Datasets:

---

bigcode

/

jupyter-parsed

like 4

Follow

BigCode 1.35k

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "import csv\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nimport json\nimport zipfile\nimport seaborn as sns", "_____no_output_____" ] ], [ [ "**READING IN KINESSO DATA**", "_____no_output_____" ] ], [ [ "imp = pd.read_csv('impressions_one_hour.csv')\nimp = imp[imp['country'] == 'Germany']\nimp = imp[~imp['zip_code'].isna()]\nimp['zip_code'] = imp['zip_code'].astype(str)", "_____no_output_____" ] ], [ [ "**READING IN 2011 GERMANY CENSUS DATA**", "_____no_output_____" ] ], [ [ "# data from: https://www.suche-postleitzahl.org/downloads\nzip_codes = pd.read_csv(\"plz_einwohner.csv\")\n", "_____no_output_____" ], [ "def add_zero(x):\n if len(x) == 4:\n return '0'+ x\n else:\n return x", "_____no_output_____" ] ], [ [ "**CORRECTING FORMATTING ERROR THAT REMOVED INITIAL '0' FROM ZIPCODES**", "_____no_output_____" ] ], [ [ "zip_codes['zipcode'] = zip_codes['zipcode'].astype(str).apply(add_zero)\nzip_codes.head()", "_____no_output_____" ] ], [ [ "Real Population of Germany is 83.02 million", "_____no_output_____" ] ], [ [ "np.sum(zip_codes['population'])", "_____no_output_____" ] ], [ [ "**CALCULATING VALUE COUNTS FROM KINESSO DATA**", "_____no_output_____" ] ], [ [ "val_cou = imp['zip_code'].value_counts()\nval_counts = pd.DataFrame(columns=['zipcode', 'count'], data=val_cou)\nval_counts['zipcode'] = val_cou.index.astype(str)\nval_counts['count'] = val_cou.values.astype(int)\nval_counts", "_____no_output_____" ] ], [ [ "**MERGING TOGETHER KINESSO VALUE COUNTS WITH KINESSO DATA**\n\n\n*ONLY 19 ZIPCODES DO NOT HAVE CENSUS DATA*", "_____no_output_____" ] ], [ [ "population_count = val_counts.merge(right=zip_codes, right_on='zipcode', left_on='zipcode', how='outer')\npopulation_count_f = population_count.dropna()", "_____no_output_____" ], [ "#only 19 zipcodes without data\nlen(population_count[population_count['population'].isna()])", "_____no_output_____" ] ], [ [ "Here count is the observed number from the Kinesso dataset and population is the expected number from census dataset", "_____no_output_____" ] ], [ [ "population_count_f", "_____no_output_____" ] ], [ [ "**CALCULATING DEVICE FREQUENCIES FOR EACH ZIPCODE**", "_____no_output_____" ] ], [ [ "imp['count'] = [1] * len(imp)\ndevice_model_make_counts = imp.groupby(['zip_code', 'device_make', 'device_model'], as_index=False).count()[['zip_code', 'device_make', 'device_model', 'count']]\ntotal_calc = device_model_make_counts.groupby(['zip_code']).sum()\n", "_____no_output_____" ], [ "percent_calc = []\nfor i in device_model_make_counts.index:\n zipc = device_model_make_counts.iloc[i]['zip_code']\n percent_calc = np.append(percent_calc, device_model_make_counts.iloc[i]['count']/total_calc.loc[zipc])", "_____no_output_____" ], [ "device_model_make_counts['device % freq']= percent_calc *100\ndevice_model_make_counts['combined'] = device_model_make_counts['device_make'] + ' ' + device_model_make_counts['device_model']\ndevice_model_make_counts['zip_code'] = device_model_make_counts['zip_code'].astype(str).apply(add_zero)\ndevice_model_make_counts\n ", "_____no_output_____" ] ], [ [ "**CALCULATING PERCENT DIFFERENCE BETWEEN EXPECTED AND OBSERVED POPULATIONS FOR EACH ZIPCODE**", "_____no_output_____" ] ], [ [ "population_count_f['population % expected'] = (population_count_f['population']/sum(population_count_f['population']))*100\npopulation_count_f['population % observed'] = (population_count_f['count']/sum(population_count_f['count']))*100\npopulation_count_f['% difference'] = population_count_f['population % observed'] - population_count_f['population % expected']\npopulation_count_f = population_count_f.rename(columns={'count':'observed population', 'population':'expected population'})\npopulation_count_f", "<ipython-input-46-cfa73adf08fd>:1: 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 population_count_f['population % expected'] = (population_count_f['population']/sum(population_count_f['population']))*100\n<ipython-input-46-cfa73adf08fd>: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 population_count_f['population % observed'] = (population_count_f['count']/sum(population_count_f['count']))*100\n<ipython-input-46-cfa73adf08fd>:3: 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 population_count_f['% difference'] = population_count_f['population % observed'] - population_count_f['population % expected']\n" ] ], [ [ "*MERGING TOGETHER WITH DEVICE FREQUENCY DATA*", "_____no_output_____" ] ], [ [ "combo = device_model_make_counts.merge(right=population_count_f, right_on='zipcode', left_on='zip_code', how='outer').drop(['count', 'device_make', 'device_model'], axis=1)\ncombined_impressions = combo.sort_values('% difference', ascending=False)\n", "_____no_output_____" ], [ "combined_impressions", "_____no_output_____" ] ], [ [ "*GROUPING TO IDENTIFY MOST COMMONLY USED DEVICE*", "_____no_output_____" ] ], [ [ "most_common_device = combined_impressions.groupby(['zip_code']).max()\nmost_common_device", "_____no_output_____" ] ], [ [ "*IDENTIFYING MOST UNDER REPRESENTED ZIP CODES*", "_____no_output_____" ] ], [ [ "underrepresented = most_common_device.sort_values('% difference').head(1000)\nunderrepresented.head(10)", "_____no_output_____" ], [ "underrepresented['combined'].value_counts()", "_____no_output_____" ] ], [ [ "*IDENTIFYING MOST OVER REPRESENTED ZIP CODES*", "_____no_output_____" ] ], [ [ "overrepresented = most_common_device.sort_values('% difference', ascending=False).head(1000)\noverrepresented.head(10)", "_____no_output_____" ], [ "overrepresented['combined'].value_counts()", "_____no_output_____" ] ], [ [ "**I actually decided not to look to closely into the device frequency numbers because for the underrepresented zipcodes there's only like 8-9 people Kinesso advertised to-- and mostly to Apple users interestingly. Instead I did some digging into the top 10 and bottom 10 in a seperate google docs titled: top 10 zipcode investigation** \n\n*quick summary: \n\n**over represented** zip codes belong to large urban cities with more industries and probably higher incomes but idk because I couldn't find zip code specific salary data\n\n**under represented:** zip codes belong to small cities with industries like coal, tourism, and power plants. Also I suspect lower incomes, but idk for sure", "_____no_output_____" ] ], [ [ "sns.distplot(most_common_device['% difference'])", "_____no_output_____" ] ] ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Fetching Twitter data\n\nThis is a simple notebook containing a simple demonstration on how to fetch tweets using a Twitter Sanbox Environment. The sample data is saved in the form of a json file, which must then be preprocessed.", "_____no_output_____" ] ], [ [ "import os\nfrom os.path import join\nfrom searchtweets import load_credentials, gen_rule_payload, ResultStream, collect_results\nimport json\n\nproject_dir = join(os.getcwd(), os.pardir)\nraw_dir = join(project_dir, 'data', 'raw')\ntwitter_creds_path = join(project_dir, 'twitter_creds.yaml')", "_____no_output_____" ], [ "search_args = load_credentials(twitter_creds_path, yaml_key='sample_tweets_api')\n\n# this should probably be moved to the configs.yaml\nquery = \"((cyclone amphan) OR amphan)\"\n\n\n##Cyclone amphan\n#Formed:16 May 2020\n#Dissipated:21 May 2020\n\nfrom_date=\"2020-05-14\"\nto_date=\"2020-06-15\"\n\n# I defined results_per_call as 100 which is default for free users. These can be 500 for paid tiers.\nrule = gen_rule_payload(query, results_per_call=100, from_date=\"2020-05-14\", to_date=\"2020-06-15\")\n\nrs = ResultStream(rule_payload=rule,\n max_results=200,\n **search_args)", "cannot read file /home/jovyan/work/notebooks/../configs.yaml\nError parsing YAML file; searching for valid environment variables\nAccount type is not specified and cannot be inferred.\n Please check your credential file, arguments, or environment variables\n for issues. The account type must be 'premium' or 'enterprise'.\n \n" ], [ "fname = f'SAMPLE_DATA_QUERY_{query}_FROMDATE_{from_date}_TODATE_{to_date}.json'\nwith open(join(raw_dir, fname), 'a', encoding='utf-8') as f:\n for tweet in rs.stream():\n json.dump(tweet, f)\n f.write('\\n')\nprint('done')", "done\n" ] ], [ [ "## Count existing tweets for a given request", "_____no_output_____" ] ], [ [ "search_args = load_credentials(twitter_creds_path, yaml_key='search_tweets_api')\nquery = \"(cyclone amphan)\"\n\ncount_rule = gen_rule_payload(query, from_date=\"2020-05-14\", to_date=\"2020-06-15\", count_bucket=\"day\", results_per_call=500)\n\ncounts = collect_results(count_rule, result_stream_args=search_args)\n\ncounts", "Grabbing bearer token from OAUTH\n" ], [ "tweets = 0\nfor day in counts:\n tweets+=day['count']\n\ntweets", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<div class=\"alert alert-success\">\n <b>Author</b>:\n\n Rashik Rahman\n [email protected]\n\n</div>\n\n# [Click here to see class lecture](https://photos.app.goo.gl/rZaFT2Ct6uETfyww6)", "_____no_output_____" ], [ "Until now we have learned uml/class diagram. Today we'll learn how to draw object diagram from class diagram. In OOP we've learned about object. In code we created object of a class with its initiated values. Here its the same thing. We have a base class like `dog` then to make an object of it we give the values and create an object.\n\n\n\nHere we create an object of `Department` class whose name is mathStat. mathStat's attribute is degree so the degree can be either undergraduate or graduate. 3\n\n\n\n\n\n", "_____no_output_____" ], [ "**USE CASE Diagram**\n\n\n\n\n\n\nHere in the example we can see that customer is actor. But in the exam they won't tell who is the actor, you have to figure it out.\n\n\n\n\n\n\n\n", "_____no_output_____" ], [ "\n\n\n\n\n\n\n`How to find actor and use cases`\n\n\n\n\n\n", "_____no_output_____" ], [ "# That's all for this lecture!", "_____no_output_____" ] ] ]

[ "markdown" ]

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

[ "CC0-1.0" ]

[ "CC0-1.0" ]

[ "CC0-1.0" ]

[ [ [ "https://pyimagesearch.com/2015/03/30/accessing-the-raspberry-pi-camera-with-opencv-and-python/\nwhy : reads image directly as np.array -> to directly image process\nNeeds : to add steps for saving (1 step for the orig capture and one for the processed image)", "_____no_output_____" ] ], [ [ "# import the necessary packages\nfrom picamera.array import PiRGBArray\nfrom picamera import PiCamera\nimport time\nimport cv2\n# initialize the camera and grab a reference to the raw camera capture\ncamera = PiCamera()\nrawCapture = PiRGBArray(camera)\n# allow the camera to warmup\ntime.sleep(0.1)\n# grab an image from the camera\ncamera.capture(rawCapture, format=\"bgr\")\nimage = rawCapture.array\n# display the image on screen and wait for a keypress\ncv2.imshow(\"Image\", image)\ncv2.waitKey(0)", "_____no_output_____" ], [ "#save the raw image, as captured\ncv2.imwrite(\"/path to storage folder/img_name_raw.tiff\", image)", "_____no_output_____" ], [ "#process\n\n\n", "_____no_output_____" ], [ "#save the processed image, ready to show\ncv2.imwrite(\"/path to slideshow folder/img_name.jpg\", img)", "_____no_output_____" ] ] ]

[ "markdown", "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "**This notebook is an exercise in the [Python](https://www.kaggle.com/learn/python) course. You can reference the tutorial at [this link](https://www.kaggle.com/colinmorris/hello-python).**\n\n---\n", "_____no_output_____" ], [ "Welcome to your first set of Python coding problems. If this is your first time using Kaggle Notebooks, welcome! \n\nNotebooks are composed of blocks (called \"cells\") of text and code. Each of these is editable, though you'll mainly be editing the code cells to answer some questions.\n\nTo get started, try running the code cell below (by pressing the ► button, or clicking on the cell and pressing ctrl+enter on your keyboard).", "_____no_output_____" ] ], [ [ "print(\"You've successfully run some Python code\")\nprint(\"Congratulations!\")", "You've successfully run some Python code\nCongratulations!\n" ] ], [ [ "Try adding another line of code in the cell above and re-running it. \n\nNow let's get a little fancier: Add a new code cell by clicking on an existing code cell, hitting the escape key, and then hitting the `a` or `b` key. The `a` key will add a cell above the current cell, and `b` adds a cell below.\n\nGreat! Now you know how to use Notebooks.\n\nEach hands-on exercise starts by setting up our feedback and code checking mechanism. Run the code cell below to do that. Then you'll be ready to move on to question 0.", "_____no_output_____" ] ], [ [ "from learntools.core import binder; binder.bind(globals())\nfrom learntools.python.ex1 import *\nprint(\"Setup complete! You're ready to start question 0.\")", "Setup complete! You're ready to start question 0.\n" ] ], [ [ "# 0.\n\n*This is a silly question intended as an introduction to the format we use for hands-on exercises throughout all Kaggle courses.*\n\n**What is your favorite color? **\n\nTo complete this question, create a variable called `color` in the cell below with an appropriate value. The function call `q0.check()` (which we've already provided in the cell below) will check your answer.", "_____no_output_____" ] ], [ [ "# create a variable called color with an appropriate value on the line below\n# (Remember, strings in Python must be enclosed in 'single' or \"double\" quotes)\ncolor = \"blue\"\n\n# Check your answer\nq0.check()", "_____no_output_____" ] ], [ [ "Didn't get the right answer? How do you not even know your own favorite color?!\n\nDelete the `#` in the line below to make one of the lines run. You can choose between getting a hint or the full answer by choosing which line to remove the `#` from. \n\nRemoving the `#` is called uncommenting, because it changes that line from a \"comment\" which Python doesn't run to code, which Python does run.", "_____no_output_____" ] ], [ [ "# q0.hint()\n# q0.solution()", "_____no_output_____" ] ], [ [ "The upcoming questions work the same way. The only thing that will change are the question numbers. For the next question, you'll call `q1.check()`, `q1.hint()`, `q1.solution()`, for question 2, you'll call `q2.check()`, and so on.", "_____no_output_____" ], [ "<hr/>\n\n# 1.\n\nComplete the code below. In case it's helpful, here is the table of available arithmetic operations:\n\n\n\n| Operator | Name | Description |\n|--------------|----------------|--------------------------------------------------------|\n| ``a + b`` | Addition | Sum of ``a`` and ``b`` |\n| ``a - b`` | Subtraction | Difference of ``a`` and ``b`` |\n| ``a * b`` | Multiplication | Product of ``a`` and ``b`` |\n| ``a / b`` | True division | Quotient of ``a`` and ``b`` |\n| ``a // b`` | Floor division | Quotient of ``a`` and ``b``, removing fractional parts |\n| ``a % b`` | Modulus | Integer remainder after division of ``a`` by ``b`` |\n| ``a ** b`` | Exponentiation | ``a`` raised to the power of ``b`` |\n| ``-a`` | Negation | The negative of ``a`` |\n\n<span style=\"display:none\"></span>\n", "_____no_output_____" ] ], [ [ "pi = 3.14159 # approximate\ndiameter = 3\n\n# Create a variable called 'radius' equal to half the diameter\nradius = diameter/2\n\n# Create a variable called 'area', using the formula for the area of a circle: pi times the radius squared\narea = pi * (radius ** 2)\n\n# Check your answer\nq1.check()", "_____no_output_____" ], [ "# Uncomment and run the lines below if you need help.\n#q1.hint()\n#q1.solution()", "_____no_output_____" ] ], [ [ "<hr/>\n\n# 2.\n\nAdd code to the following cell to swap variables `a` and `b` (so that `a` refers to the object previously referred to by `b` and vice versa).", "_____no_output_____" ] ], [ [ "########### Setup code - don't touch this part ######################\n# If you're curious, these are examples of lists. We'll talk about \n# them in depth a few lessons from now. For now, just know that they're\n# yet another type of Python object, like int or float.\na = [1, 2, 3]\nb = [3, 2, 1]\nq2.store_original_ids()\n######################################################################\n\n# Your code goes here. Swap the values to which a and b refer.\n# If you get stuck, you can always uncomment one or both of the lines in\n# the next cell for a hint, or to peek at the solution.\n\na,b = b,a\n######################################################################\n\n# Check your answer\nq2.check()", "_____no_output_____" ], [ "#q2.hint()", "_____no_output_____" ], [ "#q2.solution()", "_____no_output_____" ] ], [ [ "<hr/>\n\n# 3a.\n\nAdd parentheses to the following expression so that it evaluates to 1.", "_____no_output_____" ] ], [ [ "(5 - 3) // 2", "_____no_output_____" ], [ "#q3.a.hint()", "_____no_output_____" ], [ "# Check your answer (Run this code cell to receive credit!)\nq3.a.solution()", "_____no_output_____" ] ], [ [ "# 3b. <span title=\"A bit spicy\" style=\"color: darkgreen \">🌶️</span>\n\n<small>Questions, like this one, marked a spicy pepper are a bit harder.</small>\n\nAdd parentheses to the following expression so that it evaluates to 0.", "_____no_output_____" ] ], [ [ "(8 - (3 * 2)) - (1 + 1)", "_____no_output_____" ], [ "#q3.b.hint()", "_____no_output_____" ], [ "# Check your answer (Run this code cell to receive credit!)\nq3.b.solution()", "_____no_output_____" ] ], [ [ "<hr/>\n\n# 4. \nAlice, Bob and Carol have agreed to pool their Halloween candy and split it evenly among themselves.\nFor the sake of their friendship, any candies left over will be smashed. For example, if they collectively\nbring home 91 candies, they'll take 30 each and smash 1.\n\nWrite an arithmetic expression below to calculate how many candies they must smash for a given haul.", "_____no_output_____" ] ], [ [ "# Variables representing the number of candies collected by alice, bob, and carol\nalice_candies = 121\nbob_candies = 77\ncarol_candies = 109\n\n# Your code goes here! Replace the right-hand side of this assignment with an expression\n# involving alice_candies, bob_candies, and carol_candies\nto_smash = (alice_candies + bob_candies + carol_candies) % 3\nprint(to_smash)\n\n# Check your answer\nq4.check()", "1\n" ], [ "#q4.hint()\n#q4.solution()", "_____no_output_____" ] ], [ [ "# Keep Going\n\nNext up, you'll **[learn to write new functions and understand functions others write](https://www.kaggle.com/colinmorris/functions-and-getting-help)**. This will make you at least 10 times more productive as a Python programmer. ", "_____no_output_____" ], [ "---\n\n\n\n\n*Have questions or comments? Visit the [Learn Discussion forum](https://www.kaggle.com/learn-forum/161283) to chat with other Learners.*", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "### Assignment\n\n#### Data Description\n- covid data of daily cummulative cases of India as reported from January 2020 to 8th August 2020\n- Source: https://www.kaggle.com/sudalairajkumar/covid19-in-india\n\n#### Conduct below Insight investigation\n1. Find which state has highest mean of cummulative confirmed cases since reported from Jan 2020\n<br>- Plot line graph plotting means of top 10 States having highest daily confirmed cases\n2. Which state has highest Death Rate for the month of June, July & Aug\n<br> - Plot bar graph of Death Rates for the top 5 states\n\n#### Below key steps to be adopted to solve above Questions\n- Load Data --> Clean data / Data munging --> Grouping of Data by State --> Exploration using plots", "_____no_output_____" ], [ "#### Load Packages", "_____no_output_____" ] ], [ [ "import pandas as pd # for cleaning and loading data from csv file\nimport numpy as np\nfrom matplotlib import pyplot as plt # package for plotting graphs\nimport datetime\nimport seaborn as sns; sns.set(color_codes=True)\n%matplotlib inline", "_____no_output_____" ] ], [ [ "#### Load data", "_____no_output_____" ] ], [ [ "df = pd.read_csv(\"covid_19_india.csv\")\ndf.head() # Preview first 5 rows of dataframe", "_____no_output_____" ], [ "# Convert Date column which is a string into datetime object\ndf[\"Date\"] = pd.to_datetime(df[\"Date\"], format = \"%d/%m/%y\")\ndf.head()", "_____no_output_____" ] ], [ [ "#### Cleaning of data\n- The dataset consists of cummulative values, aim is to create columns with daily reported deaths and confirmed cases.\n- Below method is helper function to create column consisting of daily cases reported from Cummulative freq column", "_____no_output_____" ] ], [ [ "ex = np.unique(df['State/UnionTerritory'])\nex", "_____no_output_____" ] ], [ [ "From above unique values of states it is clear that Telangana is represented in multiple ways. We will change each occurrence of Telangna state with standard spelling", "_____no_output_____" ] ], [ [ "def clean_stateName(stateName):\n if stateName == 'Telangana***':\n stateName = 'Telangana'\n elif stateName == 'Telengana':\n stateName = 'Telangana'\n elif stateName == 'Telengana***':\n stateName = 'Telangana'\n \n return stateName", "_____no_output_____" ] ], [ [ "- Apply method is used to apply either user defined or builtin function across every cell of dataframe\n- Commonly lambda function is used to apply method across each cell\n- A lambda function is a small anonymous function.\n- A lambda function can take any number of arguments, but can only have one expression.", "_____no_output_____" ] ], [ [ "df[\"State/UnionTerritory\"] = df[\"State/UnionTerritory\"].apply(lambda x: clean_stateName(x))\nnp.unique(df[\"State/UnionTerritory\"]) # to identify all unique values in a column of dataframe or array", "_____no_output_____" ], [ "def daily_cases(dframe, stateColumn,dateColumn, cummColumn):\n # Sort column containing state and then by date in ascending order\n dframe.sort_values(by = [stateColumn, dateColumn], inplace = True)\n newColName = 'daily_' + cummColumn\n dframe[newColName] = dframe[cummColumn].diff() # diff is pandas method to caclucate difference between consecutive values\n# print(dframe.tail())\n '''\n Below line uses shift method of pandas to compare consecutive state names and if they are not different\n as shown by using ! symbol then create list of boolean, True for if they are different else False\n ''' \n mask = dframe[stateColumn] != dframe[stateColumn].shift(1)\n dframe[newColName][mask] = np.nan # where value of mask =True the cell value will be replaced by NaN\n dframe[newColName] = dframe[newColName].apply(lambda x: 0 if x < 0 else x) # replace negative values by 0\n# dframe.drop('diffs',axis=1, inplace = True)\n \n return dframe\n ", "_____no_output_____" ], [ "df_new = daily_cases(dframe= df, stateColumn= 'State/UnionTerritory',dateColumn= 'Date', cummColumn= 'Confirmed')", "<ipython-input-7-4efcf8e0fcc5>:12: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame\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 dframe[newColName][mask] = np.nan # where value of mask =True the cell value will be replaced by NaN\n" ], [ "df_new[df_new[\"State/UnionTerritory\"]==\"Maharashtra\"].tail(n=5)", "_____no_output_____" ] ], [ [ "#### Q1. Find which state has highest mean of cummulative confirmed cases since reported from Jan 2020", "_____no_output_____" ] ], [ [ "# Hint : Groupby state names to find their means for confirmed cases\n\ndf_group = df_new.groupby([\"State/UnionTerritory\"])['daily_Confirmed'].mean()", "_____no_output_____" ], [ "df_group = df_group.sort_values(ascending= False)[0:10]\ndf_group", "_____no_output_____" ], [ "df_group.index", "_____no_output_____" ], [ "ax = sns.lineplot(x=df_group.index, y= df_group.values)\nplt.scatter(x=df_group.index, y= df_group.values, c = 'r')\nax.figure.set_figwidth(12)\nax.figure.set_figheight(4)\nax.set_ylabel(\"Mean of Daily Confirmed Cases\")", "_____no_output_____" ] ], [ [ "#### Q2. Which state has highest Death Rate for the month of June, July & Aug", "_____no_output_____" ] ], [ [ "# Hint - explore how a datetime column of dataframe can be filtered using specific months\ndf_months = df_new['Date'].apply(lambda x: x.month in [6,7,8]) # this will create boolean basis comparison of months", "_____no_output_____" ], [ "df_final = df_new[df_months] # Filtered dataframe consisting of data from June, July & Aug", "_____no_output_____" ], [ "df_final.tail()", "_____no_output_____" ], [ "df_final['death_rate'] = df_final['Deaths'] / df_final['Confirmed'] *100", "<ipython-input-17-199836f307be>:1: 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 df_final['death_rate'] = df_final['Deaths'] / df_final['Confirmed'] *100\n" ], [ "df_final.tail()", "_____no_output_____" ], [ "df_groups_deaths = df_final.groupby([\"State/UnionTerritory\"])['death_rate'].mean()", "_____no_output_____" ], [ "top_10_deathrates = df_groups_deaths.sort_values(ascending= False)[0:10]", "_____no_output_____" ], [ "fig, ax = plt.subplots()\nfig.set_figwidth(15)\nfig.set_figheight(6)\nax.bar(x = top_10_deathrates.index, height = top_10_deathrates.values)\nax.set_xlabel('States')\nax.set_ylabel('Death Rates %')\nax.set_title('Top 10 States with Highest Death Rate since June 2020')\nfor i, v in enumerate(top_10_deathrates.values):\n ax.text(i, v, s = (\"%.2f\" % v), color='blue', fontweight='bold', fontsize = 12) # %.2f will print decimals upto 2 places\nplt.xticks(rotation=45) # this line will rotate the x axis label in 45 degrees to make it more readable", "_____no_output_____" ] ], [ [ "### Q3. Explore Trend in Confirmed Cases for the state of Maharashtra\n- Plot line graph with x axis as Date column and y axis as daily confirmed cases. - such a graph is also called\nas Time Series Plot", "_____no_output_____" ], [ "#### Hint - explore on google or in matplotlib for Time series graph from a dataframe", "_____no_output_____" ] ], [ [ "df_mah = df_new[df_new[\"State/UnionTerritory\"]=='Maharashtra']", "_____no_output_____" ], [ "fig, ax = plt.subplots()\nfig.set_figwidth(15)\nfig.set_figheight(6)\nax.plot(df_mah[\"Date\"],df_mah[\"daily_Confirmed\"])", "_____no_output_____" ], [ "df_mah = df_final[df_final[\"State/UnionTerritory\"]=='Maharashtra']\nfig, ax = plt.subplots()\nfig.set_figwidth(15)\nfig.set_figheight(6)\nax.plot(df_mah[\"Date\"],df_mah[\"death_rate\"])\nax.scatter(df_mah[\"Date\"],df_mah[\"death_rate\"])\nax.set_xlabel('Date')\nax.set_ylabel('Death Rate')\nax.set_title('Death Rate in Maharastra')", "_____no_output_____" ] ] ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/anasir514/colab/blob/main/solar-learn.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ] ], [ [ "print((4 + 8) / 2)", "6.0\n" ] ] ]

[ "markdown", "code" ]

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

[ "BSD-2-Clause" ]

[ "BSD-2-Clause" ]

[ "BSD-2-Clause" ]

[ [ [ "# Check values before feature selection in both training and test data\n\n- nan\n- different enough values", "_____no_output_____" ] ], [ [ "import pandas as pd\nimport glob\nimport os\n\ntraining_df = pd.concat(map(pd.read_csv, glob.glob(os.path.join('../Data/Train', \"*.csv\"))), ignore_index=True)\ntest_df = test_data = pd.read_csv('../data/test.csv')", "_____no_output_____" ], [ "training_df_shape = training_df.shape\ntest_df_shape = test_df.shape\nall_stations = set(training_df['station'])\n\ndef nan_analysis(column_name):\n training_with_null_df = training_df[training_df[column_name].isnull()]\n training_nan = training_with_null_df.shape\n print(f'Number of Nan for {column_name}: {training_nan} of {training_df_shape}')\n test_nan = test_df[test_df[column_name].isnull()].shape\n print(f'Number of Nan for {column_name}: {test_nan} of {test_df_shape}')\n return training_with_null_df[['station']]\n\n\ndef value_analysis(column_name):\n return pd.merge(training_df[[column_name]].describe(),\n test_df[[column_name]].describe(),\n left_index=True,\n right_index=True,\n suffixes=('training', 'test'))\n\ndef station_ids_for_non_nan(column_name):\n training_not_null = training_df[training_df[column_name].notnull()]\n training_not_null_stations = set(training_not_null['station'])\n print(f'Not nan for {column_name}: {training_not_null.shape} of {training_df_shape}')\n print(f'Station with only null values: {all_stations - training_not_null_stations}')\n", "_____no_output_____" ] ], [ [ "# Weather Data", "_____no_output_____" ] ], [ [ "precipitation = 'precipitation.l.m2'\nprecipitation_nan = nan_analysis(precipitation)\nvalue_analysis(precipitation)\n\n# -> Training data has no values for precipitation not a good feature", "Number of Nan for precipitation.l.m2: (75, 25) of (55875, 25)\nNumber of Nan for precipitation.l.m2: (0, 25) of (2250, 25)\n" ], [ "column = 'temperature.C'\ntemperature_nan = nan_analysis(column)\nvalue_analysis(column)\n\n# min temperature is quite different between training and test but there seems to be enough data", "Number of Nan for temperature.C: (75, 25) of (55875, 25)\nNumber of Nan for temperature.C: (0, 25) of (2250, 25)\n" ], [ "column = 'windMaxSpeed.m.s'\nwindmax_nan = nan_analysis(column)\nvalue_analysis(column)", "Number of Nan for windMaxSpeed.m.s: (75, 25) of (55875, 25)\nNumber of Nan for windMaxSpeed.m.s: (0, 25) of (2250, 25)\n" ], [ "column = 'windMeanSpeed.m.s'\nwindmean_nan = nan_analysis(column)\nvalue_analysis(column)", "Number of Nan for windMeanSpeed.m.s: (75, 25) of (55875, 25)\nNumber of Nan for windMeanSpeed.m.s: (0, 25) of (2250, 25)\n" ], [ "column = 'windDirection.grades'\nwinddir_nan = nan_analysis(column)\nvalue_analysis(column)", "Number of Nan for windDirection.grades: (375, 25) of (55875, 25)\nNumber of Nan for windDirection.grades: (0, 25) of (2250, 25)\n" ], [ "column = 'relHumidity.HR'\nrelhum_nan = nan_analysis(column)\nvalue_analysis(column)", "Number of Nan for relHumidity.HR: (75, 25) of (55875, 25)\nNumber of Nan for relHumidity.HR: (0, 25) of (2250, 25)\n" ], [ "column = 'airPressure.mb'\nairpressure_nan = nan_analysis(column)\nvalue_analysis(column)", "Number of Nan for airPressure.mb: (75, 25) of (55875, 25)\nNumber of Nan for airPressure.mb: (0, 25) of (2250, 25)\n" ], [ "# all weather measure are missing 75\ndiff = set(airpressure_nan.index) - set(relhum_nan.index)\n", "_____no_output_____" ], [ "diff = set(winddir_nan.index) - set(relhum_nan.index)", "_____no_output_____" ] ], [ [ "# Is Holiday\n", "_____no_output_____" ] ], [ [ "column = 'isHoliday'\nnan_analysis(column)\nvalue_analysis(column)", "Number of Nan for isHoliday: (0, 25) of (55875, 25)\nNumber of Nan for isHoliday: (0, 25) of (2250, 25)\n" ] ], [ [ "# Bikes Profile Data", "_____no_output_____" ] ], [ [ "column = 'full_profile_3h_diff_bikes'\nnan_analysis(column)\nstation_ids_for_non_nan(column)\nvalue_analysis(column)\n# each station has none null values!", "Number of Nan for full_profile_3h_diff_bikes: (12825, 25) of (55875, 25)\nNumber of Nan for full_profile_3h_diff_bikes: (0, 25) of (2250, 25)\nNot nan for full_profile_3h_diff_bikes: (43050, 25) of (55875, 25)\nStation with only null values: set()\n" ], [ "column = 'full_profile_bikes'\nnan_analysis(column)\nstation_ids_for_non_nan(column)\nvalue_analysis(column)", "Number of Nan for full_profile_bikes: (12600, 25) of (55875, 25)\nNumber of Nan for full_profile_bikes: (0, 25) of (2250, 25)\nNot nan for full_profile_bikes: (43275, 25) of (55875, 25)\nStation with only null values: set()\n" ], [ "#select the none nan", "_____no_output_____" ], [ "column = 'short_profile_3h_diff_bikes'\nnan_analysis(column)\nstation_ids_for_non_nan(column)\nvalue_analysis(column)", "Number of Nan for short_profile_3h_diff_bikes: (12825, 25) of (55875, 25)\nNumber of Nan for short_profile_3h_diff_bikes: (0, 25) of (2250, 25)\nNot nan for short_profile_3h_diff_bikes: (43050, 25) of (55875, 25)\nStation with only null values: set()\n" ], [ "column = 'short_profile_bikes'\nnan_analysis(column)\nstation_ids_for_non_nan(column)\nvalue_analysis(column)", "Number of Nan for short_profile_bikes: (12600, 25) of (55875, 25)\nNumber of Nan for short_profile_bikes: (0, 25) of (2250, 25)\nNot nan for short_profile_bikes: (43275, 25) of (55875, 25)\nStation with only null values: set()\n" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "!pip install eli5", "Collecting eli5\n\u001b[?25l Downloading https://files.pythonhosted.org/packages/97/2f/c85c7d8f8548e460829971785347e14e45fa5c6617da374711dec8cb38cc/eli5-0.10.1-py2.py3-none-any.whl (105kB)\n\r\u001b[K |███ | 10kB 14.0MB/s eta 0:00:01\r\u001b[K |██████▏ | 20kB 1.8MB/s eta 0:00:01\r\u001b[K |█████████▎ | 30kB 2.7MB/s eta 0:00:01\r\u001b[K |████████████▍ | 40kB 1.8MB/s eta 0:00:01\r\u001b[K |███████████████▌ | 51kB 2.2MB/s eta 0:00:01\r\u001b[K |██████████████████▋ | 61kB 2.6MB/s eta 0:00:01\r\u001b[K |█████████████████████▊ | 71kB 3.0MB/s eta 0:00:01\r\u001b[K |████████████████████████▊ | 81kB 2.3MB/s eta 0:00:01\r\u001b[K |███████████████████████████▉ | 92kB 2.6MB/s eta 0:00:01\r\u001b[K |███████████████████████████████ | 102kB 2.9MB/s eta 0:00:01\r\u001b[K |████████████████████████████████| 112kB 2.9MB/s \n\u001b[?25hRequirement already satisfied: graphviz in /usr/local/lib/python3.6/dist-packages (from eli5) (0.10.1)\nRequirement already satisfied: scikit-learn>=0.18 in /usr/local/lib/python3.6/dist-packages (from eli5) (0.22.1)\nRequirement already satisfied: tabulate>=0.7.7 in /usr/local/lib/python3.6/dist-packages (from eli5) (0.8.6)\nRequirement already satisfied: numpy>=1.9.0 in /usr/local/lib/python3.6/dist-packages (from eli5) (1.17.5)\nRequirement already satisfied: attrs>16.0.0 in /usr/local/lib/python3.6/dist-packages (from eli5) (19.3.0)\nRequirement already satisfied: six in /usr/local/lib/python3.6/dist-packages (from eli5) (1.12.0)\nRequirement already satisfied: scipy in /usr/local/lib/python3.6/dist-packages (from eli5) (1.4.1)\nRequirement already satisfied: jinja2 in /usr/local/lib/python3.6/dist-packages (from eli5) (2.11.1)\nRequirement already satisfied: joblib>=0.11 in /usr/local/lib/python3.6/dist-packages (from scikit-learn>=0.18->eli5) (0.14.1)\nRequirement already satisfied: MarkupSafe>=0.23 in /usr/local/lib/python3.6/dist-packages (from jinja2->eli5) (1.1.1)\nInstalling collected packages: eli5\nSuccessfully installed eli5-0.10.1\n" ], [ "import pandas as pd\nimport numpy as np\n\nfrom sklearn.tree import DecisionTreeRegressor\nfrom sklearn.model_selection import cross_val_score\nfrom sklearn.metrics import mean_absolute_error\n\nimport eli5\nfrom eli5.sklearn import PermutationImportance\nfrom sklearn.ensemble import RandomForestRegressor\n\nfrom ast import literal_eval\nfrom tqdm import tqdm_notebook\n\n", "_____no_output_____" ], [ "cd \"/content/drive/My Drive/Colab Notebooks/dw_matrix\"", "/content/drive/My Drive/Colab Notebooks/dw_matrix\n" ], [ "ls data", "men_shoes.csv\n" ], [ "df = pd.read_csv('data/men_shoes.csv', low_memory=False)", "_____no_output_____" ], [ "def run_model (feats, model = DecisionTreeRegressor(max_depth=5)):\n X= df[feats].values\n y=df['prices_amountmin'].values\n\n scores=cross_val_score(model, X, y, scoring='neg_mean_absolute_error')\n return np.mean(scores),np.std(scores)", "_____no_output_____" ], [ "df['brand_cat']=df['brand'].map(lambda x: str(x).lower()).factorize()[0]\nrun_model(['brand_cat'])", "_____no_output_____" ], [ "model = RandomForestRegressor(max_depth=5,n_estimators=100,random_state=0)\nrun_model(['brand_cat'], model)", "_____no_output_____" ], [ "df.features.head().values", "_____no_output_____" ], [ "str_dict = '[{\"key\":\"Gender\",\"value\":[\"Men\"]},{\"key\":\"Shoe Size\",\"value\":[\"M\"]},{\"key\":\"Shoe Category\",\"value\":[\"Men\\'s Shoes\"]},{\"key\":\"Color\",\"value\":[\"Multicolor\"]},{\"key\":\"Manufacturer Part Number\",\"value\":[\"8190-W-NAVY-7.5\"]},{\"key\":\"Brand\",\"value\":[\"Josmo\"]}]'\nliteral_eval(str_dict)", "_____no_output_____" ], [ "def parse_features(x):\n output_dict ={}\n if str(x)=='nan':return output_dict\n features = literal_eval(x.replace('\\\\\"','\"'))\n for item in features:\n key = item['key'].lower().strip()\n value = item['value'][0].lower().strip()\n output_dict[key]=value\n return output_dict\n\ndf['features_parsed'] = df['features'].map(parse_features)", "_____no_output_____" ], [ "keys = set()\ndf['features_parsed'].map(lambda x: keys.update(x.keys()))\nlen(keys)", "_____no_output_____" ], [ "[{'key': 'Gender', 'value': ['Men']}, {'key': 'Shoe Size', 'value': ['M']}, {'key': 'Shoe Category', 'value': [\"Men's Shoes\"]}, {'key': 'Color', 'value': ['Multicolor']}, {'key': 'Manufacturer Part Number', 'value': ['8190-W-NAVY-7.5']}, {'key': 'Brand', 'value': ['Josmo']}]", "_____no_output_____" ], [ "df.features_parsed.head().values", "_____no_output_____" ], [ "def get_name_feat(key):\n return 'feat_'+key\n\nfor key in tqdm_notebook(keys):\n df[get_name_feat(key)] = df.features_parsed.map(lambda feats: feats[key] if key in feats else np.nan)", "_____no_output_____" ], [ "df.columns", "_____no_output_____" ], [ "df[False == df['feat_athlete'].isnull()].shape[0]/df.shape[0]*100", "_____no_output_____" ], [ "df.shape[0]", "_____no_output_____" ], [ "keys_stat = {}\nfor key in keys:\n keys_stat[key] = df[False == df[get_name_feat(key)].isnull()].shape[0]/df.shape[0]*100", "_____no_output_____" ], [ "keys_stat", "_____no_output_____" ], [ "{k:v for k,v in keys_stat.items() if v > 30}", "_____no_output_____" ], [ "df['feat_brand_cat']=df['feat_brand'].factorize()[0]\ndf['feat_color_cat']=df['feat_color'].factorize()[0]\ndf['feat_gender_cat']=df['feat_gender'].factorize()[0]\ndf['feat_material_cat']=df['feat_material'].factorize()[0]\ndf['feat_manufacturer part number_cat']=df['feat_manufacturer part number'].factorize()[0]\n\nfor key in keys:\n df[get_name_feat(key)+'_cat']= df[get_name_feat(key)].factorize()[0]", "_____no_output_____" ], [ "df ['brand']=df['brand'].map(lambda x: str(x).lower())\ndf [ df.brand != df.feat_brand][['brand','feat_brand']].head()", "_____no_output_____" ], [ "feats = ['brand_cat','feat_material_cat','feat_adjustable_cat','feat_brand_cat','feat_movement_cat','feat_fabric content_cat']\n#feats +=[ ]\n\n\nmodel = RandomForestRegressor (max_depth=5, n_estimators=100)\nresult = run_model(feats,model)\nresult", "_____no_output_____" ], [ "feats_cat = [x for x in df.columns if '_cat' in x]\nfeats_cat", "_____no_output_____" ], [ "X=df[feats].values\ny=df['prices_amountmin'].values\n\nm= RandomForestRegressor(max_depth=5, n_estimators=100, random_state=0)\nm.fit(X,y)\nprint (result)\nperm = PermutationImportance(m,random_state=1).fit(X,y);\neli5.show_weights(perm, feature_names=feats)", "(-57.15104696094037, 4.179350915464255)\n" ], [ "df['brand'].value_counts()", "_____no_output_____" ], [ "ls matrix_one", "day3.ipynb day5.ipynb\n" ], [ "!git add matrix_one/day5.ipynb", "_____no_output_____" ] ] ]

