t& RNN asc-$p_T$ \t& 0.8835 $\\pm$ 0.0009 \t& 26.2 $\\pm$ 0.7 \\\\\n towers \t& RNN desc-$p_T$ \t& 0.8838 $\\pm$ 0.0010 \t& 25.1 $\\pm$ 0.6 \\\\\n towers \t& RNN $k_t$ \t& 0.8321 $\\pm$ 0.0025 \t& 12.7 $\\pm$ 0.4 \\\\\n particles \t& RNN $k_t$ (gated) \t& 0.9195 $\\pm$ 0.0009 \t& 74.3 $\\pm$ 2.4 \\\\\n particles \t& RNN C/A (gated) \t& 0.9222 $\\pm$ 0.0007 \t& 81.8 $\\pm$ 3.1 \\\\\n particles \t& RNN anti-$k_t$ (gated) \t& 0.9156 $\\pm$ 0.0012 \t& 68.3 $\\pm$ 3.2 \\\\\n particles \t& RNN random (gated) \t& 0.9106 $\\pm$ 0.0035 \t& 50.7 $\\pm$ 6.7 \\\\\n particles \t& RNN asc-$p_T$ (gated) \t& 0.9137 $\\pm$ 0.0046 \t& 54.8 $\\pm$ 11.7 \\\\\n particles \t& RNN desc-$p_T$ (gated) \t& 0.9212 $\\pm$ 0.0005 \t& 83.3 $\\pm$ 3.1 \\\\\n towers \t& RNN $k_t$ (gated) \t& 0.8822 $\\pm$ 0.0006 \t& 25.4 $\\pm$ 0.4 \\\\\n towers \t& RNN C/A (gated) \t& 0.8861 $\\pm$ 0.0014 \t& 26.2 $\\pm$ 0.8 \\\\\n towers \t& RNN anti-$k_t$ (gated) \t& 0.8804 $\\pm$ 0.0010 \t& 24.4 $\\pm$ 0.4 \\\\\n towers \t& RNN random (gated) \t& 0.8751 $\\pm$ 0.0029 \t& 22.8 $\\pm$ 1.2 \\\\\n towers \t& RNN asc-$p_T$ (gated) \t& 0.8849 $\\pm$ 0.0012 \t& 27.2 $\\pm$ 0.8 \\\\\n towers \t& RNN desc-$p_T$ (gated) \t& 0.8864 $\\pm$ 0.0007 \t& 27.5 $\\pm$ 0.6 \\\\\n towers \t& RNN $k_t$ \t& 0.8321 $\\pm$ 0.0025 \t& 12.7 $\\pm$ 0.4 \\\\\n towers \t& RNN $k_t$ (gated) \t& 0.8277 $\\pm$ 0.0028 \t& 12.4 $\\pm$ 0.3 \\\\\n" ], [ "for pattern, gated, label in [\n # Simple\n ## Particles\n (\"antikt-kt\", False, \"RNN $k_t$\"),\n ## Towers\n (\"antikt-kt-delphes\", False, \"RNN $k_t$\"),\n ## Images\n (\"antikt-kt-images\", False, \"RNN $k_t$\"),\n\n # Gated\n ## Particles\n (\"antikt-kt\", True, \"RNN $k_t$ (gated)\"),\n ## Towers\n (\"antikt-kt-delphes\", True, \"RNN $k_t$ (gated)\"),\n # Images\n (\"antikt-kt-images\", True, \"RNN $k_t$ (gated)\")\n ]:\n \n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s.pickle\" % (\"sd\" if not gated else \"gd\", pattern), \"rb\")\n r, f, t = pickle.load(fd)\n fd.close()\n \n r, f, t = remove_outliers(r, f, t)\n \n report_score(r, f, t, label=label, \n latex=True, \n input=\"particles\" if \"delphes\" not in pattern and \"images\" not in pattern else \"towers\")", " particles \t& RNN $k_t$ \t& 0.9178 $\\pm$ 0.0008 \t& 67.0 $\\pm$ 2.6 \\\\\n towers \t& RNN $k_t$ \t& 0.8801 $\\pm$ 0.0012 \t& 24.1 $\\pm$ 0.6 \\\\\n towers \t& RNN $k_t$ \t& 0.8332 $\\pm$ 0.0029 \t& 12.6 $\\pm$ 0.4 \\\\\n particles \t& RNN $k_t$ (gated) \t& 0.9187 $\\pm$ 0.0011 \t& 70.2 $\\pm$ 3.8 \\\\\n towers \t& RNN $k_t$ (gated) \t& 0.8810 $\\pm$ 0.0012 \t& 24.5 $\\pm$ 0.5 \\\\\n towers \t& RNN $k_t$ (gated) \t& 0.8255 $\\pm$ 0.0050 \t& 12.2 $\\pm$ 0.6 \\\\\n" ] ], [ [ "# Plots", "_____no_output_____" ] ], [ [ "# Simple vs gated\nfor pattern, gated, label, color in [\n (\"antikt-kt\", False, \"RNN $k_t$ (simple)\", \"r\"),\n (\"antikt-kt\", True, \"RNN $k_t$ (gated)\", \"b\")\n ]:\n \n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s.pickle\" % (\"s\" if not gated else \"g\", pattern), \"rb\")\n r, f, t = pickle.load(fd)\n fd.close()\n \n r, f, t = remove_outliers(r, f, t)\n \n plot_rocs(r, f, t, label=label, color=color)\n report_score(r, f, t, label=label)\n \nplot_show()", " RNN $k_t$ (simple)\tROC AUC=0.9185+-0.00\t1/FPR@TPR=0.5=68.32+-1.80\n RNN $k_t$ (gated)\tROC AUC=0.9195+-0.00\t1/FPR@TPR=0.5=74.34+-2.40\n" ], [ "# Topologies (particles, simple)\nfor pattern, gated, label, color in [\n (\"antikt-kt\", False, \"$k_t$\", \"r\"), \n (\"antikt-cambridge\", False, \"C/A\", \"g\"),\n (\"antikt-antikt\", False, \"anti-$k_t$\", \"b\"), \n (\"antikt-seqpt\", False, \"asc-$p_T$\", \"c\"),\n (\"antikt-seqpt-reversed\", False, \"desc-$p_T$\", \"m\"),\n (\"antikt-random\", False, \"random\", \"orange\")\n ]:\n \n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s.pickle\" % (\"s\" if not gated else \"g\", pattern), \"rb\")\n r, f, t = pickle.load(fd)\n fd.close()\n \n r, f, t = remove_outliers(r, f, t)\n \n plot_rocs(r, f, t, label=label, color=color)\n report_score(r, f, t, label=label)\n \nplot_show()", " $k_t$\tROC AUC=0.9185+-0.00\t1/FPR@TPR=0.5=68.32+-1.80\n C/A\tROC AUC=0.9192+-0.00\t1/FPR@TPR=0.5=68.29+-3.62\n anti-$k_t$\tROC AUC=0.9096+-0.00\t1/FPR@TPR=0.5=51.66+-3.49\n asc-$p_T$\tROC AUC=0.9130+-0.00\t1/FPR@TPR=0.5=52.51+-7.27\n desc-$p_T$\tROC AUC=0.9189+-0.00\t1/FPR@TPR=0.5=70.38+-3.65\n random\tROC AUC=0.9121+-0.00\t1/FPR@TPR=0.5=51.06+-2.00\n" ], [ "# Topologies (towers, simple)\nfor pattern, gated, label, color in [\n (\"antikt-kt-delphes\", False, \"RNN $k_t$\", \"r\"), \n (\"antikt-cambridge-delphes\", False, \"RNN C/A\", \"g\"),\n (\"antikt-antikt-delphes\", False, \"RNN anti-$k_t$\", \"b\"), \n (\"antikt-seqpt-delphes\", False, \"RNN asc-$p_T$\", \"c\"),\n (\"antikt-seqpt-reversed-delphes\", False, \"RNN desc-$p_T$\", \"m\"),\n (\"antikt-random-delphes\", False, \"RNN random\", \"orange\")\n ]:\n \n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s.pickle\" % (\"s\" if not gated else \"g\", pattern), \"rb\")\n r, f, t = pickle.load(fd)\n fd.close()\n \n r, f, t = remove_outliers(r, f, t)\n \n plot_rocs(r, f, t, label=label, color=color)\n report_score(r, f, t, label=label)\n \nplot_show()", " RNN $k_t$\tROC AUC=0.8807+-0.00\t1/FPR@TPR=0.5=24.07+-0.55\n RNN C/A\tROC AUC=0.8831+-0.00\t1/FPR@TPR=0.5=24.15+-0.69\n RNN anti-$k_t$\tROC AUC=0.8737+-0.00\t1/FPR@TPR=0.5=22.33+-0.85\n RNN asc-$p_T$\tROC AUC=0.8835+-0.00\t1/FPR@TPR=0.5=26.15+-0.70\n RNN desc-$p_T$\tROC AUC=0.8838+-0.00\t1/FPR@TPR=0.5=25.13+-0.58\n RNN random\tROC AUC=0.8704+-0.00\t1/FPR@TPR=0.5=20.35+-0.34\n" ], [ "# Topologies (particles, gated)\nfor pattern, gated, label, color in [\n (\"antikt-kt\", True, \"RNN $k_t$\", \"r\"), \n (\"antikt-antikt\", True, \"RNN anti-$k_t$\", \"b\"), \n (\"antikt-seqpt\", True, \"RNN asc-$p_T$\", \"c\"),\n (\"antikt-seqpt-reversed\", True, \"RNN desc-$p_T$\", \"m\"),\n ]:\n \n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s.pickle\" % (\"s\" if not gated else \"g\", pattern), \"rb\")\n r, f, t = pickle.load(fd)\n fd.close()\n \n r, f, t = remove_outliers(r, f, t)\n \n plot_rocs(r, f, t, label=label, color=color)\n report_score(r, f, t, label=label)\n \nplot_show()", " RNN $k_t$\tROC AUC=0.9195+-0.00\t1/FPR@TPR=0.5=74.34+-2.40\n RNN anti-$k_t$\tROC AUC=0.9156+-0.00\t1/FPR@TPR=0.5=68.25+-3.18\n RNN asc-$p_T$\tROC AUC=0.9137+-0.00\t1/FPR@TPR=0.5=54.79+-11.70\n RNN desc-$p_T$\tROC AUC=0.9212+-0.00\t1/FPR@TPR=0.5=83.27+-3.08\n" ], [ "# Topologies (towers, gated)\nfor pattern, gated, label, color in [\n (\"antikt-kt-delphes\", True, \"RNN $k_t$\", \"r\"), \n (\"antikt-antikt-delphes\", True, \"RNN anti-$k_t$\", \"b\"), \n (\"antikt-seqpt-delphes\", True, \"RNN asc-$p_T$\", \"c\"),\n (\"antikt-seqpt-reversed-delphes\", True, \"RNN desc-$p_T$\", \"m\"),\n ]:\n \n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s.pickle\" % (\"s\" if not gated else \"g\", pattern), \"rb\")\n r, f, t = pickle.load(fd)\n fd.close()\n \n r, f, t = remove_outliers(r, f, t)\n \n plot_rocs(r, f, t, label=label, color=color)\n report_score(r, f, t, label=label)\n \nplot_show()", " RNN $k_t$\tROC AUC=0.8822+-0.00\t1/FPR@TPR=0.5=25.41+-0.36\n RNN anti-$k_t$\tROC AUC=0.8804+-0.00\t1/FPR@TPR=0.5=24.37+-0.42\n RNN asc-$p_T$\tROC AUC=0.8849+-0.00\t1/FPR@TPR=0.5=27.16+-0.82\n RNN desc-$p_T$\tROC AUC=0.8864+-0.00\t1/FPR@TPR=0.5=27.50+-0.55\n" ], [ "# Particles vs towers vs images (simple)\nfor pattern, gated, label, color in [\n (\"antikt-kt\", False, \"particles\", \"r\"), \n (\"antikt-kt-delphes\", False, \"towers\", \"g\"),\n (\"antikt-kt-images\", False, \"images\", \"b\"), \n ]:\n \n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s.pickle\" % (\"s\" if not gated else \"g\", pattern), \"rb\")\n r, f, t = pickle.load(fd)\n fd.close()\n \n r, f, t = remove_outliers(r, f, t)\n \n plot_rocs(r, f, t, label=label, color=color)\n report_score(r, f, t, label=label)\n \nplot_show(filename=\"particles-towers-images.pdf\")", " particles\tROC AUC=0.9185+-0.00\t1/FPR@TPR=0.5=68.32+-1.80\n towers\tROC AUC=0.8807+-0.00\t1/FPR@TPR=0.5=24.07+-0.55\n images\tROC AUC=0.8321+-0.00\t1/FPR@TPR=0.5=12.67+-0.43\n" ], [ "# Particles vs towers vs images (gated)\nfor pattern, gated, label, color in [\n (\"antikt-kt\", True, \"particles\", \"r\"), \n (\"antikt-kt-delphes\", True, \"towers\", \"g\"),\n (\"antikt-kt-images\", True, \"images\", \"b\"), \n ]:\n \n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s.pickle\" % (\"s\" if not gated else \"g\", pattern), \"rb\")\n r, f, t = pickle.load(fd)\n