[ "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# 17. Module, Package, Try_except, Numpy1_20191011_014_Day4_2부", "_____no_output_____" ], [ "## Magic method 정리\n\n- 클래스 생성 후, 클래스 object의 기본 연산기능 보강할 때 활용 가능 주목!\n- object를 더할 때, plus(1,2) 함수 쓰지 않고, num1 + num2 로도 연산이 가능!\n\n- 비교\n - \\__eq__(==), \\__ne__(!=)\n - \\__lt__(<, less than), \\__gt__(>, greater than), \\__le__(<=, less or equal), \\__ge__(>=, gre or equal)\n- 연산\n - \\__add__(+), \\__sub__(-), \\__mul__(*), \\__truediv__(/)\n - \\__floordiv__(//), \\__mod__(%), \\__pow__(**)\n- 그외\n - \\__repr__(object의 그냥 represent), \\__str__(object의 print) **----> return str( ) -----> string 데이터로 리턴해줘야 함**", "_____no_output_____" ] ], [ [ "# Magic method 사용한 클래스 정의 및 object 연산\n# 예) integer 클래스 생성", "_____no_output_____" ] ], [ [ "class Integer:\n \n def __init__(self,number):\n self.num = number\n \n# def __add__(self,unit):\n# return self.num + unit.num\n\n# def __str__(self):\n# return str(self.num)\n\n# def __repr__(self):\n# return str(self.num)\n\nnum1 = Integer(1)\nnum2 = Integer(2)\n\nnum1+num2\n\n# 그냥 num1 + num2 하면,, 각 변수에는 1과 2가 들어가있으니, 당연히 + 연산 되어야 하는 거 아닌가 하겠지만,,\n# num1과 num2의 클래스(사용자정의 데이터타입), 데이터타입이 Integer라는 내가 정의 내린 타입이기 때문에,, __add__ 가 따로 없다.\n# 따라서, 저렇게 기본 연산자 활용하려면 magic method를 다시 재정의 해줘야 한다.", "_____no_output_____" ] ], [ [ "a = 1\na.__add__(2) # ====> a.num + 2.num ==== self.num + unit.num ===== def __add__(self, unit):", "_____no_output_____" ], [ "num1", "_____no_output_____" ], [ "print(num1)", "<__main__.Integer object at 0x104eb7c90>\n" ] ], [ [ "# 1. 클래스 예제\n\n- 계좌 클래스 만들기\n- 변수 : 자산(asset), 이자율(interest)\n- 함수 : 인출(draw), 입금(interest), 이자추가(add_interest)\n- 인출 시, 자산 이상의 돈을 인출할 수 없다.", "_____no_output_____" ] ], [ [ "class Account:\n def __init__(self,asset,interest=1.05):\n self.asset = asset\n self.interest = interest\n \n def draw(self,amount):\n if self.asset >= amount:\n self.asset -= amount\n print(\"{}원이 인출되었습니다.\".format(amount))\n \n else:\n print('{}원이 부족합니다.'.format((amount-self.asset)))\n \n def insert(self,amount):\n self.asset += amount\n print('{}원이 입금되었습니다.'.format(amount))\n \n def add_interest(self):\n self.asset *= self.interest\n print('{}원의 이자가 입금되었습니다.'.format((self.asset*(self.interest-1))))\n \n def __repr__(self):\n return \"asset : {}, interest : {}\".format(self.asset, self.interest)\n \nacc1 = Account(10000)\nacc1.asset", "_____no_output_____" ], [ "acc1", "_____no_output_____" ], [ "acc1.draw(12000)", "2000원이 부족합니다.\n" ], [ "acc1.draw(3000)", "3000원이 인출되었습니다.\n" ], [ "acc1", "_____no_output_____" ], [ "acc1.insert(5000)\nacc1", "5000원이 입금되었습니다.\n" ], [ "acc1.add_interest(),1\nacc1", "630.0000000000006원의 이자가 입금되었습니다.\n" ] ], [ [ "# Module package\n\n* 변수, 함수 < 클래스 < 모듈 < 패키지\n\n- 모듈 : 변수와 함수, 클래스를 모아놓은 ( .py ) 확장자를 가진 파일 ( 클래스 보다 조금 더 큰 범위 )\n- 패키지 : 모듈보다 한 단계 큰 기능. 모듈의 기능을 디렉토리 별로 정리해놓은 개념", "_____no_output_____" ], [ "1. 모듈 생성\n2. 모듈 호출", "_____no_output_____" ], [ "# 1. 모듈 생성(파일 생성)", "_____no_output_____" ] ], [ [ "!ls", "0401_Lecture_python.ipynb 0406_Lecture_python.ipynb dss.py\r\n0402_Lecture_python.ipynb 0407_Lecture_python.ipynb \u001b[34mschool\u001b[m\u001b[m\r\n0403_Lecture_python.ipynb 0408_Lecture_python.ipynb\r\n" ], [ "%%writefile dss.py\n\n# 모듈 파일 생성 (매직 커맨드 사용)\n\n# 1) %% -> 이 셀에 있는 내용에 전부다 writefile 을 적용하겠다.\n# 2) dss.py 라는 파일을 만들어서, 써있는 코드들을 이 파일에 저장하겠다.\n\n# 모듈 생성 -> 파일 저장\n# 1. 모듈 생성 (모듈 = 클래스, 함수, 변수의 set)\n\nnum = 1234\n\ndef disp1(msg):\n print(\"disp1\", msg)\n \ndef disp2(msg):\n print('disp2', msg)\n \nclass Calc:\n def plus(self, *args):\n return sum(args)\n", "Overwriting dss.py\n" ], [ "!ls", "0401_Lecture_python.ipynb 0406_Lecture_python.ipynb dss.py\r\n0402_Lecture_python.ipynb 0407_Lecture_python.ipynb \u001b[34mschool\u001b[m\u001b[m\r\n0403_Lecture_python.ipynb 0408_Lecture_python.ipynb\r\n" ], [ "%reset", "Once deleted, variables cannot be recovered. Proceed (y/[n])? \nNothing done.\n" ], [ "%whos", "Variable Type Data/Info\n------------------------------\ndss module <module 'school.dss.data1<...>업/school/dss/data1.py'>\nschool module <module 'school' (namespace)>\nurl module <module 'school.web.url' <...>수업/school/web/url.py'>\n" ] ], [ [ "# 2. 모듈 호출", "_____no_output_____" ] ], [ [ "# 모듈 호출 : import ( .py 제외한 파일명 )\n\nimport dss\n%whos", "Variable Type Data/Info\n--------------------------------\nCalc type <class 'dss.Calc'>\ncalc Calc <dss.Calc object at 0x109abb690>\ndisp1 function <function disp1 at 0x109a88ef0>\ndisp2 function <function disp2 at 0x109ab75f0>\ndss module <module 'dss' from '/User<...>ᆼ/0. 스쿨 수업/dss.py'>\nnum int 1234\nschool module <module 'school' (namespace)>\nurl module <module 'school.web.url' <...>수업/school/web/url.py'>\n" ], [ "dss.num", "_____no_output_____" ], [ "dss.disp1('안녕')", "disp1 안녕\n" ], [ "calc = dss.Calc()", "_____no_output_____" ], [ "calc.plus(1,2,3,4,5,6)", "_____no_output_____" ] ], [ [ "# 3. 모듈 내 특정 변수, 함수 호출", "_____no_output_____" ] ], [ [ "# import random --> random 모듈을 불러온 것 (random.py 라는 파일의 코드(모듈 적어놓은) 가져온 것)\n# random.randint(1,5) --> random 모듈 내 randint라는 함수를 가져온 것.\n# calc.plus --> dss 라는 모듈의 plus라는 함수 가져온 것.", "_____no_output_____" ], [ "# 모듈 안에 특정 함수, 변수, 클래스 호출\n# '모듈.변수' 로 적지 않고, '모듈' 로 바로 호출 가능\n\nfrom dss import num, disp2", "_____no_output_____" ], [ "%whos", "Variable Type Data/Info\n--------------------------------\nCalc type <class 'dss.Calc'>\ncalc Calc <dss.Calc object at 0x109baed10>\ndisp1 function <function disp1 at 0x109a88ef0>\ndisp2 function <function disp2 at 0x109ab75f0>\ndss module <module 'dss' from '/User<...>ᆼ/0. 스쿨 수업/dss.py'>\nnum int 1234\nschool module <module 'school' (namespace)>\nurl module <module 'school.web.url' <...>수업/school/web/url.py'>\n" ], [ "dss.num", "_____no_output_____" ], [ "num", "_____no_output_____" ] ], [ [ "# 4. 모듈 내 모든 변수, 함수 호출", "_____no_output_____" ] ], [ [ "%reset", "Once deleted, variables cannot be recovered. Proceed (y/[n])? \nNothing done.\n" ], [ "from dss import *\n\n%whos", "Variable Type Data/Info\n--------------------------------\nCalc type <class 'dss.Calc'>\ncalc Calc <dss.Calc object at 0x109baed10>\ndisp1 function <function disp1 at 0x109a88ef0>\ndisp2 function <function disp2 at 0x109ab75f0>\ndss module <module 'dss' from '/User<...>ᆼ/0. 스쿨 수업/dss.py'>\nnum int 1234\nschool module <module 'school' (namespace)>\nurl module <module 'school.web.url' <...>수업/school/web/url.py'>\n" ] ], [ [ "# 5. 패키지\n\n- 패키지 생성\n- 패키지 호출\n- setup.py 패키지 설치 파일 만들기 \n\n- 패키지(디렉토리) : 모듈(파일)", "_____no_output_____" ], [ "## 1) 패키지 ( 디렉토리 (dss / web) ) 생성", "_____no_output_____" ] ], [ [ "# !mkdir p- ---> school 밑에 dss 디렉토리 생성\n!mkdir -p school/dss", "_____no_output_____" ], [ "# !mkdir p- ---> school 밑에 web 디렉토리 생성\n!mkdir -p school/web", "_____no_output_____" ], [ "!tree school", "\u001b[01;34mschool\u001b[00m\r\n├── \u001b[01;34mdss\u001b[00m\r\n│   ├── __init__.py\r\n│   ├── data1.py\r\n│   └── data2.py\r\n└── \u001b[01;34mweb\u001b[00m\r\n ├── __init__.py\r\n └── url.py\r\n\r\n2 directories, 5 files\r\n" ] ], [ [ "### tree 설치\n- homebrew 설치\n - homebrew : https://brew.sh/index_ko\n - homebrew : osx 패키지 관리 설치 툴\n - /bin/bash -c \"$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install.sh)\"\n - brew install tree", "_____no_output_____" ], [ "## 2) 모듈(파일) 생성", "_____no_output_____" ] ], [ [ "# 이 단계는 파이썬 3.8버젼 이후 부터는 안해도 됨\n# !touch --> 파일 생성\n!touch school/dss/__init__.py\n!touch school/web/__init__.py", "_____no_output_____" ], [ "!tree school", "\u001b[01;34mschool\u001b[00m\r\n├── \u001b[01;34mdss\u001b[00m\r\n│   ├── __init__.py\r\n│   ├── data1.py\r\n│   └── data2.py\r\n└── \u001b[01;34mweb\u001b[00m\r\n ├── __init__.py\r\n └── url.py\r\n\r\n2 directories, 5 files\r\n" ], [ "%%writefile school/dss/data1.py\n# dss라는 패키지 안에 모듈(파일)을 추가\n# web이라는 디렉토리 안에 모듈(파일)을 추가\n\ndef plus(*args):\n print('data1')\n return sum(args)", "Overwriting school/dss/data1.py\n" ], [ "%%writefile school/dss/data2.py\n\ndef plus2(*args):\n print('data2')\n return sum(args)", "Overwriting school/dss/data2.py\n" ], [ "%%writefile school/web/url.py\n\ndef make(url):\n return url if url[:7] == 'http://' else 'http://'+url", "Overwriting school/web/url.py\n" ], [ "!tree school", "\u001b[01;34mschool\u001b[00m\r\n├── \u001b[01;34mdss\u001b[00m\r\n│   ├── __init__.py\r\n│   ├── data1.py\r\n│   └── data2.py\r\n└── \u001b[01;34mweb\u001b[00m\r\n ├── __init__.py\r\n └── url.py\r\n\r\n2 directories, 5 files\r\n" ] ], [ [ "## 3) 패키지 경로 안에 있는 모듈을 찾아들어가 사용", "_____no_output_____" ] ], [ [ "import school.dss.data1", "_____no_output_____" ], [ "%whos", "Variable Type Data/Info\n------------------------------\nschool module <module 'school' (namespace)>\n" ], [ "# school 디렉토리 - dss 디렉토리 - data1 모듈 - plus 함수 호출\nschool.dss.data1.plus(1,2,3)", "data1\n" ], [ "# 모듈 호출 명령어 너무 길다 import school.dss.data1\n# alias 로 단축명 생성\n\nimport school.dss.data1 as dss", "_____no_output_____" ], [ "dss.plus(1,2,3)", "data1\n" ], [ "# school web : 디렉토리\n# url : 모듈\nfrom school.web import url", "_____no_output_____" ], [ "url.make('google.com')", "_____no_output_____" ], [ "# 패키지의 위치 : 특정 디렉토리에 있는 패키지는 어디에서나 import 가능", "_____no_output_____" ], [ "import random", "_____no_output_____" ], [ "import sys\n\nfor path in sys.path:\n print(path)", "/Users/kimjeongseob/Desktop/0. 데이터사이언스스쿨/2. 프로그래밍/0. 스쿨 수업\n/Users/kimjeongseob/opt/anaconda3/lib/python37.zip\n/Users/kimjeongseob/opt/anaconda3/lib/python3.7\n/Users/kimjeongseob/opt/anaconda3/lib/python3.7/lib-dynload\n\n/Users/kimjeongseob/opt/anaconda3/lib/python3.7/site-packages\n/Users/kimjeongseob/opt/anaconda3/lib/python3.7/site-packages/aeosa\n/Users/kimjeongseob/opt/anaconda3/lib/python3.7/site-packages/IPython/extensions\n/Users/kimjeongseob/.ipython\n" ], [ "# !ls /Users/kimjeongseob/opt/anaconda3/lib/python3.7\n\n# 아래의 출력 결과를 변수에다 넣을 수 있음\n\nA = !ls /Users/kimjeongseob/opt/anaconda3/lib/python3.7\nlen(A), A[-5:]", "_____no_output_____" ], [ "# setup.py 를 작성해서 패키지를 설치해서 사용\n# setuptools 를 이용", "_____no_output_____" ] ] ]

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

[ [ "markdown", "markdown" ], [ "code" ], [ "raw" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code" ], [ "markdown", "markdown", "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown", "markdown" ], [ "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "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/BrunaKuntz/Python-Curso-em-Video/blob/main/Mundo01/Desafio030.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ], [ "\n\n\n# **Desafio 030**\n**Python 3 - 1º Mundo**\n\nDescrição: Crie um programa que leia um número inteiro e mostre na tela se ele é PAR ou ÍMPAR.\n\nLink: https://www.youtube.com/watch?v=4vFCzKuHOn4&t=4s", "_____no_output_____" ] ], [ [ "num = int(input('Digite um número: '))\nif num % 2 == 0:\n print('O número é par.')\nelse:\n print('O número é ímpar.')", "_____no_output_____" ] ] ]

[ "markdown", "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Example 5 - Open-loop simulation", "_____no_output_____" ], [ "An open-loop simulation is the case where no state-feedback control is used. It means that only time-dependent control is used or not control at all. This kind of simulation is mainly useful for stability analysis and for cheching the trimmed behaviior (including perturbations around the trimmed conditions).", "_____no_output_____" ], [ "### Import atmosphere model", "_____no_output_____" ] ], [ [ "from pyaat.atmosphere import atmosCOESA\natm = atmosCOESA()", "_____no_output_____" ] ], [ [ "### Import gravity model", "_____no_output_____" ] ], [ [ "from pyaat.gravity import VerticalConstant\ngrav = VerticalConstant()", "_____no_output_____" ] ], [ [ "### Import Aircraft model", "_____no_output_____" ] ], [ [ "from pyaat.aircraft import Aircraft\nairc = Aircraft()", "_____no_output_____" ] ], [ [ "### Import propulsion model", "_____no_output_____" ] ], [ [ "from pyaat.propulsion import SimpleModel\nprop = SimpleModel()", "_____no_output_____" ] ], [ [ "### Create a system", "_____no_output_____" ] ], [ [ "from pyaat.system import system\nComplete_system = system(atmosphere = atm, propulsion = prop, aircraft = airc, gravity = grav)", "_____no_output_____" ] ], [ [ "### Trimm at cruize condition", "_____no_output_____" ] ], [ [ "Xe, Ue = Complete_system.trimmer(condition='cruize', HE = 10000., VE = 200)", "_____no_output_____" ] ], [ [ "### Printing equilibrium states and controls", "_____no_output_____" ] ], [ [ "from pyaat.tools import printInfo\nprintInfo(Xe, Ue, frame ='body')", "--------------------------------\n------------ STATES ------------\n------------- BODY -------------\n--------------------------------\nx\n0.0\n-------------\ny\n0.0\n-------------\nz\n-10000.0\n-------------\nu\n199.90512762194552\n-------------\nv\n-1.4802192864744595e-21\n-------------\nw\n6.159541415857848\n-------------\nphi\n0.0\n-------------\ntheta\n1.7648577035616007\n-------------\npsi\n0.0\n-------------\np\n4.415466257582898e-22\n-------------\nq\n-8.573425743483335e-15\n-------------\nr\n3.275803472664854e-22\n" ], [ "printInfo(Xe, Ue, frame ='aero')", "--------------------------------\n------------ STATES ------------\n------------- AERO -------------\n--------------------------------\nV\n200.0\n-------------\nalpha\n1.7648577035615969\n-------------\nbeta\n-4.240515893442633e-22\n-------------\nphi\n0.0\n-------------\ntheta\n1.7648577035616007\n-------------\npsi\n0.0\n-------------\np\n4.415466257582898e-22\n-------------\nq\n-8.573425743483335e-15\n-------------\nr\n3.275803472664854e-22\n-------------\nx0\n0.0\n-------------\ny0\n0.0\n-------------\nH\n10000.0\n" ], [ "printInfo(Xe, Ue, frame='controls')", "--------------------------------\n----------- CONTROLS -----------\n--------------------------------\ndelta_p\n34.65222851433093\n-------------\ndelta_e\n-2.208294991778133\n-------------\ndelta_a\n4.978810759532202e-22\n-------------\ndelta_r\n-8.268303092392625e-22\n" ] ], [ [ "## Simulation", "_____no_output_____" ], [ "The open-loop simulation is carried out using the method 'propagate'. Mandatory inputs are the time of simulation TF, the equilibrium states Xe, the equilibrium control Ue, and a bolean variable called 'perturbation' which defines is applied during the simulation or not.", "_____no_output_____" ], [ "### Equilibrium simulation", "_____no_output_____" ] ], [ [ "solution, control = Complete_system.propagate(Xe, Ue, TF = 180, perturbation = False)", "_____no_output_____" ] ], [ [ "The outputs are two multidimentional arrays, containing the states over time and control over time.", "_____no_output_____" ] ], [ [ "print('Solution')\nsolution", "Solution\n" ], [ "print('control')\ncontrol", "control\n" ] ], [ [ "The time array can be accessed directly on the system.", "_____no_output_____" ] ], [ [ "time = Complete_system.time\ntime", "_____no_output_____" ] ], [ [ "Check out the documentation for more information about the outputs.", "_____no_output_____" ], [ "### Ploting the results\nSome plots can be generated directly using the plotter util embeeded within PyAAT.", "_____no_output_____" ] ], [ [ "from pyaat.tools import plotter\npltr = plotter()", "_____no_output_____" ], [ "pltr.states = solution\npltr.time = Complete_system.time\npltr.control = control", "_____no_output_____" ], [ "pltr.LinVel(frame = 'body')", "_____no_output_____" ], [ "pltr.LinPos()", "_____no_output_____" ], [ "pltr.Attitude()", "_____no_output_____" ], [ "pltr.AngVel()", "_____no_output_____" ], [ "pltr.Controls()", "_____no_output_____" ], [ "pltr.LinPos3D()", "_____no_output_____" ] ], [ [ "All states and controls remain constant over time, as expected.", "_____no_output_____" ], [ "## Out-of-equilibrium simulations\n\nSometimes we may be interested in verifying the behavior of the aircraft out of the equilibrium states. It can be done by applying perturbations.\n\nNote that you would obtain the same result if you input a vector Xe out of equilibrium, but consider that it may cause confusion and in more advanced simulations (considering closed-loop control) it might lead to errors.", "_____no_output_____" ], [ "### Perturbation on states", "_____no_output_____" ] ], [ [ "solution, control = Complete_system.propagate(Xe, Ue, T0 = 0.0, TF = 30.0, dt = 0.01, perturbation = True, state = {'beta':2., 'alpha':2.})", "_____no_output_____" ], [ "pltr.states = solution\npltr.time = Complete_system.time\npltr.control = control", "_____no_output_____" ], [ "pltr.LinVel(frame = 'aero')", "_____no_output_____" ], [ "pltr.LinPos()", "_____no_output_____" ], [ "pltr.Attitude()", "_____no_output_____" ], [ "pltr.AngVel()", "_____no_output_____" ], [ "pltr.Controls()", "_____no_output_____" ] ], [ [ "### open-loop control", "_____no_output_____" ], [ "Some usual control inputs are also embeeded within the toolbox, such as the doublet and step.", "_____no_output_____" ] ], [ [ "from pyaat.control import doublet, step", "_____no_output_____" ] ], [ [ "#### Doublet input on elevator", "_____no_output_____" ] ], [ [ "doub = doublet()\ndoub.command = 'elevator'\ndoub.amplitude = 3\ndoub.T = 1\ndoub.t_init = 2", "_____no_output_____" ] ], [ [ "#### Step input on aileron", "_____no_output_____" ] ], [ [ "st =step()\nst.command = 'aileron'\nst.amplitude = 1\nst.t_init = 2", "_____no_output_____" ], [ "solution, control = Complete_system.propagate(Xe, Ue, TF = 50, perturbation=True, control = [doub, st])", "_____no_output_____" ] ], [ [ "One can input as many control perturbation as we want, and we can combine it with states perturbations is desired.", "_____no_output_____" ] ], [ [ "pltr.states = solution\npltr.time = Complete_system.time\npltr.control = control", "_____no_output_____" ], [ "pltr.Controls()", "_____no_output_____" ], [ "pltr.LinVel(frame = 'aero')", "_____no_output_____" ], [ "pltr.LinPos()", "_____no_output_____" ], [ "pltr.Attitude()", "_____no_output_____" ], [ "pltr.AngVel()", "_____no_output_____" ], [ "pltr.LinPos3D()", "_____no_output_____" ] ], [ [ "#### Yes, the aircraft is falling! The lateral-directional dynamics is instable (check it out using the tools from pyaat.flight_mechanics)", "_____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", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code", "code", "code", "code", "code", "code", "code" ], [ "markdown", "markdown", "markdown" ], [ "code", "code", "code", "code", "code", "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code" ], [ "markdown" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "___\n\n<a href='http://www.pieriandata.com'> <img src='../../Pierian_Data_Logo.png' /></a>\n___\n# Pandas Built-in Data Visualization\n\nIn this lecture we will learn about pandas built-in capabilities for data visualization! It's built-off of matplotlib, but it baked into pandas for easier usage! \n\nLet's take a look!", "_____no_output_____" ], [ "## Imports", "_____no_output_____" ] ], [ [ "import numpy as np\nimport pandas as pd\n%matplotlib inline", "_____no_output_____" ] ], [ [ "## The Data\n\nThere are some fake data csv files you can read in as dataframes:", "_____no_output_____" ] ], [ [ "df1 = pd.read_csv('df1',index_col=0)\ndf2 = pd.read_csv('df2')", "_____no_output_____" ] ], [ [ "## Style Sheets\n\nMatplotlib has [style sheets](http://matplotlib.org/gallery.html#style_sheets) you can use to make your plots look a little nicer. These style sheets include plot_bmh,plot_fivethirtyeight,plot_ggplot and more. They basically create a set of style rules that your plots follow. I recommend using them, they make all your plots have the same look and feel more professional. You can even create your own if you want your company's plots to all have the same look (it is a bit tedious to create on though).\n\nHere is how to use them.\n\n**Before plt.style.use() your plots look like this:**", "_____no_output_____" ] ], [ [ "df1['A'].hist()", "_____no_output_____" ] ], [ [ "Call the style:", "_____no_output_____" ] ], [ [ "import matplotlib.pyplot as plt\nplt.style.use('ggplot')", "_____no_output_____" ] ], [ [ "Now your plots look like this:", "_____no_output_____" ] ], [ [ "df1['A'].hist()", "_____no_output_____" ], [ "plt.style.use('bmh')\ndf1['A'].hist()", "_____no_output_____" ], [ "plt.style.use('dark_background')\ndf1['A'].hist()", "_____no_output_____" ], [ "plt.style.use('fivethirtyeight')\ndf1['A'].hist()", "_____no_output_____" ], [ "plt.style.use('ggplot')", "_____no_output_____" ] ], [ [ "Let's stick with the ggplot style and actually show you how to utilize pandas built-in plotting capabilities!", "_____no_output_____" ], [ "# Plot Types\n\nThere are several plot types built-in to pandas, most of them statistical plots by nature:\n\n* df.plot.area \n* df.plot.barh \n* df.plot.density \n* df.plot.hist \n* df.plot.line \n* df.plot.scatter\n* df.plot.bar \n* df.plot.box \n* df.plot.hexbin \n* df.plot.kde \n* df.plot.pie\n\nYou can also just call df.plot(kind='hist') or replace that kind argument with any of the key terms shown in the list above (e.g. 'box','barh', etc..)\n___", "_____no_output_____" ], [ "Let's start going through them!\n\n## Area", "_____no_output_____" ] ], [ [ "df2.plot.area(alpha=0.4)", "_____no_output_____" ] ], [ [ "## Barplots", "_____no_output_____" ] ], [ [ "df2.head()", "_____no_output_____" ], [ "df2.plot.bar()", "_____no_output_____" ], [ "df2.plot.bar(stacked=True)", "_____no_output_____" ] ], [ [ "## Histograms", "_____no_output_____" ] ], [ [ "df1['A'].plot.hist(bins=50)", "_____no_output_____" ] ], [ [ "## Line Plots", "_____no_output_____" ] ], [ [ "df1.plot.line(x=df1.index,y='B',figsize=(12,3),lw=1)", "_____no_output_____" ] ], [ [ "## Scatter Plots", "_____no_output_____" ] ], [ [ "df1.plot.scatter(x='A',y='B')", "_____no_output_____" ] ], [ [ "You can use c to color based off another column value\nUse cmap to indicate colormap to use. \nFor all the colormaps, check out: http://matplotlib.org/users/colormaps.html", "_____no_output_____" ] ], [ [ "df1.plot.scatter(x='A',y='B',c='C',cmap='coolwarm')", "_____no_output_____" ] ], [ [ "Or use s to indicate size based off another column. s parameter needs to be an array, not just the name of a column:", "_____no_output_____" ] ], [ [ "df1.plot.scatter(x='A',y='B',s=df1['C']*200)", "C:\\Users\\Marcial\\Anaconda3\\lib\\site-packages\\matplotlib\\collections.py:877: RuntimeWarning: invalid value encountered in sqrt\n scale = np.sqrt(self._sizes) * dpi / 72.0 * self._factor\n" ] ], [ [ "## BoxPlots", "_____no_output_____" ] ], [ [ "df2.plot.box() # Can also pass a by= argument for groupby", "_____no_output_____" ] ], [ [ "## Hexagonal Bin Plot\n\nUseful for Bivariate Data, alternative to scatterplot:", "_____no_output_____" ] ], [ [ "df = pd.DataFrame(np.random.randn(1000, 2), columns=['a', 'b'])\ndf.plot.hexbin(x='a',y='b',gridsize=25,cmap='Oranges')", "_____no_output_____" ] ], [ [ "____", "_____no_output_____" ], [ "## Kernel Density Estimation plot (KDE)", "_____no_output_____" ] ], [ [ "df2['a'].plot.kde()", "_____no_output_____" ], [ "df2.plot.density()", "_____no_output_____" ] ], [ [ "That's it! Hopefully you can see why this method of plotting will be a lot easier to use than full-on matplotlib, it balances ease of use with control over the figure. A lot of the plot calls also accept additional arguments of their parent matplotlib plt. call. \n\n\n# Great Job!", "_____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" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code" ], [ "markdown" ] ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# Amazon Shure MV7 EDA and Sentement Analysis\n\n- toc: true\n- branch: master\n- badges: true\n- comments: true\n- categories: [Fastpages, Jupyter, Python, Selenium, Stoc]\n- annotations: true\n- hide: false\n- image: images/diagram.png\n- layout: post\n- search_exclude: true", "_____no_output_____" ], [ "### Required Packages\n\n[wordcloud](https://github.com/amueller/word_cloud), \n[geopandas](https://geopandas.org/en/stable/getting_started/install.html), \n[nbformat](https://pypi.org/project/nbformat/), \n[seaborn](https://seaborn.pydata.org/installing.html), \n[scikit-learn](https://scikit-learn.org/stable/install.html)", "_____no_output_____" ], [ "![]({{site.baseurl}}/images/diagram.png \"https://github.com/fastai/fastpages\")", "_____no_output_____" ], [ "### Now let's get started!\nFirst thing first, you need to load all the necessary libraries:", "_____no_output_____" ] ], [ [ "import pandas as pd\nfrom matplotlib import pyplot as plt\nimport numpy as np\nfrom wordcloud import WordCloud\nfrom wordcloud import STOPWORDS\nimport re\nimport plotly.graph_objects as go\nimport seaborn as sns", "_____no_output_____" ] ], [ [ "## Read the Data", "_____no_output_____" ] ], [ [ "#Import Data\ndf = pd.read_csv(\"/Users/zeyu/Desktop/DS/Ebay & Amazon/Amazon_reviews_scraping/Amazon_reviews_scraping/full_reviews.csv\")", "_____no_output_____" ] ], [ [ "", "_____no_output_____" ], [ "## Data Cleaning\n\nStep 1:\n- Splite column Date to Country and Date\n- Combine the two rating columns to one\n- Convert type of date from string to datetime", "_____no_output_____" ] ], [ [ "#Clean Data\ninfo = []\nfor i in df[\"date\"]:\n x = re.sub(\"Reviewed in \", \"\", i)\n x1 = re.sub(\" on \", \"*\", x)\n info.append(x1)\n\ndf[\"date\"] = pd.DataFrame({\"date\": info})\ndf[['country','date']] = df.date.apply(\n lambda x: pd.Series(str(x).split(\"*\")))\n\nstar = []\nstar = df.stars1.combine_first(df.stars2)\ndf[\"star\"] = pd.DataFrame({\"star\": star})\n\ndel df['stars1']\ndel df['stars2']\n\n#Convert String to Date\ndf.date = pd.to_datetime(df.date)", "_____no_output_____" ] ], [ [ "Step 2:\n- Two methods to verify if column \"star\" contain any NaN\n- Converted the type of column \"star\" from string to Int", "_____no_output_____" ] ], [ [ "\"nan\" in df['star']", "_____no_output_____" ], [ "df_no_star = df[df['star'].isna()]\ndf_no_star", "_____no_output_____" ], [ "#Convert 2.0 out of 5 stars to 2\ndf_int = []\n#df_with_star[\"stars\"] = [str(x).replace(':',' ') for x in df[\"stars\"]]\n\nfor i in df[\"star\"]:\n x = re.sub(\".0 out of 5 stars\", \"\", i)\n df_int.append(x)\n\ndf[\"rating\"] = pd.DataFrame({\"rating\": df_int})\ndf[\"rating\"] = df[\"rating\"].astype(int)\ndel df['star']", "_____no_output_____" ] ], [ [ "This is the data looks like after cleaning.\n", "_____no_output_____" ], [ "## EDA", "_____no_output_____" ] ], [ [ "temp = df['rating'].value_counts()\nfig = go.Figure(go.Bar(\n x=temp,\n y=temp.index,\n orientation='h'))\n\nfig.show()", "_____no_output_____" ], [ "df_country = df['country'].value_counts()\nfig = go.Figure(go.Bar(\n x=df_country,\n y=df_country.index,\n orientation='h'))\n\nfig.show()", "_____no_output_____" ], [ "mean_rating = df['rating'].mean()\nmean_rating", "_____no_output_____" ], [ "\"\"\"fig = px.line(df, x=temp.index, y=temp.rating, title='Life expectancy in Canada')\nfig.show()\"\"\"\nimport plotly.express as px\ntemp = df.groupby([df['date'].dt.date]).mean()\ntemp", "_____no_output_____" ], [ "#Average rating each month\ntemp = df.groupby(df['date'].dt.strftime('%B'))['rating'].mean().sort_values()\norder_temp = temp.reindex([\"January\", \"February\", \"March\", \"April\", \"May\", \"June\", \"July\", \"August\", \"September\", \"November\", \"December\"])\norder_temp.plot()", "_____no_output_____" ], [ "#Quantity of reviews in each month.\ntemp = df.groupby(df['date'].dt.strftime('%B'))['rating'].count().sort_values()\norder_temp = temp.reindex([\"January\", \"February\", \"March\", \"April\", \"May\", \"June\", \"July\", \"August\", \"September\", \"November\", \"December\"])\norder_temp.plot()", "_____no_output_____" ], [ "#Many words are useless so create a stopword list\nstopwords = set(STOPWORDS)\nstopwords.update([\"Mic\", \"Microphone\", \"using\",\"sound\",\"use\"])\n\n\ndef cleaned_visualise_word_map(x):\n words=\" \"\n for msg in x:\n msg = str(msg).lower()\n words = words+msg+\" \"\n wordcloud = WordCloud(stopwords = stopwords, width=3000, height=2500, background_color='white').generate(words)\n fig_size = plt.rcParams[\"figure.figsize\"]\n fig_size[0] = 14\n fig_size[1] = 7\n #Display image appear more smoothly\n plt.imshow(wordcloud, interpolation='bilinear')\n plt.axis(\"off\")\n plt.show(wordcloud)\ncleaned_visualise_word_map(df[\"review\"])", "_____no_output_____" ], [ "df = df[df['rating'] != 3]\ndf['sentiment'] = df['rating'].apply(lambda rating : +1 if rating > 3 else -1)", "_____no_output_____" ], [ "positive = df[df['sentiment'] == 1]\nnegative = df[df['sentiment'] == -1]", "_____no_output_____" ], [ "df['sentimentt'] = df['sentiment'].replace({-1 : 'negative'})\ndf['sentimentt'] = df['sentimentt'].replace({1 : 'positive'})\nfig = px.histogram(df, x=\"sentimentt\")\nfig.update_traces(marker_color=\"indianred\",marker_line_color='rgb(8,48,107)',\n marker_line_width=1.5)\nfig.update_layout(title_text='Product Sentiment')\nfig.show()", "_____no_output_____" ], [ "stopwords = set(STOPWORDS)\n#stopwords.update([\"Mic\", \"Microphone\", \"using\", \"sound\", \"use\"]) \n\n## good and great removed because they were included in negative sentiment\npos = \" \".join(review for review in positive.title)\nwordcloud2 = WordCloud(stopwords=stopwords).generate(pos)\nplt.imshow(wordcloud2, interpolation='bilinear')\nplt.axis(\"off\")\nplt.show()", "_____no_output_____" ], [ "pos = \" \".join(review for review in negative.title)\nwordcloud2 = WordCloud(stopwords=stopwords).generate(pos)\nplt.imshow(wordcloud2, interpolation='bilinear')\nplt.axis(\"off\")\nplt.show()", "_____no_output_____" ] ], [ [ "## Sentiment Analysis", "_____no_output_____" ] ], [ [ "def remove_punctuation(text):\n final = \"\".join(u for u in text if u not in (\"?\", \".\", \";\", \":\", \"!\",'\"'))\n return final\n\ndf['review'] = df['review'].apply(remove_punctuation)\ndf = df.dropna(subset=['title'])\ndf['title'] = df['title'].apply(remove_punctuation)", "_____no_output_____" ], [ "dfNew = df[['title','sentiment']]\ndfNew.head()", "_____no_output_____" ], [ "dfLong = df[['review','sentiment']]\ndfLong.head()", "_____no_output_____" ], [ "index = df.index\ndf['random_number'] = np.random.randn(len(index))\ntrain = df[df['random_number'] <= 0.8]\ntest = df[df['random_number'] > 0.8]", "_____no_output_____" ], [ "#change df frame to a bag of words\nfrom sklearn.feature_extraction.text import CountVectorizer\nvectorizer = CountVectorizer(token_pattern=r'\\b\\w+\\b')", "_____no_output_____" ] ], [ [ "[Vectorizer](https://towardsdatascience.com/hacking-scikit-learns-vectorizers-9ef26a7170af) &\n[Bag-of-Words](https://towardsdatascience.com/hacking-scikit-learns-vectorizers-9ef26a7170af)\n", "_____no_output_____" ] ], [ [ "train_matrix = vectorizer.fit_transform(train['title'])\ntest_matrix = vectorizer.transform(test['title'])", "_____no_output_____" ], [ "train_matrix_l = vectorizer.fit_transform(train['review'])\ntest_matrix_l = vectorizer.transform(test['review'])", "_____no_output_____" ], [ "from sklearn.linear_model import LogisticRegression\nlr = LogisticRegression()", "_____no_output_____" ], [ "X_train = train_matrix\nX_test = test_matrix\ny_train = train['sentiment']\ny_test = test['sentiment']", "_____no_output_____" ], [ "X_train_l = train_matrix_l\nX_test_l = test_matrix_l\ny_train_l = train['sentiment']\ny_test_l = test['sentiment']", "_____no_output_____" ], [ "lr.fit(X_train,y_train)", "_____no_output_____" ], [ "lr.fit(X_train_l,y_train_l)", "_____no_output_____" ], [ "predictions = lr.predict(X_test)", "_____no_output_____" ], [ "predictions_l = lr.predict(X_test_l)", "_____no_output_____" ], [ "# find accuracy, precision, recall:\nfrom sklearn.metrics import confusion_matrix,classification_report\nnew = np.asarray(y_test)\nconfusion_matrix(predictions,y_test)", "_____no_output_____" ], [ "long = np.asarray(y_test_l)\nconfusion_matrix(predictions_l,y_test_l)", "_____no_output_____" ], [ "print(classification_report(predictions,y_test))\n#0.88 Accuracy", " precision recall f1-score support\n\n -1 0.00 0.00 0.00 0\n 1 1.00 0.89 0.94 116\n\n accuracy 0.89 116\n macro avg 0.50 0.44 0.47 116\nweighted avg 1.00 0.89 0.94 116\n\n" ], [ "print(classification_report(predictions_l,y_test_l))", " precision recall f1-score support\n\n -1 0.00 0.00 0.00 0\n 1 1.00 0.89 0.94 116\n\n accuracy 0.89 116\n macro avg 0.50 0.44 0.47 116\nweighted avg 1.00 0.89 0.94 116\n\n" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "#Setup\n%matplotlib inline\nimport numpy as np\nimport matplotlib\nimport matplotlib.pyplot as plt\nimport scipy\nfrom scipy import stats\nimport pickle\nimport pandas as pd\nfrom sklearn.neural_network import MLPClassifier\nfrom sklearn.model_selection import train_test_split\nfrom sklearn import tree\nplt.rcParams[\"figure.figsize\"] = (20,10)", "_____no_output_____" ], [ "data = open('higgs_100000_pt_250_500.pkl','rb')\nnew_dict = pickle.load(data)\ndata2 = open('qcd_100000_pt_250_500.pkl','rb')\nqcd_dict = pickle.load(data2)", "_____no_output_____" ], [ "high_lumi = pd.read_hdf('data_highLumi_pt_250_500.h5')", "_____no_output_____" ], [ "low_lumi = pd.read_hdf('data_lowLumi_pt_250_500.h5')", "_____no_output_____" ], [ "#Random Samples using expected number of events for a given run\nstate = 123\nhiggs_events = new_dict.sample(n=100, random_state = state)\nqcd_events = qcd_dict.sample(n=20000, random_state = state)", "_____no_output_____" ], [ "#Making lists of labels\nhiggs_labels = [1]*100000\nqcd_labels = [0]*100000", "_____no_output_____" ], [ "#Labeling and combining sampled data\nnew_dict['label'] = higgs_labels\nqcd_dict['label'] = qcd_labels\nsample = pd.concat([new_dict,qcd_dict])", "_____no_output_____" ] ], [ [ "## **Part 1:** Event Selection Optimization", "_____no_output_____" ], [ "#### 1) Make a stacked histogram plot for the feature variable: mass", "_____no_output_____" ] ], [ [ "fig, ax = plt.subplots(1,1)\nax.hist(higgs_events['mass'],density = True,alpha = 0.8, label = 'higgs')\nax.hist(qcd_events['mass'],density = True,alpha = 0.8, label = 'qcd')\nplt.legend(fontsize = 18)\nplt.show()", "_____no_output_____" ] ], [ [ "Expected events in background is 20,000 and is poisson distirbuted", "_____no_output_____" ], [ "#### $\\cdot$ Use Poisson statistics for significance calculation", "_____no_output_____" ] ], [ [ "np.random.seed(123)\ndist = stats.poisson.rvs(20000, size = 10000)\nplt.hist(dist,density = True, bins = np.linspace(19450,20550,50), label = 'Expected Yield Distribution')\nplt.axvline(20100,color = 'red',label = 'Observed Yield')\nplt.legend(fontsize = 18)\nplt.show()", "_____no_output_____" ], [ "print('Significance of 20100 events:', np.round(stats.norm.isf(stats.poisson.sf(20100,20000)),3),'sigma')", "Significance of 20100 events: 0.711 sigma\n" ] ], [ [ "$\\frac{\\textbf{N}_{Higgs}}{\\sqrt{\\textbf{N}_{QCD}}} = \\frac{100}{\\sqrt{20000}} = 0.707$\n\nThis value is different than the value obtained in the previous calculation. This is because the value $\\frac{\\textbf{N}_{Higgs}}{\\sqrt{\\textbf{N}_{QCD}}}$ is the number of standard deviations away from the mean the measurment is, while the number from the above calculation is how the probability of the background producing a value larger than the observed value corresponds to the standard normal distributions $\\sigma$.", "_____no_output_____" ] ], [ [ "def mult_cut(qcd,higgs,features,cuts):\n '''\n Parameters:\n qcd - qcd data dictionary\n higgs - higgs data dictionary\n features (list) - the features to apply cuts to\n cuts (list of touples) - in format ((min,max),(min,max)) \n Returns:\n number of qcd and higgs events\n cut min and max\n significance\n '''\n qcd_factor = 20000/len(qcd)\n higgs_factor = 100/len(higgs)\n mu = qcd\n signal = higgs\n for i in range(0,len(features)):\n a = np.array(mu[features[i]])\n b = np.array(signal[features[i]])\n mu = mu[:][np.logical_and(a>cuts[i][0], a<cuts[i][1])]\n signal = signal[:][np.logical_and(b>cuts[i][0], b<cuts[i][1])]\n mu = len(mu)*qcd_factor\n signal = len(signal)*higgs_factor\n sig = np.round(stats.norm.isf(stats.poisson.sf(mu + signal,mu)),3)\n print(features,'cuts', cuts ,'leaves',mu,'expected qcd events and',signal,'expected higgs events')\n print('Significance of', mu+signal ,'events:',sig,'sigma')\n print('---------------------------------------------\\n')", "_____no_output_____" ] ], [ [ "#### 2) Identify mass cuts to optimize the expected significance", "_____no_output_____" ] ], [ [ "s = 120\nfor n in range(0,7):\n mult_cut(qcd_dict,new_dict,['mass'],[(s,150)])\n s+=1", "['mass'] cuts [(120, 150)] leaves 2381.6 expected qcd events and 78.065 expected higgs events\nSignificance of 2459.665 events: 1.591 sigma\n---------------------------------------------\n\n['mass'] cuts [(121, 150)] leaves 2262.4 expected qcd events and 77.038 expected higgs events\nSignificance of 2339.438 events: 1.615 sigma\n---------------------------------------------\n\n['mass'] cuts [(122, 150)] leaves 2157.4 expected qcd events and 75.605 expected higgs events\nSignificance of 2233.005 events: 1.632 sigma\n---------------------------------------------\n\n['mass'] cuts [(123, 150)] leaves 2052.6 expected qcd events and 73.643 expected higgs events\nSignificance of 2126.243 events: 1.625 sigma\n---------------------------------------------\n\n['mass'] cuts [(124, 150)] leaves 1953.0 expected qcd events and 70.109 expected higgs events\nSignificance of 2023.109 events: 1.59 sigma\n---------------------------------------------\n\n['mass'] cuts [(125, 150)] leaves 1852.6000000000001 expected qcd events and 59.393 expected higgs events\nSignificance of 1911.9930000000002 events: 1.365 sigma\n---------------------------------------------\n\n['mass'] cuts [(126, 150)] leaves 1755.2 expected qcd events and 31.146 expected higgs events\nSignificance of 1786.346 events: 0.749 sigma\n---------------------------------------------\n\n" ], [ "s = 132\nfor n in range(0,7):\n mult_cut(qcd_dict,new_dict,['mass'],[(124,s)])\n s-=1", "['mass'] cuts [(124, 132)] leaves 724.6 expected qcd events and 69.554 expected higgs events\nSignificance of 794.154 events: 2.563 sigma\n---------------------------------------------\n\n['mass'] cuts [(124, 131)] leaves 640.6 expected qcd events and 68.992 expected higgs events\nSignificance of 709.592 events: 2.682 sigma\n---------------------------------------------\n\n['mass'] cuts [(124, 130)] leaves 551.6 expected qcd events and 67.891 expected higgs events\nSignificance of 619.491 events: 2.842 sigma\n---------------------------------------------\n\n['mass'] cuts [(124, 129)] leaves 469.20000000000005 expected qcd events and 65.21600000000001 expected higgs events\nSignificance of 534.416 events: 2.956 sigma\n---------------------------------------------\n\n['mass'] cuts [(124, 128)] leaves 382.8 expected qcd events and 60.361000000000004 expected higgs events\nSignificance of 443.161 events: 3.034 sigma\n---------------------------------------------\n\n['mass'] cuts [(124, 127)] leaves 291.40000000000003 expected qcd events and 53.394 expected higgs events\nSignificance of 344.79400000000004 events: 3.032 sigma\n---------------------------------------------\n\n['mass'] cuts [(124, 126)] leaves 197.8 expected qcd events and 38.963 expected higgs events\nSignificance of 236.763 events: 2.68 sigma\n---------------------------------------------\n\n" ] ], [ [ "Cut optimization was performed on the unsampled data in order to not overfit the cuts to the sample selected. The optimal cuts kept data with a mass between 124 and 128, and with those cuts yielded a measurement significance of 3.034 sigma.", "_____no_output_____" ], [ "#### 3) Make stacked histogram plots for the rest of the features\n##### With and without optimal mass cuts", "_____no_output_____" ] ], [ [ "plt.rcParams[\"figure.figsize\"] = (20,50)\nfig, ((ax1,ax2),(ax3,ax4),(ax5,ax6),(ax7,ax8),(ax9,ax10),(ax11,ax12),(ax13,ax14),(ax15,ax16),(ax17,ax18),(ax19,ax20),(ax21,ax22),(ax23,ax24),(ax25,ax26),(ax27,ax28)) = plt.subplots(14,2)\naxes = ((ax1,ax2),(ax3,ax4),(ax5,ax6),(ax7,ax8),(ax9,ax10),(ax11,ax12),(ax13,ax14),(ax15,ax16),(ax17,ax18),(ax19,ax20),(ax21,ax22),(ax23,ax24),(ax25,ax26),(ax27,ax28))\nlabels = ['pt', 'eta', 'phi', 'mass', 'ee2', 'ee3', 'd2', 'angularity', 't1', 't2', 't3', 't21', 't32', 'KtDeltaR']\na = np.array(new_dict['mass'])\nb = np.array(qcd_dict['mass'])\nfor i in range(0,14):\n axes[i][0].hist(new_dict[labels[i]],density = True, alpha = 0.7,label = 'higgs')\n axes[i][0].hist(qcd_dict[labels[i]],density = True, alpha = 0.7,label = 'qcd')\n axes[i][0].set_xlabel(labels[i])\n axes[i][0].legend()\n axes[i][1].hist(new_dict[labels[i]][np.logical_and(a<135, a>124)],density = True, alpha = 0.7,label = 'higgs with mass cuts')\n axes[i][1].hist(qcd_dict[labels[i]][np.logical_and(b<135, b>124)],density = True, alpha = 0.7,label = 'qcd with mass cuts')\n axes[i][1].set_xlabel(labels[i])\n axes[i][1].legend()\nplt.show()", "_____no_output_____" ] ], [ [ "#### 4) Optimize event selections using multiple features", "_____no_output_____" ] ], [ [ "mult_cut(qcd_dict,new_dict,['d2'],[(0,1.42)])", "['d2'] cuts [(0, 1.42)] leaves 2029.6000000000001 expected qcd events and 64.296 expected higgs events\nSignificance of 2093.896 events: 1.415 sigma\n---------------------------------------------\n\n" ], [ "mult_cut(qcd_dict,new_dict,['t3'],[(0,0.17)])\nmult_cut(qcd_dict,new_dict,['KtDeltaR'],[(0.48,0.93)])\nmult_cut(qcd_dict,new_dict,['ee2'],[(0.11,0.21)])\nmult_cut(qcd_dict,new_dict,['d2'],[(0,1.42)])", "['t3'] cuts [(0, 0.17)] leaves 712.6 expected qcd events and 46.548 expected higgs events\nSignificance of 759.148 events: 1.744 sigma\n---------------------------------------------\n\n['KtDeltaR'] cuts [(0.48, 0.93)] leaves 4809.2 expected qcd events and 72.474 expected higgs events\nSignificance of 4881.674 events: 1.042 sigma\n---------------------------------------------\n\n['ee2'] cuts [(0.11, 0.21)] leaves 4373.6 expected qcd events and 73.07300000000001 expected higgs events\nSignificance of 4446.673000000001 events: 1.102 sigma\n---------------------------------------------\n\n['d2'] cuts [(0, 1.42)] leaves 2029.6000000000001 expected qcd events and 64.296 expected higgs events\nSignificance of 2093.896 events: 1.415 sigma\n---------------------------------------------\n\n" ], [ "mult_cut(qcd_events,higgs_events,['mass','d2'],[(124,128),(0,1.42)])\nmult_cut(qcd_events,higgs_events,['mass','KtDeltaR'],[(124,128),(0.48,0.93)])\nmult_cut(qcd_events,higgs_events,['mass','ee2'],[(124,128),(0.11,0.21)])\nmult_cut(qcd_events,higgs_events,['mass','t3'],[(124,128),(0,0.17)])", "['mass', 'd2'] cuts [(124, 128), (0, 1.42)] leaves 28.0 expected qcd events and 55.0 expected higgs events\nSignificance of 83.0 events: 8.48 sigma\n---------------------------------------------\n\n['mass', 'KtDeltaR'] cuts [(124, 128), (0.48, 0.93)] leaves 150.0 expected qcd events and 60.0 expected higgs events\nSignificance of 210.0 events: 4.666 sigma\n---------------------------------------------\n\n['mass', 'ee2'] cuts [(124, 128), (0.11, 0.21)] leaves 266.0 expected qcd events and 61.0 expected higgs events\nSignificance of 327.0 events: 3.648 sigma\n---------------------------------------------\n\n['mass', 't3'] cuts [(124, 128), (0, 0.17)] leaves 18.0 expected qcd events and 39.0 expected higgs events\nSignificance of 57.0 events: 7.418 sigma\n---------------------------------------------\n\n" ], [ "mult_cut(qcd_events,higgs_events,['mass','d2','KtDeltaR'],[(124,128),(0,1.42),(0.48,0.93)])", "['mass', 'd2', 'KtDeltaR'] cuts [(124, 128), (0, 1.42), (0.48, 0.93)] leaves 18.0 expected qcd events and 51.0 expected higgs events\nSignificance of 69.0 events: 9.238 sigma\n---------------------------------------------\n\n" ] ], [ [ "#### 5) Plot 2-dimensional scattering plots between top two most discriminative features", "_____no_output_____" ] ], [ [ "plt.rcParams[\"figure.figsize\"] = (20,10)\nfig, (ax1,ax2) = plt.subplots(1,2)\nax1.plot(qcd_dict['mass'],qcd_dict['d2'],color = 'red', label = 'QCD',ls='',marker='.',alpha=0.5)\nax1.plot(new_dict['mass'],qcd_dict['d2'],color = 'blue',label = 'Higgs',ls='',marker='.',alpha=0.5)\nax1.legend(fontsize = 18)\nax1.set_xlabel('mass',fontsize = 18)\nax1.set_ylabel('d2',fontsize = 18)\n\nax2.plot(qcd_dict['mass'],qcd_dict['KtDeltaR'],color = 'red', label = 'QCD',ls='',marker='.',alpha=0.5)\nax2.plot(new_dict['mass'],qcd_dict['KtDeltaR'],color = 'blue',label = 'Higgs',ls='',marker='.',alpha=0.5)\nax2.legend(fontsize = 18)\nax2.set_xlabel('mass',fontsize = 18)\nax2.set_ylabel('KtDeltaR',fontsize = 18)\nplt.show()", "_____no_output_____" ] ], [ [ "Using Maching Learning to predict", "_____no_output_____" ] ], [ [ "sample_train, sample_test = train_test_split(sample,test_size = 0.2)\n\nX_train = sample_train.drop('label',axis = 1)\ny_train = sample_train['label']\n\nX_test = sample_test.drop('label',axis = 1)\ny_test = sample_test['label']", "_____no_output_____" ], [ "mdl = MLPClassifier(hidden_layer_sizes = (8,20,20,8,8,4),max_iter=200,alpha = 10**-6,learning_rate = 'invscaling')\nmdl.fit(X_train,y_train)", "_____no_output_____" ], [ "sum(mdl.predict(X_test) == y_test)/len(y_test)", "_____no_output_____" ], [ "from sklearn.metrics import confusion_matrix\nconf = confusion_matrix(y_test,mdl.predict(X_test))\nprint([conf[1]*100/sum(y_test == 1),conf[0]*20000/sum(y_test == 0)])", "[array([10.91109111, 89.08890889]), array([18293.17068293, 1706.82931707])]\n" ], [ "true_higgs = conf[1][1]*100/sum(y_test == 1)\nfalse_higgs = conf[0][1]*20000/sum(y_test == 0)\nprint(false_higgs,true_higgs)", "1706.8293170682932 89.08890889088909\n" ], [ "sig = stats.norm.isf(stats.poisson.sf(k = true_higgs+false_higgs, mu = false_higgs))\nprint(\"significance using neural network is\",np.round(sig,3),'sigma')", "significance using neural network is 2.132 sigma\n" ] ], [ [ "Machine learning model chosen was less effective than the cuts that I had determined. With a more optimized loss function I'm sure machine learning would out perform manually selected cuts, but in this instance it didn't.", "_____no_output_____" ], [ "## **Part 2:** Pseudo-experiment data analysis", "_____no_output_____" ] ], [ [ "#Defining a function to make cuts and return the cut data, not calculating significance like previous function\ndef straight_cut(data,features,cuts):\n for i in range(0,len(features)):\n a = np.array(data[features[i]])\n data = data[:][np.logical_and(a>cuts[i][0], a<cuts[i][1])]\n return data", "_____no_output_____" ] ], [ [ "#### 1) High Luminosity", "_____no_output_____" ] ], [ [ "plt.rcParams[\"figure.figsize\"] = (20,30)\nfig, ((ax1,ax2),(ax3,ax4),(ax5,ax6)) = plt.subplots(3,2)\naxes = (ax1,ax2,ax3,ax4,ax5,ax6)\nfeatures = ['mass','d2','KtDeltaR','ee2','t3','ee3']\nfor i in range(0,6):\n counts,bins = np.histogram(new_dict[features[i]],bins = 50)\n axes[i].hist(bins[:-1],bins, weights = counts*40344/100000, color = 'red',label = 'Higgs',alpha = 0.7)\n counts,bins = np.histogram(qcd_dict[features[i]],bins = 50)\n axes[i].hist(bins[:-1],bins, weights = counts*40344/100000, color = 'blue',label = 'QCD',alpha = 0.7)\n axes[i].hist(high_lumi[features[i]], color = 'green',label = 'data', bins = 50,alpha = 0.7)\n axes[i].legend()\nplt.show()", "_____no_output_____" ], [ "plt.rcParams[\"figure.figsize\"] = (20,30)\nfig, ((ax1,ax2),(ax3,ax4),(ax5,ax6)) = plt.subplots(3,2)\naxes = (ax1,ax2,ax3,ax4,ax5,ax6)\nfeatures = ['mass','d2','KtDeltaR','ee2','t3','ee3']\ncut_higgs = straight_cut(new_dict,['mass','d2','KtDeltaR'],[(124,128),(0,1.42),(0.48,0.93)])\ncut_qcd = straight_cut(qcd_dict,['mass','d2','KtDeltaR'],[(124,128),(0,1.42),(0.48,0.93)])\ncut_high = straight_cut(high_lumi,['mass','d2','KtDeltaR'],[(124,128),(0,1.42),(0.48,0.93)])\nfor i in range(0,6):\n counts,bins = np.histogram(cut_higgs[features[i]])\n axes[i].hist(bins[:-1],bins, weights = counts*40344/100000, color = 'red',label = 'Higgs',alpha = 0.7)\n counts,bins = np.histogram(cut_qcd[features[i]])\n axes[i].hist(bins[:-1],bins, weights = counts*40344/100000, color = 'blue',label = 'QCD',alpha = 0.7)\n axes[i].hist(cut_high[features[i]], color = 'green',label = 'data',alpha = 0.7)\n axes[i].legend()\n axes[i].set_yscale('log')\nplt.show()", "_____no_output_____" ], [ "n_qcd = len(cut_qcd)*40344/100000\nn_observed = len(cut_high)\nsig = np.round(stats.norm.isf(stats.poisson.sf(n_observed,n_qcd)),3)\nprint('Significance of', n_observed ,'events:',sig,'sigma')", "Significance of 128 events: 10.724 sigma\n" ] ], [ [ "The same cuts made on the simulated data gave a lower significance of $9.2\\sigma$", "_____no_output_____" ], [ "#### 2) Low Luminosity", "_____no_output_____" ] ], [ [ "plt.rcParams[\"figure.figsize\"] = (20,30)\nfig, ((ax1,ax2),(ax3,ax4),(ax5,ax6)) = plt.subplots(3,2)\naxes = (ax1,ax2,ax3,ax4,ax5,ax6)\nfeatures = ['mass','d2','KtDeltaR','ee2','t3','ee3']\nfor i in range(0,6):\n counts,bins = np.histogram(new_dict[features[i]],bins = 50)\n axes[i].hist(bins[:-1],bins, weights = counts*4060/100000, color = 'red',label = 'Higgs',alpha = 0.7)\n counts,bins = np.histogram(qcd_dict[features[i]],bins = 50)\n axes[i].hist(bins[:-1],bins, weights = counts*4060/100000, color = 'blue',label = 'QCD',alpha = 0.7)\n axes[i].hist(low_lumi[features[i]], color = 'green',label = 'data', bins = 50,alpha = 0.7)\n axes[i].legend()\nplt.show()", "_____no_output_____" ], [ "plt.rcParams[\"figure.figsize\"] = (20,30)\nfig, ((ax1,ax2),(ax3,ax4),(ax5,ax6)) = plt.subplots(3,2)\naxes = (ax1,ax2,ax3,ax4,ax5,ax6)\nfeatures = ['mass','d2','KtDeltaR','ee2','t3','ee3']\ncut_low = straight_cut(low_lumi,['mass','d2','KtDeltaR'],[(124,128),(0,1.42),(0.48,0.93)])\nfor i in range(0,6):\n counts,bins = np.histogram(cut_higgs[features[i]])\n axes[i].hist(bins[:-1],bins, weights = counts*4060/100000, color = 'red',label = 'Higgs',alpha = 0.7)\n counts,bins = np.histogram(cut_qcd[features[i]])\n axes[i].hist(bins[:-1],bins, weights = counts*4060/100000, color = 'blue',label = 'QCD',alpha = 0.7)\n axes[i].hist(cut_low[features[i]], color = 'green',label = 'data',alpha = 0.7)\n axes[i].legend()\n axes[i].set_yscale('log')\nplt.show()", "_____no_output_____" ], [ "n_qcd = len(cut_qcd)*4060/100000\nn_observed = len(cut_low)\nsig = np.round(stats.norm.isf(stats.poisson.sf(n_observed,n_qcd)),3)\nprint('Significance of', n_observed ,'events:',sig,'sigma')", "Significance of 9 events: 2.273 sigma\n" ] ], [ [ "#### 3) Confidence Levels of signal yield\n\n95% Upper limit for signal yield low luminosity\n\n$$\\sum_{k = 9}^{\\infty}P(\\mu,k) = 0.95$$\n$$P(\\mu,k) = \\frac{e^{-\\mu}\\mu^k}{k!}$$\n$$\\sum_{k = 0}^{9}\\frac{e^{-\\mu}\\mu^k}{k!} = 0.05$$\n$$\\mu = 15.71$$", "_____no_output_____" ] ], [ [ "print('With a true signal of 15.71, the probability seeing something stronger than 9 events is:',np.round(stats.poisson.sf(9,15.71),4))", "With a true signal of 15.71, the probability seeing something stronger than 9 events is: 0.9501\n" ] ], [ [ "This means that 95% of the time would see more than 9 events if there were a true signal strength of 15.71 events.", "_____no_output_____" ], [ "For the low luminosity data we expected to see 4.22 events, since the data is poisson distributed we will round up to 5 events in order to get more than 95%\n\n$$\\sum_{k = 5}^{\\infty}P(\\mu,k) = 0.95$$\n$$P(\\mu,k) = \\frac{e^{-\\mu}\\mu^k}{k!}$$\n$$\\sum_{k = 0}^{5}\\frac{e^{-\\mu}\\mu^k}{k!} = 0.05$$\n$$\\mu = 10.51$$", "_____no_output_____" ] ], [ [ "prob = 0\nmu = 128\nwhile prob>0.05:\n prob = stats.poisson.cdf(128,mu)\n mu+=0.02\nprint(mu,prob)", "128 0\n" ], [ "print('With a true signal of 10.513, the probability seeing something stronger than 4.22 events is:',np.round(stats.poisson.sf(4.22,10.513),4))", "With a true signal of 10.513, the probability seeing something stronger than 4.22 events is: 0.9791\n" ] ], [ [ "The expected upper limit of 10.513 is lower than the observed upper limit of 15.71, this means that while there were more observed events than expected, we cannot say with 95% certainty that there was no signal present, but we cannot also say with certainty that we have seen a signal.", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Weighting in taxcalc_helpers\n\n## Setup", "_____no_output_____" ] ], [ [ "import numpy as np\nimport pandas as pd\n\nimport taxcalc as tc\nimport microdf as mdf", "_____no_output_____" ], [ "tc.__version__", "_____no_output_____" ] ], [ [ "## Load data\n\nStart with a `DataFrame` with `nu18` and `XTOT`, and also calculate `XTOT_m`.", "_____no_output_____" ] ], [ [ "df = mdf.calc_df(group_vars=['nu18'], metric_vars=['XTOT'])\ndf.columns", "_____no_output_____" ] ], [ [ "From this we can calculate the number of people and tax units by the tax unit's number of children.", "_____no_output_____" ] ], [ [ "df.groupby('nu18')[['s006_m', 'XTOT_m']].sum()", "_____no_output_____" ] ], [ [ "What if we also want to calculate the total number of *children* by the tax unit's number of children?\n\nFor this we can use `add_weighted_metrics`, the function called within `calc_df`.", "_____no_output_____" ] ], [ [ "mdf.add_weighted_metrics(df, ['nu18'])", "_____no_output_____" ] ], [ [ "Now we can do the same thing as before, with the new `nu18_m` column.", "_____no_output_____" ] ], [ [ "df.groupby('nu18')[['nu18_m']].sum()", "_____no_output_____" ] ], [ [ "We can also calculate weighted sums without adding the weighted metric.", "_____no_output_____" ] ], [ [ "total_children = mdf.weighted_sum(df, 'nu18', 's006')\n# Fix this decimal.\n'Total children: ' + str(round(total_children / 1e6)) + 'M.'", "_____no_output_____" ] ], [ [ "We can also calculate the weighted mean and median.", "_____no_output_____" ] ], [ [ "mdf.weighted_mean(df, 'nu18', 's006')", "_____no_output_____" ], [ "mdf.weighted_median(df, 'nu18', 's006')", "_____no_output_____" ] ], [ [ "We can also look at more quantiles.\n\n*Note that weighted quantiles have a different interface.*", "_____no_output_____" ] ], [ [ "decile_bounds = np.arange(0, 1.1, 0.1)\ndeciles = mdf.weighted_quantile(df, 'nu18', 's006', decile_bounds)\npd.DataFrame(deciles, index=decile_bounds)", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Natural and artificial perturbations", "_____no_output_____" ] ], [ [ "import functools\n\nimport numpy as np\nimport matplotlib.pyplot as plt\nplt.ion()\n\nfrom astropy import units as u\nfrom astropy.time import Time\nfrom astropy.coordinates import solar_system_ephemeris\n\nfrom poliastro.twobody.propagation import propagate, cowell\nfrom poliastro.ephem import build_ephem_interpolant\nfrom poliastro.core.elements import rv2coe\n\nfrom poliastro.core.util import norm\nfrom poliastro.util import time_range\nfrom poliastro.core.perturbations import (\n atmospheric_drag, third_body, J2_perturbation\n)\nfrom poliastro.bodies import Earth, Moon\nfrom poliastro.twobody import Orbit\nfrom poliastro.plotting import OrbitPlotter2D, OrbitPlotter3D", "_____no_output_____" ] ], [ [ "### Atmospheric drag ###\nThe poliastro package now has several commonly used natural perturbations. One of them is atmospheric drag! See how one can monitor decay of the near-Earth orbit over time using our new module poliastro.twobody.perturbations!", "_____no_output_____" ] ], [ [ "R = Earth.R.to(u.km).value\nk = Earth.k.to(u.km**3 / u.s**2).value\n\norbit = Orbit.circular(Earth, 250 * u.km, epoch=Time(0.0, format='jd', scale='tdb'))\n\n# parameters of a body\nC_D = 2.2 # dimentionless (any value would do)\nA = ((np.pi / 4.0) * (u.m**2)).to(u.km**2).value # km^2\nm = 100 # kg\nB = C_D * A / m\n\n# parameters of the atmosphere\nrho0 = Earth.rho0.to(u.kg / u.km**3).value # kg/km^3\nH0 = Earth.H0.to(u.km).value\ntof = (100000 * u.s).to(u.day).value\ntr = time_range(0.0, periods=2000, end=tof, format='jd', scale='tdb')\ncowell_with_ad = functools.partial(cowell, ad=atmospheric_drag,\n R=R, C_D=C_D, A=A, m=m, H0=H0, rho0=rho0)\n\nrr = propagate(\n orbit, (tr - orbit.epoch).to(u.s), method=cowell_with_ad\n)", "_____no_output_____" ], [ "plt.ylabel('h(t)')\nplt.xlabel('t, days')\nplt.plot(tr.value, rr.data.norm() - Earth.R);", "_____no_output_____" ] ], [ [ "### Evolution of RAAN due to the J2 perturbation ###\nWe can also see how the J2 perturbation changes RAAN over time!", "_____no_output_____" ] ], [ [ "r0 = np.array([-2384.46, 5729.01, 3050.46]) * u.km\nv0 = np.array([-7.36138, -2.98997, 1.64354]) * u.km / u.s\n\norbit = Orbit.from_vectors(Earth, r0, v0)\n\ntof = 48.0 * u.h\n\n# This will be easier with propagate\n# when this is solved:\n# https://github.com/poliastro/poliastro/issues/257\nrr, vv = cowell(\n Earth.k,\n orbit.r,\n orbit.v,\n np.linspace(0, tof, 2000),\n ad=J2_perturbation,\n J2=Earth.J2.value,\n R=Earth.R.to(u.km).value\n)\n\nk = Earth.k.to(u.km**3 / u.s**2).value\nrr = rr.to(u.km).value\nvv = vv.to(u.km / u.s).value\n\nraans = [rv2coe(k, r, v)[3] for r, v in zip(rr, vv)]\nplt.ylabel('RAAN(t)')\nplt.xlabel('t, h')\nplt.plot(np.linspace(0, tof, 2000), raans);", "_____no_output_____" ] ], [ [ "### 3rd body ###\nApart from time-independent perturbations such as atmospheric drag, J2/J3, we have time-dependend perturbations. Lets's see how Moon changes the orbit of GEO satellite over time!", "_____no_output_____" ] ], [ [ "# database keeping positions of bodies in Solar system over time\nsolar_system_ephemeris.set('de432s')\n\nj_date = 2454283.0 * u.day # setting the exact event date is important\n\ntof = (60 * u.day).to(u.s).value\n\n# create interpolant of 3rd body coordinates (calling in on every iteration will be just too slow)\nbody_r = build_ephem_interpolant(Moon, 28 * u.day, (j_date, j_date + 60 * u.day), rtol=1e-2)\n\nepoch = Time(j_date, format='jd', scale='tdb')\ninitial = Orbit.from_classical(Earth, 42164.0 * u.km, 0.0001 * u.one, 1 * u.deg, \n 0.0 * u.deg, 0.0 * u.deg, 0.0 * u.rad, epoch=epoch)\n\n# multiply Moon gravity by 400 so that effect is visible :)\ncowell_with_3rdbody = functools.partial(cowell, rtol=1e-6, ad=third_body,\n k_third=400 * Moon.k.to(u.km**3 / u.s**2).value, \n third_body=body_r)\n\ntr = time_range(j_date.value, periods=1000, end=j_date.value + 60, format='jd', scale='tdb')\n\nrr = propagate(\n initial, (tr - initial.epoch).to(u.s), method=cowell_with_3rdbody\n)", "_____no_output_____" ], [ "frame = OrbitPlotter3D()\n\nframe.set_attractor(Earth)\nframe.plot_trajectory(rr, label='orbit influenced by Moon')", "_____no_output_____" ] ], [ [ "### Thrusts ###\nApart from natural perturbations, there are artificial thrusts aimed at intentional change of orbit parameters. One of such changes is simultaineous change of eccenricy and inclination.", "_____no_output_____" ] ], [ [ "from poliastro.twobody.thrust import change_inc_ecc\n\necc_0, ecc_f = 0.4, 0.0\na = 42164 # km\ninc_0 = 0.0 # rad, baseline\ninc_f = (20.0 * u.deg).to(u.rad).value # rad\nargp = 0.0 # rad, the method is efficient for 0 and 180\nf = 2.4e-6 # km / s2\n\nk = Earth.k.to(u.km**3 / u.s**2).value\ns0 = Orbit.from_classical(\n Earth,\n a * u.km, ecc_0 * u.one, inc_0 * u.deg,\n 0 * u.deg, argp * u.deg, 0 * u.deg,\n epoch=Time(0, format='jd', scale='tdb')\n)\n \na_d, _, _, t_f = change_inc_ecc(s0, ecc_f, inc_f, f)\n\ncowell_with_ad = functools.partial(cowell, rtol=1e-6, ad=a_d)\n\ntr = time_range(0.0, periods=1000, end=(t_f * u.s).to(u.day).value, format='jd', scale='tdb')\n\nrr2 = propagate(\n s0, (tr - s0.epoch).to(u.s), method=cowell_with_ad\n)", "_____no_output_____" ], [ "frame = OrbitPlotter3D()\n\nframe.set_attractor(Earth)\nframe.plot_trajectory(rr2, label='orbit with artificial thrust')", "_____no_output_____" ] ] ]

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

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

[ "CC-BY-4.0" ]

[ "CC-BY-4.0" ]

[ "CC-BY-4.0" ]

[ [ [ "#### Reason for these tests\nA PR is raised in [ISSUE_1](https://github.com/frankaging/Reason-SCAN/issues/1), the reporter finds some discrepancies in split numbers. Specifically, the `test` split in our main data frame, is not matching up with our sub-test splits as `p1`, `p2` and `p3`. This PR further exposes another issue with our documentations about the splits (i.e., how we generate our splits). Thus, we use this live debug notebook to address these comments.", "_____no_output_____" ], [ "#### The Issue", "_____no_output_____" ] ], [ [ "import os, json\np1_test_path_to_data = \"../../ReaSCAN-v1.0/ReaSCAN-compositional-p1-test/data-compositional-splits.txt\"\nprint(f\"Reading dataset from file: {p1_test_path_to_data}...\")\np1_test_data = json.load(open(p1_test_path_to_data, \"r\"))\nprint(len(p1_test_data[\"examples\"][\"test\"]))\n\np2_test_path_to_data = \"../../ReaSCAN-v1.0/ReaSCAN-compositional-p2-test/data-compositional-splits.txt\"\nprint(f\"Reading dataset from file: {p2_test_path_to_data}...\")\np2_test_data = json.load(open(p2_test_path_to_data, \"r\"))\nprint(len(p2_test_data[\"examples\"][\"test\"]))\n\np3_test_path_to_data = \"../../ReaSCAN-v1.0/ReaSCAN-compositional-p3-test/data-compositional-splits.txt\"\nprint(f\"Reading dataset from file: {p3_test_path_to_data}...\")\np3_test_data = json.load(open(p3_test_path_to_data, \"r\"))\nprint(len(p3_test_data[\"examples\"][\"test\"]))", "Reading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-p1-test/data-compositional-splits.txt...\n921\nReading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-p2-test/data-compositional-splits.txt...\n2120\nReading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-p3-test/data-compositional-splits.txt...\n2712\n" ], [ "len(p1_test_data[\"examples\"][\"test\"]) + len(p2_test_data[\"examples\"][\"test\"]) + len(p3_test_data[\"examples\"][\"test\"])", "_____no_output_____" ], [ "ReaSCAN_path_to_data = \"../../ReaSCAN-v1.0/ReaSCAN-compositional/data-compositional-splits.txt\"\nprint(f\"Reading dataset from file: {ReaSCAN_path_to_data}...\")\nReaSCAN_data = json.load(open(ReaSCAN_path_to_data, \"r\"))", "Reading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional/data-compositional-splits.txt...\n" ], [ "p1_test_example_filtered = []\np2_test_example_filtered = []\np3_test_example_filtered = []\nfor example in ReaSCAN_data[\"examples\"][\"test\"]:\n if example['derivation'] == \"$OBJ_0\":\n p1_test_example_filtered += [example]\n elif example['derivation'] == \"$OBJ_0 ^ $OBJ_1\":\n p2_test_example_filtered += [example]\n elif example['derivation'] == \"$OBJ_0 ^ $OBJ_1 & $OBJ_2\":\n p3_test_example_filtered += [example]\nprint(f\"p1 test example count={len(p1_test_example_filtered)}\")\nprint(f\"p2 test example count={len(p2_test_example_filtered)}\")\nprint(f\"p3 test example count={len(p3_test_example_filtered)}\")", "p1 test example count=907\np2 test example count=2122\np3 test example count=2724\n" ], [ "len(p1_test_example_filtered) + len(p2_test_example_filtered) + len(p3_test_example_filtered)", "_____no_output_____" ] ], [ [ "For instance, as you can see `p1 test example count` should be equal to `921`, but it is not. However, you can see that the total number of test examples matches up. The **root cause** potentially is that our sub-test splits are created asynchronously with the test split in the main data. \n\nBefore confirming the **root cause**, we need to first analyze what is the actual **impact** on performance numbers? Are they changing our results qualitatively? or just quantitatively? We come up with some tests around this issue starting from basic to more complex.", "_____no_output_____" ], [ "#### Test-1: Validity\nWe need to ensure our sub-test splits **only** contain commands appear in the training set. Otherwise, our test splits become compositional splits.", "_____no_output_____" ] ], [ [ "train_command_set = set([])\nfor example in ReaSCAN_data[\"examples\"][\"train\"]:\n train_command_set.add(example[\"command\"])", "_____no_output_____" ], [ "for example in p1_test_data[\"examples\"][\"test\"]:\n assert example[\"command\"] in train_command_set\nfor example in p2_test_data[\"examples\"][\"test\"]:\n assert example[\"command\"] in train_command_set\nfor example in p3_test_data[\"examples\"][\"test\"]:\n assert example[\"command\"] in train_command_set\nprint(\"Test-1 Passed\")", "Test-1 Passed\n" ] ], [ [ "#### Test-2: Overestimating?\nWhat about the shape world? Are there overlaps between train and test?", "_____no_output_____" ] ], [ [ "import hashlib\ntrain_example_hash = set([])\nfor example in ReaSCAN_data[\"examples\"][\"train\"]:\n example_hash_object = hashlib.md5(json.dumps(example).encode('utf-8'))\n train_example_hash.add(example_hash_object.hexdigest())\nassert len(train_example_hash) == len(ReaSCAN_data[\"examples\"][\"train\"])", "_____no_output_____" ], [ "p1_test_example_hash = set([])\nfor example in p1_test_data[\"examples\"][\"test\"]:\n example_hash_object = hashlib.md5(json.dumps(example).encode('utf-8'))\n p1_test_example_hash.add(example_hash_object.hexdigest())\nassert len(p1_test_example_hash) == len(p1_test_data[\"examples\"][\"test\"])\n\np2_test_example_hash = set([])\nfor example in p2_test_data[\"examples\"][\"test\"]:\n example_hash_object = hashlib.md5(json.dumps(example).encode('utf-8'))\n p2_test_example_hash.add(example_hash_object.hexdigest())\nassert len(p2_test_example_hash) == len(p2_test_data[\"examples\"][\"test\"])\n\np3_test_example_hash = set([])\nfor example in p3_test_data[\"examples\"][\"test\"]:\n example_hash_object = hashlib.md5(json.dumps(example).encode('utf-8'))\n p3_test_example_hash.add(example_hash_object.hexdigest())\nassert len(p3_test_example_hash) == len(p3_test_data[\"examples\"][\"test\"])", "_____no_output_____" ], [ "p1_test_dup_count = 0\nfor hash_str in p1_test_example_hash:\n if hash_str in train_example_hash:\n p1_test_dup_count += 1\n \np2_test_dup_count = 0\nfor hash_str in p2_test_example_hash:\n if hash_str in train_example_hash:\n p2_test_dup_count += 1\n\np3_test_dup_count = 0\nfor hash_str in p3_test_example_hash:\n if hash_str in train_example_hash:\n p3_test_dup_count += 1", "_____no_output_____" ], [ "print(f\"p1_test_dup_count={p1_test_dup_count}\")\nprint(f\"p2_test_dup_count={p2_test_dup_count}\")\nprint(f\"p3_test_dup_count={p3_test_dup_count}\")", "p1_test_dup_count=858\np2_test_dup_count=1982\np3_test_dup_count=2548\n" ], [ "main_p1_test_example_hash = set([])\nfor example in p1_test_example_filtered:\n example_hash_object = hashlib.md5(json.dumps(example).encode('utf-8'))\n main_p1_test_example_hash.add(example_hash_object.hexdigest())\nassert len(main_p1_test_example_hash) == len(p1_test_example_filtered)", "_____no_output_____" ], [ "main_p1_test_dup_count = 0\nfor hash_str in main_p1_test_example_hash:\n if hash_str in train_example_hash:\n main_p1_test_dup_count += 1", "_____no_output_____" ], [ "print(f\"main_p1_test_dup_count={main_p1_test_dup_count}\")", "main_p1_test_dup_count=0\n" ] ], [ [ "**Conclusion**: Yes. As you can see, we have many duplicated examples in our random tests. This means that, we need to use updated testing splits for evaluating performance. As a result, the **table 3** in the paper needs to be updated since it is now overestimating model performance for non-generalizing test splits (e.g., `p1`, `p2` nad `p3`).", "_____no_output_____" ], [ "**Action Required**: Need to re-evaluation model performance on those splits.", "_____no_output_____" ], [ "#### Test-3: Does this issue affect any other generalization splits?\nDoes our generalization splits containing duplicates?", "_____no_output_____" ] ], [ [ "def get_example_hash_set(split):\n split_test_path_to_data = f\"../../ReaSCAN-v1.0/ReaSCAN-compositional-{split}/data-compositional-splits.txt\"\n print(f\"Reading dataset from file: {split_test_path_to_data}...\")\n split_test_data = json.load(open(split_test_path_to_data, \"r\"))\n split_test_data_test_example_hash = set([])\n for example in split_test_data[\"examples\"][\"test\"]:\n example_hash_object = hashlib.md5(json.dumps(example).encode('utf-8'))\n split_test_data_test_example_hash.add(example_hash_object.hexdigest())\n assert len(split_test_data_test_example_hash) == len(split_test_data[\"examples\"][\"test\"])\n return split_test_data_test_example_hash\n ", "_____no_output_____" ], [ "a1_hash = get_example_hash_set(\"a1\")\na2_hash = get_example_hash_set(\"a2\")\na3_hash = get_example_hash_set(\"a3\")\n\nb1_hash = get_example_hash_set(\"b1\")\nb2_hash = get_example_hash_set(\"b2\")\n\nc1_hash = get_example_hash_set(\"c1\")\nc2_hash = get_example_hash_set(\"c2\")", "Reading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-a1/data-compositional-splits.txt...\nReading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-a2/data-compositional-splits.txt...\nReading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-a3/data-compositional-splits.txt...\nReading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-b1/data-compositional-splits.txt...\nReading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-b2/data-compositional-splits.txt...\nReading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-c1/data-compositional-splits.txt...\nReading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-c2/data-compositional-splits.txt...\n" ], [ "a1_dup_count = 0\nfor hash_str in a1_hash:\n if hash_str in train_example_hash:\n a1_dup_count += 1\na2_dup_count = 0\nfor hash_str in a2_hash:\n if hash_str in train_example_hash:\n a2_dup_count += 1\na3_dup_count = 0\nfor hash_str in a3_hash:\n if hash_str in train_example_hash:\n a3_dup_count += 1", "_____no_output_____" ], [ "print(f\"a1_dup_count={a1_dup_count}\")\nprint(f\"a2_dup_count={a2_dup_count}\")\nprint(f\"a3_dup_count={a3_dup_count}\")", "a1_dup_count=0\na2_dup_count=0\na3_dup_count=0\n" ], [ "b1_dup_count = 0\nfor hash_str in b1_hash:\n if hash_str in train_example_hash:\n b1_dup_count += 1\nb2_dup_count = 0\nfor hash_str in b2_hash:\n if hash_str in train_example_hash:\n b2_dup_count += 1", "_____no_output_____" ], [ "print(f\"b1_dup_count={b1_dup_count}\")\nprint(f\"b2_dup_count={b2_dup_count}\")", "b1_dup_count=0\nb2_dup_count=0\n" ], [ "c1_dup_count = 0\nfor hash_str in c1_hash:\n if hash_str in train_example_hash:\n c1_dup_count += 1\nc2_dup_count = 0\nfor hash_str in c2_hash:\n if hash_str in train_example_hash:\n c2_dup_count += 1", "_____no_output_____" ], [ "print(f\"c1_dup_count={c1_dup_count}\")\nprint(f\"c2_dup_count={c2_dup_count}\")", "c1_dup_count=0\nc2_dup_count=0\n" ] ], [ [ "**Conclusion**: No.", "_____no_output_____" ], [ "#### Test-4: What about correctness of generalization splits in general?\nWe see there is no duplicate, but what about general correctness? Are their created correctly? In this section, we add more sanity checks to show correctness of each generalization split.\n\nFor each split, we verify two things:\n* the generalization split can ONLY contain test examples that it is designed to test.\n* the training split DOES NOT contain examples that are aligned with the generalization split.", "_____no_output_____" ], [ "A1:novel color modifier", "_____no_output_____" ] ], [ [ "split_test_path_to_data = f\"../../ReaSCAN-v1.0/ReaSCAN-compositional-a1/data-compositional-splits.txt\"\nprint(f\"Reading dataset from file: {split_test_path_to_data}...\")\nsplit_test_data = json.load(open(split_test_path_to_data, \"r\"))", "Reading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-a1/data-compositional-splits.txt...\n" ], [ "for example in split_test_data[\"examples\"][\"test\"]:\n assert \"yellow,square\" in example[\"command\"]", "_____no_output_____" ], [ "for example in ReaSCAN_data[\"examples\"][\"train\"]:\n assert \"yellow,square\" not in example[\"command\"]", "_____no_output_____" ] ], [ [ "A2: novel color attribute", "_____no_output_____" ] ], [ [ "# this test may be a little to weak for now. maybe improve it to verify the shape world?\nsplit_test_path_to_data = f\"../../ReaSCAN-v1.0/ReaSCAN-compositional-a2/data-compositional-splits.txt\"\nprint(f\"Reading dataset from file: {split_test_path_to_data}...\")\nsplit_test_data = json.load(open(split_test_path_to_data, \"r\"))", "Reading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-a2/data-compositional-splits.txt...\n" ], [ "for example in ReaSCAN_data[\"examples\"][\"train\"]:\n assert \"red,square\" not in example[\"command\"]", "_____no_output_____" ], [ "for example in split_test_data[\"examples\"][\"test\"]:\n if \"red,square\" not in example[\"command\"]:\n # then, some background object referred in the command needs to be a red square!!\n if example[\"derivation\"] == \"$OBJ_0\":\n assert example['situation']['placed_objects']['0']['object']['shape'] == \"square\"\n assert example['situation']['placed_objects']['0']['object']['color'] == \"red\"\n elif example[\"derivation\"] == \"$OBJ_0 ^ $OBJ_1\":\n assert example['situation']['placed_objects']['0']['object']['shape'] == \"square\" or example['situation']['placed_objects']['1']['object']['shape'] == \"square\"\n assert example['situation']['placed_objects']['0']['object']['color'] == \"red\" or example['situation']['placed_objects']['1']['object']['color'] == \"red\"\n elif example[\"derivation\"] == \"$OBJ_0 ^ $OBJ_1 & $OBJ_2\":\n assert example['situation']['placed_objects']['0']['object']['shape'] == \"square\" or example['situation']['placed_objects']['1']['object']['shape'] == \"square\" or example['situation']['placed_objects']['2']['object']['shape'] == \"square\"\n assert example['situation']['placed_objects']['0']['object']['color'] == \"red\" or example['situation']['placed_objects']['1']['object']['color'] == \"red\" or example['situation']['placed_objects']['2']['object']['color'] == \"red\"\n else:\n pass", "_____no_output_____" ] ], [ [ "A3: novel size attribute", "_____no_output_____" ] ], [ [ "# this test may be a little to weak for now. maybe improve it to verify the shape world?\nsplit_test_path_to_data = f\"../../ReaSCAN-v1.0/ReaSCAN-compositional-a3/data-compositional-splits.txt\"\nprint(f\"Reading dataset from file: {split_test_path_to_data}...\")\nsplit_test_data = json.load(open(split_test_path_to_data, \"r\"))", "Reading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-a3/data-compositional-splits.txt...\n" ], [ "for example in split_test_data[\"examples\"][\"test\"]:\n assert \"small,cylinder\" in example['command'] or \\\n \"small,red,cylinder\" in example['command'] or \\\n \"small,blue,cylinder\" in example['command'] or \\\n \"small,yellow,cylinder\" in example['command'] or \\\n \"small,green,cylinder\" in example['command']", "_____no_output_____" ], [ "for example in ReaSCAN_data[\"examples\"][\"train\"]:\n assert not (\"small,cylinder\" in example['command'] or \\\n \"small,red,cylinder\" in example['command'] or \\\n \"small,blue,cylinder\" in example['command'] or \\\n \"small,yellow,cylinder\" in example['command'] or \\\n \"small,green,cylinder\" in example['command'])", "_____no_output_____" ] ], [ [ "B1: novel co-occurrence of objects", "_____no_output_____" ] ], [ [ "# this test may be a little to weak for now. maybe improve it to verify the shape world?\nsplit_test_path_to_data = f\"../../ReaSCAN-v1.0/ReaSCAN-compositional-b1/data-compositional-splits.txt\"\nprint(f\"Reading dataset from file: {split_test_path_to_data}...\")\nsplit_test_data = json.load(open(split_test_path_to_data, \"r\"))", "Reading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-b1/data-compositional-splits.txt...\n" ], [ "from collections import namedtuple, OrderedDict\nseen_command_structs = {}\nseen_concepts = {} # add in seen concepts, so we can select concepts that are seen, but new composites!\nseen_object_co = set([])\nseen_rel_co = set([])\n\nfor example_selected in ReaSCAN_data[\"examples\"][\"train\"]:\n rel_map = OrderedDict({})\n for ele in example_selected[\"relation_map\"]:\n rel_map[tuple(ele[0])] = ele[1]\n example_struct = OrderedDict({\n 'obj_pattern_map': example_selected[\"object_pattern_map\"],\n 'rel_map': rel_map,\n 'obj_map': example_selected[\"object_expression\"],\n 'grammer_pattern': example_selected['grammer_pattern'],\n 'adverb': example_selected['adverb_in_command'],\n 'verb': example_selected['verb_in_command']\n })\n obj_co = []\n for k, v in example_selected[\"object_expression\"].items():\n if v not in seen_concepts:\n seen_concepts[v] = 1\n else:\n seen_concepts[v] += 1\n obj_co += [v]\n obj_co.sort()\n seen_object_co.add(tuple(obj_co))\n \n rel_co = []\n for k, v in rel_map.items():\n if v not in seen_concepts:\n seen_concepts[v] = 1\n else:\n seen_concepts[v] += 1\n rel_co += [v]\n rel_co.sort()\n seen_rel_co.add(tuple(rel_co))", "_____no_output_____" ], [ "test_seen_command_structs = {}\ntest_seen_concepts = {} # add in seen concepts, so we can select concepts that are seen, but new composites!\ntest_seen_object_co = set([])\ntest_seen_rel_co = set([])\n\nfor example_selected in split_test_data[\"examples\"][\"test\"]:\n rel_map = OrderedDict({})\n for ele in example_selected[\"relation_map\"]:\n rel_map[tuple(ele[0])] = ele[1]\n example_struct = OrderedDict({\n 'obj_pattern_map': example_selected[\"object_pattern_map\"],\n 'rel_map': rel_map,\n 'obj_map': example_selected[\"object_expression\"],\n 'grammer_pattern': example_selected['grammer_pattern'],\n 'adverb': example_selected['adverb_in_command'],\n 'verb': example_selected['verb_in_command']\n })\n obj_co = []\n for k, v in example_selected[\"object_expression\"].items():\n if v not in test_seen_concepts:\n test_seen_concepts[v] = 1\n else:\n test_seen_concepts[v] += 1\n obj_co += [v]\n obj_co.sort()\n test_seen_object_co.add(tuple(obj_co))\n \n rel_co = []\n for k, v in rel_map.items():\n if v not in test_seen_concepts:\n test_seen_concepts[v] = 1\n else:\n test_seen_concepts[v] += 1\n rel_co += [v]\n rel_co.sort()\n test_seen_rel_co.add(tuple(rel_co))", "_____no_output_____" ], [ "test_seen_object_co.intersection(seen_object_co)", "_____no_output_____" ] ], [ [ "B2: novel co-occurrence of relations", "_____no_output_____" ] ], [ [ "# this test may be a little to weak for now. maybe improve it to verify the shape world?\nsplit_test_path_to_data = f\"../../ReaSCAN-v1.0/ReaSCAN-compositional-b2/data-compositional-splits.txt\"\nprint(f\"Reading dataset from file: {split_test_path_to_data}...\")\nsplit_test_data = json.load(open(split_test_path_to_data, \"r\"))", "Reading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-b2/data-compositional-splits.txt...\n" ], [ "test_seen_command_structs = {}\ntest_seen_concepts = {} # add in seen concepts, so we can select concepts that are seen, but new composites!\ntest_seen_object_co = set([])\ntest_seen_rel_co = set([])\n\nfor example_selected in split_test_data[\"examples\"][\"test\"]:\n rel_map = OrderedDict({})\n for ele in example_selected[\"relation_map\"]:\n rel_map[tuple(ele[0])] = ele[1]\n example_struct = OrderedDict({\n 'obj_pattern_map': example_selected[\"object_pattern_map\"],\n 'rel_map': rel_map,\n 'obj_map': example_selected[\"object_expression\"],\n 'grammer_pattern': example_selected['grammer_pattern'],\n 'adverb': example_selected['adverb_in_command'],\n 'verb': example_selected['verb_in_command']\n })\n obj_co = []\n for k, v in example_selected[\"object_expression\"].items():\n if v not in test_seen_concepts:\n test_seen_concepts[v] = 1\n else:\n test_seen_concepts[v] += 1\n obj_co += [v]\n obj_co.sort()\n test_seen_object_co.add(tuple(obj_co))\n \n rel_co = []\n for k, v in rel_map.items():\n if v not in test_seen_concepts:\n test_seen_concepts[v] = 1\n else:\n test_seen_concepts[v] += 1\n rel_co += [v]\n rel_co.sort()\n test_seen_rel_co.add(tuple(rel_co))", "_____no_output_____" ], [ "test_seen_rel_co", "_____no_output_____" ] ], [ [ "C1:novel conjunctive clause length", "_____no_output_____" ] ], [ [ "# this test may be a little to weak for now. maybe improve it to verify the shape world?\nsplit_test_path_to_data = f\"../../ReaSCAN-v1.0/ReaSCAN-compositional-c1/data-compositional-splits.txt\"\nprint(f\"Reading dataset from file: {split_test_path_to_data}...\")\nsplit_test_data = json.load(open(split_test_path_to_data, \"r\"))", "Reading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-c1/data-compositional-splits.txt...\n" ], [ "for example in split_test_data[\"examples\"][\"test\"]:\n assert example[\"derivation\"] == \"$OBJ_0 ^ $OBJ_1 & $OBJ_2 & $OBJ_3\"\n assert (example[\"command\"].count(\"and\")) == 2", "_____no_output_____" ] ], [ [ "C2:novel relative clauses", "_____no_output_____" ] ], [ [ "# this test may be a little to weak for now. maybe improve it to verify the shape world?\nsplit_test_path_to_data = f\"../../ReaSCAN-v1.0/ReaSCAN-compositional-c2/data-compositional-splits.txt\"\nprint(f\"Reading dataset from file: {split_test_path_to_data}...\")\nsplit_test_data = json.load(open(split_test_path_to_data, \"r\"))", "Reading dataset from file: ../../ReaSCAN-v1.0/ReaSCAN-compositional-c2/data-compositional-splits.txt...\n" ], [ "for example in split_test_data[\"examples\"][\"test\"]:\n assert example[\"derivation\"] == \"$OBJ_0 ^ $OBJ_1 ^ $OBJ_2\"\n assert (example[\"command\"].count(\"that,is\")) == 2", "_____no_output_____" ] ], [ [ "**Conclusion**: No.", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import time\nimport torch\nfrom torch.autograd import Variable\nimport torch.nn.functional as F\nfrom wavenet_model import *\nfrom audio_data import WavenetDataset\nfrom wavenet_training import *\nfrom model_logging import *\nfrom scipy.io import wavfile\n\ndtype = torch.FloatTensor\nltype = torch.LongTensor\n\nuse_cuda = torch.cuda.is_available()\nif use_cuda:\n print('use gpu')\n dtype = torch.cuda.FloatTensor\n ltype = torch.cuda.LongTensor\nelse: \n print(\"no gpu found\")", "no gpu found\n" ], [ "model = WaveNetModel(layers=10,\n blocks=4,\n dilation_channels=32,\n residual_channels=32,\n skip_channels=32,\n output_length=64,\n dtype=dtype)\n#model = load_latest_model_from('snapshots', use_cuda=use_cuda)\nif use_cuda:\n model.cuda()\n\nprint('model: ', model)\nprint('receptive field: ', model.receptive_field)\nprint('parameter count: ', model.parameter_count())", "model: WaveNetModel(\n (filter_convs): ModuleList(\n (0): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (1): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (2): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (3): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (4): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (5): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (6): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (7): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (8): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (9): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (10): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (11): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (12): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (13): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (14): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (15): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (16): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (17): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (18): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (19): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (20): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (21): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (22): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (23): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (24): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (25): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (26): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (27): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (28): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (29): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (30): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (31): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (32): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (33): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (34): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (35): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (36): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (37): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (38): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (39): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n )\n (gate_convs): ModuleList(\n (0): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (1): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (2): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (3): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (4): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (5): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (6): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (7): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (8): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (9): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (10): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (11): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (12): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (13): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (14): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (15): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (16): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (17): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (18): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (19): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (20): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (21): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (22): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (23): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (24): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (25): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (26): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (27): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (28): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (29): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (30): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (31): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (32): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (33): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (34): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (35): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (36): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (37): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (38): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n (39): Conv1d (32, 32, kernel_size=(2,), stride=(1,), bias=False)\n )\n (residual_convs): ModuleList(\n (0): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (1): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (2): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (3): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (4): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (5): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (6): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (7): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (8): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (9): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (10): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (11): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (12): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (13): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (14): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (15): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (16): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (17): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (18): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (19): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (20): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (21): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (22): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (23): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (24): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (25): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (26): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (27): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (28): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (29): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (30): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (31): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (32): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (33): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (34): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (35): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (36): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (37): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (38): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (39): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n )\n (skip_convs): ModuleList(\n (0): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (1): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (2): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (3): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (4): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (5): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (6): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (7): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (8): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (9): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (10): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (11): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (12): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (13): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (14): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (15): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (16): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (17): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (18): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (19): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (20): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (21): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (22): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (23): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (24): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (25): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (26): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (27): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (28): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (29): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (30): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (31): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (32): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (33): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (34): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (35): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (36): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (37): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (38): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n (39): Conv1d (32, 32, kernel_size=(1,), stride=(1,), bias=False)\n )\n (start_conv): Conv1d (1, 32, kernel_size=(1,), stride=(1,), bias=False)\n (end_conv): Conv1d (32, 256, kernel_size=(1,), stride=(1,))\n)\nreceptive field: 4093\nparameter count: 254240\n" ], [ "batch_size = 32\ninput_data = Variable(torch.zeros([batch_size, 1, model.receptive_field + model.output_length - 1]))\nprint(input_data)", "Variable containing:\n( 0 ,.,.) = \n 0 0 0 ... 0 0 0\n\n( 1 ,.,.) = \n 0 0 0 ... 0 0 0\n\n( 2 ,.,.) = \n 0 0 0 ... 0 0 0\n ... \n\n( 29 ,.,.) = \n 0 0 0 ... 0 0 0\n\n( 30 ,.,.) = \n 0 0 0 ... 0 0 0\n\n( 31 ,.,.) = \n 0 0 0 ... 0 0 0\n[torch.FloatTensor of size 32x1x4156]\n\n" ], [ "with torch.autograd.profiler.profile(enabled=True, use_cuda=True) as prof:\n out = model(input_data)\n loss = F.cross_entropy(out.squeeze(), Variable(torch.zeros([batch_size * model.output_length]).type(ltype)))\n loss.backward()\nprint(prof.key_averages().table(sort_by='cpu_time_total'))", "_____no_output_____" ], [ "prof.export_chrome_trace('profiling/latest_trace.json')", "_____no_output_____" ], [ "with torch.autograd.profiler.profile() as prof:\n model.generate_fast(num_samples=100)\nprint(prof.key_averages().table(sort_by='cpu_time_total'))", "one generating step does take approximately 0.012284069061279297 seconds)\n--------------- --------------- --------------- --------------- --------------- ---------------\nName CPU time CUDA time Calls CPU total CUDA total\n--------------- --------------- --------------- --------------- --------------- ---------------\nview 6.082us 0.000us 1 6.082us 0.000us\nsqueeze 1.857us 0.000us 100 185.680us 0.000us\nthreshold 3.185us 0.000us 100 318.454us 0.000us\ndiv 3.349us 0.000us 100 334.918us 0.000us\nsoftmax 6.903us 0.000us 100 690.307us 0.000us\nunsqueeze 1.704us 0.000us 4000 6816.902us 0.000us\nmul 2.180us 0.000us 4000 8721.320us 0.000us\nsigmoid 2.209us 0.000us 4000 8837.406us 0.000us\ntanh 2.279us 0.000us 4000 9115.436us 0.000us\nadd 2.339us 0.000us 8000 18713.918us 0.000us\ncat 5.684us 0.000us 3508 19939.675us 0.000us\nSetItem 10.601us 0.000us 4000 42403.212us 0.000us\nIndex 8.634us 0.000us 15408 133035.871us 0.000us\nConvForward 16.638us 0.000us 16200 269537.283us 0.000us\n\n" ] ] ]