fd.close()\n \n r, f, t = remove_outliers(r, f, t)\n \n plot_rocs(r, f, t, label=label, color=color)\n report_score(r, f, t, label=label)\n \nplot_show()", " particles\tROC AUC=0.9195+-0.00\t1/FPR@TPR=0.5=74.34+-2.40\n towers\tROC AUC=0.8822+-0.00\t1/FPR@TPR=0.5=25.41+-0.36\n images\tROC AUC=0.8277+-0.00\t1/FPR@TPR=0.5=12.36+-0.30\n" ] ], [ [ "# Trimming", "_____no_output_____" ] ], [ [ "for pattern_train, pattern_test, gated in [\n (\"antikt-kt\", \"antikt-kt\", False),\n (\"antikt-kt\", \"antikt-kt-trimmed\", False),\n (\"antikt-kt-trimmed\", \"antikt-kt-trimmed\", False),\n (\"antikt-kt-trimmed\", \"antikt-kt\", False),\n ]:\n r, f, t = build_rocs(pattern_train, pattern_test, pattern_train, gated=gated)\n \n # Save\n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s-%s.pickle\" % \n (\"s\" if not gated else \"g\", pattern_train, pattern_test), \"wb\")\n pickle.dump((r, f, t), fd)\n fd.close()", "_____no_output_____" ], [ "for pattern_train, pattern_test, gated, label, color in [\n (\"antikt-kt\", \"antikt-kt\", False, \"$k_t$ on $k_t$\", \"b\"),\n (\"antikt-kt\", \"antikt-kt-trimmed\", False, \"$k_t$ on $k_t$-trimmed\", \"c\"),\n (\"antikt-kt-trimmed\", \"antikt-kt-trimmed\", False, \"$k_t$-trimmed on $k_t$-trimmed\", \"r\"),\n (\"antikt-kt-trimmed\", \"antikt-kt\", False, \"$k_t$-trimmed on $k_t$\", \"orange\"),\n ]:\n \n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s-%s.pickle\" % \n (\"s\" if not gated else \"g\", pattern_train, pattern_test), \"rb\")\n r, f, t = pickle.load(fd)\n fd.close()\n \n r, f, t = remove_outliers(r, f, t)\n \n plot_rocs(r, f, t, label=label, color=color)\n report_score(r, f, t, label=label)\n \nplot_show()", " $k_t$ on $k_t$\tROC AUC=0.9185+-0.00\t1/FPR@TPR=0.5=68.32+-1.80\n $k_t$ on $k_t$-trimmed\tROC AUC=0.8952+-0.00\t1/FPR@TPR=0.5=28.70+-1.30\n $k_t$-trimmed on $k_t$-trimmed\tROC AUC=0.9034+-0.00\t1/FPR@TPR=0.5=33.05+-1.85\n $k_t$-trimmed on $k_t$\tROC AUC=0.8893+-0.01\t1/FPR@TPR=0.5=36.78+-5.88\n" ] ], [ [ "# Colinear splits", "_____no_output_____" ] ], [ [ "from functools import partial\nfrom recnn.preprocessing import sequentialize_by_pt\n\npreprocess_seqpt = partial(sequentialize_by_pt, reverse=False)\npreprocess_seqpt_rev = partial(sequentialize_by_pt, reverse=True)\n\nfor pattern_train, pattern_test, gated, preprocess in [\n # kt\n (\"antikt-kt\", \"antikt-kt-colinear1\", False, None),\n (\"antikt-kt\", \"antikt-kt-colinear10\", False, None),\n (\"antikt-kt\", \"antikt-kt-colinear1-max\", False, None),\n (\"antikt-kt\", \"antikt-kt-colinear10-max\", False, None),\n \n # asc-pt\n (\"antikt-seqpt\", \"antikt-kt-colinear1\", False, preprocess_seqpt),\n (\"antikt-seqpt\", \"antikt-kt-colinear10\", False, preprocess_seqpt),\n (\"antikt-seqpt\", \"antikt-kt-colinear1-max\", False, preprocess_seqpt),\n (\"antikt-seqpt\", \"antikt-kt-colinear10-max\", False, preprocess_seqpt),\n \n # desc-pt\n (\"antikt-seqpt-reversed\", \"antikt-kt-colinear1\", False, preprocess_seqpt_rev),\n (\"antikt-seqpt-reversed\", \"antikt-kt-colinear10\", False, preprocess_seqpt_rev),\n (\"antikt-seqpt-reversed\", \"antikt-kt-colinear1-max\", False, preprocess_seqpt_rev),\n (\"antikt-seqpt-reversed\", \"antikt-kt-colinear10-max\", False, preprocess_seqpt_rev),\n ]:\n \n r, f, t = build_rocs(pattern_train, pattern_test, pattern_train, gated=gated, preprocess=preprocess)\n \n # Save\n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s-%s.pickle\" % \n (\"s\" if not gated else \"g\", pattern_train, pattern_test), \"wb\")\n pickle.dump((r, f, t), fd)\n fd.close()", "_____no_output_____" ], [ "for pattern_train, pattern_test, gated, label in [\n # kt\n (\"antikt-kt\", \"antikt-kt-colinear1\", False, \"$k_t$ colinear1\"),\n (\"antikt-kt\", \"antikt-kt-colinear10\", False, \"$k_t$ colinear10\"),\n (\"antikt-kt\", \"antikt-kt-colinear1-max\", False, \"$k_t$ colinear1-max\"),\n (\"antikt-kt\", \"antikt-kt-colinear10-max\", False, \"$k_t$ colinear10-max\"),\n \n # asc-pt\n (\"antikt-seqpt\", \"antikt-kt-colinear1\", False, \"asc-$p_T$ colinear1\"),\n (\"antikt-seqpt\", \"antikt-kt-colinear10\", False, \"asc-$p_T$ colinear10\"),\n (\"antikt-seqpt\", \"antikt-kt-colinear1-max\", False, \"asc-$p_T$ colinear1-max\"),\n (\"antikt-seqpt\", \"antikt-kt-colinear10-max\", False, \"asc-$p_T$ colinear10-max\"),\n \n # desc-pt\n (\"antikt-seqpt-reversed\", \"antikt-kt-colinear1\", False, \"desc-$p_T$ colinear1\"),\n (\"antikt-seqpt-reversed\", \"antikt-kt-colinear10\", False, \"desc-$p_T$ colinear10\"),\n (\"antikt-seqpt-reversed\", \"antikt-kt-colinear1-max\", False, \"desc-$p_T$ colinear1-max\"),\n (\"antikt-seqpt-reversed\", \"antikt-kt-colinear10-max\", False, \"desc-$p_T$ colinear10-max\"),\n ]:\n \n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s-%s.pickle\" % \n (\"s\" if not gated else \"g\", pattern_train, pattern_test), \"rb\")\n r, f, t = pickle.load(fd)\n fd.close()\n \n r, f, t = remove_outliers(r, f, t)\n \n report_score(r, f, t, label=label,\n latex=True, short=True)", " $k_t$ colinear1 \t& 0.9183 $\\pm$ 0.0006 \t& 68.7 $\\pm$ 2.0 \\\\\n $k_t$ colinear10 \t& 0.9174 $\\pm$ 0.0006 \t& 67.5 $\\pm$ 2.6 \\\\\n $k_t$ colinear1-max \t& 0.9184 $\\pm$ 0.0006 \t& 68.5 $\\pm$ 2.8 \\\\\n $k_t$ colinear10-max \t& 0.9159 $\\pm$ 0.0009 \t& 65.7 $\\pm$ 2.7 \\\\\n asc-$p_T$ colinear1 \t& 0.9129 $\\pm$ 0.0031 \t& 53.0 $\\pm$ 7.5 \\\\\n asc-$p_T$ colinear10 \t& 0.9126 $\\pm$ 0.0033 \t& 53.6 $\\pm$ 7.2 \\\\\n asc-$p_T$ colinear1-max \t& 0.9129 $\\pm$ 0.0032 \t& 54.2 $\\pm$ 7.9 \\\\\n asc-$p_T$ colinear10-max \t& 0.9090 $\\pm$ 0.0039 \t& 50.2 $\\pm$ 8.2 \\\\\n desc-$p_T$ colinear1 \t& 0.9188 $\\pm$ 0.0010 \t& 70.7 $\\pm$ 4.0 \\\\\n desc-$p_T$ colinear10 \t& 0.9178 $\\pm$ 0.0011 \t& 67.9 $\\pm$ 4.3 \\\\\n desc-$p_T$ colinear1-max \t& 0.9191 $\\pm$ 0.0010 \t& 72.4 $\\pm$ 4.3 \\\\\n desc-$p_T$ colinear10-max \t& 0.9140 $\\pm$ 0.0016 \t& 63.5 $\\pm$ 5.2 \\\\\n" ] ], [ [ "# Soft particles", "_____no_output_____" ] ], [ [ "from functools import partial\nfrom recnn.preprocessing import sequentialize_by_pt\n\npreprocess_seqpt = partial(sequentialize_by_pt, reverse=False)\npreprocess_seqpt_rev = partial(sequentialize_by_pt, reverse=True)\n\nfor pattern_train, pattern_test, gated, preprocess in [\n (\"antikt-kt\", \"antikt-kt-soft\", False, None),\n (\"antikt-seqpt\", \"antikt-kt-soft\", False, preprocess_seqpt),\n (\"antikt-seqpt-reversed\", \"antikt-kt-soft\", False, preprocess_seqpt_rev),\n ]:\n \n r, f, t = build_rocs(pattern_train, pattern_test, pattern_train, gated=gated, preprocess=preprocess)\n \n # Save\n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s-%s.pickle\" % \n (\"s\" if not gated else \"g\", pattern_train, pattern_test), \"wb\")\n pickle.dump((r, f, t), fd)\n fd.close()", "_____no_output_____" ], [ "for pattern_train, pattern_test, gated, label in [\n (\"antikt-kt\", \"antikt-kt-soft\", False, \"$k_t$ soft\"),\n (\"antikt-seqpt\", \"antikt-kt-soft\", False, \"asc-$p_T$ soft\"),\n (\"antikt-seqpt-reversed\", \"antikt-kt-soft\", False, \"desc-$p_T$ soft\"),\n ]:\n \n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s-%s.pickle\" % \n (\"s\" if not gated else \"g\", pattern_train, pattern_test), \"rb\")\n r, f, t = pickle.load(fd)\n fd.close()\n \n r, f, t = remove_outliers(r, f, t)\n \n report_score(r, f, t, label=label, latex=True, short=True)", " $k_t$ soft \t& 0.9179 $\\pm$ 0.0006 \t& 68.2 $\\pm$ 2.3 \\\\\n asc-$p_T$ soft \t& 0.9121 $\\pm$ 0.0032 \t& 51.3 $\\pm$ 6.0 \\\\\n desc-$p_T$ soft \t& 0.9188 $\\pm$ 0.0009 \t& 70.2 $\\pm$ 3.7 \\\\\n" ] ], [ [ "# Learning curve", "_____no_output_____" ] ], [ [ "for pattern, gated, n_events in [\n# (\"antikt-kt\", False, 6000),\n# (\"antikt-seqpt-reversed\", False, 6000),\n (\"antikt-kt\", True, 6000),\n (\"antikt-seqpt-reversed\", True, 6000),\n# (\"antikt-kt\", False, 15000),\n# (\"antikt-seqpt-reversed\", False, 15000),\n (\"antikt-kt\", True, 15000),\n (\"antikt-seqpt-reversed\", True, 15000),\n ]:\n \n tf = load_tf(\"../data/w-vs-qcd/final/%s-train.pickle\" % pattern, n_events_train=n_events)\n X, y, w = load_test(tf, \"../data/w-vs-qcd/final/%s-test.pickle\" % pattern) \n \n if not gated:\n rocs, fprs, tprs = evaluate_models(X, y, w, \n \"../models/jet-study-2/model-w-s-%s-%d-[0-9]*.pickle\" % (pattern, n_events))\n else:\n rocs, fprs, tprs = evaluate_models(X, y, w, \n \"../models/jet-study-2/model-w-g-%s-%d-[0-9]*.pickle\" % (pattern, n_events), func=grnn_predict_gated)\n \n # Save\n fd = open(\"../models/jet-study-2/rocs/rocs-%s-%s-%d.pickle\" % (\"s\" if not