[ "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "from scipy.ndimage.filters import gaussian_filter1d\nimport pandas as pd\nimport seaborn as sn\nimport sys\nsys.path.insert(0, '../build')\n\nimport pyabcranger\nimport sys\nimport elfi\nimport matplotlib.pyplot as plt\nfrom statsmodels.tsa.stattools import acf, pacf\nimport math\nimport numpy as np\n\n\ndef MAq(t, n_obs=10000, batch_size=1, random_state=None):\n # Make inputs 2d arrays for numpy broadcasting with w\n s = t.shape[1]\n assert t.shape[0] == batch_size\n random_state = random_state or np.random\n w = random_state.randn(batch_size, n_obs+s) # i.i.d. sequence ~ N(0,1)\n x = w[:, s:]\n for p in range(s):\n x = x + np.repeat(np.reshape(t[:, p], (batch_size, 1)),\n n_obs, axis=1)*w[:, (s-p-1):(-p-1)]\n return x\n\ndef generate_maq_priors(q, tq , batch_size=1, random_state=None):\n assert tq.shape[0] == batch_size\n d = q // 2\n if (q % 2) == 0:\n d = d - 1\n random_state = random_state or np.random\n nc = random_state.randint(q, size=batch_size)\n nd = random_state.randint(d, size=batch_size)\n #r = np.random.uniform(min, max, (batch_size, 1))\n genr = np.exp(random_state.dirichlet(\n np.ones(q), batch_size)*np.log(np.abs(1/tq[:,np.newaxis])))\n # genr = genr * randSign(q,(r <= 0),batch_size)\n genr[:, -1] = -genr[:, -1]\n alphas = np.zeros((batch_size, q))\n for i in range(batch_size):\n gen = random_state.uniform(0, math.pi, nd[i])\n d2 = (q - (2*nd[i])) // 2\n if (q % 2) == 0:\n d2 = d2 - 1\n nq = random_state.randint(d2)\n alphas[i, :nd[i]] = gen\n alphas[i, nd[i]:(2*nd[i])] = -gen\n alphas[i, -(2*nq+1):] = -1\n roots = np.zeros((batch_size, q), dtype=complex)\n roots.real = np.cos(alphas)\n roots.imag = np.sin(alphas)\n if (q % 2) != 0:\n roots[:, nc] = -roots[:, nc]\n roots = roots / genr\n assert np.min(np.abs(roots)) > 1, str(roots) # Prior constraint checking\n poly = np.apply_along_axis(\n np.polynomial.polynomial.polyfromroots, 1, roots).real[:, 1:]\n return poly * np.reshape(tq, (batch_size, 1))\n\nNcovmult=4\n\ndef pautocorr(x, to=1):\n C = np.zeros((x.shape[0], to*Ncovmult))\n for i in range(x.shape[0]):\n C[i, 0::Ncovmult] = acf(x[i][1:], True, nlags=to, fft=True)[1:]\n res = pacf(x[i][1:], nlags=to, method='ols', alpha=0.05)\n C[i, 1::Ncovmult] = res[0][1:]\n C[i, 2::Ncovmult] = res[1][1:, 0]\n C[i, 3::Ncovmult] = res[1][1:, 1]\n return C\n\nclass ClassPrior(elfi.Distribution):\n def rvs(n, size=1, random_state=None):\n random_state = random_state or np.random\n return random_state.choice(n,size,p=np.arange(n,0,-1)/(n*(n+1)/2))\n \nclass GlobalPrior(elfi.Distribution):\n def rvs(qp, tq, qpriors, size=1, random_state=None):\n class_count = np.zeros(qpriors.shape[0], dtype='int')\n res = np.zeros((size[0], maxt))\n for q in range(qpriors.shape[0]):\n qr = qpriors[q]\n class_count[q] = np.sum(qp == q)\n if (class_count[q] > 0):\n res[qp == q, :qr] = generate_maq_priors(\n qr, tq[qp == q], class_count[q],random_state)\n return res\n \ndef listvar(prefix, s):\n return [prefix+str(i) for i in range(1, s+1)]\n\ndef listvarautocorr(s):\n arr = []\n for i in range(1, s//Ncovmult+1):\n arr.append(\"acf\"+str(i))\n arr.append(\"pacf\"+str(i))\n arr.append(\"pacfq1_\"+str(i))\n arr.append(\"pacfq2_\"+str(i))\n return arr", "_____no_output_____" ], [ "minprior = 1\nmaxprior = 2\nntree = 500\nNy = 200 # Length of the serie\nNcov = 20 # Maximum of autocorrelation lag\nq = 10 # Our chosen q for the observed data\nnref = 2000 # Number of expected simulated data from ABC\nbatchsize = 100\n\n#qpriors = np.array([6,7,8,9,10,11,12,13,14,15,16])\nqpriors = np.arange(6,17,dtype=np.int)\nnclasses = qpriors.shape[0]\nmaxt = np.max(qpriors)", "_____no_output_____" ], [ "tq = elfi.Prior('uniform',1,1)\nqp = elfi.Prior(ClassPrior, nclasses)\nt = elfi.Prior(GlobalPrior, qp, tq, qpriors)\n\nY = elfi.Simulator(MAq, t)\nS = elfi.Summary(pautocorr, Y, Ncov)\nd = elfi.Distance('euclidean', S)\n\nelfi.set_client('multiprocessing')\nrej = elfi.Rejection(d, batch_size=batchsize, output_names=['S'])", "_____no_output_____" ], [ "from tqdm.notebook import tqdm, trange\npredicted = []\npostproba = []\n\n\nwith trange(100) as tr:\n for k in tr:\n # Generation of the observed data\n modsimple = generate_maq_priors(q, np.random.uniform(low=1.0,high=2.0,size=(1)))\n y_obs = MAq(modsimple, Ny)\n\n Y.become(elfi.Simulator(MAq,t,observed=y_obs))\n result = rej.sample(nref, quantile=1.0,bar=False)\n\n rf = pyabcranger.reftable(\n nref,\n [np.sum(result.samples['qp'] == i) for i in range(nclasses)],\n qpriors,\n listvar('t', maxt),\n listvarautocorr(result.outputs['S'].shape[1]),\n result.outputs['S'],\n result.samples['t'],\n result.samples['qp']+1\n )\n\n postres = pyabcranger.modelchoice(\n rf, S.observed[0], \"--ntree \"+str(ntree), True)\n\n tr.set_postfix(model=qpriors[postres.predicted_model])\n predicted.append(qpriors[postres.predicted_model])\n postproba.append(postres.post_proba)", "_____no_output_____" ], [ "plt.figure()\nplt.hist(predicted,np.arange(6,18),weights=postproba,align='left')\nplt.xticks(np.arange(6,17));", "_____no_output_____" ] ] ]

[ "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "### Simulating From the Null Hypothesis\n\nLoad in the data below, and follow the questions to assist with answering the quiz questions below.", "_____no_output_____" ] ], [ [ "import pandas as pd\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n%matplotlib inline\nnp.random.seed(42)\n\nfull_data = pd.read_csv('../data/coffee_dataset.csv')\nsample_data = full_data.sample(200)", "_____no_output_____" ] ], [ [ "`1.` If you were interested in if the average height for coffee drinkers is the same as for non-coffee drinkers, what would the null and alternative be? Place them in the cell below, and use your answer to answer the first quiz question below.", "_____no_output_____" ], [ "**Since there is no directional component associated with this statement, a not equal to seems most reasonable.**\n\n$$H_0: \\mu_{coff} - \\mu_{no} = 0$$\n\n\n$$H_0: \\mu_{coff} - \\mu_{no} \\neq 0$$\n\n\n**$\\mu_{coff}$ and $\\mu_{no}$ are the population mean values for coffee drinkers and non-coffee drinkers, respectivley.**", "_____no_output_____" ], [ "`2.` If you were interested in if the average height for coffee drinkers is less than non-coffee drinkers, what would the null and alternative be? Place them in the cell below, and use your answer to answer the second quiz question below.", "_____no_output_____" ], [ "**In this case, there is a question associated with a direction - that is the average height for coffee drinkers is less than non-coffee drinkers. Below is one of the ways you could write the null and alternative. Since the mean for coffee drinkers is listed first here, the alternative would suggest that this is negative.**\n\n$$H_0: \\mu_{coff} - \\mu_{no} \\geq 0$$\n\n\n$$H_0: \\mu_{coff} - \\mu_{no} < 0$$\n\n\n**$\\mu_{coff}$ and $\\mu_{no}$ are the population mean values for coffee drinkers and non-coffee drinkers, respectivley.**", "_____no_output_____" ], [ "`3.` For 10,000 iterations: bootstrap the sample data, calculate the mean height for coffee drinkers and non-coffee drinkers, and calculate the difference in means for each sample. You will want to have three arrays at the end of the iterations - one for each mean and one for the difference in means. Use the results of your sampling distribution, to answer the third quiz question below.", "_____no_output_____" ] ], [ [ "nocoff_means, coff_means, diffs = [], [], []\n\nfor _ in range(10000):\n bootsamp = sample_data.sample(200, replace = True)\n coff_mean = bootsamp[bootsamp['drinks_coffee'] == True]['height'].mean()\n nocoff_mean = bootsamp[bootsamp['drinks_coffee'] == False]['height'].mean()\n # append the info \n coff_means.append(coff_mean)\n nocoff_means.append(nocoff_mean)\n diffs.append(coff_mean - nocoff_mean) \n ", "_____no_output_____" ], [ "np.std(nocoff_means) # the standard deviation of the sampling distribution for nocoff", "_____no_output_____" ], [ "np.std(coff_means) # the standard deviation of the sampling distribution for coff", "_____no_output_____" ], [ "np.std(diffs) # the standard deviation for the sampling distribution for difference in means", "_____no_output_____" ], [ "plt.hist(nocoff_means, alpha = 0.5);\nplt.hist(coff_means, alpha = 0.5); # They look pretty normal to me!", "_____no_output_____" ], [ "plt.hist(diffs, alpha = 0.5); # again normal - this is by the central limit theorem", "_____no_output_____" ] ], [ [ "`4.` Now, use your sampling distribution for the difference in means and [the docs](https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.random.normal.html) to simulate what you would expect if your sampling distribution were centered on zero. Also, calculate the observed sample mean difference in `sample_data`. Use your solutions to answer the last questions in the quiz below.", "_____no_output_____" ], [ "** We would expect the sampling distribution to be normal by the Central Limit Theorem, and we know the standard deviation of the sampling distribution of the difference in means from the previous question, so we can use this to simulate draws from the sampling distribution under the null hypothesis. If there is truly no difference, then the difference between the means should be zero.**", "_____no_output_____" ] ], [ [ "null_vals = np.random.normal(0, np.std(diffs), 10000) # Here are 10000 draws from the sampling distribution under the null", "_____no_output_____" ], [ "plt.hist(null_vals); #Here is the sampling distribution of the difference under the null", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Iterators, Generators, and Uncertainty", "_____no_output_____" ], [ "Suppose you are working on a Python API that provides access to a real-time data stream (perhaps from an array of sensors or from a web service that handles user requests). You would like to deliver to the consumers of your API a simple but flexible abstraction that allows them to operate on new items from the stream when they choose to do so. Furthermore, you would like the API to allow users to do the following three things:\n\n* specify fall-back or default data streams (*e.g.*, if their first choice of stream is exhausted);\n* interleave items coming from multiple streams (presenting them as a single, new stream); and\n* process the items from a stream in parallel using multiprocessing.\n\nWhat abstraction should you use? How much of it must be custom-built and how much can be done using native Python features? When working with data streams, state spaces, and other abstractions that represent large or unbounded structures, it can be tempting to custom-build solutions that may become increasingly complex and difficult to maintain. Understanding a range of features that are already available in a language or its built-in libraries can help mitigate this while saving significant time and effort (both your own and that of others who build upon your work).", "_____no_output_____" ], [ "Iterators and generators are powerful tools in the Python language that have compelling applications in a number of contexts. This article reviews how they are defined, how they can be used, how they are related to one another, and how they can help you work in an elegant and flexible way with data structures and data streams of an unknown or infinite size.", "_____no_output_____" ], [ "## Iterables, Iterators, and Generators", "_____no_output_____" ], [ "When discussing Python, the terms *iterable*, *iterator*, and *generator* often appear in similar contexts or are even used interchangeably. These language features also solve similar problems. This can lead to some confusion, but there is a reason that on occasion these are conflated.", "_____no_output_____" ], [ "One way to understand the term *iterator* is that it refers to *any* Python data structure that *has an interface* that supports iteration over objects one at a time. A data structure is *iterable* if there is *at least one way* to construct an iterator that traverses it in some way. On the other hand, a *generator* is a particular kind of data structure, defined in a specific way within Python, that maintains an internal state and constructs or retrieves zero or more objects or values one at a time.", "_____no_output_____" ], [ "Thus, a generator by virtue of its characteristics can have an interface that allows it to qualify as an iterator, which consequently also makes it an iterable. In fact, all generators are iterators and iterable. However, not all iterators or iterable data structures are generators because there exist other approaches for building a Python object that possesses the kind of interface an iterator or iterable is expected to have.", "_____no_output_____" ], [ "## Iterators", "_____no_output_____" ], [ "If you want to implement an iterator data structure directly, you need to include a special method `__next__` in the class definition, which will be invoked whenever the built-in [`next`](https://docs.python.org/3/library/functions.html#next) function is applied to an instance of that data structure. The `skips` data structure below can emit every other positive integer via its definition of a `__next__` method.", "_____no_output_____" ] ], [ [ "class skips:\n def __init__(self):\n self.integer = 0\n\n def __next__(self):\n self.integer += 2\n return self.integer", "_____no_output_____" ] ], [ [ "Now it is possible to use the built-in [`next`](https://docs.python.org/3/library/functions.html#next) function to retrieve each item one at a time from an instance of the `skips` data structure.", "_____no_output_____" ] ], [ [ "ns = skips()\n[next(ns), next(ns), next(ns)]", "_____no_output_____" ] ], [ [ "The number of items over which the data structure will iterate can be limited by raising the `StopIteration` exception when more items can not (or should not) be returned.", "_____no_output_____" ] ], [ [ "class skips:\n def __init__(self, start, end):\n self.integer = start-2\n self.end = end\n\n def __next__(self):\n self.integer += 2\n if self.integer > self.end:\n raise StopIteration\n return self.integer", "_____no_output_____" ] ], [ [ "It is then the responsibility of any code that uses an instance of this iterator to catch this exception and handle it appropriately. It is worth acknowledging that this is a somewhat unusual use of a language feature normally associated with catching errors (because an iterator being exhausted is not always an error condition).", "_____no_output_____" ] ], [ [ "ns = skips(0, 10)\nwhile True:\n try:\n print(next(ns))\n except StopIteration:\n break", "0\n2\n4\n6\n8\n10\n" ] ], [ [ "## Iterables", "_____no_output_____" ], [ "In Python, there is a distinction between an *iterator* and an *iterable data structure*. This distinction is useful to maintain for a variety of reasons, including the ones below.\n\n* You may not want to clutter a data structure (as it may represent a spreadsheet, a database table, a large graph, and so on) with the state necessary to keep track of an iteration process.\n* You may want the data structure to support *multiple* iterators, either semantically (*e.g.*, iteration over rows versus over columns) or in terms of implementation (*e.g.*, breadth-first search versus depth-first search).\n* You may want to make it easy to *reset* iteration without fiddling with the internal state of a data structure instance (*i.e.*, resetting a traversal of the data structure instance could involve simply creating a fresh iterator).", "_____no_output_____" ], [ "As an example, consider a data structure `interval` that represents all positive integers in some range. Users might be allowed to obtain two different kinds of iterators for an instance of this data structure: those that iterate over only the even integers and those that iterate over only the odd integers.", "_____no_output_____" ] ], [ [ "class interval:\n def __init__(self, lower, upper):\n self.lower = lower\n self.upper = upper\n\n def evens(self):\n return skips(\n self.lower + (0 if (self.lower % 2) == 0 else 1),\n self.upper\n )\n \n def odds(self):\n return skips(\n self.lower + (0 if (self.lower % 2) == 1 else 1),\n self.upper\n )", "_____no_output_____" ] ], [ [ "The example below illustrates how an iterator returned by one of the methods in the definition of `interval` can be used.", "_____no_output_____" ] ], [ [ "ns = interval(0, 10).odds()\nwhile True: # Keep iterating and printing until exhaustion.\n try:\n print(next(ns))\n except StopIteration:\n break", "1\n3\n5\n7\n9\n" ] ], [ [ "So far in this article, the distinction between *iterators* and *iterable data structures* has been explicit for clarity. However, the convention that is supported (and sometimes expected) throughout Python is that an iterable data structure has a *single* iterator that can be used to iterate over it. This iterator is returned by a special method [`__iter__`](https://docs.python.org/3/reference/datamodel.html#object.__iter__) that is included in the class definition. In the example below, the `interval` class supports the creation of an iterator that visits every integer in the interval.", "_____no_output_____" ] ], [ [ "class every:\n def __init__(self, start, end):\n self.integer = start - 1\n self.end = end\n\n def __next__(self):\n self.integer += 1\n if self.integer > self.end:\n raise StopIteration\n return self.integer\n\nclass interval:\n def __init__(self, lower, upper):\n self.lower = lower\n self.upper = upper\n\n def __iter__(self):\n return every(self.lower, self.upper)", "_____no_output_____" ] ], [ [ "Python's built-in [`iter`](https://docs.python.org/3/library/functions.html#iter) function can be used to invoke `__iter__` for an instance of this data structure.", "_____no_output_____" ] ], [ [ "ns = iter(interval(1, 3))\nwhile True: # Keep iterating and printing until exhaustion.\n try:\n print(next(ns))\n except StopIteration:\n break", "1\n2\n3\n" ] ], [ [ "Including a definition for an `__iter__` method also makes it possible to use many of Python's built-in functions and language constructs that expect an iterable data structure. This includes functions such as [`list`](https://docs.python.org/3/library/functions.html#func-list) and [`set`](https://docs.python.org/3/library/functions.html#func-set), which use `iter` to obtain an iterator for their inputs.", "_____no_output_____" ] ], [ [ "list(interval(0, 10)), set(interval(0, 10))", "_____no_output_____" ] ], [ [ "This also includes comprehensions and `for` loops.", "_____no_output_____" ] ], [ [ "for n in interval(1, 4):\n print([k for k in interval(1, n)])", "[1]\n[1, 2]\n[1, 2, 3]\n[1, 2, 3, 4]\n" ] ], [ [ "There is nothing stopping you from making the iterator itself an iterable by having it return itself, as in the variant below.", "_____no_output_____" ] ], [ [ "class every:\n def __init__(self, start, end):\n self.integer = start - 1\n self.end = end\n\n def __next__(self):\n self.integer += 1\n if self.integer > self.end:\n raise StopIteration\n return self.integer\n\n def __iter__(self):\n return self", "_____no_output_____" ] ], [ [ "This approach ensures that there is no ambiguity (from a programmer's perspective) about what will happen when built-in functions such as `list` are applied to an instance of the data structure.", "_____no_output_____" ] ], [ [ "list(every(0, 10))", "_____no_output_____" ] ], [ [ "This practice is common and is the cause of some of the confusion and conflation that occurs between iterators and iterables. In addition to the potential for confusion, users of such a data structure must be careful to use the iterator as an iterable only once (or, alternatively, the object must reset its internal state every time `__iter__` is invoked).", "_____no_output_____" ] ], [ [ "ns = every(0, 10)\nlist(ns), list(ns) # Only returns contents the first time.", "_____no_output_____" ] ], [ [ "Nevertheless, this can also be a useful practice. Going back to the example with `evens` and `odds`, ensuring the iterators returned by these methods are also iterable means they can be fed directly into contexts that expect an iterable.", "_____no_output_____" ] ], [ [ "class skips:\n def __init__(self, start, end):\n self.integer = start - 2\n self.end = end\n\n def __next__(self):\n self.integer += 2\n if self.integer > self.end:\n raise StopIteration\n return self.integer\n\n def __iter__(self):\n return self\n\nclass interval:\n def __init__(self, lower, upper):\n self.lower = lower\n self.upper = upper\n\n def evens(self):\n return skips(\n self.lower + (0 if (self.lower % 2) == 0 else 1),\n self.upper\n )\n \n def odds(self):\n return skips(\n self.lower + (0 if (self.lower % 2) == 1 else 1),\n self.upper\n )", "_____no_output_____" ] ], [ [ "The example below illustrates how this kind of interface can be used.", "_____no_output_____" ] ], [ [ "i = interval(0, 10)\nlist(i.evens()), set(i.odds())", "_____no_output_____" ] ], [ [ "## Generators", "_____no_output_____" ], [ "Generators are data structures defined using either the `yield` statement or comprehension notation (also known as a [generator expression](https://docs.python.org/3/glossary.html#term-generator-expression)). The example below defines a generator `skips` using both approaches. ", "_____no_output_____" ] ], [ [ "def skips(start, end):\n integer = start\n while integer <= end:\n yield integer\n integer += 2\n\ndef skips(start, end):\n return (\n integer\n for integer in range(start, end)\n if (integer - start) % 2 == 0\n )", "_____no_output_____" ] ], [ [ "When it is evaluated, a generator returns an iterator (more precisely called a [generator iterator](https://docs.python.org/3/glossary.html#term-generator-iterator)). These are technically both iterators and iterables. For example, as with any iterator, `next` can be applied directly to instances of this data structure.", "_____no_output_____" ] ], [ [ "ns = skips(0, 10)\nnext(ns), next(ns), next(ns)", "_____no_output_____" ] ], [ [ "As with any iterator, exhaustion can be detected by catching the `StopIteration` exception.", "_____no_output_____" ] ], [ [ "ns = skips(0, 2)\ntry:\n next(ns), next(ns), next(ns)\nexcept StopIteration:\n print(\"Exhausted generator iterator.\")", "Exhausted generator iterator.\n" ] ], [ [ "Finally, an instance of the data structure can be used in any context that expects an iterable.", "_____no_output_____" ] ], [ [ "list(skips(0, 10))", "_____no_output_____" ] ], [ [ "It is possible to confirm that the result of evaluating `skips` is indeed a generator by checking its type.\n", "_____no_output_____" ] ], [ [ "import types\nisinstance(skips(0, 10), types.GeneratorType)", "_____no_output_____" ] ], [ [ "It is also possible to inspect its type to confirm that `skips` indeed evaluates to an iterator.", "_____no_output_____" ] ], [ [ "import collections\nisinstance(skips(0, 10), collections.abc.Iterator)", "_____no_output_____" ] ], [ [ "## Data Structures of Infinite or Unknown Size", "_____no_output_____" ], [ "Among the use cases that demonstrate how iterators/generators serve as a powerful language feature are scenarios involving data structures whose size is unknown or unbounded/infinite (such as streams, very large files, databases, and so on). You have already seen that you can define an iterable that can produce new objects or values indefinitely, so iterables are an effective way to represent and encapsulate such structures.", "_____no_output_____" ], [ "Returning to the example described at the beginning of the article, recall that you are faced with creating a Python API for working with data streams that might (or might not) run out of items that can be drawn from them. The advantages of leveraging iterables and generators should be evident at this point, so suppose you move ahead with this option and implement an iterable to represent a data stream. How can you address the three specific requirements (*i.e.*, default/fall-back streams, interleaving, and splitting for parallelism) using these features?", "_____no_output_____" ], [ "To satisfy the first requirement, you must allow a user to exhaust one iterable and then switch to another one. This is straightforward to do by constructing a generator that concatenates two iterables.", "_____no_output_____" ] ], [ [ "def concatenate(xs, ys):\n for x in xs:\n yield x\n for y in ys:\n yield y", "_____no_output_____" ] ], [ [ "Concatenating two instances of an iterable data structure is now straightforward.", "_____no_output_____" ] ], [ [ "list(concatenate(skips(0,5), skips(6,11)))", "_____no_output_____" ] ], [ [ "Notice that if the first iterable is never exhausted, the second one will never be used.", "_____no_output_____" ], [ "To address the second requirement, first consider a simpler scenario. What if you would like to \"line up\" or \"pair up\" entries in two or more iterables? You can use the built-in [`zip`](https://docs.python.org/3/library/functions.html#zip) function.", "_____no_output_____" ] ], [ [ "list(zip(skips(0,5), skips(6,11)))", "_____no_output_____" ] ], [ [ "Notice that the result of evaluating `zip` is indeed an iterator.", "_____no_output_____" ] ], [ [ "import collections\nisinstance(\n zip(skips(0,5), skips(6,11)),\n collections.abc.Iterator\n)", "_____no_output_____" ] ], [ [ "Combining `zip` with comprehension syntax, you can now define a generator that *interleaves* two iterables (switching back and forth between emitting an item from one and then the other).", "_____no_output_____" ] ], [ [ "def interleave(xs, ys):\n return (\n z \n for (x, y) in zip(xs, ys) \n for z in (x, y)\n )", "_____no_output_____" ] ], [ [ "As with concatenation, interleaving is now concise and straightforward.", "_____no_output_____" ] ], [ [ "list(interleave(skips(0,5), skips(6,11)))", "_____no_output_____" ] ], [ [ "Finally, how can you help users process items from a stream in parallel? Because you are already using iterables, users have some options available to them from the built-in [`itertools`](https://docs.python.org/3/library/itertools.html) library.", "_____no_output_____" ], [ "One option is [`islice`](https://docs.python.org/3/library/itertools.html#itertools.islice), which behaves in a similar manner to Python [slice notation](https://docs.python.org/3/library/functions.html?highlight=slice#slice) (such as `xs[0:10]` to extract the first ten entries from a list `xs`). Users can use this function to extract items in batches and (1) pass each item in a batch to its own separate thread or (2) pass batches of items to separate threads. A basic batching method is presented below.", "_____no_output_____" ] ], [ [ "from itertools import islice\ndef batch(xs, size):\n ys = list(islice(xs, 0, size))\n while len(ys) > 0:\n yield ys\n ys = list(islice(xs, 0, size))", "_____no_output_____" ] ], [ [ "Notice that this method inherits the graceful behavior of slice notation when the boundaries of the slices do not line up exactly with the number entries in the data structure instance.", "_____no_output_____" ] ], [ [ "list(batch(skips(0,21), 3))", "_____no_output_____" ] ], [ [ "Can you define a generator that returns batches of batches (*e.g.*, at most `n` batches each of size at most `k`)?", "_____no_output_____" ], [ "Another option is to use the [`tee`](https://docs.python.org/3/library/itertools.html#itertools.tee) function, which can duplicate a single iterable into multiple iterables. However, this function is really only simulating this effect by storing a large amount of auxiliary information from one of the iterables. Thus, it may use a significant amount of memory and is not safe to use with multiprocessing. It is best suited for situations in which the iterables are known to have a small number of items, as in the example below.", "_____no_output_____" ] ], [ [ "from itertools import tee\n(a, b) = tee(skips(0,11), 2)\n(list(a), list(b))", "_____no_output_____" ] ], [ [ "The example above is arguably implemented in a more clear and familiar way by simply wrapping the iterables using `list`.", "_____no_output_____" ] ], [ [ "ns = list(skips(0,11))\n(ns, ns)", "_____no_output_____" ] ], [ [ "## Further Reading", "_____no_output_____" ], [ "This article presents the relationships and distinctions between a number of related Python language constructs, and illustrates via the use case of data streams how these constructs can be leveraged. You can visit the Python Wiki if you are looking for additional discussion and examples of both [iterators](https://wiki.python.org/moin/Iterator) and [generators](https://wiki.python.org/moin/Generators). The Python documentation also contains a [Functional Programming HOWTO](https://docs.python.org/3/howto/functional.html) that discusses how iterators and generators offer new kinds of modularity and composability. Furthermore, the Python documentation entry for the built-in [itertools](https://docs.python.org/3/library/itertools.html) library contains many built-in functions and recommended patterns that are specialized to particular scenarios.", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "## TAREA GRAVEDAD", "_____no_output_____" ] ], [ [ "%matplotlib inline\nimport matplotlib.pyplot as plt\nimport numpy as np", "_____no_output_____" ], [ "m = 1 \nx_0 = .5\nx_0_dot = .1", "_____no_output_____" ], [ "t = np.linspace(0, 50, 300)", "_____no_output_____" ], [ "gravedad=np.array([9.81,2.78,8.87,3.72,22.88])\ngravedad", "_____no_output_____" ], [ "plt.figure(figsize = (7, 4))\nfor indx, g in enumerate (gravedad):\n omega_0 = np.sqrt(g/m)\n x_t = x_0 *np.cos(omega_0 *t) + (x_0_dot/omega_0) * np.sin(omega_0 *t)\n x_t_dot = -omega_0 * x_0 * np.sin(omega_0 * t) + x_0_dot * np.cos(omega_0 * t)\n plt.plot(x_t, x_t_dot/omega_0, 'ro', ms = 2)\n plt.legend(loc='best', bbox_to_anchor=(1.01, 0.5), prop={'size': 14})\n plt.scatter (x_t , (x_t_dot/omega_0), cmap = \"viridis\", label = g)\nplt.show()", "C:\\Users\\MaríaEsther\\Anaconda3\\lib\\site-packages\\matplotlib\\axes\\_axes.py:545: UserWarning: No labelled objects found. Use label='...' kwarg on individual plots.\n warnings.warn(\"No labelled objects found. \"\n" ] ] ]

[ "markdown", "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Modeling and Simulation in Python-Project 1\n\n\nDhara Patel and Corinne Wilklow \n\nCopyright 2018 Allen Downey\n\nLicense: [Creative Commons Attribution 4.0 International](https://creativecommons.org/licenses/by/4.0)\n", "_____no_output_____" ] ], [ [ "# Configure Jupyter so figures appear in the notebook\n%matplotlib inline\n\n# Configure Jupyter to display the assigned value after an assignment\n%config InteractiveShell.ast_node_interactivity='last_expr_or_assign'\n\n# import functions from the modsim library\nfrom modsim import *\n\nfrom pandas import read_html\n\nprint('done')", "done\n" ], [ "# Importing Population Data\nfilename = 'data/US_Population_data.html'\ntables = read_html(filename, header=0, index_col=0, decimal='M')\ntable = tables[3]\ntable1 = table[1910.0:2010.0]\ntable1.columns = ['population']\nprint(table1)", " population\nCensusyear \n1910.0 92228496.0\n1920.0 106021537.0\n1930.0 123202624.0\n1940.0 132164569.0\n1950.0 151325798.0\n1960.0 179323175.0\n1970.0 203211926.0\n1980.0 226545805.0\n1990.0 248709873.0\n2000.0 281421906.0\n2010.0 308745538.0\n" ] ], [ [ "<bk> \n The state: initial child population, initial United States population\n\nThe system: birth rates, child mortality rates, mature rates(birth rates 18 years prior)\n\nMetrics: annual child population", "_____no_output_____" ] ], [ [ "def plot_results(population, childseries, title):\n \"\"\"Plot the estimates and the model.\n \n population: TimeSeries of historical population data\n childseries: TimeSeries of child population estimates\n title: string\n \"\"\"\n plot(population, ':', label='US Population')\n if len(childseries):\n plot(childseries, color='gray', label='US Children')\n # plot(ratioseries, label='Ratio of children')\n decorate(xlabel='Year', \n ylabel='Population (million)',\n title=title)", "_____no_output_____" ], [ "def plot_ratio(ratioseries, title):\n \"\"\"Plot the estimates and the model.\n \n population: TimeSeries of historical population data\n childseries: TimeSeries of child population estimates\n title: string\n \"\"\"\n if len(ratioseries):\n plot(ratioseries, color='gray', label='Ratio of Children')\n # plot(ratioseries, label='Ratio of children')\n decorate(xlabel='Year', \n ylabel='Population (million)',\n title=title)", "_____no_output_____" ], [ "population = table1.population / 1e6\nchildseries = TimeSeries()\nratioseries = TimeSeries()\nplot_results(population, childseries, 'U.S. population')", "_____no_output_____" ] ], [ [ "## Why is the proportion of children in the United States decreasing?\n<bk>\n Over the past two decades, the United States population grew by about 20%. During the same time frame, the nation’s child population grew by only 5%. The population all around the world is aging, and children represent a smaller and smaller share of it. There are other countries in which this decrease is more dramatic, such as Germany or Japan which no longer have a positive natural increase in population. A decreasing proportion of children is a problem because the issue will only compound over time, until the population as a whole begins to decline. \n<bk>\nThe decreasing ratio of children could be due to several factors: declining fertility rates, an aging population, and a drop in net immigration levels. Our model focuses on the effects of fertility rates and child mortality rates on proportions of children in the US. Specifically, if we sweep birthrates and child mortality rates, what effects does that have on the population as a whole? Could changing birth rates and death rates account for the entirety of the changing demographics? We will use US Census data from 1910-2010 to compare to our results.\n", "_____no_output_____" ] ], [ [ "#sweeping both the mortality rate and the birth rate will make the model more accurate \nbirthrate = [29.06, 25.03, 19.22, 22.63, 24.86, 20.33, 15.57, 15.83, 15.08, 13.97]\ndeathrate = linspace(0.0065, 0.0031, 10)\nmaturerate = [31.5, 29.06, 25.03, 19.22, 22.63, 24.86, 20.33, 15.57, 15.83, 15.08]\nprint(birthrate)\nprint(deathrate)\nprint(maturerate)", "[29.06, 25.03, 19.22, 22.63, 24.86, 20.33, 15.57, 15.83, 15.08, 13.97]\n[0.0065 0.00612222 0.00574444 0.00536667 0.00498889 0.00461111\n 0.00423333 0.00385556 0.00347778 0.0031 ]\n[31.5, 29.06, 25.03, 19.22, 22.63, 24.86, 20.33, 15.57, 15.83, 15.08]\n" ], [ "state = State(children = 47.3, t_pop= 151325798.0/1e6, ratio = 47.3/151325798.0/1e6)", "_____no_output_____" ] ], [ [ "Parameters:", "_____no_output_____" ] ], [ [ "system = System(birthrate = birthrate,\n maturerate = maturerate,\n deathrate = deathrate,\n t_0 = 1910.0,\n t_end = 2010.0,\n state=state)", "_____no_output_____" ] ], [ [ "Our update function computes the updated state of these parameters at the end of each ten year increment. ", "_____no_output_____" ] ], [ [ "def update_func1(state, t, system):\n t_pop=151325798.0\n\n if t == 1910:\n i = int((t-1910)/10)\n else: \n i = int((t-1910)/10 - 1)\n \n mrate = system.maturerate \n brate = system.birthrate\n drate = system.deathrate\n\n \n births = brate[i]/100 * state.children #metric\n maturings = mrate[i]/100 * state.children #metric\n deaths = drate[i]/100 * state.children #metric\n\n population = state.children + births - maturings - deaths\n #print('children',children)\n \n return State(children=population)", "_____no_output_____" ] ], [ [ "To test our update function, we'll input the initial condition:", "_____no_output_____" ] ], [ [ "update_func1(state,system.t_0,system)", "_____no_output_____" ], [ "def run_simulation(state, system, update_func):\n \"\"\"Simulate the system using any update function.\n \n state: initial State object\n system: System object\n update_func: function that computes the population next year\n \n returns: TimeSeries of Ratios\n \"\"\"\n #t_pop=151325798.0\n results = TimeSeries()\n state = system.state\n results[system.t_0] = state.children\n \n \n for t in linrange(1910.0, 2020.0):\n if t%10 == 0:\n '''if t == 1910:\n i = int((t-1910)/10)\n else: \n i = int((t-1910)/10 - 1)'''\n state.children = update_func1(state, t, system)\n results[t] = state.children\n \n return results", "_____no_output_____" ], [ "print(population[1910])", "92.228496\n" ], [ "'''def update_ratio(state, t, system):\n childpop = state.children\n popu = population[t]\n \n ratio = childpop/popu \n return State(ratio = ratio)\n \ndef run_ratio(state, system, update_ratio):\n results = TimeSeries()\n results[system.t_0] = state.ratio\n \n for t in linrange(1910.0, 2020.0):\n if t%10 == 0:\n results[t] = update_ratio(state, t, system)'''", "_____no_output_____" ], [ "childseries = run_simulation(state, system, update_func1)\n\nfor t in linrange(1910.0, 2020.0):\n if t%10 == 0:\n ratioseries[t] = childseries[t]/population[t]\n\nprint(ratioseries)", "1910 0.500310\n1920 0.424573\n1930 0.350618\n1940 0.307835\n1950 0.278010\n1960 0.239825\n1970 0.202035\n1980 0.172592\n1990 0.157614\n2000 0.138243\n2010 0.124606\ndtype: float64\n" ], [ "empty = TimeSeries()\nfig1 = plot_results(population, childseries, 'Population of children in U.S.')\n", "_____no_output_____" ], [ "fig2 = plot_ratio(ratioseries, 'Ratio of children in the U.S.')", "_____no_output_____" ] ], [ [ "## Interpretation \n<bk> \n The model above uses birth rates and child mortality rates to model the decline of child population in the United States. According to the model, the population of children shrunk from 46.14 million in 1910 to 38.47 million in 2010. The first model predicts that even as the population as a whole grows, the lower birth rate will cause the population of children to decrease. The second model, showing a ratio of the overall population to the child population, allows us to conclude that a decreased birth rate over time could easily lead to the declining population that we see today. \n<bk>\nAlthough this model allows us to conclude that birth rate and mortality rate has a definite effect on the child population as a whole, it does not allow us to conclude decisively why the population of children is decreasing. \n<bk>\nIn the future, we could build a model that takes into account more parameters, such as immigration rate, and attempt to build a model that fits all facets of the data. We could also explore how different portions of the population, not just children, are aging and changing the demographics of our society. \n", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "corpus = [\n 'this is the first document',\n 'this document is the second document',\n 'and this is the third one',\n 'is this the first document',\n]", "_____no_output_____" ], [ "from gensim.models import TfidfModel\nfrom gensim.corpora import Dictionary\n\ncorp = [x.split(\" \") for x in corpus]\ndct = Dictionary(corp) ", "_____no_output_____" ], [ "for id in dct:\n print(f\"{id}: {dct[id]}\")", "0: document\n1: first\n2: is\n3: the\n4: this\n5: second\n6: and\n7: one\n8: third\n" ], [ "c = [dct.doc2bow(line) for line in corp] \n\nmodel = TfidfModel(c) ", "_____no_output_____" ], [ "\"this is a test document\".split(\" \")", "_____no_output_____" ], [ "dct.doc2bow(['EXPERT', 'WARN', 'BACKLASH', 'DONALD', 'TRUMPS', 'CHINA', 'TRADE', 'POLICIES', 'NEW', 'YORK', 'TIMES'])", "_____no_output_____" ], [ "[(dct[i[0]] + \": \" + str(i[1])) for i in model.__getitem__(dct.doc2bow(\"this is a test document\".split(\" \")), eps=-1)]", "_____no_output_____" ], [ "model.__getitem__(dct.doc2bow(\"this is a test document\".split(\" \")), eps=-1)", "_____no_output_____" ], [ "[(dct[i[0]] + \": \" + str(i[1])) for i in model.__getitem__(dct.doc2bow(\"this is test\".split(\" \")), eps=0)]", "_____no_output_____" ], [ "[(dct[i[0]] + \": \" + str(i[1])) for i in model[c[0]]]\n", "_____no_output_____" ], [ "from elasticsearch import Elasticsearch\nfrom elasticsearch import helpers\nimport spacy\nimport re\n\nclass ElasticCorpus:\n \n def __init__(self, host, port, username, password):\n self.elastic = Elasticsearch([{'host':\"localhost\" , 'port': 9200}], http_auth=(\"elastic\",\"elastic\"))\n self.space = spacy.load('en_core_web_sm')\n self.dictonary = Dictionary()\n \n def __iter__(self):\n for entry in helpers.scan(self.elastic, query={\"query\": {\"match_all\": {}}}, _source=[\"title\"], index=\"documents\", size=2000):\n text = entry[\"_source\"][\"title\"]\n text = re.sub('[^A-Za-z0-9 ]+', '', text)\n text = re.sub(' +', ' ', text)\n \n doc = self.space(text)\n tokens = [t.lemma_.upper() for t in doc if not t.is_stop]\n self.dictonary.add_documents([tokens])\n \n print(self.counter)\n \n yield self.dictonary.doc2bow(tokens)", "_____no_output_____" ], [ "ec = ElasticCorpus(\"localhost\", 9200, \"elastic\", \"elastic\")", "_____no_output_____" ], [ "model = TfidfModel(ec) ", "_____no_output_____" ], [ "from elasticsearch import Elasticsearch\nfrom elasticsearch import helpers\n\nes = Elasticsearch([{'host':\"localhost\" , 'port': 9200}], http_auth=(\"elastic\",\"elastic\"))\n\ni=0\nfor entry in helpers.scan(es, query={\"query\": {\"match_all\": {}}}, _source=[\"title\"], index=\"documents\", size=2000):\n i = i +1\n print(entry[\"_source\"][\"title\"])\n break\n \nprint(i)", "Experts Warn of Backlash in Donald Trump’s China Trade Policies - The New York Times\n1\n" ], [ "numList = [12, 13, 14, 15, 16]\nnumList[:3]", "_____no_output_____" ], [ "es.count(index='documents', body={'query': {\"match_all\": {}}})[\"count\"]", "_____no_output_____" ] ] ]

[ "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import os\nos.environ['CUDA_VISIBLE_DEVICES']='1'\nimport sys\n\nsys.path.append('../')\n\nimport csv\nimport numpy as np\nimport sys\nimport scipy.ndimage as nd\nimport json\nimport pickle\nimport torch\nimport torch.nn as nn\nimport torchvision\nfrom torch.utils.data import Dataset, DataLoader\nfrom models.resnet import *\nimport torch.optim as optim\nfrom torch.autograd import Variable\nimport torch.backends.cudnn as cudnn\nimport time\nimport math\nfrom utils.utils import AverageMeter\nfrom datasets.FattyLiverDatasets import FattyLiverClsDatasetsDiff3D \nfrom train.train_3d_cls2 import test\n\nimport torch.nn.functional as F\n\nimport scipy.ndimage as nd\nimport json\nimport pickle\nimport pandas as pd\nimport matplotlib.pyplot as plt", "_____no_output_____" ], [ "config_file = '../config/config_diff_3d.json'\nwith open(config_file,encoding='gb2312') as f:\n config = json.load(f)\nconfig", "_____no_output_____" ], [ "batch_size = 2\nnum_workers = 4\nphase = 'train'\nepochs = 10000\ndisplay = 2\ncrop_size = [16, 384, 512]", "_____no_output_____" ], [ "model = resnet34(num_classes=2, \n shortcut_type=True, \n sample_size_y=crop_size[1], \n sample_size_x=crop_size[2], sample_duration=crop_size[0])\n\n\n# pretrained_weights = '../data/experiment_0/9.model_cls2_exp1/ct_pos_recogtion_20200819110631/ct_pos_recognition_0047_best.pth'\n# pretrained_weights = '../data/experiment_0/9.model_cls2_exp1/ct_pos_recogtion_20200820103204/ct_pos_recognition_0007_best.pth'\n# pretrained_weights = '../data/experiment_0/9.model_cls2_exp1/ct_pos_recogtion_20200820135922/ct_pos_recognition_0022_best.pth'\n\npretrained_weights = '../data/z16_zhenni_Fattyliver_v3_cls2/raw_diff_Fattyliver/raw_diff_0.7647058823529411_55_Fattyliver.pth'\n# pretrained_weights = '../data/z16_zhenni_Fattyliver_v3_cls2/raw_diff_Fattyliver/raw_diff_0.7058823529411765_27_Fattyliver.pth'\n# pretrained_weights = '/home/zhangwd/code/work/FattyLiver_Solution/data/experiment_Oct_cls2/fattyliver_task_raw_diff_best.pth/fattyliver_z16_raw_diff_9.pth'\n\n# pretrained_weights = '../data/z16_zhenni_Fattyliver_v3_cls2/cut_diff_Fattyliver/cut_diff_0.7058823529411765_13_Fattyliver.pth'\n\nmodel.load_state_dict(torch.load(pretrained_weights))", "../models/resnet.py:233: UserWarning: nn.init.kaiming_normal is now deprecated in favor of nn.init.kaiming_normal_.\n m.weight = nn.init.kaiming_normal(m.weight, mode='fan_out')\n" ], [ "data_root = '../data/experiment_0/0.ori'\nconfig_test = '../data/config/config_train.txt'\ntest_ds = FattyLiverClsDatasetsDiff3D(data_root, config_test,crop_size)\ntest_dataloader = DataLoader(test_ds, batch_size=batch_size, shuffle=False,\n num_workers=num_workers, pin_memory=False)", "====> fatty liver count is:115\n" ], [ "criterion = nn.CrossEntropyLoss().cuda()\nacc, logger, tot_pred, tot_label, tot_prob = test(test_dataloader, nn.DataParallel(model).cuda(), criterion, 0, 20)\nprint(acc)\n# print(tot_prob)", "../train/train_3d_cls2.py:115: UserWarning: Implicit dimension choice for softmax has been deprecated. Change the call to include dim=X as an argument.\n tot_prob = np.append(tot_prob, F.softmax(output).cpu().detach().numpy()[:,1])\n" ], [ "def acu_curve(y,prob):\n from sklearn import metrics\n fpr,tpr,threshold = metrics.roc_curve(y,prob) ###计算真正率和假正率\n roc_auc = metrics.auc(fpr,tpr) ###计算auc的值\n \n plt.figure()\n lw = 2\n plt.figure(figsize=(10,10))\n plt.plot(fpr, tpr, color='darkorange',\n lw=lw, label='ROC curve (area = %0.3f)' % roc_auc) ###假正率为横坐标，真正率为纵坐标做曲线\n plt.plot([0, 1], [0, 1], color='navy', lw=lw, linestyle='--')\n plt.xlim([0.0, 1.0])\n plt.ylim([0.0, 1.05])\n plt.xlabel('False Positive Rate')\n plt.ylabel('True Positive Rate')\n plt.title('Receiver operating characteristic train')\n plt.legend(loc=\"lower right\")\n \n plt.show()\n\n\ndef plot_roc(y_true, y_pred, class_name='dr'):\n print('\\n====> plot {} info:\\n'.format(class_name))\n log = []\n from sklearn import metrics\n def calc_metrics_table(y_true, y_pred, thresholds):\n metrics_list = list()\n for threshold in thresholds:\n y_pred_binary = np.zeros(y_pred.shape, dtype=np.uint8)\n y_pred_binary[y_pred>threshold] = 1\n tn, fp, fn, tp = metrics.confusion_matrix(y_true, y_pred_binary).ravel()\n# print('tn:{:.3f}\\tfp:{:.3f}\\tfn:{:.3f}\\ttp:{:.3f}\\t'.format(tn, fp, fn, tp))\n accuracy = (tp+tn)/(tn+fp+fn+tp)\n sensitivity = tp/(tp+fn)\n specificity = tn/(fp+tn)\n ppv = tp/(tp+fp)\n npv = tn/(tn+fn)\n metrics_list.append([threshold, accuracy, sensitivity, specificity, ppv, npv])\n metrics_table = pd.DataFrame(np.array(metrics_list), columns=['threshold','accuracy','sensitivity','specificity','ppv','npv'])\n return metrics_table\n\n\n fpr, tpr, thres = metrics.roc_curve(y_true, y_pred)\n# print('fpr\\t\\t\\t','tpr')\n# for i in range(len(fpr)):\n# print(fpr[i],'\\t',tpr[i])\n\n auc = metrics.auc(fpr, tpr)\n\n thresholds = np.arange(0.01, 1., 0.01)\n metrics_table = calc_metrics_table(y_true, y_pred, thresholds)\n\n print('\\nAUC:%.4f\\n'% auc)\n log.append('AUC:%.4f'% auc)\n\n# plt.figure()\n# plt.title('{} roc curve'.format(class_name))\n# plt.plot(fpr, tpr, 'r')\n# plt.xlabel('fpr')\n# plt.ylabel('tpr')\n# plt.xticks(np.arange(0, 1.1, step=0.1))\n# plt.yticks(np.arange(0, 1.1, step=0.1))\n# plt.grid(ls='--')\n# plt.show()\n acu_curve(y_true, y_pred)\n\n print(metrics_table)\n log.append(metrics_table)\n \n metrics_table.to_csv('/home/zhangwd/code/work/FattyLiver_Solution/train/Train_3D_0.54.csv')\n\n return log", "_____no_output_____" ], [ "log = plot_roc(np.array(tot_label, dtype=np.float32), np.array(tot_prob), 'fatty liver classification 2')", "\n====> plot fatty liver classification 2 info:\n\n\nAUC:0.5421\n\n" ], [ "# from sklearn import metrics\n# ?metrics.roc_curve", "_____no_output_____" ] ] ]

[ "code" ]

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# Big docs example 9", "_____no_output_____" ], [ "## reasonable paragraph \n\n### reasonable sub paragraph\n\n## reasonable paragraph \n\n### reasonable sub paragraph\n\n## paragraph with long text\n\n### sub paragraph with long text\n\n## paragraph with long text\n\n### sub paragraph with long text\n\n## paragraph with long text\n\n### sub paragraph with long text\n\n## paragraph with extra super long text\n\n### sub paragraph with extra super long text\n\n## paragraph with extra super long text\n\n### sub paragraph with extra super long text\n\n## paragraph with extra super long text\n\n### sub paragraph with extra super long text\n\n## paragraph with extra super long text\n\n### sub paragraph with extra super long text\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph", "_____no_output_____" ], [ "## reasonable paragraph \n\n### reasonable sub paragraph\n\n## reasonable paragraph \n\n### reasonable sub paragraph\n\n## paragraph with long text\n\n### sub paragraph with long text\n\n## paragraph with long text\n\n### sub paragraph with long text\n\n## paragraph with long text\n\n### sub paragraph with long text\n\n## paragraph with extra super long text\n\n### sub paragraph with extra super long text\n\n## paragraph with extra super long text\n\n### sub paragraph with extra super long text\n\n## paragraph with extra super long text\n\n### sub paragraph with extra super long text\n\n## paragraph with extra super long text\n\n### sub paragraph with extra super long text\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph\n\n## paragraph\n\n### sub paragraph", "_____no_output_____" ] ] ]

[ "markdown" ]

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

[ "CNRI-Python", "RSA-MD", "Xnet", "Net-SNMP", "X11" ]

[ "CNRI-Python", "RSA-MD", "Xnet", "Net-SNMP", "X11" ]

[ "CNRI-Python", "RSA-MD", "Xnet", "Net-SNMP", "X11" ]

[ [ [ "from skimage import data\nimport napari\n", "_____no_output_____" ], [ "directory = '/Volumes/lowegrp/Data/Kristina/MDCK_90WT_10Sc_NoComp/17_07_24/pos0/density_matlab_python_comparison/s_0_before.tif'\n\nwith napari.gui_qt():\n viewer = napari.Viewer()\n viewer = napari.view_image(directory, rgb=False)\n\n ", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "empty" ] ] ]

[ "empty" ]