gated else \"g\", pattern, n_events), \"wb\")\n pickle.dump((rocs, fprs, tprs), fd)\n fd.close()", "_____no_output_____" ], [ "for pattern, label, color in [\n (\"s-antikt-kt\", \"$k_t$ 100k\", \"r\"),\n (\"s-antikt-kt-15000\", \"$k_t$ 10k\", \"g\"),\n (\"s-antikt-kt-6000\", \"$k_t$ 1k\", \"b\"),\n (\"s-antikt-seqpt-reversed\", \"desc-$p_T$ 100k\", \"r--\"),\n (\"s-antikt-seqpt-reversed-15000\", \"desc-$p_T$ 10k\", \"g--\"),\n (\"s-antikt-seqpt-reversed-6000\", \"desc-$p_T$ 1k\", \"b--\"),\n ]:\n \n fd = open(\"../models/jet-study-2/rocs/rocs-%s.pickle\" % pattern, \"rb\")\n r, f, t = pickle.load(fd)\n fd.close()\n \n r, f, t = remove_outliers(r, f, t)\n plot_rocs(r, f, t, label=label, color=color)\n report_score(r, f, t, label=label)\n \nplot_show()", " $k_t$ 100k\tROC AUC=0.9185+-0.00\t1/FPR@TPR=0.5=68.32+-1.80\n $k_t$ 10k\tROC AUC=0.9038+-0.00\t1/FPR@TPR=0.5=40.66+-4.58\n $k_t$ 1k\tROC AUC=0.8758+-0.01\t1/FPR@TPR=0.5=23.74+-1.95\n desc-$p_T$ 100k\tROC AUC=0.9189+-0.00\t1/FPR@TPR=0.5=70.38+-3.65\n desc-$p_T$ 10k\tROC AUC=0.9030+-0.00\t1/FPR@TPR=0.5=36.71+-4.00\n desc-$p_T$ 1k\tROC AUC=0.8768+-0.00\t1/FPR@TPR=0.5=22.04+-2.38\n" ], [ "for pattern, label, color in [\n (\"g-antikt-kt\", \"$k_t$ 100k\", \"r\"),\n (\"g-antikt-kt-15000\", \"$k_t$ 10k\", \"g\"),\n (\"g-antikt-kt-6000\", \"$k_t$ 1k\", \"b\"),\n (\"g-antikt-seqpt-reversed\", \"desc-$p_T$ 100k\", \"r--\"),\n (\"g-antikt-seqpt-reversed-15000\", \"desc-$p_T$ 10k\", \"g--\"),\n (\"g-antikt-seqpt-reversed-6000\", \"desc-$p_T$ 1k\", \"b--\"),\n ]:\n \n fd = open(\"../models/jet-study-2/rocs/rocs-%s.pickle\" % pattern, \"rb\")\n r, f, t = pickle.load(fd)\n fd.close()\n \n r, f, t = remove_outliers(r, f, t)\n plot_rocs(r, f, t, label=label, color=color)\n report_score(r, f, t, label=label)\n \nplot_show()", " $k_t$ 100k\tROC AUC=0.9195+-0.00\t1/FPR@TPR=0.5=74.34+-2.40\n $k_t$ 10k\tROC AUC=0.9042+-0.00\t1/FPR@TPR=0.5=46.42+-5.09\n $k_t$ 1k\tROC AUC=0.8726+-0.01\t1/FPR@TPR=0.5=22.63+-2.97\n desc-$p_T$ 100k\tROC AUC=0.9212+-0.00\t1/FPR@TPR=0.5=83.27+-3.08\n desc-$p_T$ 10k\tROC AUC=0.9123+-0.00\t1/FPR@TPR=0.5=56.12+-4.75\n desc-$p_T$ 1k\tROC AUC=0.8860+-0.00\t1/FPR@TPR=0.5=27.35+-2.51\n" ] ], [ [ "# Tau21", "_____no_output_____" ] ], [ [ "import h5py", "_____no_output_____" ], [ "f = h5py.File(\"../data/w-vs-qcd/h5/w_100000_j1p0_sj0p30_delphes_jets_images.h5\", \"r\")[\"auxvars\"]\ntau1 = f[\"tau_1\"]\ntau2 = f[\"tau_2\"]\ntau21 = np.true_divide(tau2, tau1)\npt = f[\"pt_trimmed\"]\nmass = f[\"mass_trimmed\"]\nmask = (f[\"mass_trimmed\"] < 110) & (f[\"mass_trimmed\"] > 50) & (f[\"pt_trimmed\"] < 300) & (f[\"pt_trimmed\"] > 250)\n#mask = mask & np.isfinite(tau21) & (tau21 != 0.)