[ [ "empty" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Jupyter (iPython) Notebookを使って技術ノート環境を構築する方法\n\nmyenigma.hatenablog.com", "_____no_output_____" ] ], [ [ "from sympy import *", "_____no_output_____" ], [ "x=Symbol('x')", "_____no_output_____" ], [ "%matplotlib inline", "_____no_output_____" ], [ "init_printing()", "_____no_output_____" ], [ "expand((x - 3)**5)", "_____no_output_____" ] ], [ [ "しかし、下記のJupyter Notebooksのextensionをインストールすることで、\n\n様々な拡張機能が使えるようになり、\n\n画像のドラッグアンドドロップもできるようになります。\n\nインストールは、READMEにある通り、\n\n下記のコマンドでOKです。\n\n(下記のインストールをする場合は\n\nJupyter Notebookのプロセスはkillしておきましょう)\n\n\n$ pip install https://github.com/ipython-contrib/IPython-notebook-extensions/archive/master.zip –user\n\n\n続いて、Jupyter Notebookを起動し、", "_____no_output_____" ], [ "URLに/nbextensions/を追加して、移動すると\n\n下記のような拡張機能の追加画面に移動します。\n\nあとは、Drag and DropのチェックボタンをONして下さい。\n\nすると、Jupyter Notebookで画像をドラッグアンドドロップすると、\n\n画像が挿入されるようになります。\n\nちなみに画像ファイルはD＆Dすると\n\n.ipynbファイルと\n\n同じ場所にコピーされます。\n\n目次を見出し情報から自動生成する\n上記のIPython-notebook-extensionsの一機能を使うことで、\n\nMarkdownの見出し情報から、\n\n上記のように自動で目次を生成することができます。\n\n使い方としては、\n\n先ほどの/nbextensions/の設定画面で、\n\nTable Contents(2)にチェックをいれます。\n\n加えて、同じ設定画面で、\n\n“Add a Table of Contents at the top of the notebook”\n\nにチェックをいれます。\n\n最後に下記のように\n\n目次ウインドウを表示させて、\n\ntボタンを押すと、\n\n自動的に目次情報が先頭のセルに追加されます。\n\nレポートタイトルを入力する方法\n残念ながら、今のところ\n\nレポートのタイトルのようなものを\n\n入力するツールは見つかっていませんが、\n\nJupyterのMarkdownのモードでは、\n\nHTMLを入力するとマークアップされた文字を入力できます。\n\nこの機能を使うことで、レポートタイトルっぽい文字を表示できました。\n\n例えば、下記のようなhtmlを\n\nJupyterのMarkdownモードのセルに入力すると、\n\n<br />\n\n<div style=\"text-align: center;\">\n<font size=\"7\">Jupyter Report</font>\n</div>\n<br />\n<div style=\"text-align: right;\">\n<font size=\"4\">Atsushi Sakai</font>\n</div>\n\n<br />\n下記のように表示されます。\n\nPDFに出力する\n上記の方法である程度のレポート形式の技術ノートを作ることは\n\nできるはずなので、あとはPDFなどに変換すれば、\n\nプログラマ以外の他の人と資料を共有することができます。\n\nMacの場合\nJupyter NotebookのデータをPDFに変換する方法は色々ありますが、\n\nMacで一番シンプルなのは、Jupyter Notebookをブラウザで開いた状態で、\n\nCtrl-Pで、Macの標準機能でPDFに変換するのが一番簡単だと思います。\n\n下記のように詳細設定の部分で、ヘッダーとフッターに\n\n時刻や、ファイル名、ページなどを入力する設定ができます。\n\nあとは普通にPDFを出力すると、\n\n下記のようにそれなりのレポートがPDFで作成できます。\n\nJupyterでプレゼン資料を作る方法\n下記を参照下さい。\n\nmyenigma.hatenablog.com\n\nJupyterのサーバをHerokuの無料枠で構築する方法\n下記を参照下さい。\n\nmyenigma.hatenablog.com\n\nJupyter Markdown数式の入力テンプレート\n最近、数式の入力はすべてJupyterで実施していますが、\n\n複雑でいつもググっている数式コマンドをメモとして残しておきます。\n\n行列\n\\begin{equation*}\n\\begin{bmatrix}\n1 & 0 & \\cdots \\\\\n0 & 1 & \\cdots \\\\\n\\vdots & \\vdots & \\ddots \\\\\n-1 & 0 & \\cdots \\\\\n0 & -1 & \\cdots \\\\\n\\vdots & \\vdots & \\ddots \\\\\n\\end{bmatrix}\n\\end{equation*}\n数式参考リンク\nできれば改善してもらいたい部分\nもう少しで完全にWordをおさらばできそうなので、\n\n下記のような機能を実現してもらいたいですね。\n\n参考文献管理\n\n図のタイトル入力\n\n自動図番号割り当て\n\n上記のような機能を実現できる拡張などを知っている方は\n\nコメントなどで教えて頂けると幸いです。\n\n(Python使えるから、自分で作ればいいのか。。。)\n\nvimユーザのためのJupyterプラグイン\n下記を参照下さい\n\nlambdalisue.hatenablog.com\n\nなにかおかしいので云々。\nbookmark", "_____no_output_____" ], [ "上記のプラグインを使うことで、\n\nJupyterのエディタの部分でvimのキーバインドが使えます。\n\nJupyter Notebook上でコードの処理時間を計測する\nJupyter notebookでは、\n\n%timeit\n\nの後にコードを記述すると、そのコードの計算時間を計測してくれます。\n\nコードやアルゴリズムのレポートに便利ですね。\n\nmoqada.hatenablog.com\n\nJupyter上でモジュールや関数のdocstringを確認する\nJupyter notebookでは、\n\nモジュールや関数の後ろに?(はてな)をつけると、\n\ndocstringを表示してくれるので、\n\nマニュアル確認も簡単です。\n\n更にJupyterを学びたい人は\n下記の資料がおすすめです。\n\n\n\nまた、どうしても解決できない問題などが\n\nある場合は下記のようなQ＆Aサイトで質問してみると、\n\nかなりの確率で回答がもらえると思います。\n\n参考資料\nmyenigma.hatenablog.com\n\nmyenigma.hatenablog.com\n\nmyenigma.hatenablog.com\n\nmyenigma.hatenablog.com\n\nmyenigma.hatenablog.com\n\nMyEnigma Supporters\nもしこの記事が参考になり、\n\nブログをサポートしたいと思われた方は、\n\nこちらからよろしくお願いします。", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<a href=\"http://cocl.us/pytorch_link_top\">\n <img src=\"https://s3-api.us-geo.objectstorage.softlayer.net/cf-courses-data/CognitiveClass/DL0110EN/notebook_images%20/Pytochtop.png\" width=\"750\" alt=\"IBM Product \" />\n</a> ", "_____no_output_____" ], [ "<img src=\"https://s3-api.us-geo.objectstorage.softlayer.net/cf-courses-data/CognitiveClass/DL0110EN/notebook_images%20/cc-logo-square.png\" width=\"200\" alt=\"cognitiveclass.ai logo\" />", "_____no_output_____" ], [ "<h1>Linear Regression Multiple Outputs</h1> ", "_____no_output_____" ], [ "<h2>Table of Contents</h2>\n<p>In this lab, you will create a model the PyTroch way. This will help you more complicated models.</p>\n\n<ul>\n <li><a href=\"#Makeup_Data\">Make Some Data</a></li>\n <li><a href=\"#Model_Cost\">Create the Model and Cost Function the PyTorch way</a></li>\n <li><a href=\"#BGD\">Train the Model: Batch Gradient Descent</a></li>\n</ul>\n<p>Estimated Time Needed: <strong>20 min</strong></p>\n\n<hr>", "_____no_output_____" ], [ "<h2>Preparation</h2>", "_____no_output_____" ], [ "We'll need the following libraries:", "_____no_output_____" ] ], [ [ "# Import the libraries we need for this lab\n\nfrom torch import nn,optim\nimport torch\nimport numpy as np\nimport matplotlib.pyplot as plt\n\nfrom mpl_toolkits.mplot3d import Axes3D\nfrom torch.utils.data import Dataset, DataLoader", "_____no_output_____" ] ], [ [ "Set the random seed:", "_____no_output_____" ] ], [ [ "# Set the random seed to 1. \n\ntorch.manual_seed(1)", "_____no_output_____" ] ], [ [ "Use this function for plotting: ", "_____no_output_____" ] ], [ [ "# The function for plotting 2D\n\ndef Plot_2D_Plane(model, dataset, n=0):\n w1 = model.state_dict()['linear.weight'].numpy()[0][0]\n w2 = model.state_dict()['linear.weight'].numpy()[0][1]\n b = model.state_dict()['linear.bias'].numpy()\n\n # Data\n x1 = data_set.x[:, 0].view(-1, 1).numpy()\n x2 = data_set.x[:, 1].view(-1, 1).numpy()\n y = data_set.y.numpy()\n\n # Make plane\n X, Y = np.meshgrid(np.arange(x1.min(), x1.max(), 0.05), np.arange(x2.min(), x2.max(), 0.05))\n yhat = w1 * X + w2 * Y + b\n\n # Plotting\n fig = plt.figure()\n ax = fig.gca(projection='3d')\n\n ax.plot(x1[:, 0], x2[:, 0], y[:, 0],'ro', label='y') # Scatter plot\n \n ax.plot_surface(X, Y, yhat) # Plane plot\n \n ax.set_xlabel('x1 ')\n ax.set_ylabel('x2 ')\n ax.set_zlabel('y')\n plt.title('estimated plane iteration:' + str(n))\n ax.legend()\n\n plt.show()", "_____no_output_____" ] ], [ [ "<!--Empty Space for separating topics-->", "_____no_output_____" ], [ "<h2 id=\" #Makeup_Data\" > Make Some Data </h2>", "_____no_output_____" ], [ "Create a dataset class with two-dimensional features:", "_____no_output_____" ] ], [ [ "# Create a 2D dataset\n\nclass Data2D(Dataset):\n \n # Constructor\n def __init__(self):\n self.x = torch.zeros(20, 2)\n self.x[:, 0] = torch.arange(-1, 1, 0.1)\n self.x[:, 1] = torch.arange(-1, 1, 0.1)\n self.w = torch.tensor([[1.0], [1.0]])\n self.b = 1\n self.f = torch.mm(self.x, self.w) + self.b \n self.y = self.f + 0.1 * torch.randn((self.x.shape[0],1))\n self.len = self.x.shape[0]\n\n # Getter\n def __getitem__(self, index): \n return self.x[index], self.y[index]\n \n # Get Length\n def __len__(self):\n return self.len", "_____no_output_____" ] ], [ [ "Create a dataset object:", "_____no_output_____" ] ], [ [ "# Create the dataset object\n\ndata_set = Data2D()", "_____no_output_____" ] ], [ [ "<h2 id=\"Model_Cost\">Create the Model, Optimizer, and Total Loss Function (Cost)</h2>", "_____no_output_____" ], [ "Create a customized linear regression module: ", "_____no_output_____" ] ], [ [ "# Create a customized linear\n\nclass linear_regression(nn.Module):\n \n # Constructor\n def __init__(self, input_size, output_size):\n super(linear_regression, self).__init__()\n self.linear = nn.Linear(input_size, output_size)\n \n # Prediction\n def forward(self, x):\n yhat = self.linear(x)\n return yhat", "_____no_output_____" ] ], [ [ "Create a model. Use two features: make the input size 2 and the output size 1: ", "_____no_output_____" ] ], [ [ "# Create the linear regression model and print the parameters\n\nmodel = linear_regression(2,1)\nprint(\"The parameters: \", list(model.parameters()))", "The parameters: [Parameter containing:\ntensor([[ 0.6209, -0.1178]], requires_grad=True), Parameter containing:\ntensor([0.3026], requires_grad=True)]\n" ] ], [ [ "Create an optimizer object. Set the learning rate to 0.1. <b>Don't forget to enter the model parameters in the constructor.</b>", "_____no_output_____" ], [ "<img src=\"https://s3-api.us-geo.objectstorage.softlayer.net/cf-courses-data/CognitiveClass/DL0110EN/notebook_images%20/chapter2/2.6.2paramater_hate.png\" width = \"100\" alt=\"How the optimizer works\" />", "_____no_output_____" ] ], [ [ "# Create the optimizer\n\noptimizer = optim.SGD(model.parameters(), lr=0.1)", "_____no_output_____" ] ], [ [ "Create the criterion function that calculates the total loss or cost:", "_____no_output_____" ] ], [ [ "# Create the cost function\n\ncriterion = nn.MSELoss()", "_____no_output_____" ] ], [ [ "Create a data loader object. Set the batch_size equal to 2: ", "_____no_output_____" ] ], [ [ "# Create the data loader\n\ntrain_loader = DataLoader(dataset=data_set, batch_size=2)", "_____no_output_____" ] ], [ [ "<!--Empty Space for separating topics-->", "_____no_output_____" ], [ "<h2 id=\"BGD\">Train the Model via Mini-Batch Gradient Descent</h2>", "_____no_output_____" ], [ "Run 100 epochs of Mini-Batch Gradient Descent and store the total loss or cost for every iteration. Remember that this is an approximation of the true total loss or cost:", "_____no_output_____" ] ], [ [ "# Train the model\n\nLOSS = []\nprint(\"Before Training: \")\nPlot_2D_Plane(model, data_set) \nepochs = 100\n \ndef train_model(epochs): \n for epoch in range(epochs):\n for x,y in train_loader:\n yhat = model(x)\n loss = criterion(yhat, y)\n LOSS.append(loss.item())\n optimizer.zero_grad()\n loss.backward()\n optimizer.step() \ntrain_model(epochs)\nprint(\"After Training: \")\nPlot_2D_Plane(model, data_set, epochs) ", "Before Training: \n" ], [ "# Plot out the Loss and iteration diagram\n\nplt.plot(LOSS)\nplt.xlabel(\"Iterations \")\nplt.ylabel(\"Cost/total loss \")", "_____no_output_____" ] ], [ [ "<h3>Practice</h3>", "_____no_output_____" ], [ "Create a new <code>model1</code>. Train the model with a batch size 30 and learning rate 0.1, store the loss or total cost in a list <code>LOSS1</code>, and plot the results.", "_____no_output_____" ] ], [ [ "# Practice create model1. Train the model with batch size 30 and learning rate 0.1, store the loss in a list <code>LOSS1</code>. Plot the results.\n\ndata_set = Data2D()", "_____no_output_____" ] ], [ [ "Double-click <b>here</b> for the solution.\n\n<!-- Your answer is below:\ntrain_loader = DataLoader(dataset = data_set, batch_size = 30)\nmodel1 = linear_regression(2, 1)\noptimizer = optim.SGD(model1.parameters(), lr = 0.1)\nLOSS1 = []\nepochs = 100\ndef train_model(epochs): \n for epoch in range(epochs):\n for x,y in train_loader:\n yhat = model1(x)\n loss = criterion(yhat,y)\n LOSS1.append(loss.item())\n optimizer.zero_grad()\n loss.backward()\n optimizer.step() \ntrain_model(epochs)\nPlot_2D_Plane(model1 , data_set) \nplt.plot(LOSS1)\nplt.xlabel(\"iterations \")\nplt.ylabel(\"Cost/total loss \")\n-->", "_____no_output_____" ], [ "Use the following validation data to calculate the total loss or cost for both models:", "_____no_output_____" ] ], [ [ "torch.manual_seed(2)\n\nvalidation_data = Data2D()\nY = validation_data.y\nX = validation_data.x", "_____no_output_____" ] ], [ [ "Double-click <b>here</b> for the solution.\n<!-- Your answer is below:\nprint(\"total loss or cost for model: \",criterion(model(X),Y))\nprint(\"total loss or cost for model: \",criterion(model1(X),Y))\n-->", "_____no_output_____" ], [ "<!--Empty Space for separating topics-->", "_____no_output_____" ], [ "<a href=\"http://cocl.us/pytorch_link_bottom\">\n <img src=\"https://s3-api.us-geo.objectstorage.softlayer.net/cf-courses-data/CognitiveClass/DL0110EN/notebook_images%20/notebook_bottom%20.png\" width=\"750\" alt=\"PyTorch Bottom\" />\n</a>", "_____no_output_____" ], [ "<h2>About the Authors:</h2> \n\n<a href=\"https://www.linkedin.com/in/joseph-s-50398b136/\">Joseph Santarcangelo</a> has a PhD in Electrical Engineering, his research focused on using machine learning, signal processing, and computer vision to determine how videos impact human cognition. Joseph has been working for IBM since he completed his PhD. ", "_____no_output_____" ], [ "Other contributors: <a href=\"https://www.linkedin.com/in/michelleccarey/\">Michelle Carey</a>, <a href=\"www.linkedin.com/in/jiahui-mavis-zhou-a4537814a\">Mavis Zhou</a>", "_____no_output_____" ], [ "<hr>", "_____no_output_____" ], [ "Copyright &copy; 2018 <a href=\"cognitiveclass.ai?utm_source=bducopyrightlink&utm_medium=dswb&utm_campaign=bdu\">cognitiveclass.ai</a>. This notebook and its source code are released under the terms of the <a href=\"https://bigdatauniversity.com/mit-license/\">MIT License</a>.", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "from candidates.models import SavedCandidate\nfrom jobs.models import Job\nfrom groups.models import Group", "_____no_output_____" ] ], [ [ "## Creating groups\nWe are going to create a simple group.", "_____no_output_____" ] ], [ [ "group = Group.objects.create(name='My First Group')\ngroup.pk", "_____no_output_____" ] ], [ [ "Now lets create a group that is a child of the first group.", "_____no_output_____" ] ], [ [ "group_child = Group.objects.create(name='Child of (My First Group)', parent_group=group)", "_____no_output_____" ], [ "group_child.parent_group.name", "_____no_output_____" ] ], [ [ "### Creating jobs\nNow that we have a few groups lets create some jobs to add to the groups.", "_____no_output_____" ] ], [ [ "job_1 = Job.objects.create(title='Job 1')\njob_2 = Job.objects.create(title='Job 2')", "_____no_output_____" ] ], [ [ "### Adding jobs to a group", "_____no_output_____" ] ], [ [ "group.jobs.add(job_1)\ngroup_child.jobs.add(job_2)", "_____no_output_____" ] ], [ [ "### Ok now lets add some saved candidates to a new group", "_____no_output_____" ] ], [ [ "candidate_1 = SavedCandidate.objects.create(name='Candidate 1')\ncandidate_2 = SavedCandidate.objects.create(name='Candidate 2')\ngroup_2 = Group.objects.create(name='Group 2')\ngroup_2_child = Group.objects.create(name='Group 2 Child', parent_group=group_2)\ngroup_2_child.saved_candidates.add(candidate_1)\ngroup_2_child.saved_candidates.add(candidate_2)", "_____no_output_____" ] ], [ [ "Lets loop all the groups and display there names, jobs and saved candiates for each.", "_____no_output_____" ] ], [ [ "for group in Group.objects.all():\n print(\"Group: {}\".format(group.name))\n print(\"jobs: {}\".format(group.jobs.count()))\n for job in group.jobs.all():\n print(job.title)\n \n print(\"saved candidates: {}\".format(group.saved_candidates.count()))\n for candidate in group.saved_candidates.all():\n print(candidate.name)\n print(\"\\n\")", "Group: My First Group\njobs: 1\nJob 1\nsaved candidates: 0\n\n\nGroup: Child of (My First Group)\njobs: 1\nJob 2\nsaved candidates: 0\n\n\nGroup: Group 2\njobs: 0\nsaved candidates: 0\n\n\nGroup: Group 2 Child\njobs: 0\nsaved candidates: 2\nCandidate 1\nCandidate 2\n\n\nGroup: My First Group\njobs: 1\nJob 1\nsaved candidates: 0\n\n\nGroup: Child of (My First Group)\njobs: 1\nJob 2\nsaved candidates: 0\n\n\nGroup: Group 2\njobs: 0\nsaved candidates: 0\n\n\nGroup: Group 2 Child\njobs: 0\nsaved candidates: 2\nCandidate 1\nCandidate 2\n\n\nGroup: My First Group\njobs: 1\nJob 1\nsaved candidates: 0\n\n\nGroup: Child of (My First Group)\njobs: 1\nJob 2\nsaved candidates: 0\n\n\nGroup: Group 2\njobs: 0\nsaved candidates: 0\n\n\nGroup: Group 2 Child\njobs: 0\nsaved candidates: 0\n\n\nGroup: My First Group\njobs: 1\nJob 1\nsaved candidates: 0\n\n\nGroup: Child of (My First Group)\njobs: 1\nJob 2\nsaved candidates: 0\n\n\nGroup: Group 2\njobs: 0\nsaved candidates: 0\n\n\nGroup: Group 2 Child\njobs: 0\nsaved candidates: 0\n\n\nGroup: My First Group\njobs: 1\nJob 1\nsaved candidates: 0\n\n\nGroup: Child of (My First Group)\njobs: 1\nJob 2\nsaved candidates: 0\n\n\nGroup: Group 2\njobs: 0\nsaved candidates: 0\n\n\nGroup: Group 2 Child\njobs: 0\nsaved candidates: 2\nCandidate 1\nCandidate 2\n\n\n" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Title: Alert Investigation (Windows Process Alerts)\n\n**Notebook Version:** 1.0<br>\n**Python Version:** Python 3.6 (including Python 3.6 - AzureML)<br>\n**Required Packages**: kqlmagic, msticpy, pandas, numpy, matplotlib, networkx, ipywidgets, ipython, scikit_learn<br>\n**Platforms Supported**:<br>\n- Azure Notebooks Free Compute\n- Azure Notebooks DSVM\n- OS Independent\n\n**Data Sources Required**:<br>\n- Log Analytics - SecurityAlert, SecurityEvent (EventIDs 4688 and 4624/25)\n- (Optional) - VirusTotal (with API key)\n\n## Description:\nThis notebook is intended for triage and investigation of security alerts. It is specifically targeted at alerts triggered by suspicious process activity on Windows hosts. Some of the sections will work on other types of alerts but this is not guaranteed.\n", "_____no_output_____" ], [ "<a id='toc'></a>\n## Table of Contents\n- [Setup and Authenticate](#setup)\n\n- [Get Alerts List](#getalertslist)\n- [Choose an Alert to investigate](#enteralertid)\n - [Extract Properties and entities from alert](#extractalertproperties)\n - [Entity Graph](#entitygraph)\n- [Related Alerts](#related_alerts)\n- [Session Process Tree](#processtree)\n - [Process Timeline](#processtimeline)\n- [Other Process on Host](#process_clustering)\n- [Check for IOCs in Commandline](#cmdlineiocs)\n - [VirusTotal lookup](#virustotallookup)\n- [Alert command line - Occurrence on other hosts in subscription](#cmdlineonotherhosts)\n- [Host Logons](#host_logons)\n - [Alert Account](#logonaccount)\n - [Failed Logons](#failed_logons)\n- [Appendices](#appendices)\n - [Saving data to Excel](#appendices)\n", "_____no_output_____" ], [ "<a id='setup'></a>[Contents](#toc)\n# Setup\n\n1. Make sure that you have installed packages specified in the setup (uncomment the lines to execute)\n2. There are some manual steps up to selecting the alert ID. After this most of the notebook can be executed sequentially\n3. Major sections should be executable independently (e.g. Alert Command line and Host Logons can be run skipping Session Process Tree)\n\n## Install Packages\nThe first time this cell runs for a new Azure Notebooks project or local Python environment it will take several minutes to download and install the packages. In subsequent runs it should run quickly and confirm that package dependencies are already installed. Unless you want to upgrade the packages you can feel free to skip execution of the next cell.\n\nIf you see any import failures (```ImportError```) in the notebook, please re-run this cell and answer 'y', then re-run the cell where the failure occurred. \n\nNote you may see some warnings about package incompatibility with certain packages. This does not affect the functionality of this notebook but you may need to upgrade the packages producing the warnings to a more recent version.", "_____no_output_____" ] ], [ [ "import sys\nimport warnings\n\nwarnings.filterwarnings(\"ignore\",category=DeprecationWarning)\n\nMIN_REQ_PYTHON = (3,6)\nif sys.version_info < MIN_REQ_PYTHON:\n print('Check the Kernel->Change Kernel menu and ensure that Python 3.6')\n print('or later is selected as the active kernel.')\n sys.exit(\"Python %s.%s or later is required.\\n\" % MIN_REQ_PYTHON)\n \n# Package Installs - try to avoid if they are already installed\ntry:\n import msticpy.sectools as sectools\n import Kqlmagic\n print('If you answer \"n\" this cell will exit with an error in order to avoid the pip install calls,')\n print('This error can safely be ignored.')\n resp = input('msticpy and Kqlmagic packages are already loaded. Do you want to re-install? (y/n)')\n if resp.strip().lower() != 'y':\n sys.exit('pip install aborted - you may skip this error and continue.')\n else:\n print('After installation has completed, restart the current kernel and run '\n 'the notebook again skipping this cell.')\nexcept ImportError:\n pass\n\nprint('\\nPlease wait. Installing required packages. This may take a few minutes...')\n!pip install git+https://github.com/microsoft/msticpy --upgrade --user\n!pip install Kqlmagic --no-cache-dir --upgrade --user\n\nprint('\\nTo ensure that the latest versions of the installed libraries '\n 'are used, please restart the current kernel and run '\n 'the notebook again skipping this cell.')", "_____no_output_____" ] ], [ [ "### Import Python Packages", "_____no_output_____" ], [ "### Get WorkspaceId\nTo find your Workspace Id go to [Log Analytics](https://ms.portal.azure.com/#blade/HubsExtension/Resources/resourceType/Microsoft.OperationalInsights%2Fworkspaces). Look at the workspace properties to find the ID.", "_____no_output_____" ] ], [ [ "# Imports\nimport sys\nMIN_REQ_PYTHON = (3,6)\nif sys.version_info < MIN_REQ_PYTHON:\n print('Check the Kernel->Change Kernel menu and ensure that Python 3.6')\n print('or later is selected as the active kernel.')\n sys.exit(\"Python %s.%s or later is required.\\n\" % MIN_REQ_PYTHON)\n\nimport numpy as np\nfrom IPython import get_ipython\nfrom IPython.display import display, HTML, Markdown\nimport ipywidgets as widgets\n\nimport matplotlib.pyplot as plt\nimport seaborn as sns\nimport networkx as nx\nsns.set()\nimport pandas as pd\npd.set_option('display.max_rows', 500)\npd.set_option('display.max_columns', 50)\npd.set_option('display.max_colwidth', 100)\n \nimport msticpy.sectools as sectools\nimport msticpy.nbtools as mas\nimport msticpy.nbtools.kql as qry\nimport msticpy.nbtools.nbdisplay as nbdisp\n\n# Some of our dependencies (networkx) still use deprecated Matplotlib\n# APIs - we can't do anything about it so suppress them from view\nfrom matplotlib import MatplotlibDeprecationWarning\nwarnings.simplefilter(\"ignore\", category=MatplotlibDeprecationWarning)\n", "_____no_output_____" ], [ "import os\nfrom msticpy.nbtools.wsconfig import WorkspaceConfig\nws_config_file = 'config.json'\n\nWORKSPACE_ID = None\nTENANT_ID = None\ntry:\n ws_config = WorkspaceConfig(ws_config_file)\n display(Markdown(f'Read Workspace configuration from local config.json for workspace **{ws_config[\"workspace_name\"]}**'))\n for cf_item in ['tenant_id', 'subscription_id', 'resource_group', 'workspace_id', 'workspace_name']:\n display(Markdown(f'**{cf_item.upper()}**: {ws_config[cf_item]}'))\n \n if ('cookiecutter' not in ws_config['workspace_id'] or\n 'cookiecutter' not in ws_config['tenant_id']):\n WORKSPACE_ID = ws_config['workspace_id']\n TENANT_ID = ws_config['tenant_id']\nexcept:\n pass\n\nif not WORKSPACE_ID or not TENANT_ID:\n display(Markdown('**Workspace configuration not found.**\\n\\n'\n 'Please go to your Log Analytics workspace, copy the workspace ID'\n ' and/or tenant Id and paste here.<br> '\n 'Or read the workspace_id from the config.json in your Azure Notebooks project.'))\n ws_config = None\n ws_id = mas.GetEnvironmentKey(env_var='WORKSPACE_ID',\n prompt='Please enter your Log Analytics Workspace Id:', auto_display=True)\n ten_id = mas.GetEnvironmentKey(env_var='TENANT_ID',\n prompt='Please enter your Log Analytics Tenant Id:', auto_display=True)\n\n ", "_____no_output_____" ] ], [ [ "### Authenticate to Log Analytics\nIf you are using user/device authentication, run the following cell. \n- Click the 'Copy code to clipboard and authenticate' button.\n- This will pop up an Azure Active Directory authentication dialog (in a new tab or browser window). The device code will have been copied to the clipboard. \n- Select the text box and paste (Ctrl-V/Cmd-V) the copied value. \n- You should then be redirected to a user authentication page where you should authenticate with a user account that has permission to query your Log Analytics workspace.\n\nUse the following syntax if you are authenticating using an Azure Active Directory AppId and Secret:\n```\n%kql loganalytics://tenant(aad_tenant).workspace(WORKSPACE_ID).clientid(client_id).clientsecret(client_secret)\n```\ninstead of\n```\n%kql loganalytics://code().workspace(WORKSPACE_ID)\n```\n\nNote: you may occasionally see a JavaScript error displayed at the end of the authentication - you can safely ignore this.<br>\nOn successful authentication you should see a ```popup schema``` button.", "_____no_output_____" ] ], [ [ "if not WORKSPACE_ID or not TENANT_ID:\n try:\n WORKSPACE_ID = ws_id.value\n TENANT_ID = ten_id.value\n except NameError:\n raise ValueError('No workspace or Tenant Id.')\n\nmas.kql.load_kql_magic()\n%kql loganalytics://code().tenant(TENANT_ID).workspace(WORKSPACE_ID)", "_____no_output_____" ] ], [ [ "<a id='getalertslist'></a>[Contents](#toc)\n# Get Alerts List\n\nSpecify a time range to search for alerts. One this is set run the following cell to retrieve any alerts in that time window.\nYou can change the time range and re-run the queries until you find the alerts that you want.", "_____no_output_____" ] ], [ [ "alert_q_times = mas.QueryTime(units='day', max_before=20, before=5, max_after=1)\nalert_q_times.display()", "_____no_output_____" ], [ "alert_counts = qry.list_alerts_counts(provs=[alert_q_times])\nalert_list = qry.list_alerts(provs=[alert_q_times])\nprint(len(alert_counts), ' distinct alert types')\nprint(len(alert_list), ' distinct alerts')\ndisplay(HTML('<h2>Alert Timeline</h2>'))\nnbdisp.display_timeline(data=alert_list, source_columns = ['AlertName', 'CompromisedEntity'], title='Alerts', height=200)\ndisplay(HTML('<h2>Top alerts</h2>'))\nalert_counts.head(20) # remove '.head(20)'' to see the full list grouped by AlertName", "12 distinct alert types\n51 distinct alerts\n" ] ], [ [ "<a id='enteralertid'></a>[Contents](#toc)\n# Choose Alert to Investigate\nEither pick an alert from a list of retrieved alerts or paste the SystemAlertId into the text box in the following section.", "_____no_output_____" ], [ "### Select alert from list\nAs you select an alert, the main properties will be shown below the list.\n\nUse the filter box to narrow down your search to any substring in the AlertName.", "_____no_output_____" ] ], [ [ "alert_select = mas.AlertSelector(alerts=alert_list, action=nbdisp.display_alert)\nalert_select.display()", "_____no_output_____" ] ], [ [ "### Or paste in an alert ID and fetch it\n**Skip this if you selected from the above list**", "_____no_output_____" ] ], [ [ "# Allow alert to be selected\n# Allow subscription to be selected\nget_alert = mas.GetSingleAlert(action=nbdisp.display_alert)\nget_alert.display()", "_____no_output_____" ] ], [ [ "<a id='extractalertproperties'></a>[Contents](#toc)\n## Extract properties and entities from Alert\nThis section extracts the alert information and entities into a SecurityAlert object allowing us to query the properties more reliably. \n\nIn particular, we use the alert to automatically provide parameters for queries and UI elements.\nSubsequent queries will use properties like the host name and derived properties such as the OS family (Linux or Windows) to adapt the query. Query time selectors like the one above will also default to an origin time that matches the alert selected.\n\nThe alert view below shows all of the main properties of the alert plus the extended property dictionary (if any) and JSON representations of the Entity.", "_____no_output_____" ] ], [ [ "# Extract entities and properties into a SecurityAlert class\nif alert_select.selected_alert is None and get_alert.selected_alert is None:\n sys.exit(\"Please select an alert before executing remaining cells.\")\n\nif get_alert.selected_alert is not None:\n security_alert = mas.SecurityAlert(get_alert.selected_alert)\nelif alert_select.selected_alert is not None:\n security_alert = mas.SecurityAlert(alert_select.selected_alert)\n \nmas.disp.display_alert(security_alert, show_entities=True)", "_____no_output_____" ] ], [ [ "<a id='entitygraph'></a>[Contents](#toc)\n## Entity Graph\nDepending on the type of alert there may be one or more entities attached as properties. Entities are things like Host, Account, IpAddress, Process, etc. - essentially the 'nouns' of security investigation. Events and alerts are the things that link them in actions so can be thought of as the verbs. Entities are often related to other entities - for example a process will usually have a related file entity (the process image) and an Account entity (the context in which the process was running). Endpoint alerts typically always have a host entity (which could be a physical or virtual machine).", "_____no_output_____" ], [ "### Plot using Networkx/Matplotlib", "_____no_output_____" ] ], [ [ "# Draw the graph using Networkx/Matplotlib\n%matplotlib inline\nalertentity_graph = mas.create_alert_graph(security_alert)\nnbdisp.draw_alert_entity_graph(alertentity_graph, width=15)", "_____no_output_____" ] ], [ [ "<a id='related_alerts'></a>[Contents](#toc)\n# Related Alerts\nFor a subset of entities in the alert we can search for any alerts that have that entity in common. Currently this query looks for alerts that share the same Host, Account or Process and lists them below. \n**Notes:**\n- Some alert types do not include all of these entity types.\n- The original alert will be included in the \"Related Alerts\" set if it occurs within the query time boundary set below.\n\nThe query time boundaries default to a longer period than when searching for the alert. You can extend the time boundary searched before or after the alert time. If the widget doesn't support the time boundary that you want you can change the max_before and max_after parameters in the call to QueryTime below to extend the possible time boundaries.", "_____no_output_____" ] ], [ [ "# set the origin time to the time of our alert\nquery_times = mas.QueryTime(units='day', origin_time=security_alert.TimeGenerated, \n max_before=28, max_after=1, before=5)\nquery_times.display()", "_____no_output_____" ], [ "related_alerts = qry.list_related_alerts(provs=[query_times, security_alert])\n\nif related_alerts is not None and not related_alerts.empty:\n host_alert_items = related_alerts\\\n .query('host_match == @True')[['AlertType', 'StartTimeUtc']]\\\n .groupby('AlertType').StartTimeUtc.agg('count').to_dict()\n acct_alert_items = related_alerts\\\n .query('acct_match == @True')[['AlertType', 'StartTimeUtc']]\\\n .groupby('AlertType').StartTimeUtc.agg('count').to_dict()\n proc_alert_items = related_alerts\\\n .query('proc_match == @True')[['AlertType', 'StartTimeUtc']]\\\n .groupby('AlertType').StartTimeUtc.agg('count').to_dict()\n\n def print_related_alerts(alertDict, entityType, entityName):\n if len(alertDict) > 0:\n print('Found {} different alert types related to this {} (\\'{}\\')'\n .format(len(alertDict), entityType, entityName))\n for (k,v) in alertDict.items():\n print(' {}, Count of alerts: {}'.format(k, v))\n else:\n print('No alerts for {} entity \\'{}\\''.format(entityType, entityName))\n\n print_related_alerts(host_alert_items, 'host', security_alert.hostname)\n print_related_alerts(acct_alert_items, 'account', \n security_alert.primary_account.qualified_name \n if security_alert.primary_account\n else None)\n print_related_alerts(proc_alert_items, 'process', \n security_alert.primary_process.ProcessFilePath \n if security_alert.primary_process\n else None)\n nbdisp.display_timeline(data=related_alerts, source_columns = ['AlertName'], title='Alerts', height=100)\nelse:\n display(Markdown('No related alerts found.'))", "Found 8 different alert types related to this host ('msticalertswin1')\n Detected potentially suspicious use of Telegram tool, Count of alerts: 2\n Detected the disabling of critical services, Count of alerts: 2\n Digital currency mining related behavior detected, Count of alerts: 2\n Potential attempt to bypass AppLocker detected, Count of alerts: 4\n Security incident detected, Count of alerts: 2\n Security incident with shared process detected, Count of alerts: 3\n Suspicious system process executed, Count of alerts: 2\n Suspiciously named process detected, Count of alerts: 2\nFound 13 different alert types related to this account ('msticalertswin1\\msticadmin')\n An history file has been cleared, Count of alerts: 12\n Azure Security Center test alert (not a threat), Count of alerts: 13\n Detected potentially suspicious use of Telegram tool, Count of alerts: 2\n Detected the disabling of critical services, Count of alerts: 2\n Digital currency mining related behavior detected, Count of alerts: 2\n New SSH key added, Count of alerts: 13\n Possible credential access tool detected, Count of alerts: 11\n Possible suspicious scheduling tasks access detected, Count of alerts: 1\n Potential attempt to bypass AppLocker detected, Count of alerts: 3\n Suspicious Download Then Run Activity, Count of alerts: 13\n Suspicious binary detected, Count of alerts: 13\n Suspicious system process executed, Count of alerts: 2\n Suspiciously named process detected, Count of alerts: 2\nFound 2 different alert types related to this process ('c:\\w!ndows\\system32\\suchost.exe')\n Digital currency mining related behavior detected, Count of alerts: 2\n Suspiciously named process detected, Count of alerts: 2\n" ] ], [ [ "### Show these related alerts on a graph\nThis should indicate which entities the other alerts are related to.\n\nThis can be unreadable with a lot of alerts. Use the matplotlib interactive zoom control to zoom in to part of the graph.", "_____no_output_____" ] ], [ [ "# Draw a graph of this (add to entity graph)\n%matplotlib notebook\n%matplotlib inline\n\nif related_alerts is not None and not related_alerts.empty:\n rel_alert_graph = mas.add_related_alerts(related_alerts=related_alerts,\n alertgraph=alertentity_graph)\n nbdisp.draw_alert_entity_graph(rel_alert_graph, width=15)\nelse:\n display(Markdown('No related alerts found.'))", "_____no_output_____" ] ], [ [ "### Browse List of Related Alerts\nSelect an Alert to view details. \n\nIf you want to investigate that alert - copy its *SystemAlertId* property and open a new instance of this notebook to investigate this alert.", "_____no_output_____" ] ], [ [ "\ndef disp_full_alert(alert):\n global related_alert\n related_alert = mas.SecurityAlert(alert)\n nbdisp.display_alert(related_alert, show_entities=True)\n\nif related_alerts is not None and not related_alerts.empty:\n related_alerts['CompromisedEntity'] = related_alerts['Computer']\n print('Selected alert is available as \\'related_alert\\' variable.')\n rel_alert_select = mas.AlertSelector(alerts=related_alerts, action=disp_full_alert)\n rel_alert_select.display()\nelse:\n display(Markdown('No related alerts found.'))", "Selected alert is available as 'related_alert' variable.\n" ] ], [ [ "<a id='processtree'></a>[Contents](#toc)\n# Get Process Tree\nIf the alert has a process entity this section tries to retrieve the entire process tree to which that process belongs.\n\nNotes:\n- The alert must have a process entity\n- Only processes started within the query time boundary will be included\n- Ancestor and descented processes are retrieved to two levels (i.e. the parent and grandparent of the alert process plus any child and grandchild processes).\n- Sibling processes are the processes that share the same parent as the alert process\n- This can be a long-running query, especially if a wide time window is used! Caveat Emptor!\n\nThe source (alert) process is shown in red.\n\nWhat's shown for each process:\n- Each process line is indented according to its position in the tree hierarchy\n- Top line fields:\n - \\[relationship to source process:lev# - where # is the hops away from the source process\\]\n - Process creation date-time (UTC)\n - Process Image path\n - PID - Process Id\n - SubjSess - the session Id of the process spawning the new process\n - TargSess - the new session Id if the process is launched in another context/session. If 0/0x0 then the process is launched in the same session as its parent\n- Second line fields:\n - Process command line\n - Account - name of the account context in which the process is running", "_____no_output_____" ] ], [ [ "# set the origin time to the time of our alert\nquery_times = mas.QueryTime(units='minute', origin_time=security_alert.origin_time)\nquery_times.display()", "_____no_output_____" ], [ "from msticpy.nbtools.query_defns import DataFamily\n\nif security_alert.data_family != DataFamily.WindowsSecurity:\n raise ValueError('The remainder of this notebook currently only supports Windows. '\n 'Linux support is in development but not yet implemented.')\n\ndef extract_missing_pid(security_alert):\n for pid_ext_name in ['Process Id', 'Suspicious Process Id']:\n pid = security_alert.ExtendedProperties.get(pid_ext_name, None)\n if pid:\n return pid\n\ndef extract_missing_sess_id(security_alert):\n sess_id = security_alert.ExtendedProperties.get('Account Session Id', None)\n if sess_id:\n return sess_id\n for session in [e for e in security_alert.entities if\n e['Type'] == 'host-logon-session' or e['Type'] == 'hostlogonsession']:\n return session['SessionId']\n \nif (security_alert.primary_process):\n # Do some patching up if the process entity doesn't have a PID\n pid = security_alert.primary_process.ProcessId\n if not pid:\n pid = extract_missing_pid(security_alert)\n if pid:\n security_alert.primary_process.ProcessId = pid\n else:\n raise ValueError('Could not find the process Id for the alert process.')\n \n # Do the same if we can't find the account logon ID\n if not security_alert.get_logon_id():\n sess_id = extract_missing_sess_id(security_alert)\n if sess_id and security_alert.primary_account:\n security_alert.primary_account.LogonId = sess_id\n else:\n raise ValueError('Could not find the session Id for the alert process.')\n \n # run the query\n process_tree = qry.get_process_tree(provs=[query_times, security_alert])\n\n if len(process_tree) > 0:\n # Print out the text view of the process tree\n nbdisp.display_process_tree(process_tree)\n else:\n display(Markdown('No processes were returned so cannot obtain a process tree.'\n '\\n\\nSkip to [Other Processes](#process_clustering) later in the'\n ' notebook to retrieve all processes'))\nelse:\n display(Markdown('This alert has no process entity so cannot obtain a process tree.'\n '\\n\\nSkip to [Other Processes](#process_clustering) later in the'\n ' notebook to retrieve all processes'))\n process_tree = None\n", "_____no_output_____" ] ], [ [ "<a id='processtimeline'></a>[Contents](#toc)\n## Process TimeLine\nThis shows each process in the process tree on a timeline view.\n\nLabelling of individual process is very performance intensive and often results in nothing being displayed at all! Besides, for large numbers of processes it would likely result in an unreadable mess. \n\nYour main tools for negotiating the timeline are the Hover tool (toggled on and off by the speech bubble icon) and the wheel-zoom and pan tools (the former is an icon with an elipse and a magnifying glass, the latter is the crossed-arrows icon). The wheel zoom is particularly useful.\n\nAs you hover over each process it will display the image name, PID and commandline.\n\nAlso shown on the graphic is the timestamp line of the source/alert process.", "_____no_output_____" ] ], [ [ "# Show timeline of events\nif process_tree is not None and not process_tree.empty:\n nbdisp.display_timeline(data=process_tree, alert=security_alert, \n title='Alert Process Session', height=250)", "_____no_output_____" ] ], [ [ "<a id='process_clustering'></a>[Contents](#toc)\n# Other Processes on Host - Clustering\nSometimes you don't have a source process to work with. Other times it's just useful to see what else is going on on the host. This section retrieves all processes on the host within the time bounds\nset in the query times widget.\n\nYou can display the raw output of this by looking at the *processes_on_host* dataframe. Just copy this into a new cell and hit Ctrl-Enter.\n\nUsually though, the results return a lot of very repetitive and unintersting system processes so we attempt to cluster these to make the view easier to negotiate. \nTo do this we process the raw event list output to extract a few features that render strings (such as commandline)into numerical values. The default below uses the following features:\n- commandLineTokensFull - this is a count of common delimiters in the commandline \n (given by this regex r'[\\s\\-\\\\/\\.,\"\\'|&:;%$()]'). The aim of this is to capture the commandline structure while ignoring variations on what is essentially the same pattern (e.g. temporary path GUIDs, target IP or host names, etc.)\n- pathScore - this sums the ordinal (character) value of each character in the path (so /bin/bash and /bin/bosh would have similar scores).\n- isSystemSession - 1 if this is a root/system session, 0 if anything else.\n\nThen we run a clustering algorithm (DBScan in this case) on the process list. The result groups similar (noisy) processes together and leaves unique process patterns as single-member clusters.", "_____no_output_____" ], [ "### Clustered Processes (i.e. processes that have a cluster size > 1)", "_____no_output_____" ] ], [ [ "from msticpy.sectools.eventcluster import dbcluster_events, add_process_features\n\nprocesses_on_host = qry.list_processes(provs=[query_times, security_alert])\n\nif processes_on_host is not None and not processes_on_host.empty:\n feature_procs = add_process_features(input_frame=processes_on_host,\n path_separator=security_alert.path_separator)\n\n # you might need to play around with the max_cluster_distance parameter.\n # decreasing this gives more clusters.\n (clus_events, dbcluster, x_data) = dbcluster_events(data=feature_procs,\n cluster_columns=['commandlineTokensFull', \n 'pathScore', \n 'isSystemSession'],\n max_cluster_distance=0.0001)\n print('Number of input events:', len(feature_procs))\n print('Number of clustered events:', len(clus_events))\n clus_events[['ClusterSize', 'processName']][clus_events['ClusterSize'] > 1].plot.bar(x='processName', \n title='Process names with Cluster > 1', \n figsize=(12,3));\nelse:\n display(Markdown('Unable to obtain any processes for this host. This feature'\n ' is currently only supported for Windows hosts.'\n '\\n\\nIf this is a Windows host skip to [Host Logons](#host_logons)'\n ' later in the notebook to examine logon events.'))", "Number of input events: 190\nNumber of clustered events: 24\n" ] ], [ [ "### Variability in Command Lines and Process Names\nThe top chart shows the variability of command line content for a give process name. The wider the box, the more instances were found with different command line structure \n\nNote, the 'structure' in this case is measured by the number of tokens or delimiters in the command line and does not look at content differences. This is done so that commonly varying instances of the same command line are grouped together.<br>\nFor example `updatepatch host1.mydom.com` and `updatepatch host2.mydom.com` will be grouped together.\n\nThe second chart shows the variability in executable path. This does compare content so `c:\\windows\\system32\\net.exe` and `e:\\windows\\system32\\net.exe` are treated as distinct. You would normally not expect to see any variability in this chart unless you have multiple copies of the same name executable or an executable is trying masquerade as another well-known binary.", "_____no_output_____" ] ], [ [ "# Looking at the variability of commandlines and process image paths\nimport seaborn as sns\nsns.set(style=\"darkgrid\")\n\nif processes_on_host is not None and not processes_on_host.empty:\n proc_plot = sns.catplot(y=\"processName\", x=\"commandlineTokensFull\", \n data=feature_procs.sort_values('processName'),\n kind='box', height=10)\n proc_plot.fig.suptitle('Variability of Commandline Tokens', x=1, y=1)\n\n proc_plot = sns.catplot(y=\"processName\", x=\"pathLogScore\", \n data=feature_procs.sort_values('processName'),\n kind='box', height=10, hue='isSystemSession')\n proc_plot.fig.suptitle('Variability of Path', x=1, y=1);", "_____no_output_____" ] ], [ [ "The top graph shows that, for a given process, some have a wide variability in their command line content while the majority have little or none. Looking at a couple of examples - like cmd.exe, powershell.exe, reg.exe, net.exe - we can recognize several common command line tools.\n\nThe second graph shows processes by full process path content. We wouldn't normally expect to see variation here - as is the cast with most. There is also quite a lot of variance in the score making it a useful proxy feature for unique path name (this means that proc1.exe and proc2.exe that have the same commandline score won't get collapsed into the same cluster).\n\nAny process with a spread of values here means that we are seeing the same process name (but not necessarily the same file) is being run from different locations.", "_____no_output_____" ] ], [ [ "if not clus_events.empty:\n resp = input('View the clustered data? y/n')\n if resp == 'y':\n display(clus_events.sort_values('TimeGenerated')[['TimeGenerated', 'LastEventTime',\n 'NewProcessName', 'CommandLine', \n 'ClusterSize', 'commandlineTokensFull',\n 'pathScore', 'isSystemSession']])", "View the clustered data? y/ny\n" ], [ "# Look at clusters for individual process names\ndef view_cluster(exe_name):\n display(clus_events[['ClusterSize', 'processName', 'CommandLine', 'ClusterId']][clus_events['processName'] == exe_name])\n \ndisplay(Markdown('You can view the cluster members for individual processes'\n 'by inserting a new cell and entering:<br>'\n '`>>> view_cluster(process_name)`<br></div>'\n 'where process_name is the unqualified process binary. E.g<br>'\n '`>>> view_cluster(\\'reg.exe\\')`'))", "_____no_output_____" ] ], [ [ "### Time showing clustered vs. original data", "_____no_output_____" ] ], [ [ "# Show timeline of events - clustered events\nif not clus_events.empty:\n nbdisp.display_timeline(data=clus_events, \n overlay_data=processes_on_host, \n alert=security_alert, \n title='Distinct Host Processes (top) and All Proceses (bottom)')", "_____no_output_____" ] ], [ [ "<a id='cmdlineiocs'></a>[Contents](#toc)\n# Base64 Decode and Check for IOCs\nThis section looks for Indicators of Compromise (IoC) within the data sets passed to it.\n\nThe first section looks at the commandline for the alert process (if any). It also looks for base64 encoded strings within the data - this is a common way of hiding attacker intent. It attempts to decode any strings that look like base64. Additionally, if the base64 decode operation returns any items that look like a base64 encoded string or file, a gzipped binary sequence, a zipped or tar archive, it will attempt to extract the contents before searching for potentially interesting items.", "_____no_output_____" ] ], [ [ "process = security_alert.primary_process\nioc_extractor = sectools.IoCExtract()\n\nif process:\n # if nothing is decoded this just returns the input string unchanged\n base64_dec_str, _ = sectools.b64.unpack_items(input_string=process[\"CommandLine\"])\n if base64_dec_str and '<decoded' in base64_dec_str:\n print('Base64 encoded items found.')\n print(base64_dec_str)\n \n # any IoCs in the string?\n iocs_found = ioc_extractor.extract(base64_dec_str)\n \n if iocs_found:\n print('\\nPotential IoCs found in alert process:')\n display(iocs_found)\nelse:\n print('Nothing to process')\n", "\nPotential IoCs found in alert process:\n" ] ], [ [ "### If we have a process tree, look for IoCs in the whole data set\nYou can replace the data=process_tree parameter to ioc_extractor.extract() to pass other data frames.\nuse the columns parameter to specify which column or columns that you want to search.", "_____no_output_____" ] ], [ [ "ioc_extractor = sectools.IoCExtract()\n\ntry:\n if not process_tree.empty:\n source_processes = process_tree\n else:\n source_processes = clus_events\nexcept NameError:\n source_processes = None\nif source_processes is not None: \n\n ioc_df = ioc_extractor.extract(data=source_processes, \n columns=['CommandLine'], \n os_family=security_alert.os_family,\n ioc_types=['ipv4', 'ipv6', 'dns', 'url',\n 'md5_hash', 'sha1_hash', 'sha256_hash'])\n if len(ioc_df):\n display(HTML(\"<h3>IoC patterns found in process tree.</h3>\"))\n display(ioc_df)\nelse:\n ioc_df = None", "_____no_output_____" ] ], [ [ "### If any Base64 encoded strings, decode and search for IoCs in the results.\nFor simple strings the Base64 decoded output is straightforward. However for nested encodings this can get a little complex and difficult to represent in a tabular format.\n\n**Columns**\n - reference - The index of the row item in dotted notation in depth.seq pairs (e.g. 1.2.2.3 would be the 3 item at depth 3 that is a child of the 2nd item found at depth 1). This may not always be an accurate notation - it is mainly use to allow you to associate an individual row with the reference value contained in the full_decoded_string column of the topmost item).\n - original_string - the original string before decoding.\n - file_name - filename, if any (only if this is an item in zip or tar file).\n - file_type - a guess at the file type (this is currently elementary and only includes a few file types).\n - input_bytes - the decoded bytes as a Python bytes string.\n - decoded_string - the decoded string if it can be decoded as a UTF-8 or UTF-16 string. Note: binary sequences may often successfully decode as UTF-16 strings but, in these cases, the decodings are meaningless.\n - encoding_type - encoding type (UTF-8 or UTF-16) if a decoding was possible, otherwise 'binary'.\n - file_hashes - collection of file hashes for any decoded item.\n - md5 - md5 hash as a separate column.\n - sha1 - sha1 hash as a separate column.\n - sha256 - sha256 hash as a separate column.\n - printable_bytes - printable version of input_bytes as a string of \\xNN values\n - src_index - the index of the row in the input dataframe from which the data came.\n - full_decoded_string - the full decoded string with any decoded replacements. This is only really useful for top-level items, since nested items will only show the 'full' string representing the child fragment.", "_____no_output_____" ] ], [ [ "if source_processes is not None:\n dec_df = sectools.b64.unpack_items(data=source_processes, column='CommandLine')\n \nif source_processes is not None and not dec_df.empty:\n display(HTML(\"<h3>Decoded base 64 command lines</h3>\"))\n display(HTML(\"Warning - some binary patterns may be decodable as unicode strings\"))\n display(dec_df[['full_decoded_string', 'original_string', 'decoded_string', 'input_bytes', 'file_hashes']])\n\n ioc_dec_df = ioc_extractor.extract(data=dec_df, columns=['full_decoded_string'])\n if len(ioc_dec_df):\n display(HTML(\"<h3>IoC patterns found in base 64 decoded data</h3>\"))\n display(ioc_dec_df)\n if ioc_df is not None:\n ioc_df = ioc_df.append(ioc_dec_df ,ignore_index=True)\n else:\n ioc_df = ioc_dec_df\nelse:\n print(\"No base64 encodings found.\")\n ioc_df = None", "_____no_output_____" ] ], [ [ "<a id='virustotallookup'></a>[Contents](#toc)\n## Virus Total Lookup\nThis section uses the popular Virus Total service to check any recovered IoCs against VTs database.\n\nTo use this you need an API key from virus total, which you can obtain here: https://www.virustotal.com/.\n\nNote that VT throttles requests for free API keys to 4/minute. If you are unable to process the entire data set, try splitting it and submitting smaller chunks.\n\n**Things to note:**\n- Virus Total lookups include file hashes, domains, IP addresses and URLs.\n- The returned data is slightly different depending on the input type\n- The VTLookup class tries to screen input data to prevent pointless lookups. E.g.:\n - Only public IP Addresses will be submitted (no loopback, private address space, etc.)\n - URLs with only local (unqualified) host parts will not be submitted.\n - Domain names that are unqualified will not be submitted.\n - Hash-like strings (e.g 'AAAAAAAAAAAAAAAAAA') that do not appear to have enough entropy to be a hash will not be submitted.\n\n**Output Columns**\n - Observable - The IoC observable submitted\n - IoCType - the IoC type\n - Status - the status of the submission request\n - ResponseCode - the VT response code\n - RawResponse - the entire raw json response\n - Resource - VT Resource\n - SourceIndex - The index of the Observable in the source DataFrame. You can use this to rejoin to your original data.\n - VerboseMsg - VT Verbose Message\n - ScanId - VT Scan ID if any\n - Permalink - VT Permanent URL describing the resource\n - Positives - If this is not zero, it indicates the number of malicious reports that VT holds for this observable.\n - MD5 - The MD5 hash, if any\n - SHA1 - The MD5 hash, if any\n - SHA256 - The MD5 hash, if any\n - ResolvedDomains - In the case of IP Addresses, this contains a list of all domains that resolve to this IP address\n - ResolvedIPs - In the case Domains, this contains a list of all IP addresses resolved from the domain.\n - DetectedUrls - Any malicious URLs associated with the observable.", "_____no_output_____" ] ], [ [ "vt_key = mas.GetEnvironmentKey(env_var='VT_API_KEY',\n help_str='To obtain an API key sign up here https://www.virustotal.com/',\n prompt='Virus Total API key:')\nvt_key.display()", "_____no_output_____" ], [ "if vt_key.value and ioc_df is not None and not ioc_df.empty:\n vt_lookup = sectools.VTLookup(vt_key.value, verbosity=2)\n\n print(f'{len(ioc_df)} items in input frame')\n supported_counts = {}\n for ioc_type in vt_lookup.supported_ioc_types:\n supported_counts[ioc_type] = len(ioc_df[ioc_df['IoCType'] == ioc_type])\n print('Items in each category to be submitted to VirusTotal')\n print('(Note: items have pre-filtering to remove obvious erroneous '\n 'data and false positives, such as private IPaddresses)')\n print(supported_counts)\n print('-' * 80)\n vt_results = vt_lookup.lookup_iocs(data=ioc_df, type_col='IoCType', src_col='Observable')\n \n pos_vt_results = vt_results.query('Positives > 0')\n if len(pos_vt_results) > 0:\n display(HTML(f'<h3>{len(pos_vt_results)} Positive Results Found</h3>'))\n display(pos_vt_results[['Observable', 'IoCType','Permalink', \n 'ResolvedDomains', 'ResolvedIPs', \n 'DetectedUrls', 'RawResponse']])\n display(HTML('<h3>Other results</h3>'))\n display(vt_results.query('Status == \"Success\"'))", "5 items in input frame\nItems in each category to be submitted to VirusTotal\n(Note: items have pre-filtering to remove obvious erroneous data and false positives, such as private IPaddresses)\n{'ipv4': 0, 'dns': 2, 'url': 2, 'md5_hash': 0, 'sha1_hash': 0, 'sh256_hash': 0}\n--------------------------------------------------------------------------------\nInvalid observable format: \"wh401k.org\", type \"dns\", status: Observable does not match expected pattern for dns - skipping. (Source index 4)\nInvalid observable format: \"wh401k.org\", type \"dns\", status: Observable does not match expected pattern for dns - skipping. (Source index 0)\nSubmitting observables: \"http://wh401k.org/getps\"\", type \"url\" to VT. (Source index 4)\nError in response submitting observables: \"http://wh401k.org/getps\"\", type \"url\" http status is 403. Response: None (Source index 4)\nSubmitting observables: \"http://wh401k.org/getps\"</decoded>\", type \"url\" to VT. (Source index 0)\nError in response submitting observables: \"http://wh401k.org/getps\"</decoded>\", type \"url\" http status is 403. Response: None (Source index 0)\nSubmission complete. 4 responses from 5 input rows\n" ] ], [ [ "To view the raw response for a specific row.\n```\nimport json\nrow_idx = 0 # The row number from one of the above dataframes\nraw_response = json.loads(pos_vt_results['RawResponse'].loc[row_idx])\nraw_response\n```", "_____no_output_____" ], [ "<a id='cmdlineonotherhosts'></a>[Contents](#toc)\n# Alert command line - Occurrence on other hosts in workspace\nTo get a sense of whether the alert process is something that is occuring on other hosts, run this section.\n\nThis might tell you that the alerted process is actually a commonly-run process and the alert is a false positive. Alternatively, it may tell you that a real infection or attack is happening on other hosts in your environment.", "_____no_output_____" ] ], [ [ "# set the origin time to the time of our alert\nquery_times = mas.QueryTime(units='day', before=5, max_before=20,\n after=1, max_after=10,\n origin_time=security_alert.origin_time)\nquery_times.display()", "_____no_output_____" ], [ "# API ILLUSTRATION - Find the query to use\nqry.list_queries()", "_____no_output_____" ], [ "# API ILLUSTRATION - What does the query look like?\nqry.query_help('list_hosts_matching_commandline')", "Query: list_hosts_matching_commandline\nRetrieves processes on other hosts with matching commandline\nDesigned to be executed with data_source: process_create\nSupported data families: DataFamily.WindowsSecurity, DataFamily.LinuxSecurity\nSupported data environments: DataEnvironment.LogAnalytics\nQuery parameters:\n['add_query_items', 'subscription_filter', 'process_name', 'start', 'end', 'host_filter_neq', 'commandline']\nOptional parameters:\nadd_query_items\nQuery:\n{table}\n{query_project}\n| where {subscription_filter}\n| where {host_filter_neq}\n| where TimeGenerated >= datetime({start})\n| where TimeGenerated <= datetime({end})\n| where NewProcessName endswith '{process_name}'\n| where CommandLine =~ '{commandline}'\n{add_query_items}\n" ], [ "# This query needs a commandline parameter which isn't supplied\n# by default from the the alert \n# - so extract and escape this from the process\nif not security_alert.primary_process:\n raise ValueError('This alert has no process entity. This section is not applicable.')\n\nproc_match_in_ws = None\ncommandline = security_alert.primary_process.CommandLine\ncommandline = mas.utility.escape_windows_path(commandline)\n\nif commandline.strip():\n proc_match_in_ws = qry.list_hosts_matching_commandline(provs=[query_times, security_alert],\n commandline=commandline)\nelse:\n print('process has empty commandline')\n# Check the results\nif proc_match_in_ws is None or proc_match_in_ws.empty:\n print('No proceses with matching commandline found in on other hosts in workspace')\n print('between', query_times.start, 'and', query_times.end)\nelse:\n hosts = proc_match_in_ws['Computer'].drop_duplicates().shape[0]\n processes = proc_match_in_ws.shape[0]\n print('{numprocesses} proceses with matching commandline found on {numhosts} hosts in workspace'\\\n .format(numprocesses=processes, numhosts=hosts))\n print('between', query_times.start, 'and', query_times.end)\n print('To examine these execute the dataframe \\'{}\\' in a new cell'.format('proc_match_in_ws'))\n print(proc_match_in_ws[['TimeCreatedUtc','Computer', 'NewProcessName', 'CommandLine']].head())\n ", "No proceses with matching commandline found in on other hosts in workspace\nbetween 2019-02-08 22:04:16 and 2019-02-14 22:04:16\n" ] ], [ [ "<a id='host_logons'></a>[Contents](#toc)\n# Host Logons\nThis section retrieves the logon events on the host in the alert.\n\nYou may want to use the query times to search over a broader range than the default.", "_____no_output_____" ] ], [ [ "# set the origin time to the time of our alert\nquery_times = mas.QueryTime(units='day', origin_time=security_alert.origin_time,\n before=1, after=0, max_before=20, max_after=1)\nquery_times.display()", "_____no_output_____" ] ], [ [ "<a id='logonaccount'></a>[Contents](#toc)\n## Alert Logon Account\nThe logon associated with the process in the alert.", "_____no_output_____" ] ], [ [ "logon_id = security_alert.get_logon_id()\n\nif logon_id:\n if logon_id in ['0x3e7', '0X3E7', '-1', -1]:\n print('Cannot retrieve single logon event for system logon id '\n '- please continue with All Host Logons below.')\n else:\n logon_event = qry.get_host_logon(provs=[query_times, security_alert])\n nbdisp.display_logon_data(logon_event, security_alert)\nelse:\n print('No account entity in the source alert or the primary account had no logonId value set.')", "### Account Logon\nAccount: MSTICAdmin\nAccount Domain: MSTICAlertsWin1\nLogon Time: 2019-02-13 22:03:42.283000\nLogon type: 4 (Batch)\nUser Id/SID: S-1-5-21-996632719-2361334927-4038480536-500\n SID S-1-5-21-996632719-2361334927-4038480536-500 is administrator\n SID S-1-5-21-996632719-2361334927-4038480536-500 is local machine or domain account\nSession id '0x1e821b5' \nSubject (source) account: WORKGROUP/MSTICAlertsWin1$\nLogon process: Advapi \nAuthentication: Negotiate\nSource IpAddress: -\nSource Host: MSTICAlertsWin1\nLogon status: \n\n" ] ], [ [ "### All Host Logons\nSince the number of logon events may be large and, in the case of system logons, very repetitive, we use clustering to try to identity logons with unique characteristics.\n\nIn this case we use the numeric score of the account name and the logon type (i.e. interactive, service, etc.). The results of the clustered logons are shown below along with a more detailed, readable printout of the logon event information. The data here will vary depending on whether this is a Windows or Linux host.", "_____no_output_____" ] ], [ [ "from msticpy.sectools.eventcluster import dbcluster_events, add_process_features, _string_score\n\nhost_logons = qry.list_host_logons(provs=[query_times, security_alert])\nif host_logons is not None and not host_logons.empty:\n logon_features = host_logons.copy()\n logon_features['AccountNum'] = host_logons.apply(lambda x: _string_score(x.Account), axis=1)\n logon_features['LogonHour'] = host_logons.apply(lambda x: x.TimeGenerated.hour, axis=1)\n\n # you might need to play around with the max_cluster_distance parameter.\n # decreasing this gives more clusters.\n (clus_logons, _, _) = dbcluster_events(data=logon_features, time_column='TimeGenerated',\n cluster_columns=['AccountNum',\n 'LogonType'],\n max_cluster_distance=0.0001)\n print('Number of input events:', len(host_logons))\n print('Number of clustered events:', len(clus_logons))\n print('\\nDistinct host logon patterns:')\n display(clus_logons.sort_values('TimeGenerated'))\nelse:\n print('No logon events found for host.')", "Number of input events: 22\nNumber of clustered events: 3\n\nDistinct host logon patterns:\n" ], [ "# Display logon details\nnbdisp.display_logon_data(clus_logons, security_alert)", "### Account Logon\nAccount: MSTICAdmin\nAccount Domain: MSTICAlertsWin1\nLogon Time: 2019-02-13 22:03:42.283000\nLogon type: 4 (Batch)\nUser Id/SID: S-1-5-21-996632719-2361334927-4038480536-500\n SID S-1-5-21-996632719-2361334927-4038480536-500 is administrator\n SID S-1-5-21-996632719-2361334927-4038480536-500 is local machine or domain account\nSession id '0x1e821b5' \nSubject (source) account: WORKGROUP/MSTICAlertsWin1$\nLogon process: Advapi \nAuthentication: Negotiate\nSource IpAddress: -\nSource Host: MSTICAlertsWin1\nLogon status: \n\n### Account Logon\nAccount: SYSTEM\nAccount Domain: NT AUTHORITY\nLogon Time: 2019-02-13 21:10:58.540000\nLogon type: 5 (Service)\nUser Id/SID: S-1-5-18\n SID S-1-5-18 is LOCAL_SYSTEM\nSession id '0x3e7' System logon session\n\nSubject (source) account: WORKGROUP/MSTICAlertsWin1$\nLogon process: Advapi \nAuthentication: Negotiate\nSource IpAddress: -\nSource Host: -\nLogon status: \n\n### Account Logon\nAccount: DWM-2\nAccount Domain: Window Manager\nLogon Time: 2019-02-12 22:22:21.240000\nLogon type: 2 (Interactive)\nUser Id/SID: S-1-5-90-0-2\nSession id '0x106b458' \nSubject (source) account: WORKGROUP/MSTICAlertsWin1$\nLogon process: Advapi \nAuthentication: Negotiate\nSource IpAddress: -\nSource Host: -\nLogon status: \n\n" ] ], [ [ "### Comparing All Logons with Clustered results relative to Alert time line", "_____no_output_____" ] ], [ [ "# Show timeline of events - all logons + clustered logons\nif host_logons is not None and not host_logons.empty:\n nbdisp.display_timeline(data=host_logons, overlay_data=clus_logons,\n alert=security_alert, \n source_columns=['Account', 'LogonType'],\n title='All Host Logons')", "_____no_output_____" ] ], [ [ "### View Process Session and Logon Events in Timelines\nThis shows the timeline of the clustered logon events with the process tree obtained earlier. This allows you to get a sense of which logon was responsible for the process tree session whether any additional logons (e.g. creating a process as another user) might be associated with the alert timeline.\n\n*Note you should use the pan and zoom tools to align the timelines since the data may be over different time ranges.*", "_____no_output_____" ] ], [ [ "# Show timeline of events - all events\nif host_logons is not None and not host_logons.empty:\n nbdisp.display_timeline(data=clus_logons, source_columns=['Account', 'LogonType'],\n alert=security_alert,\n title='Clustered Host Logons', height=200)\n try:\n nbdisp.display_timeline(data=process_tree, alert=security_alert, title='Alert Process Session', height=200)\n except NameError:\n print('process_tree not available for this alert.')", "_____no_output_____" ], [ "# Counts of Logon types by Account\nif host_logons is not None and not host_logons.empty:\n display(host_logons[['Account', 'LogonType', 'TimeGenerated']]\n .groupby(['Account','LogonType']).count()\n .rename(columns={'TimeGenerated': 'LogonCount'}))", "_____no_output_____" ] ], [ [ "<a id='failed logons'></a>[Contents](#toc)\n## Failed Logons", "_____no_output_____" ] ], [ [ "failedLogons = qry.list_host_logon_failures(provs=[query_times, security_alert])\nif failedLogons.shape[0] == 0:\n display(print('No logon failures recorded for this host between {security_alert.start} and {security_alert.start}'))\n\nfailedLogons", "_____no_output_____" ] ], [ [ "<a id='appendices'></a>[Contents](#toc)\n# Appendices", "_____no_output_____" ], [ "## Available DataFrames", "_____no_output_____" ] ], [ [ "print('List of current DataFrames in Notebook')\nprint('-' * 50)\ncurrent_vars = list(locals().keys())\nfor var_name in current_vars:\n if isinstance(locals()[var_name], pd.DataFrame) and not var_name.startswith('_'):\n print(var_name)", "List of current DataFrames in Notebook\n--------------------------------------------------\nmydf\nalert_counts\nalert_list\nrelated_alerts\nprocess_tree\nprocesses_on_host\nfeature_procs\nclus_events\nsource_processes\nioc_df\ndec_df\nioc_dec_df\nvt_results\npos_vt_results\nproc_match_in_ws\nlogon_event\nhost_logons\nlogon_features\nclus_logons\nfailedLogons\n" ] ], [ [ "## Saving Data to CSV\nTo save the contents of a pandas DataFrame to an CSV\nuse the following syntax\n```\nhost_logons.to_csv('host_logons.csv')\n```", "_____no_output_____" ], [ "## Saving Data to Excel\nTo save the contents of a pandas DataFrame to an Excel spreadsheet\nuse the following syntax\n```\nwriter = pd.ExcelWriter('myWorksheet.xlsx')\nmy_data_frame.to_excel(writer,'Sheet1')\nwriter.save()\n```", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import numpy as np\nglobal np\nimport scipy as sp\nimport scipy.signal as signal\nimport matplotlib.pyplot as plt\nimport IPython.display as ipd\nfrom ipywidgets import interact\nimport sys\nimport wave\nsys.path.append(\"../backend/\")\n%matplotlib inline\n\ndef load_wav(filepath, t_start = 0, t_end = 2**32) :\n \"\"\"Load a wave file, which must be 22050Hz and 16bit and must be either\n mono or stereo. \n Inputs:\n filepath: audio file\n t_start, t_end: (optional) subrange of file to load (in seconds)\n Returns:\n a numpy floating-point array with a range of [-1, 1]\n \"\"\"\n wf = wave.open(filepath)\n num_channels, sampwidth, fs, end, comptype, compname = wf.getparams()\n \n # for now, we will only accept 16 bit files at 22k\n assert(sampwidth == 2)\n assert(fs == 22050)\n\n # start frame, end frame, and duration in frames\n f_start = int(t_start * fs)\n f_end = min(int(t_end * fs), end)\n frames = f_end - f_start\n\n wf.setpos(f_start)\n raw_bytes = wf.readframes(frames)\n # convert raw data to numpy array, assuming int16 arrangement\n samples = np.fromstring(raw_bytes, dtype = np.int16)\n\n # convert from integer type to floating point, and scale to [-1, 1]\n samples = samples.astype(np.float)\n samples *= (1 / 32768.0)\n\n if num_channels == 1:\n return samples\n\n elif num_channels == 2:\n return 0.5 * (samples[0::2] + samples[1::2])\n\n else:\n raise('Can only handle mono or stereo wave files') ", "_____no_output_____" ] ], [ [ "Today, in preparation for our final projects, we are going to talk about problems that can arise with hardware implementations, and how we can avoid this with good design.", "_____no_output_____" ], [ "## Eliminating 60Hz Noise", "_____no_output_____" ], [ "## Impedance Mathing", "_____no_output_____" ], [ "## Ground Feedback Loops", "_____no_output_____" ], [ " ", "_____no_output_____" ] ] ]

[ "code", "markdown" ]

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "#hide\n%load_ext autoreload\n%autoreload 2", "_____no_output_____" ], [ "# default_exp email", "_____no_output_____" ] ], [ [ "# Compose and send emails\n\n> Compose and send html emails through an SMTP server using TLS.", "_____no_output_____" ] ], [ [ "#hide\nfrom nbdev.showdoc import *", "_____no_output_____" ], [ "#export\nimport smtplib\nfrom email.message import EmailMessage\nimport mimetypes\nfrom pathlib2 import Path\nimport re", "_____no_output_____" ] ], [ [ "## Complose a message", "_____no_output_____" ] ], [ [ "#export\ndef create_html_message(from_address, to_addresses, subject, html, text = None, image_path = Path.cwd()):\n msg = EmailMessage()\n msg['From'] = from_address\n msg['To'] = to_addresses\n msg['Subject'] = subject\n if text is not None:\n msg.set_content(text)\n msg.add_alternative(html, subtype='html')\n \n cid_images = list(re.findall(fr'<img src=\"cid:(.*?)\"', html))\n cid_images.extend(list(re.findall(fr'url\\(cid:(.*?)\\)', html)))\n cid_images = list(set(cid_images))\n for cid_img in cid_images:\n with open(image_path / cid_img, 'rb') as img:\n msg.get_payload()[-1].add_related(img.read(),'image', 'jpeg', cid = cid_img)\n return msg", "_____no_output_____" ] ], [ [ "### Add an attachment to a message", "_____no_output_____" ] ], [ [ "#export\ndef add_attachmet(msg, path):\n \"Add an attachment with location `path` to the cunnet message `msg`.\"\n # Guess the content type based on the file's extension. Encoding\n # will be ignored, although we should check for simple things like\n # gzip'd or compressed files.\n ctype, encoding = mimetypes.guess_type(path)\n if ctype is None or encoding is not None:\n # No guess could be made, or the file is encoded (compressed), so\n # use a generic bag-of-bits type.\n ctype = 'application/octet-stream'\n maintype, subtype = ctype.split('/', 1)\n with open(path, 'rb') as f:\n msg.add_attachment(f.read(), maintype = maintype, subtype = subtype, filename = path.name)\n return msg", "_____no_output_____" ] ], [ [ "## Send a message using SMTP and TLS", "_____no_output_____" ] ], [ [ "#export\ndef send_smtp_email(server, tls_port, user, pw, msg):\n \"Send the message `msg` using the specified `server` and `port` - login using `user` and `pw`.\"\n # Create a secure SSL context\n try:\n smtp = smtplib.SMTP(server, tls_port)\n smtp.starttls()\n smtp.login(user, pw)\n smtp.send_message(msg)\n except Exception as e:\n print(e)\n finally:\n smtp.quit()", "_____no_output_____" ] ], [ [ "## Examples", "_____no_output_____" ], [ "The following is an example on how to compose and send a html-email message.", "_____no_output_____" ] ], [ [ "## set user and password of the smtp server\n#user = ''\n#pw = ''\n#\n## send email from and to myself\n#from_email = user\n#to_email = ''\n#\n#html = \"\"\"\n#Hello, this is a test message!\n#<h1>Hello 22!</h1>\n#<img src=\"cid:email.jpg\">\n#<h1>Hello 23!</h1>\n#<img src=\"cid:iceberg.jpg\">\n#\"\"\"\n#\n#msg = create_html_message(from_email, to_email, 'subject', html, image_path=Path(''))\n#add_attachmet(msg, Path(''))\n#\n## uncomment after setting user and pw above\n##send_smtp_email('', 587, user, pw, msg)", "_____no_output_____" ] ] ]

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

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

[ "Unlicense" ]

[ "Unlicense" ]

[ "Unlicense" ]

[ [ [ "# DQN With Prioritized Replay Buffer\nUse prioritized replay buffer to train a DQN agent.", "_____no_output_____" ] ], [ [ "import random\nimport numpy as np\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\n\nimport utils\n\nimport gym\nimport numpy as np\n\nfrom gym.core import ObservationWrapper\nfrom gym.spaces import Box\nimport cv2\nimport os\n\nimport atari_wrappers # adjust env\nfrom framebuffer import FrameBuffer # stack 4 consec images \n", "_____no_output_____" ], [ "ENV_NAME = \"BreakoutNoFrameskip-v4\"\n\n# create break-out env\nenv = gym.make(ENV_NAME)\nenv.reset()", "_____no_output_____" ] ], [ [ "## Preprocessing\nCrop the important part of the image, then resize to 64 x 64", "_____no_output_____" ] ], [ [ "class PreprocessAtariObs(ObservationWrapper):\n def __init__(self, env):\n \"\"\"A gym wrapper that crops, scales image into the desired shapes and grayscales it.\"\"\"\n ObservationWrapper.__init__(self, env)\n\n self.image_size = (1, 64, 64)\n self.observation_space = Box(0.0, 1.0, self.image_size)\n\n def observation(self, img):\n \"\"\"what happens to each observation\"\"\"\n\n # Here's what you need to do:\n # * crop image, remove irrelevant parts\n # * resize image to self.img_size\n # (use imresize from any library you want,\n # e.g. opencv, skimage, PIL, keras)\n # * cast image to grayscale\n # * convert image pixels to (0,1) range, float32 type\n\n # crop the image \n # remove the top part\n img = img[50:]\n\n # resize the image\n img = cv2.resize(img, dsize=(self.image_size[1], self.image_size[2]))\n\n # gray scale\n img = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)\n\n # normalize to (0, 1)\n img = img.astype(np.float32) / 255.0\n\n # add channel dimension\n return img[None]\n\n# adjust the env by some wrappers\ndef PrimaryAtariWrap(env, clip_rewards=True):\n assert 'NoFrameskip' in env.spec.id\n\n # This wrapper holds the same action for <skip> frames and outputs\n # the maximal pixel value of 2 last frames (to handle blinking\n # in some envs)\n env = atari_wrappers.MaxAndSkipEnv(env, skip=4)\n\n # This wrapper sends done=True when each life is lost\n # (not all the 5 lives that are givern by the game rules).\n # It should make easier for the agent to understand that losing is bad.\n env = atari_wrappers.EpisodicLifeEnv(env)\n\n # This wrapper laucnhes the ball when an episode starts.\n # Without it the agent has to learn this action, too.\n # Actually it can but learning would take longer.\n env = atari_wrappers.FireResetEnv(env)\n\n # This wrapper transforms rewards to {-1, 0, 1} according to their sign\n if clip_rewards:\n env = atari_wrappers.ClipRewardEnv(env)\n\n # This wrapper is yours :)\n env = PreprocessAtariObs(env)\n return env\n \ndef make_env(clip_rewards=True, seed=None):\n env = gym.make(ENV_NAME) # create raw env\n if seed is not None:\n env.seed(seed)\n env = PrimaryAtariWrap(env, clip_rewards)\n env = FrameBuffer(env, n_frames=4, dim_order='pytorch')\n return env\n\nenv = make_env()\nenv.reset()\nn_actions = env.action_space.n\nstate_shape = env.observation_space.shape\nprint(\"adjusted env with 4 consec images stacked can be created\")", "adjusted env with 4 consec images stacked can be created\n" ] ], [ [ "## Model", "_____no_output_____" ] ], [ [ "device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n\ndef conv2d_size_out(size, kernel_size, stride):\n \"\"\"\n common use case:\n cur_layer_img_w = conv2d_size_out(cur_layer_img_w, kernel_size, stride)\n cur_layer_img_h = conv2d_size_out(cur_layer_img_h, kernel_size, stride)\n to understand the shape for dense layer's input\n \"\"\"\n return (size - (kernel_size - 1) - 1) // stride + 1\n\nclass DuelingDQNAgent(nn.Module):\n def __init__(self, state_shape, n_actions, epsilon=0):\n super().__init__()\n self.epsilon = epsilon\n self.n_actions = n_actions\n self.state_shape = state_shape\n\n # Define your network body here. Please make sure agent is fully contained here\n # nn.Flatten() can be useful\n kernel_size = 3\n stride = 2\n self.conv1 = nn.Conv2d(4, 16, kernel_size, stride)\n out_size = conv2d_size_out(state_shape[1], kernel_size, stride)\n self.conv2 = nn.Conv2d(16, 32, kernel_size, stride)\n out_size = conv2d_size_out(out_size, kernel_size, stride)\n self.conv3 = nn.Conv2d(32, 64, kernel_size, stride)\n out_size = conv2d_size_out(out_size, kernel_size, stride)\n\n # size of the output tensor after convolution batch_size x 64 x out_size x out_size\n self.linear = nn.Linear(64*out_size*out_size, 256)\n \n # advantage\n self.advantage = nn.Sequential(\n nn.Linear(256, 512),\n nn.ReLU(),\n nn.Linear(512, self.n_actions)\n )\n \n # state value\n self.value = nn.Sequential(\n nn.Linear(256, 512),\n nn.ReLU(),\n nn.Linear(512, 1)\n )\n \n def forward(self, state_t):\n \"\"\"\n takes agent's observation (tensor), returns qvalues (tensor)\n :param state_t: a batch of 4-frame buffers, shape = [batch_size, 4, h, w]\n \"\"\"\n # Use your network to compute qvalues for given state\n # qvalues = <YOUR CODE>\n t = self.conv1(state_t)\n t = F.relu(t)\n t = self.conv2(t)\n t = F.relu(t)\n t = self.conv3(t)\n t = F.relu(t)\n\n t = t.view(state_t.shape[0], -1)\n t = self.linear(t)\n t = F.relu(t)\n \n # compute advantage and state value as different heads\n advantage = self.advantage(t)\n value = self.value(t)\n \n qvalues = value + advantage - advantage.mean(dim=1, keepdim=True)\n\n assert qvalues.requires_grad, \"qvalues must be a torch tensor with grad\"\n assert len(\n qvalues.shape) == 2 and qvalues.shape[0] == state_t.shape[0] and qvalues.shape[1] == n_actions\n\n return qvalues\n\n def get_qvalues(self, states):\n \"\"\"\n like forward, but works on numpy arrays, not tensors\n \"\"\"\n model_device = next(self.parameters()).device\n states = torch.tensor(states, device=model_device, dtype=torch.float)\n qvalues = self.forward(states)\n return qvalues.data.cpu().numpy()\n\n def sample_actions(self, qvalues):\n \"\"\"pick actions given qvalues. Uses epsilon-greedy exploration strategy. \"\"\"\n epsilon = self.epsilon\n batch_size, n_actions = qvalues.shape\n\n random_actions = np.random.choice(n_actions, size=batch_size)\n best_actions = qvalues.argmax(axis=-1)\n\n should_explore = np.random.choice(\n [0, 1], batch_size, p=[1-epsilon, epsilon])\n return np.where(should_explore, random_actions, best_actions)\n \n# Evaluate the agent\ndef evaluate(env, agent, n_games=1, greedy=False, t_max=10000):\n rewards = []\n for _ in range(n_games):\n reward = 0.0\n s = env.reset()\n for _ in range(t_max):\n qvalues = agent.get_qvalues([s])\n action = qvalues.argmax(axis=-1)[0] if greedy else agent.sample_actions(\n qvalues)[0]\n s, r, done, _ = env.step(action)\n reward += r\n if done:\n break\n \n rewards.append(reward)\n return np.mean(rewards)\n", "_____no_output_____" ] ], [ [ "## Compute TD loss", "_____no_output_____" ] ], [ [ "def compute_td_loss(states, actions, rewards, next_states, is_done,\n agent, target_network,\n gamma=0.99,\n device=device, check_shapes=False):\n \"\"\" Compute td loss using torch operations only. Use the formulae above. '''\n \n objective of agent is \n \\hat Q(s_t, a_t) = r_t + \\gamma Target(s_{t+1}, argmax_{a} Q(s_{t+1}, a)) \n \"\"\"\n states = torch.tensor(states, device=device, dtype=torch.float) # shape: [batch_size, *state_shape]\n\n # for some torch reason should not make actions a tensor\n actions = torch.tensor(actions, device=device, dtype=torch.long) # shape: [batch_size]\n rewards = torch.tensor(rewards, device=device, dtype=torch.float) # shape: [batch_size]\n # shape: [batch_size, *state_shape]\n next_states = torch.tensor(next_states, device=device, dtype=torch.float)\n \n is_done = torch.tensor(\n is_done,\n device=device,\n dtype=torch.float\n ) # shape: [batch_size]\n \n is_not_done = 1 - is_done\n \n # get q-values for all actions in current states\n predicted_qvalues = agent(states)\n \n # compute q-values for all actions in next states\n predicted_next_qvalues = target_network(next_states)\n \n # best action in next state\n next_best_actions = torch.argmax(agent(states), dim=1)\n\n # select q-values for chosen actions\n predicted_qvalues_for_actions = predicted_qvalues[range(\n len(actions)), actions]\n \n # compute the objective of the agent\n next_state_values = predicted_next_qvalues[range(\n len(actions)), next_best_actions] \n \n # assert next_state_values.dim(\n # == 1 and next_state_values.shape[0] == states.shape[0], \"must predict one value per state\"\n\n # compute \"target q-values\" for loss - it's what's inside square parentheses in the above formula.\n # at the last state use the simplified formula: Q(s,a) = r(s,a) since s' doesn't exist\n # you can multiply next state values by is_not_done to achieve this.\n # target_qvalues_for_actions = <YOUR CODE>\n\n target_qvalues_for_actions = rewards + next_state_values * is_not_done\n\n # mean squared error loss adjusted by importance sampling weights to minimize\n #loss = torch.mean(\n # weights * torch.pow(predicted_qvalues_for_actions - target_qvalues_for_actions.detach(), 2)\n #)\n \n # return the TD-loss\n \n if check_shapes:\n assert predicted_next_qvalues.data.dim(\n ) == 2, \"make sure you predicted q-values for all actions in next state\"\n assert next_state_values.data.dim(\n ) == 1, \"make sure you computed V(s') as maximum over just the actions axis and not all axes\"\n assert target_qvalues_for_actions.data.dim(\n ) == 1, \"there's something wrong with target q-values, they must be a vector\"\n \n return target_qvalues_for_actions - predicted_qvalues_for_actions", "_____no_output_____" ] ], [ [ "## Test the memory need of the replay buffer", "_____no_output_____" ], [ "Init DQN agent and play a total 10^4 time steps", "_____no_output_____" ] ], [ [ "def play_and_record(initial_state, agent, env, exp_replay, n_steps=1):\n \"\"\"\n Play the game for exactly n steps, record every (s,a,r,s', done) to replay buffer. \n Whenever game ends, add record with done=True and reset the game.\n It is guaranteed that env has done=False when passed to this function.\n\n PLEASE DO NOT RESET ENV UNLESS IT IS \"DONE\"\n\n :returns: return sum of rewards over time and the state in which the env stays\n \"\"\"\n s = initial_state\n sum_rewards = 0\n\n # Play the game for n_steps as per instructions above\n sum_rewards = 0.0 \n for _ in range(n_steps):\n qvalues = agent.get_qvalues([s])\n action = agent.sample_actions(qvalues)[0] \n next_s, r, done, _ = env.step(action)\n\n exp_replay.add((s, action, r, next_s, done))\n sum_rewards += r\n if done:\n s = env.reset()\n else:\n s = next_s\n\n return sum_rewards, s\n\n", "_____no_output_____" ], [ "import utils\nimport imp\nimport replay_buffer\nimp.reload(replay_buffer)\n\nfrom replay_buffer import PrioritizedReplayBuffer\n\n\n#n_actions = env.action_space.n\n#state_shape = env.observation_space.shape\n\nagent = DuelingDQNAgent(state_shape=state_shape, n_actions=n_actions)\nexp_replay = PrioritizedReplayBuffer(10**4)\n\n'''\nfor i in range(100):\n state = env.reset()\n if not utils.is_enough_ram(min_available_gb=0.1):\n print(\"\"\"\n Less than 100 Mb RAM available. \n Make sure the buffer size in not too huge.\n Also check, maybe other processes consume RAM heavily.\n \"\"\"\n )\n break\n play_and_record(state, agent, env, exp_replay, n_steps=10**2)\n if len(exp_replay) == 10**4:\n break\nprint(len(exp_replay))\n\ndel exp_replay\n'''\n", "_____no_output_____" ], [ "seed = 42\n\n# env\nn_lives = 5\n\n\n# training params\nT = 1 # number of experiences to get from env before each update\nbatch_size = 16\ntotal_steps = 3 * 10**1 # total steps to train the agent\ndecay_steps = 10**1 # steps to decay the epsilon, \n # after the decay_steps, epsilon stops decaying\n # and the agent explores with a fixed probability\nmax_grad_norm = 50 \n\n \nrefresh_target_network_freq = 5000 # freqency to update the target network\nlearning_rate = 1e-4\n\n\n# agent \ngamma = 0.99 # discount factor\ninit_epsilon = 1.0\nfinal_epsilon = 0.1\n\n# buffer\nbuffer_size = 10**4\n\n# eval\nloss_freq = 50 \neval_freq = 5000\n\n# logs \nckpt_dir = 'logs'\nckpt_file = 'prioritized_experience_replay_ckpt.pth'\nmetrics_file = 'prioritized_experience_replay_metrics.pth'\nckpt_freq = 10*5000 # Debug param", "_____no_output_____" ], [ "# main loop\n\nenv = make_env(seed)\n\nstate_shape = env.observation_space.shape\nn_actions = env.action_space.n\nstate = env.reset()\n\nagent = DuelingDQNAgent(state_shape, n_actions, epsilon=1).to(device)\ntarget_network = DuelingDQNAgent(state_shape, n_actions).to(device)\ntarget_network.load_state_dict(agent.state_dict())\n\nexp_replay = PrioritizedReplayBuffer(buffer_size)\n\n\n\nopt = torch.optim.Adam(agent.parameters(), lr=learning_rate)\n\nmean_rw_history = []\ntd_loss_history = []\ngrad_norm_history = []\ninitial_state_v_history = []\n\nprint(\"Starts training on {}\".format(next(agent.parameters()).device))\n\n# populate the buffer with 128 samples\ninit_size = 128\nplay_and_record(state, agent, env, exp_replay, init_size)\n\nfor step in range(total_steps):\n agent.epsilon = utils.linear_decay(init_epsilon, final_epsilon, step, decay_steps)\n \n # play for $T time steps and cache the exprs to the buffer\n _, state = play_and_record(state, agent, env, exp_replay, T)\n \n b_idx, obses_t, actions, rewards, obses_tp1, dones, weights = exp_replay.sample(\n batch_size)\n \n # td loss for each sample\n td_loss = compute_td_loss(\n states=obses_t, \n actions=actions, \n rewards=rewards, \n next_states=obses_tp1, \n is_done=dones,\n agent=agent,\n target_network=target_network,\n gamma=gamma,\n device=device,\n check_shapes=True)\n \n '''\n A batch of samples from prioritized replay looks like:\n (states, actions, rewards, next_states, weights, is_done)\n weights here are importance sampling weights\n\n Basically:\n Loss = weights * MSE\n \n '''\n \n # compute MSE adjusted by importance sampling weights\n # and backprop\n weights = torch.tensor(weights, dtype=torch.float32)\n #print(weights, torch.pow(td_loss, 2))\n loss = torch.mean(weights * torch.pow(td_loss, 2))\n loss.backward()\n grad_norm = nn.utils.clip_grad_norm_(agent.parameters(), max_grad_norm)\n opt.step()\n opt.zero_grad()\n \n # update the priorities of sampled exprs\n exp_replay.batch_update(b_idx, np.abs(td_loss.detach().cpu().numpy()))\n \n # increase the importance sampling hyperparameter b gradually to 1\n exp_replay.increment_b()\n \n if step % loss_freq == 0:\n # save MSE without importance sampling\n loss = torch.mean(torch.pow(td_loss, 2))\n td_loss_history.append(loss.cpu().item())\n \n if step % refresh_target_network_freq == 0:\n target_network.load_state_dict(agent.state_dict())\n \n if step % eval_freq == 0:\n mean_rw_history.append(evaluate(\n make_env(clip_rewards=True, seed=step),\n agent, n_games=3*n_lives, greedy=True\n ))\n \n initial_state_q_values = agent.get_qvalues(\n [make_env(seed=step).reset()]\n )\n \n initial_state_v_history.append(np.max(initial_state_q_values))\n \n print(\"buffer size = %i, epsilon: %.5f\" % \n (len(exp_replay), agent.epsilon))\n \n \n # TODO \n # checkpointing\n if step % ckpt_freq == 0:\n print(\"checkpointing ...\")\n \n if not os.path.exists(ckpt_dir):\n os.makedirs(ckpt_dir)\n \n # check point model and optimizer\n checkpoint = {\n \"step\": step,\n \"agent\": agent.state_dict(),\n \"epsilon\": agent.epsilon,\n \"target_network\": target_network.state_dict(),\n \"optimizer\": opt.state_dict(),\n \"replay_buffer\": exp_replay\n }\n \n torch.save(checkpoint, os.path.join(ckpt_dir, ckpt_file))\n \n # save the performance metric \n metrics = {\n \"mean_rw_history\": mean_rw_history,\n \"td_loss_history\": td_loss_history,\n \"grad_norm_history\": grad_norm_history,\n \"initial_state_v_history\": initial_state_v_history\n }\n \n torch.save(metrics, os.path.join(ckpt_dir, metrics_file))\n \n \n# check point model and optimizer\ncheckpoint = {\n \"step\": step,\n \"agent\": agent.state_dict(),\n \"epsilon\": agent.epsilon,\n \"target_network\": target_network.state_dict(),\n \"optimizer\": opt.state_dict(),\n \"replay_buffer\": exp_replay\n}\n\ntorch.save(checkpoint, os.path.join(ckpt_dir, ckpt_file))\n\n# save the performance metric \nmetrics = {\n \"mean_rw_history\": mean_rw_history,\n \"td_loss_history\": td_loss_history,\n \"grad_norm_history\": grad_norm_history,\n \"initial_state_v_history\": initial_state_v_history\n}\n\ntorch.save(metrics, os.path.join(ckpt_dir, metrics_file))", "Starts training on cpu\nbuffer size = 129, epsilon: 1.00000\ncheckpointing ...\n" ] ] ]

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

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# Image Segmentation\n\n* This code is **only** `tensorflow API` version for [TensorFlow tutorials/Image Segmentation](https://github.com/tensorflow/models/blob/master/samples/outreach/blogs/segmentation_blogpost/image_segmentation.ipynb) which is made of `tf.keras`.\n* You can see the detail description [tutorial link](https://github.com/tensorflow/models/blob/master/samples/outreach/blogs/segmentation_blogpost/image_segmentation.ipynb) \n\n* I use below dataset instead of [carvana-image-masking-challenge dataset](https://www.kaggle.com/c/carvana-image-masking-challenge/rules) in TensorFlow Tutorials which is a kaggle competition dataset.\n * carvana-image-masking-challenge dataset: Too large dataset (14GB)\n* [Gastrointestinal Image ANAlys Challenges (GIANA)](https://giana.grand-challenge.org) Dataset (345MB)\n * Train data: 300 images with RGB channels (bmp format)\n * Train lables: 300 images with 1 channels (bmp format)\n * Image size: 574 x 500", "_____no_output_____" ] ], [ [ "from __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport os\nimport time\nimport functools\n\nimport numpy as np\nimport matplotlib.pyplot as plt\n%matplotlib inline\nimport matplotlib as mpl\nmpl.rcParams['axes.grid'] = False\nmpl.rcParams['figure.figsize'] = (12,12)\n\nfrom sklearn.model_selection import train_test_split\nfrom PIL import Image\nfrom IPython.display import clear_output\n\nimport tensorflow as tf\nslim = tf.contrib.slim\n\ntf.logging.set_verbosity(tf.logging.INFO)\n\nsess_config = tf.ConfigProto(gpu_options=tf.GPUOptions(allow_growth=True))\nos.environ[\"CUDA_VISIBLE_DEVICES\"]=\"0\"", "/home/lab4all/anaconda3/lib/python3.6/site-packages/h5py/__init__.py:34: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n from ._conv import register_converters as _register_converters\n" ] ], [ [ "# Get all the files \nSince this tutorial will be using a dataset from [Giana Dataset](https://giana.grand-challenge.org/Dates/).", "_____no_output_____" ] ], [ [ "# Unfortunately you cannot downlaod GIANA dataset from website\n# So I upload zip file on my dropbox\n# if you want to download from my dropbox uncomment below\n#!wget https://goo.gl/mxikqa\n#!mv mxikqa sd_train.zip\n#!unzip sd_train.zip\n#!mkdir ../../datasets\n#!mv sd_train ../../datasets\n#!rm sd_train.zip", "_____no_output_____" ], [ "dataset_dir = '../../datasets/sd_train'\nimg_dir = os.path.join(dataset_dir, \"train\")\nlabel_dir = os.path.join(dataset_dir, \"train_labels\")", "_____no_output_____" ], [ "x_train_filenames = [os.path.join(img_dir, filename) for filename in os.listdir(img_dir)]\nx_train_filenames.sort()\ny_train_filenames = [os.path.join(label_dir, filename) for filename in os.listdir(label_dir)]\ny_train_filenames.sort()", "_____no_output_____" ], [ "x_train_filenames, x_test_filenames, y_train_filenames, y_test_filenames = \\\n train_test_split(x_train_filenames, y_train_filenames, test_size=0.2, random_state=219)", "_____no_output_____" ], [ "num_train_examples = len(x_train_filenames)\nnum_test_examples = len(x_test_filenames)\n\nprint(\"Number of training examples: {}\".format(num_train_examples))\nprint(\"Number of test examples: {}\".format(num_test_examples))", "Number of training examples: 240\nNumber of test examples: 60\n" ] ], [ [ "### Here's what the paths look like", "_____no_output_____" ] ], [ [ "x_train_filenames[:10]", "_____no_output_____" ], [ "y_train_filenames[:10]", "_____no_output_____" ], [ "y_test_filenames[:10]", "_____no_output_____" ] ], [ [ "# Visualize\nLet's take a look at some of the examples of different images in our dataset. ", "_____no_output_____" ] ], [ [ "display_num = 5\n\nr_choices = np.random.choice(num_train_examples, display_num)\n\nplt.figure(figsize=(10, 15))\nfor i in range(0, display_num * 2, 2):\n img_num = r_choices[i // 2]\n x_pathname = x_train_filenames[img_num]\n y_pathname = y_train_filenames[img_num]\n \n plt.subplot(display_num, 2, i + 1)\n plt.imshow(Image.open(x_pathname))\n plt.title(\"Original Image\")\n \n example_labels = Image.open(y_pathname)\n label_vals = np.unique(example_labels)\n \n plt.subplot(display_num, 2, i + 2)\n plt.imshow(example_labels)\n plt.title(\"Masked Image\")\n \nplt.suptitle(\"Examples of Images and their Masks\")\nplt.show()", "_____no_output_____" ] ], [ [ "# Set up", "_____no_output_____" ], [ "Let’s begin by setting up some parameters. We’ll standardize and resize all the shapes of the images. We’ll also set up some training parameters: ", "_____no_output_____" ] ], [ [ "# Set hyperparameters\nimage_size = 128\nimg_shape = (image_size, image_size, 3)\nbatch_size = 8\nmax_epochs = 100\nprint_steps = 50\nsave_epochs = 50\ntrain_dir = 'train/exp1'", "_____no_output_____" ] ], [ [ "# Build our input pipeline with `tf.data`\nSince we begin with filenames, we will need to build a robust and scalable data pipeline that will play nicely with our model. If you are unfamiliar with **tf.data** you should check out my other tutorial introducing the concept! \n\n### Our input pipeline will consist of the following steps:\n1. Read the bytes of the file in from the filename - for both the image and the label. Recall that our labels are actually images with each pixel annotated as car or background (1, 0). \n2. Decode the bytes into an image format\n3. Apply image transformations: (optional, according to input parameters)\n * `resize` - Resize our images to a standard size (as determined by eda or computation/memory restrictions)\n * The reason why this is optional is that U-Net is a fully convolutional network (e.g. with no fully connected units) and is thus not dependent on the input size. However, if you choose to not resize the images, you must use a batch size of 1, since you cannot batch variable image size together\n * Alternatively, you could also bucket your images together and resize them per mini-batch to avoid resizing images as much, as resizing may affect your performance through interpolation, etc.\n * `hue_delta` - Adjusts the hue of an RGB image by a random factor. This is only applied to the actual image (not our label image). The `hue_delta` must be in the interval `[0, 0.5]` \n * `horizontal_flip` - flip the image horizontally along the central axis with a 0.5 probability. This transformation must be applied to both the label and the actual image. \n * `width_shift_range` and `height_shift_range` are ranges (as a fraction of total width or height) within which to randomly translate the image either horizontally or vertically. This transformation must be applied to both the label and the actual image. \n * `rescale` - rescale the image by a certain factor, e.g. 1/ 255.\n4. Shuffle the data, repeat the data (so we can iterate over it multiple times across epochs), batch the data, then prefetch a batch (for efficiency).\n\nIt is important to note that these transformations that occur in your data pipeline must be symbolic transformations. ", "_____no_output_____" ], [ "#### Why do we do these image transformations?\nThis is known as **data augmentation**. Data augmentation \"increases\" the amount of training data by augmenting them via a number of random transformations. During training time, our model would never see twice the exact same picture. This helps prevent [overfitting](https://developers.google.com/machine-learning/glossary/#overfitting) and helps the model generalize better to unseen data.", "_____no_output_____" ], [ "## Processing each pathname", "_____no_output_____" ] ], [ [ "def _process_pathnames(fname, label_path):\n # We map this function onto each pathname pair\n img_str = tf.read_file(fname)\n img = tf.image.decode_bmp(img_str, channels=3)\n\n label_img_str = tf.read_file(label_path)\n label_img = tf.image.decode_bmp(label_img_str, channels=1)\n \n resize = [image_size, image_size]\n img = tf.image.resize_images(img, resize)\n label_img = tf.image.resize_images(label_img, resize)\n \n scale = 1 / 255.\n img = tf.to_float(img) * scale\n label_img = tf.to_float(label_img) * scale\n \n return img, label_img", "_____no_output_____" ], [ "def get_baseline_dataset(filenames,\n labels,\n threads=5,\n batch_size=batch_size,\n max_epochs=max_epochs,\n shuffle=True):\n num_x = len(filenames)\n # Create a dataset from the filenames and labels\n dataset = tf.data.Dataset.from_tensor_slices((filenames, labels))\n # Map our preprocessing function to every element in our dataset, taking\n # advantage of multithreading\n dataset = dataset.map(_process_pathnames, num_parallel_calls=threads)\n \n if shuffle:\n dataset = dataset.shuffle(num_x * 10)\n \n # It's necessary to repeat our data for all epochs \n dataset = dataset.repeat(max_epochs).batch(batch_size)\n return dataset", "_____no_output_____" ] ], [ [ "## Set up train and test datasets\nNote that we apply image augmentation to our training dataset but not our validation dataset.", "_____no_output_____" ] ], [ [ "train_ds = get_baseline_dataset(x_train_filenames,\n y_train_filenames)\ntest_ds = get_baseline_dataset(x_test_filenames,\n y_test_filenames,\n shuffle=False)", "_____no_output_____" ], [ "train_ds", "_____no_output_____" ] ], [ [ "### Plot some train data", "_____no_output_____" ] ], [ [ "temp_ds = get_baseline_dataset(x_train_filenames,\n y_train_filenames,\n batch_size=1,\n max_epochs=1,\n shuffle=False)\n# Let's examine some of these augmented images\ntemp_iter = temp_ds.make_one_shot_iterator()\nnext_element = temp_iter.get_next()\nwith tf.Session() as sess:\n batch_of_imgs, label = sess.run(next_element)\n\n # Running next element in our graph will produce a batch of images\n plt.figure(figsize=(10, 10))\n img = batch_of_imgs[0]\n\n plt.subplot(1, 2, 1)\n plt.imshow(img)\n\n plt.subplot(1, 2, 2)\n plt.imshow(label[0, :, :, 0])\n plt.show()", "_____no_output_____" ] ], [ [ "# Build the model\nWe'll build the U-Net model. U-Net is especially good with segmentation tasks because it can localize well to provide high resolution segmentation masks. In addition, it works well with small datasets and is relatively robust against overfitting as the training data is in terms of the number of patches within an image, which is much larger than the number of training images itself. Unlike the original model, we will add batch normalization to each of our blocks. \n\nThe Unet is built with an encoder portion and a decoder portion. The encoder portion is composed of a linear stack of [`Conv`](https://developers.google.com/machine-learning/glossary/#convolution), `BatchNorm`, and [`Relu`](https://developers.google.com/machine-learning/glossary/#ReLU) operations followed by a [`MaxPool`](https://developers.google.com/machine-learning/glossary/#pooling). Each `MaxPool` will reduce the spatial resolution of our feature map by a factor of 2. We keep track of the outputs of each block as we feed these high resolution feature maps with the decoder portion. The Decoder portion is comprised of UpSampling2D, Conv, BatchNorm, and Relus. Note that we concatenate the feature map of the same size on the decoder side. Finally, we add a final Conv operation that performs a convolution along the channels for each individual pixel (kernel size of (1, 1)) that outputs our final segmentation mask in grayscale. ", "_____no_output_____" ] ], [ [ "def conv_block(inputs, num_outputs, is_training, scope):\n batch_norm_params = {'decay': 0.9,\n 'epsilon': 0.001,\n 'is_training': is_training,\n 'scope': 'batch_norm'}\n with tf.variable_scope(scope) as scope:\n with slim.arg_scope([slim.conv2d],\n num_outputs=num_outputs,\n kernel_size=[3, 3],\n normalizer_fn=slim.batch_norm,\n normalizer_params=batch_norm_params):\n encoder = slim.conv2d(inputs, scope='conv1')\n encoder = slim.conv2d(encoder, scope='conv2')\n return encoder\n\ndef encoder_block(inputs, num_outputs, is_training, scope):\n with tf.variable_scope(scope) as scope:\n encoder = conv_block(inputs, num_outputs, is_training, scope)\n encoder_pool = slim.max_pool2d(encoder, kernel_size=[2, 2], scope='pool')\n \n return encoder_pool, encoder\n\ndef decoder_block(inputs, concat_tensor, num_outputs, is_training, scope):\n batch_norm_params = {'decay': 0.9,\n 'epsilon': 0.001,\n 'is_training': is_training,\n 'scope': 'batch_norm'}\n with tf.variable_scope(scope) as scope:\n decoder = slim.conv2d_transpose(inputs, num_outputs,\n kernel_size=[2, 2], stride=[2, 2],\n activation_fn=None, scope='convT')\n decoder = tf.concat([concat_tensor, decoder], axis=-1)\n decoder = slim.batch_norm(decoder, **batch_norm_params)\n decoder = tf.nn.relu(decoder)\n with slim.arg_scope([slim.conv2d],\n num_outputs=num_outputs,\n kernel_size=[3, 3],\n stride=[1, 1],\n normalizer_fn=slim.batch_norm,\n normalizer_params=batch_norm_params):\n decoder = slim.conv2d(decoder, scope='conv1')\n decoder = slim.conv2d(decoder, scope='conv2')\n return decoder", "_____no_output_____" ], [ "class UNet(object):\n def __init__(self, train_ds, test_ds):\n self.train_ds = train_ds\n self.test_ds = test_ds\n \n def build_images(self):\n # tf.data.Iterator.from_string_handle의 output_shapes는 default = None이지만 꼭 값을 넣는 게 좋음\n self.handle = tf.placeholder(tf.string, shape=[])\n self.iterator = tf.data.Iterator.from_string_handle(self.handle,\n self.train_ds.output_types,\n self.train_ds.output_shapes)\n self.input_images, self.targets = self.iterator.get_next()\n \n def inference(self, inputs, is_training, reuse=False):\n with tf.variable_scope('', reuse=reuse) as scope:\n # inputs: [128, 128, 3]\n encoder0_pool, encoder0 = encoder_block(inputs, 32, is_training, 'encoder0')\n # encoder0_pool: [64, 64, 32], encoder0: [128, 128, 32]\n encoder1_pool, encoder1 = encoder_block(encoder0_pool, 64, is_training, 'encoder1')\n # encoder1_pool: [32, 32, 64], encoder1: [64, 64, 64]\n encoder2_pool, encoder2 = encoder_block(encoder1_pool, 128, is_training, 'encoder2')\n # encoder2_pool: [16, 16, 128], encoder2: [32, 32, 128]\n encoder3_pool, encoder3 = encoder_block(encoder2_pool, 256, is_training, 'encoder3')\n # encoder3_pool: [8, 8, 256], encoder3: [16, 16, 256]\n center = conv_block(encoder3_pool, 512, is_training, 'center')\n # center: [8, 8, 512]\n decoder3 = decoder_block(center, encoder3, 256, is_training, 'decoder3')\n # decoder3 = [16, 16, 256]\n decoder2 = decoder_block(decoder3, encoder2, 128, is_training, 'decoder2')\n # decoder2 = [32, 32, 128]\n decoder1 = decoder_block(decoder2, encoder1, 64, is_training, 'decoder1')\n # decoder1 = [64, 64, 64]\n decoder0 = decoder_block(decoder1, encoder0, 32, is_training, 'decoder0')\n # decoder0 = [128, 128, 32]\n logits = slim.conv2d(decoder0, 1, [1, 1], activation_fn=None, scope='outputs')\n # logits = [128, 128, 1]\n\n return logits\n \n def dice_coeff(self, y_true, y_logits):\n smooth = 1.\n # Flatten\n y_true_f = tf.reshape(y_true, [-1])\n y_pred_f = tf.reshape(tf.nn.sigmoid(y_logits), [-1])\n intersection = tf.reduce_sum(y_true_f * y_pred_f)\n score = (2. * intersection + smooth) / (tf.reduce_sum(y_true_f) + tf.reduce_sum(y_pred_f) + smooth)\n return score\n \n def dice_loss(self, y_true, y_logits):\n loss = 1 - self.dice_coeff(y_true, y_logits)\n return loss\n \n def bce_dice_loss(self, y_true, y_logits):\n loss = tf.losses.sigmoid_cross_entropy(y_true, y_logits) + self.dice_loss(y_true, y_logits)\n return loss\n \n def build(self):\n self.global_step = tf.train.get_or_create_global_step()\n \n self.build_images()\n self.logits = self.inference(self.input_images, is_training=True)\n self.logits_val = self.inference(self.input_images, is_training=False, reuse=True)\n self.predicted_images = tf.nn.sigmoid(self.logits_val)\n\n self.loss = self.bce_dice_loss(self.targets, self.logits)\n \n print(\"complete model build.\")", "_____no_output_____" ] ], [ [ "### Create a model (UNet)", "_____no_output_____" ] ], [ [ "model = UNet(train_ds=train_ds, test_ds=test_ds)\nmodel.build()\n\n# show info for trainable variables\nt_vars = tf.trainable_variables()\nslim.model_analyzer.analyze_vars(t_vars, print_info=True)", "complete model build.\n---------\nVariables: name (type shape) [size]\n---------\nencoder0/conv1/weights:0 (float32_ref 3x3x3x32) [864, bytes: 3456]\nencoder0/conv1/batch_norm/beta:0 (float32_ref 32) [32, bytes: 128]\nencoder0/conv2/weights:0 (float32_ref 3x3x32x32) [9216, bytes: 36864]\nencoder0/conv2/batch_norm/beta:0 (float32_ref 32) [32, bytes: 128]\nencoder1/conv1/weights:0 (float32_ref 3x3x32x64) [18432, bytes: 73728]\nencoder1/conv1/batch_norm/beta:0 (float32_ref 64) [64, bytes: 256]\nencoder1/conv2/weights:0 (float32_ref 3x3x64x64) [36864, bytes: 147456]\nencoder1/conv2/batch_norm/beta:0 (float32_ref 64) [64, bytes: 256]\nencoder2/conv1/weights:0 (float32_ref 3x3x64x128) [73728, bytes: 294912]\nencoder2/conv1/batch_norm/beta:0 (float32_ref 128) [128, bytes: 512]\nencoder2/conv2/weights:0 (float32_ref 3x3x128x128) [147456, bytes: 589824]\nencoder2/conv2/batch_norm/beta:0 (float32_ref 128) [128, bytes: 512]\nencoder3/conv1/weights:0 (float32_ref 3x3x128x256) [294912, bytes: 1179648]\nencoder3/conv1/batch_norm/beta:0 (float32_ref 256) [256, bytes: 1024]\nencoder3/conv2/weights:0 (float32_ref 3x3x256x256) [589824, bytes: 2359296]\nencoder3/conv2/batch_norm/beta:0 (float32_ref 256) [256, bytes: 1024]\ncenter/conv1/weights:0 (float32_ref 3x3x256x512) [1179648, bytes: 4718592]\ncenter/conv1/batch_norm/beta:0 (float32_ref 512) [512, bytes: 2048]\ncenter/conv2/weights:0 (float32_ref 3x3x512x512) [2359296, bytes: 9437184]\ncenter/conv2/batch_norm/beta:0 (float32_ref 512) [512, bytes: 2048]\ndecoder3/convT/weights:0 (float32_ref 2x2x256x512) [524288, bytes: 2097152]\ndecoder3/convT/biases:0 (float32_ref 256) [256, bytes: 1024]\ndecoder3/batch_norm/beta:0 (float32_ref 512) [512, bytes: 2048]\ndecoder3/conv1/weights:0 (float32_ref 3x3x512x256) [1179648, bytes: 4718592]\ndecoder3/conv1/batch_norm/beta:0 (float32_ref 256) [256, bytes: 1024]\ndecoder3/conv2/weights:0 (float32_ref 3x3x256x256) [589824, bytes: 2359296]\ndecoder3/conv2/batch_norm/beta:0 (float32_ref 256) [256, bytes: 1024]\ndecoder2/convT/weights:0 (float32_ref 2x2x128x256) [131072, bytes: 524288]\ndecoder2/convT/biases:0 (float32_ref 128) [128, bytes: 512]\ndecoder2/batch_norm/beta:0 (float32_ref 256) [256, bytes: 1024]\ndecoder2/conv1/weights:0 (float32_ref 3x3x256x128) [294912, bytes: 1179648]\ndecoder2/conv1/batch_norm/beta:0 (float32_ref 128) [128, bytes: 512]\ndecoder2/conv2/weights:0 (float32_ref 3x3x128x128) [147456, bytes: 589824]\ndecoder2/conv2/batch_norm/beta:0 (float32_ref 128) [128, bytes: 512]\ndecoder1/convT/weights:0 (float32_ref 2x2x64x128) [32768, bytes: 131072]\ndecoder1/convT/biases:0 (float32_ref 64) [64, bytes: 256]\ndecoder1/batch_norm/beta:0 (float32_ref 128) [128, bytes: 512]\ndecoder1/conv1/weights:0 (float32_ref 3x3x128x64) [73728, bytes: 294912]\ndecoder1/conv1/batch_norm/beta:0 (float32_ref 64) [64, bytes: 256]\ndecoder1/conv2/weights:0 (float32_ref 3x3x64x64) [36864, bytes: 147456]\ndecoder1/conv2/batch_norm/beta:0 (float32_ref 64) [64, bytes: 256]\ndecoder0/convT/weights:0 (float32_ref 2x2x32x64) [8192, bytes: 32768]\ndecoder0/convT/biases:0 (float32_ref 32) [32, bytes: 128]\ndecoder0/batch_norm/beta:0 (float32_ref 64) [64, bytes: 256]\ndecoder0/conv1/weights:0 (float32_ref 3x3x64x32) [18432, bytes: 73728]\ndecoder0/conv1/batch_norm/beta:0 (float32_ref 32) [32, bytes: 128]\ndecoder0/conv2/weights:0 (float32_ref 3x3x32x32) [9216, bytes: 36864]\ndecoder0/conv2/batch_norm/beta:0 (float32_ref 32) [32, bytes: 128]\noutputs/weights:0 (float32_ref 1x1x32x1) [32, bytes: 128]\noutputs/biases:0 (float32_ref 1) [1, bytes: 4]\nTotal size of variables: 7761057\nTotal bytes of variables: 31044228\n" ], [ "opt = tf.train.AdamOptimizer(learning_rate=2e-4)\nwith tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):\n opt_op = opt.minimize(model.loss, global_step=model.global_step)", "_____no_output_____" ], [ "saver = tf.train.Saver(tf.global_variables(), max_to_keep=1000)", "_____no_output_____" ] ], [ [ "### Train a model", "_____no_output_____" ] ], [ [ "%%time\nsess = tf.Session(config=sess_config)\nsess.run(tf.global_variables_initializer())\ntf.logging.info('Start Session.')\n\ntrain_iterator = train_ds.make_one_shot_iterator()\ntrain_handle = sess.run(train_iterator.string_handle())\ntest_iterator = test_ds.make_one_shot_iterator()\ntest_handle = sess.run(test_iterator.string_handle())\n\n# save loss values for plot\nloss_history = []\npre_epochs = 0\nwhile True:\n try:\n start_time = time.time()\n _, global_step_, loss = sess.run([opt_op,\n model.global_step,\n model.loss],\n feed_dict={model.handle: train_handle})\n\n epochs = global_step_ * batch_size / float(num_train_examples)\n duration = time.time() - start_time\n\n if global_step_ % print_steps == 0:\n clear_output(wait=True)\n examples_per_sec = batch_size / float(duration)\n print(\"Epochs: {:.2f} global_step: {} loss: {:.3f} ({:.2f} examples/sec; {:.3f} sec/batch)\".format(\n epochs, global_step_, loss, examples_per_sec, duration))\n\n loss_history.append([epochs, loss])\n\n # print sample image\n img, label, predicted_label = sess.run([model.input_images, model.targets, model.predicted_images],\n feed_dict={model.handle: test_handle})\n plt.figure(figsize=(10, 20))\n plt.subplot(1, 3, 1)\n plt.imshow(img[0,: , :, :])\n plt.title(\"Input image\")\n \n plt.subplot(1, 3, 2)\n plt.imshow(label[0, :, :, 0])\n plt.title(\"Actual Mask\")\n \n plt.subplot(1, 3, 3)\n plt.imshow(predicted_label[0, :, :, 0])\n plt.title(\"Predicted Mask\")\n plt.show()\n\n # save model checkpoint periodically\n if int(epochs) % save_epochs == 0 and pre_epochs != int(epochs):\n tf.logging.info('Saving model with global step {} (= {} epochs) to disk.'.format(global_step_, int(epochs)))\n saver.save(sess, train_dir + 'model.ckpt', global_step=global_step_)\n pre_epochs = int(epochs)\n\n except tf.errors.OutOfRangeError:\n print(\"End of dataset\") # ==> \"End of dataset\"\n tf.logging.info('Saving model with global step {} (= {} epochs) to disk.'.format(global_step_, int(epochs)))\n saver.save(sess, train_dir + 'model.ckpt', global_step=global_step_)\n break\n\ntf.logging.info('complete training...')", "Epochs: 98.33 global_step: 2950 loss: 0.090 (248.52 examples/sec; 0.032 sec/batch)\n" ] ], [ [ "### Plot the loss", "_____no_output_____" ] ], [ [ "loss_history = np.asarray(loss_history)\nplt.plot(loss_history[:,0], loss_history[:,1])\nplt.show()", "_____no_output_____" ] ], [ [ "### Evaluate the test dataset and Plot", "_____no_output_____" ] ], [ [ "test_ds_eval = get_baseline_dataset(x_test_filenames,\n y_test_filenames,\n batch_size=num_test_examples,\n max_epochs=1,\n shuffle=False)\n\ntest_iterator_eval = test_ds_eval.make_one_shot_iterator()\ntest_handle_eval = sess.run(test_iterator_eval.string_handle())", "_____no_output_____" ], [ "mean_iou, mean_iou_op = tf.metrics.mean_iou(labels=tf.to_int32(tf.round(model.targets)),\n predictions=tf.to_int32(tf.round(model.predicted_images)),\n num_classes=2,\n name='mean_iou')\nsess.run(tf.local_variables_initializer())\n\nsess.run(mean_iou_op, feed_dict={model.handle: test_handle_eval})\nprint(\"mean iou:\", sess.run(mean_iou))", "mean iou: 0.87998605\n" ] ], [ [ "#### Visualize testset", "_____no_output_____" ] ], [ [ "test_ds_visual = get_baseline_dataset(x_test_filenames, \n y_test_filenames,\n batch_size=1,\n max_epochs=1,\n shuffle=False)\n\ntest_iterator_visual = test_ds_visual.make_one_shot_iterator()\ntest_handle_visual = sess.run(test_iterator_visual.string_handle())", "_____no_output_____" ], [ "# Let's visualize some of the outputs \n\n# Running next element in our graph will produce a batch of images\nplt.figure(figsize=(10, 20))\nfor i in range(5):\n #img, label, predicted_label = sess.run([model.input_images, model.targets, model.predicted_images],\n img, label, predicted_label = sess.run([model.input_images,\n tf.to_int32(tf.round(model.targets)),\n tf.to_int32(tf.round(model.predicted_images))],\n feed_dict={model.handle: test_handle_visual})\n\n plt.subplot(5, 3, 3 * i + 1)\n plt.imshow(img[0,: , :, :])\n plt.title(\"Input image\")\n\n plt.subplot(5, 3, 3 * i + 2)\n plt.imshow(label[0, :, :, 0])\n plt.title(\"Actual Mask\")\n\n plt.subplot(5, 3, 3 * i + 3)\n plt.imshow(predicted_label[0, :, :, 0])\n plt.title(\"Predicted Mask\")\nplt.suptitle(\"Examples of Input Image, Label, and Prediction\")\nplt.show()", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "Green's function\n==============", "_____no_output_____" ], [ "Fundamental solution\n-------------------------------", "_____no_output_____" ] ], [ [ "from sympy import *\ninit_printing()", "_____no_output_____" ], [ "x1, x2, xi1, xi2 = symbols('x_1 x_2 xi_1 xi_2')\nE = -1/(2*pi) * log(sqrt((x1-xi1)**2 + (x2-xi2)**2))\nE", "_____no_output_____" ] ], [ [ "**Task**: Check that $\\nabla^2_\\xi E = 0$ for $x \\neq \\xi$.\n\n*Hint*: https://docs.sympy.org/latest/tutorial/calculus.html#derivatives", "_____no_output_____" ] ], [ [ "diff(E,x,2)", "_____no_output_____" ] ], [ [ "Directional derivative\n------------------------------", "_____no_output_____" ] ], [ [ "n1, n2 = symbols('n_1 n_2')", "_____no_output_____" ] ], [ [ "**Task**: Compute the directional derivative $\\frac{\\partial E}{\\partial n}$.", "_____no_output_____" ], [ "**Task** (optional): Write a function which returns the directional derivative of an expression.", "_____no_output_____" ] ], [ [ "def ddn(expr):\n pass", "_____no_output_____" ] ], [ [ "Reflection principle\n----------------------------", "_____no_output_____" ], [ "For simple geometries Green's function can sometimes be found by reflecting the fundamental solution at the boundary and linearly combining the fundamental solution with its reflection. ", "_____no_output_____" ], [ "", "_____no_output_____" ], [ "**Task**: Based on $E$, find the solution for the half-space problem\n\\begin{align*}\n \\nabla^2G(x,\\xi) &= -\\delta(x-\\xi), & \\xi\\in\\Omega \\\\\n G(x,\\xi) &= 0, & \\xi\\in\\partial\\Omega \\\\\n \\Omega &= \\{\\xi\\in\\mathbb{R}^2 : \\xi_2 > 0\\}\n\\end{align*}\n\n*Hint*: https://docs.sympy.org/latest/tutorial/basic_operations.html#substitution", "_____no_output_____" ], [ "**Task**: Based on $E$, find the solution for the half-space problem\n\\begin{align*}\n \\nabla^2G(x,\\xi) &= -\\delta(x-\\xi), & \\xi\\in\\Omega \\\\\n \\frac{\\partial G(x,\\xi)}{\\partial n} &= 0, & \\xi\\in\\partial\\Omega \\\\\n \\Omega &= \\{\\xi\\in\\mathbb{R}^2 : \\xi_2 > 0\\}\n\\end{align*}", "_____no_output_____" ] ] ]