\nsignal_tau21 = tau21[mask]\nsignal_pt = pt[mask]\nsignal_mass = mass[mask]\n\nf = h5py.File(\"../data/w-vs-qcd/h5/qcd_100000_j1p0_sj0p30_delphes_jets_images.h5\", \"r\")[\"auxvars\"]\ntau1 = f[\"tau_1\"]\ntau2 = f[\"tau_2\"]\ntau21 = np.true_divide(tau2, tau1)\npt = f[\"pt_trimmed\"]\nmass = f[\"mass_trimmed\"]\nmask = (f[\"mass_trimmed\"] < 110) & (f[\"mass_trimmed\"] > 50) & (f[\"pt_trimmed\"] < 300) & (f[\"pt_trimmed\"] > 250)\n#mask = mask & np.isfinite(tau21) & (tau21 != 0.)\nbkg_tau21 = tau21[mask]\nbkg_pt = pt[mask]\nbkg_mass = mass[mask]", "/home/gilles/anaconda3/envs/hep/lib/python2.7/site-packages/ipykernel/__main__.py:4: RuntimeWarning: invalid value encountered in true_divide\n/home/gilles/anaconda3/envs/hep/lib/python2.7/site-packages/ipykernel/__main__.py:16: RuntimeWarning: invalid value encountered in true_divide\n" ], [ "plt.hist(bkg_mass, histtype=\"step\", bins=40, normed=1)\nplt.hist(signal_mass, histtype=\"step\", bins=40, normed=1)", "_____no_output_____" ], [ "tau21 = np.concatenate((signal_tau21, bkg_tau21))\npts = np.concatenate((signal_pt, bkg_pt))\nmasss = np.concatenate((signal_mass, bkg_mass))\n\nX = np.hstack([tau21.reshape(-1,1), masss.reshape(-1,1)])\ny = np.concatenate((np.ones(len(signal_tau21)), np.zeros(len(bkg_tau21))))\n\nw = np.zeros(len(y)) \n\npdf, edges = np.histogram(pts[y == 0], density=True, range=[250, 300], bins=50)\nindices = np.searchsorted(edges, pts[y == 0]) - 1\ninv_w = 1. / pdf[indices]\ninv_w /= inv_w.sum()\nw[y==0] = inv_w\n\npdf, edges = np.histogram(pts[y == 1], density=True, range=[250, 300], bins=50)\nindices = np.searchsorted(edges, pts[y == 1]) - 1\ninv_w = 1. / pdf[indices]\ninv_w /= inv_w.sum()\nw[y==1] = inv_w", "_____no_output_____" ], [ "X_train, X_test, y_train, y_test, w_train, w_test = train_test_split(X, y, w, train_size=0.5)", "_____no_output_____" ], [ "def evaluate_models(X, y, w):\n rocs = []\n fprs = []\n tprs = []\n \n y_pred = X\n\n # Roc\n rocs.append(roc_auc_score(y, y_pred, sample_weight=w))\n fpr, tpr, _ = roc_curve(y, y_pred, sample_weight=w)\n\n fprs.append(fpr)\n tprs.append(tpr)\n \n return rocs, fprs, tprs\n\nr, f, t = evaluate_models(-tau21, y, w)\nplot_rocs(r, f, t, label=\"tau21\")\nreport_score(r, f, t, label=\"tau21\")\n\nr, f, t = evaluate_models(masss, y, w)\nplot_rocs(r, f, t, label=\"mass\")\nreport_score(r, f, t, label=\"mass\")\n\nplot_show()", " tau21\tROC AUC=0.7644+-0.00\t1/FPR@TPR=0.5=6.79+-0.00\n mass\tROC AUC=0.6194+-0.00\t1/FPR@TPR=0.5=2.80+-0.00\n" ], [ "clf = ExtraTreesClassifier(n_estimators=1000, min_samples_leaf=100, max_features=1)\nclf.fit(X_train, y_train)", "_____no_output_____" ], [ "r, f, t = evaluate_models(-clf.predict_proba(X_test)[:, 0], y_test, w_test)\nplot_rocs(r, f, t, label=\"tau21+mass\")\nreport_score(r, f, t, label=\"tau21+mass\")\n\nplot_show()", " tau21+mass\tROC AUC=0.8207+-0.00\t1/FPR@TPR=0.5=11.06+-0.00\n" ] ] ]

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

[ "ADSL" ]

[ "ADSL" ]

[ "ADSL" ]