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

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "import pandas as pd\nimport scipy\nimport numpy\nfrom scipy import stats", "_____no_output_____" ], [ "data1=pd.read_csv(\"Cutlets.csv\")", "_____no_output_____" ], [ "data1.head()", "_____no_output_____" ], [ "unit_A=pd.Series(data1.iloc[:,0])\nunit_A", "_____no_output_____" ], [ "unit_B=pd.Series(data1.iloc[:,1])\nunit_B", "_____no_output_____" ], [ "p_value=stats.ttest_ind(unit_A,unit_B)[1]\np_value", "_____no_output_____" ], [ "#compare p value with 0.05\n#p value is > than 0.05 accept null hypothesis(there is no difffrence between two unit)", "_____no_output_____" ] ] ]

[ "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Multidimensional Pattern identification with SAX\n## In-site view\n\nThis script performs pattern identification over the {time, attribute} cuboid, that covers the intra-building frame. It serves for within-site exploration on how a given building operates across time and building-specific attributes.\n\nThe data is first normalized then transformed using SAX over normalized daily sequences. Motifs are identified across buildings, and a final clustering phase is executed over the reduced counts of sequences. \n\nResults are presented visually allowing interpretable analytics.", "_____no_output_____" ] ], [ [ "# Import modules\nimport pandas as pd\nimport numpy as np\nimport time\nfrom sklearn.cluster import KMeans\nimport sklearn.metrics as metrics\nfrom collections import Counter\nfrom sklearn.preprocessing import StandardScaler, MinMaxScaler\n# Plotting modules\nfrom plotly.offline import init_notebook_mode\ninit_notebook_mode(connected = True)\nimport matplotlib.pyplot as plt\nplt.rcdefaults()\n# Importing utility script\nimport utils as ut\n\n# Version\nversion = \"v1.0\"\n\n# Path definition\npath_data = \"..\\\\data\\\\cube\\\\\"\npath_fig_out = \"..\\\\figures\\\\insite_view\\\\\"", "_____no_output_____" ] ], [ [ "## Read", "_____no_output_____" ] ], [ [ "# Read Cuboid\nblg_id = \"Fox_education_Melinda\"\ndf = pd.read_csv(path_data + \"cuboid_B_\"+blg_id+\".csv\", index_col=\"timestamp\")\n\n# Format index to datetime object\ndf.index = pd.to_datetime(df.index, format='%Y-%m-%d %H:%M:%S')\ndf.head()", "_____no_output_____" ] ], [ [ "# Pre-Mining\n## Motifs identification", "_____no_output_____" ] ], [ [ "# SAX Parameters\nday_number_of_pieces = 4\nalphabet_size = 3\nscaler_function = StandardScaler()\n\n# Normalize per attribute\ndf_normalized = ut.scale_df_columns_NanRobust(df, df.columns, scaler=scaler_function)\n\n# Perform SAX transformation\nsax_dict, counts, sax_data = ut.SAX_mining(df_normalized, W=day_number_of_pieces, A=alphabet_size)\n\n# Plot the sequence counts per attribute\nfor meter in df.columns.values:\n fig = ut.counter_plot(counts[meter], title=meter)\n fig.savefig(path_fig_out+\"SAXcounts_StandardScaler_blg_\"+blg_id+\"_meter_\"+meter+\"_\"+version+\".jpg\", dpi=300, bbox_inches='tight')", "_____no_output_____" ], [ "# Reformating sax results for plotting\nsax_dict_data, index_map_dictionary = dict(), dict()\nfor meter in sax_data:\n sax_dict_data[meter], index_map_dictionary[meter] = ut.sax_df_reformat(sax_data, sax_dict, meter)\n\n# Plotting all SAX sequences and saving figure\nfig = ut.SAX_dailyhm_visualization(sax_dict_data, sax_dict, index_map_dictionary)\nut.png_output([len(sax_dict.keys())*250, 800])\nfig.show()\nfig.write_image(path_fig_out+\"SAX_blg_\"+blg_id+\"_\"+version+\".png\")", "_____no_output_____" ], [ "# Filter discords from established threshold\nthreshold = 10 # motif number threshold\nindexes = dict()\nfor meter in df.columns.values:\n df_count = pd.DataFrame.from_dict(Counter(sax_dict[meter]), orient='index').rename(columns={0:'count'})\n df_count.fillna(0)\n motifs = df_count[df_count.values > threshold]\n indexes[meter] = [i for i,x in enumerate(sax_dict[meter]) if x in list(motifs.index)] # returns all indexes", "_____no_output_____" ] ], [ [ "# Mining\n## Attribute motifs clustering\nAttribute daily profile motifs are clustered together resulting in a reduced number of typical patterns from the previous motif identification thanks to SAX trasnformation.", "_____no_output_____" ] ], [ [ "# Identify optimal cluster number\nwcss, sil = [], []\nfor meter in sax_data:\n wcss_l, sil_l = ut.elbow_method(sax_data[meter].iloc[indexes[meter]].interpolate(method='linear').transpose(), n_cluster_max=20)\n wcss.append(wcss_l)\n sil.append(sil_l)\n# Get similarity index quantiles (cross attributes)\narr_sil, arr_wcss = np.array(sil), np.array(wcss)\nwcss_med = np.quantile(arr_wcss, .5, axis=0)\nsil_med = np.quantile(arr_sil, .5, axis=0)\nerr_wcss = [np.quantile(arr_wcss, .25, axis=0), np.quantile(arr_wcss, .75, axis=0)]\nerr_sil = [np.quantile(arr_sil, .25, axis=0), np.quantile(arr_sil, .75, axis=0)]\n\n# Plots\nplt.rcParams.update({'font.size': 12})\nplt.rcParams['font.sans-serif'] = ['Times New Roman']\nfig = ut.similarity_index_werror_plot(wcss_med, sil_med, err_wcss, err_sil)\nfig.savefig(path_fig_out+\"blg_\"+blg_id+\"_cluster_SimilarityIndex_\"+version+\".jpg\", dpi=300, bbox_inches='tight')", "_____no_output_____" ], [ "## Clustering identified motifs\n\n# Cluster the identified motifs\nnb_clusters_opt = 4\nkmeans = KMeans(n_clusters=nb_clusters_opt, init='k-means++', max_iter=300, n_init=10, random_state=0)\nkmeans_pred_y, clust_sax_data = dict(), dict()\nfor meter in df.columns.values:\n clust_sax_data[meter] = sax_data[meter].iloc[indexes[meter]]\n kmeans_pred_y[meter] = kmeans.fit_predict(clust_sax_data[meter].interpolate(method='linear', limit_direction='both'))\n\n# Reformating cluster results for plotting\nclust_dict_data, index_map_dictionary = dict(), dict()\nmax_shape = 0\nfor meter in sax_data:\n clust_dict_data[meter], index_map_dictionary[meter] = ut.sax_df_reformat(clust_sax_data, kmeans_pred_y, meter)\n max_shape = max(max_shape, max(np.shape(clust_dict_data[meter])))\n# Adjusting reformaating from variable attribute motifs lengths\nfor meter in sax_data:\n # Defining width of empty dataframe to add\n space_btw_saxseq = max_shape - max(np.shape(clust_dict_data[meter]))\n # Creating empty frame\n empty_sax_df = pd.DataFrame(columns=sax_data[meter].columns, index=[' ']*space_btw_saxseq)\n # Adding empty frame to the df\n clust_dict_data[meter] = clust_dict_data[meter].append(empty_sax_df)\n\n\n# Plotting cluster results results\nfig = ut.SAX_dailyhm_visualization(clust_dict_data, sax_dict, index_map_dictionary)\nut.png_output([len(clust_dict_data.keys())*250, 800])\nfig.show()\nfig.write_image(path_fig_out+\"clust_blg_\"+blg_id+\"_\"+version+\".png\")", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import numpy as np\nimport matplotlib.pyplot as plt\nimport h5py", "_____no_output_____" ], [ "from tensorflow.keras.utils import to_categorical\nfrom sklearn.preprocessing import LabelEncoder", "_____no_output_____" ], [ "data_loc = '/gpfs/slac/atlas/fs1/d/rafaeltl/public/ML/L1RNN/datasets_2020_ff/'", "_____no_output_____" ], [ "file_str = 'Jan06_FlavFix_smear_1_std_xtd_zst.h5'", "_____no_output_____" ], [ "f5 = h5py.File(data_loc+file_str, 'r')", "_____no_output_____" ], [ "x_train = np.array( f5['x_train'] )\ny_train = to_categorical ( np.array( f5['y_train'] ) )\nw_train = np.array( f5['w_train'] )", "_____no_output_____" ], [ "y_train", "_____no_output_____" ], [ "from tensorflow.keras.layers import Dense, Activation, BatchNormalization, LSTM, Masking, Input, GRU\nfrom tensorflow.keras.models import Model\nfrom tensorflow.keras.optimizers import Adam\nfrom tensorflow.keras.regularizers import l1", "_____no_output_____" ], [ "from tensorflow.keras import regularizers", "_____no_output_____" ], [ "def lstmmodel(max_len, n_var, rec_units, ndense=[10], l1_reg=0,\n l2_reg=0, rec_act='sigmoid', extra_lab='none', rec_kernel_init='VarianceScaling',\n dense_kernel_init='lecun_uniform', domask=False):\n \n rec_layer = 'LSTM'\n \n track_inputs = Input(shape=(max_len, n_var,))\n \n if domask:\n hidden = Masking( mask_value=0, name=\"masking_1\")(track_inputs)\n else:\n hidden = track_inputs\n \n if l1_reg > 1e-6 and l2_reg > 1e-6:\n hidden = LSTM(units=rec_units,\n recurrent_activation = rec_act,\n kernel_initializer = rec_kernel_init, \n kernel_regularizer = regularizers.l1_l2(l1 = l1_reg, l2 = l2_reg),\n name = 'lstm1_l1l2')(hidden)\n elif l1_reg > 1e-6:\n hidden = LSTM(units=rec_units,\n recurrent_activation = rec_act,\n kernel_initializer = rec_kernel_init, \n kernel_regularizer = regularizers.l1(l1 = l1_reg),\n name = 'lstm1_l1')(hidden)\n elif l2_reg > 1e-6:\n hidden = LSTM(units=rec_units,\n recurrent_activation = rec_act,\n kernel_initializer = rec_kernel_init, \n kernel_regularizer = regularizers.l2(l2 = l2_reg),\n name = 'lstm1_l2')(hidden)\n else:\n hidden = LSTM(units=rec_units,\n recurrent_activation = rec_act,\n kernel_initializer = rec_kernel_init, \n name = 'lstm1')(hidden)\n\n for ind,nd in enumerate(ndense):\n hidden = Dense(nd, activation='relu', kernel_initializer=dense_kernel_init, name=f'dense_{ind}' )(hidden)\n \n output = Dense(3, activation='softmax', kernel_initializer=dense_kernel_init, name = 'output_softmax')(hidden)\n \n model = Model(inputs=track_inputs, outputs=output)\n \n d_layers = ''.join([ str(dl) for dl in ndense ])\n \n if domask:\n mname = f'MASKED_rnn_{rec_layer}.{rec_units}_Dense.{d_layers}_'\n else:\n mname = f'rnn_{rec_layer}.{rec_units}_Dense.{d_layers}_'\n mname += f'LSTMKernelInit.{rec_kernel_init}_DenseKernelInit.{dense_kernel_init}'\n mname += f'KRl1.{l1_reg}_KRl2.{l2_reg}_recAct.{rec_act}' #LSTM kernel regularizer\n \n if 'none' not in extra_lab:\n mname += f'_{extra_lab}'\n \n return model, mname\n\n# mask = Masking( mask_value=0, name=\"masking_1\")(track_inputs)\n##########################################\n# use_bias=False,\n# activation='relu',\n# recurrent_activation='relu',\n# kernel_regularizer = regularizers.l1_l2(l1= 0.001, l2 = 0.0001), \n# bias_regularizer = regularizers.l1_l2(l1= 1, l2 = 1), \n# activity_regularizer=regularizers.l1_l2(l1= 0.001, l2 = 0.0001),\n##########################################\n", "_____no_output_____" ], [ "def grumodel(max_len, n_var, rec_units, ndense=[10], l1_reg=0,\n l2_reg=0, rec_act='sigmoid', extra_lab='none', rec_kernel_init='VarianceScaling',\n dense_kernel_init='lecun_uniform', domask=False):\n \n rec_layer = 'GRU'\n \n track_inputs = Input(shape=(max_len, n_var,))\n \n if domask:\n hidden = Masking( mask_value=0, name=\"masking_1\")(track_inputs)\n else:\n hidden = track_inputs\n \n\n if l1_reg > 1e-6 and l2_reg > 1e-6:\n hidden = GRU(units=rec_units,\n kernel_initializer = rec_kernel_init, \n kernel_regularizer = regularizers.l1_l2(l1 = l1_reg, l2 = l2_reg),\n name = 'gru_l1l2')(hidden)\n elif l1_reg > 1e-6:\n hidden = GRU(units=rec_units,\n recurrent_activation = rec_act,\n kernel_initializer = rec_kernel_init, \n kernel_regularizer = regularizers.l1(l1 = l1_reg),\n name = 'gru_l1')(hidden)\n elif l2_reg > 1e-6:\n hidden = GRU(units=rec_units,\n recurrent_activation = rec_act,\n kernel_initializer = rec_kernel_init, \n kernel_regularizer = regularizers.l2(l2 = l2_reg),\n name = 'gru_l2')(hidden)\n else:\n hidden = GRU(units=rec_units,\n recurrent_activation = rec_act,\n kernel_initializer = rec_kernel_init, \n name = 'gru')(hidden)\n \n\n for ind,nd in enumerate(ndense):\n hidden = Dense(nd, activation='relu', kernel_initializer=dense_kernel_init, name=f'dense_{ind}' )(hidden)\n \n output = Dense(3, activation='softmax', kernel_initializer=dense_kernel_init, name = 'output_softmax')(hidden)\n \n model = Model(inputs=track_inputs, outputs=output)\n \n d_layers = ''.join([ str(dl) for dl in ndense ])\n \n if domask:\n mname = f'MASKED_rnn_{rec_layer}.{rec_units}_Dense.{d_layers}_'\n else:\n mname = f'rnn_{rec_layer}.{rec_units}_Dense.{d_layers}_'\n mname += f'LSTMKernelInit.{rec_kernel_init}_DenseKernelInit.{dense_kernel_init}'\n mname += f'KRl1.{l1_reg}_KRl2.{l2_reg}_recAct.{rec_act}' #LSTM kernel regularizer\n \n if 'none' not in extra_lab:\n mname += f'_{extra_lab}'\n \n return model, mname\n\n# mask = Masking( mask_value=0, name=\"masking_1\")(track_inputs)\n##########################################\n# use_bias=False,\n# activation='relu',\n# recurrent_activation='relu',\n# kernel_regularizer = regularizers.l1_l2(l1= 0.001, l2 = 0.0001), \n# bias_regularizer = regularizers.l1_l2(l1= 1, l2 = 1), \n# activity_regularizer=regularizers.l1_l2(l1= 0.001, l2 = 0.0001),\n##########################################\n", "_____no_output_____" ], [ "l1_reg = 0\nl2_reg = 0\n\n## GRU Model\n\n# model, model_name = grumodel(15, 6, 120, [50, 10], l1_reg=l1_reg, l2_reg=l2_reg)\n\n## LSTM Model\n\nmodel, model_name = lstmmodel(15, 6, 120, [50, 10], l1_reg=l1_reg, l2_reg=l2_reg)\n\n# Masked model\n\n# model, model_name = lstmmodel(15, 6, 20, [10], l1_reg=l1_reg, l2_reg=l2_reg, domask=True)\n\n\n# ## Very very tiny model\n\n# model, model_name = lstmmodel(15, 6, 2, [], l1_reg=l1_reg, l2_reg=l2_reg)\n\n# ## Very tiny model\n\n# model, model_name = lstmmodel(15, 6, 10, [], l1_reg=l1_reg, l2_reg=l2_reg)\n\n# ## Tiny model\n\n# model, model_name = lstmmodel(15, 6, 10, [10], l1_reg=l1_reg, l2_reg=l2_reg)\n\n# ## Small model\n\n# model, model_name = lstmmodel(15, 6, 20, [10], l1_reg=l1_reg, l2_reg=l2_reg)\n\n# ## Little model\n\n# model, model_name = lstmmodel(15, 6, 50, [10], l1_reg=l1_reg, l2_reg=l2_reg)\n\n# ## Intermediate model\n\n# model, model_name = lstmmodel(15, 6, 50, [10, 10], l1_reg=l1_reg, l2_reg=l2_reg)\n\n# ## Large model\n\n# model, model_name = lstmmodel(15, 6, 100, [50, 10], l1_reg=l1_reg, l2_reg=l2_reg)\n\n# model, model_name = lstmmodel(15, 6, 100, [10], l1_reg=l1_reg, l2_reg=l2_reg)\n\n# model, model_name = lstmmodel(15, 6, 100, [10], l1_reg=l1_reg, l2_reg=l2_reg)", "_____no_output_____" ], [ "model.summary()\nprint(model_name)", "Model: \"model\"\n_________________________________________________________________\nLayer (type) Output Shape Param # \n=================================================================\ninput_1 (InputLayer) [(None, 15, 6)] 0 \n_________________________________________________________________\nlstm1 (LSTM) (None, 120) 60960 \n_________________________________________________________________\ndense_0 (Dense) (None, 50) 6050 \n_________________________________________________________________\ndense_1 (Dense) (None, 10) 510 \n_________________________________________________________________\noutput_softmax (Dense) (None, 3) 33 \n=================================================================\nTotal params: 67,553\nTrainable params: 67,553\nNon-trainable params: 0\n_________________________________________________________________\nrnn_LSTM.120_Dense.5010_LSTMKernelInit.VarianceScaling_DenseKernelInit.lecun_uniformKRl1.0_KRl2.0_recAct.sigmoid\n" ], [ "model_json = model.to_json()\nwith open(f'keras/model_{model_name}_arch.json', \"w\") as json_file:\n json_file.write(model_json)", "_____no_output_____" ], [ "# adam = Adam(learning_rate=0.01)\nmodel.compile(optimizer='adam', loss=['categorical_crossentropy'], metrics=['accuracy'])", "_____no_output_____" ], [ "from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint", "_____no_output_____" ], [ "model_output = f'keras/model_{model_name}_weights.h5'", "_____no_output_____" ], [ "train = True", "_____no_output_____" ], [ "if train:\n history = model.fit( x_train , y_train,\n batch_size=2**14,\n # epochs=10,\n epochs=150,\n validation_split=0.1,\n shuffle = True,\n sample_weight= w_train,\n callbacks = [\n EarlyStopping(verbose=True, patience=20, monitor='val_accuracy'),\n ModelCheckpoint(model_output, monitor='val_accuracy', verbose=True, save_best_only=True)\n ],\n verbose=True\n )\n \nmodel.load_weights(model_output)", "_____no_output_____" ], [ "model.summary()", "_____no_output_____" ], [ "x_test = np.array( f5['x_test'] )\ny_test = to_categorical ( np.array( f5['y_test'] ) )", "_____no_output_____" ], [ "import plotting\nimport matplotlib.pyplot as plt\nfrom sklearn.metrics import accuracy_score", "_____no_output_____" ], [ "import importlib", "_____no_output_____" ], [ "importlib.reload(plotting)", "_____no_output_____" ], [ "pred_test = model.predict(x_test, batch_size=2**10)", "_____no_output_____" ], [ "print(\"Accuracy: {}\".format(accuracy_score(np.argmax(y_test, axis=1), np.argmax(pred_test, axis=1))))", "_____no_output_____" ], [ "pb_b = pred_test[:,0] [y_test[:,0] == 1]\npc_b = pred_test[:,1] [y_test[:,0] == 1]\npl_b = pred_test[:,2] [y_test[:,0] == 1]\n \npc_c = pred_test[:,1] [y_test[:,1] == 1]\npb_c = pred_test[:,0] [y_test[:,1] == 1]\n \npl_l = pred_test[:,2] [y_test[:,2] == 1]\npb_l = pred_test[:,0] [y_test[:,2] == 1]\n\nplt.Figure()\n\nplt.hist( pb_b/(pb_b+pl_b), range=(0,1), bins=1000, histtype='step' )\nplt.hist( pb_l/(pb_l+pl_l), range=(0,1), bins=1000, histtype='step' )\n\nplt.show()\n\n\nplt.Figure()\n\nplt.hist( pb_b/(pb_b+pc_b), range=(0,1), bins=1000, histtype='step' )\nplt.hist( pb_c/(pb_c+pc_c), range=(0,1), bins=1000, histtype='step' )\n\nplt.show()", "_____no_output_____" ], [ "plt.figure(figsize=(9,9))\n_ = plotting.makeRoc(y_test, pred_test)", "_____no_output_____" ], [ "for layer in model.layers:\n print(layer.name)\n# plt.Figure()\n \n this_wgts = layer.get_weights()\n# if len(this_wgts) < 1: continue\n print(layer.get_config())\n \n for wgt in this_wgts:\n print(wgt)\n print()\n# max_wgts = np.max(this_wgts)\n# min_wgts = np.min(this_wgts)\n# plt.hist(this_wgts, bins=100, range=(min_wgts, max_wgts))\n# plt.xlabel(f'{layer.name}')\n# plt.show", "_____no_output_____" ] ] ]

[ "code" ]

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# Practical Session 2: Classification algorithms\n\n*Notebook by Ekaterina Kochmar*", "_____no_output_____" ], [ "## 0.1 Your task\n\nIn practical 1, you worked with the housing prices and bike sharing datasets on the tasks that required you to predict some value (e.g., price of a house) or amount (e.g., the count of rented bikes, or the number of registered users) based on a number of attributes – age of the house, number of rooms, income level of the house owners for the house price prediction (or weather conditions and time of the day for the prediction of the number of rented bikes). That is, you were predicting some continuous value.\n\nThis time, your task is to predict a particular category the instance belongs to based on its characteristics. This type of tasks is called *classification*.", "_____no_output_____" ], [ "# Assignment: Handwritten digits dataset\n\nThe dataset that you will use in this assignment is the [*digits* dataset](http://scikit-learn.org/stable/modules/generated/sklearn.datasets.load_digits.html) which contains $1797$ images of $10$ hand-written digits. The digits have been preprocessed so that $32 \\times 32$ bitmaps are divided into non-overlapping blocks of $4 \\times 4$ and the number of on pixels are counted in each block. This generates an input matrix of $8 \\times 8$ where each element is an integer in the range of $[0, ..., 16]$. This reduces dimensionality and gives invariance to small distortions.\n\nFor further information on NIST preprocessing routines applied to this data, see M. D. Garris, J. L. Blue, G. T. Candela, D. L. Dimmick, J. Geist, P. J. Grother, S. A. Janet, and C. L. Wilson, *NIST Form-Based Handprint Recognition System*, NISTIR 5469, 1994.\n\nAs before, use the `sklearn`'s data uploading routines to load the dataset and get the data fields:", "_____no_output_____" ] ], [ [ "from sklearn import datasets\ndigits = datasets.load_digits()\nlist(digits.keys())", "_____no_output_____" ], [ "digits", "_____no_output_____" ], [ "X, y = digits[\"data\"], digits[\"target\"]\nX.shape", "_____no_output_____" ], [ "y.shape", "_____no_output_____" ] ], [ [ "You can access the digits and visualise them using the following code (feel free to select another digit):", "_____no_output_____" ] ], [ [ "import matplotlib\nfrom matplotlib import pyplot as plt\n\nsome_digit = X[3]\nsome_digit_image = some_digit.reshape(8, 8)\n\nplt.imshow(some_digit_image, cmap=matplotlib.cm.binary, interpolation=\"nearest\")\nplt.axis(\"off\")\nplt.show()", "_____no_output_____" ], [ "y[3]", "_____no_output_____" ] ], [ [ "For the rest of the practical, apply the data preprocessing techniques, implement and evaluate the classification models on the digits dataset using the steps that you applied above to the iris dataset.", "_____no_output_____" ], [ "## Step 2: Splitting the data into training and test subsets", "_____no_output_____" ] ], [ [ "from sklearn.model_selection import StratifiedShuffleSplit\n\nsplit = StratifiedShuffleSplit(n_splits=1, test_size=0.2, random_state=42)\nsplit.get_n_splits(X, y)\nprint(split) \n\nfor train_index, test_index in split.split(X, y):\n print(\"TRAIN:\", len(train_index), \"TEST:\", len(test_index))\n X_train, X_test = X[train_index], X[test_index]\n y_train, y_test = y[train_index], y[test_index]\n\nprint(X_train.shape, y_train.shape, X_test.shape, y_test.shape)", "StratifiedShuffleSplit(n_splits=1, random_state=42, test_size=0.2,\n train_size=None)\nTRAIN: 1437 TEST: 360\n(1437, 64) (1437,) (360, 64) (360,)\n" ] ], [ [ "Check proportions", "_____no_output_____" ] ], [ [ "import pandas as pd\n\n# def original_proportions(data):\n# props = {}\n# for value in set(data[\"target\"]):\n# data_value = [i for i in data[\"target\"] if i==value]\n# props[value] = len(data_value) / len(data[\"target\"])\n# return props\n\ndef subset_proportions(subset):\n props = {}\n for value in set(subset):\n data_value = [i for i in subset if i==value]\n props[value] = len(data_value) / len(subset)\n return props\n\n \ncompare_props = pd.DataFrame({\n \"Overall\": subset_proportions(digits[\"target\"]),\n \"Stratified tr\": subset_proportions(y_train),\n \"Stratified ts\": subset_proportions(y_test),\n})\ncompare_props[\"Strat. tr %error\"] = 100 * compare_props[\"Stratified tr\"] / compare_props[\"Overall\"] - 100\ncompare_props[\"Strat. ts %error\"] = 100 * compare_props[\"Stratified ts\"] / compare_props[\"Overall\"] - 100\n\ncompare_props.sort_index()", "_____no_output_____" ] ], [ [ "# Case 1: Binary Classification", "_____no_output_____" ] ], [ [ "y_train_zero = (y_train == 0) # will return True when the label is 0 (i.e., zero)\ny_test_zero = (y_test == 0)\ny_test_zero", "_____no_output_____" ], [ "zero_example = X_test[10]\n", "_____no_output_____" ] ], [ [ "## Perceptron", "_____no_output_____" ] ], [ [ "from sklearn.linear_model import SGDClassifier\n\nsgd = SGDClassifier(max_iter=5, tol=None, random_state=42,\n loss=\"perceptron\", eta0=1, learning_rate=\"constant\", penalty=None)\nsgd.fit(X_train, y_train_zero)\nsgd.predict([zero_example])", "_____no_output_____" ] ], [ [ "Trying it for label 1", "_____no_output_____" ] ], [ [ "y_train_one = (y_train == 1) # True when the label is 1 (i.e., versicolor)\ny_test_one = (y_test == 1)\ny_test_one", "_____no_output_____" ], [ "one_example = X_test[40]\nprint(\"Class\", y_test[40], \"(\", digits.target_names[y_test[40]], \")\")\n\nsgd.fit(X_train, y_train_one)\nprint(sgd.predict([one_example]))", "Class 1 ( 1 )\n[ True]\n" ] ], [ [ "Perceptron did well", "_____no_output_____" ], [ "## Logistic Regression", "_____no_output_____" ] ], [ [ "from sklearn.linear_model import LogisticRegression\n\nlog_reg = LogisticRegression()\nlog_reg.fit(X_train, y_train_zero)\nprint(log_reg.predict([zero_example]))", "[ True]\n" ], [ "log_reg.fit(X_train, y_train_one)\nlog_reg.predict([one_example])", "_____no_output_____" ] ], [ [ "Looks like Logistic regression didn't get the 1", "_____no_output_____" ], [ "## Naive Bayes", "_____no_output_____" ] ], [ [ "from sklearn.naive_bayes import GaussianNB, MultinomialNB\n\ngnb = MultinomialNB() # or:\ngnb = GaussianNB() \ngnb.fit(X_train, y_train_zero)\ngnb.predict([zero_example])", "_____no_output_____" ], [ "gnb.fit(X_train, y_train_one)\n\ngnb.predict([one_example])", "_____no_output_____" ] ], [ [ "Naive Bayes did good", "_____no_output_____" ], [ "# Step 3: Evaluation\n## Performance measures\n- Acc for cross-val", "_____no_output_____" ] ], [ [ "from sklearn.model_selection import cross_val_score\n\nprint(cross_val_score(log_reg, X_train, y_train_zero, cv=5, scoring=\"accuracy\"))\nprint(cross_val_score(gnb, X_train, y_train_zero, cv=5, scoring=\"accuracy\"))\nprint(cross_val_score(sgd, X_train, y_train_zero, cv=5, scoring=\"accuracy\"))", "[1. 1. 0.99651568 1. 1. ]\n[0.98958333 0.99305556 0.99651568 1. 0.9825784 ]\n[0.99652778 0.99652778 0.99651568 1. 1. ]\n" ], [ "print(cross_val_score(log_reg, X_train, y_train_one, cv=5, scoring=\"accuracy\"))\nprint(cross_val_score(gnb, X_train, y_train_one, cv=5, scoring=\"accuracy\"))\nprint(cross_val_score(sgd, X_train, y_train_one, cv=5, scoring=\"accuracy\"))", "[0.97916667 0.97916667 0.96515679 0.97212544 0.95470383]\n[0.61805556 0.62847222 0.61324042 0.66550523 0.51916376]\n[0.97222222 0.95486111 0.95470383 0.95470383 0.95818815]\n" ] ], [ [ "Brute force predicting only non-ones", "_____no_output_____" ] ], [ [ "from sklearn.base import BaseEstimator\nimport numpy as np\nnp.random.seed(42)\n\nclass NotXClassifier(BaseEstimator):\n def fit(self, X, y=None):\n pass\n def predict(self, X):\n return np.zeros((len(X), 1), dtype=bool)\n \nnotone_clf = NotXClassifier()\ncross_val_score(notone_clf, X_train, y_train_one, cv=5, scoring=\"accuracy\")", "_____no_output_____" ] ], [ [ "- Confusion Matrix", "_____no_output_____" ] ], [ [ "from sklearn.model_selection import cross_val_predict\nfrom sklearn.metrics import confusion_matrix\n\ny_train_pred = cross_val_predict(log_reg, X_train, y_train_zero, cv=5)\nconfusion_matrix(y_train_zero, y_train_pred)", "_____no_output_____" ], [ "y_train_pred = cross_val_predict(gnb, X_train, y_train_zero, cv=5)\nconfusion_matrix(y_train_zero, y_train_pred)", "_____no_output_____" ], [ "y_train_pred = cross_val_predict(log_reg, X_train, y_train_one, cv=5)\nconfusion_matrix(y_train_one, y_train_pred)", "_____no_output_____" ], [ "y_train_pred = cross_val_predict(gnb, X_train, y_train_one, cv=5)\nconfusion_matrix(y_train_one, y_train_pred)", "_____no_output_____" ] ], [ [ "- precision, recall, f1", "_____no_output_____" ] ], [ [ "from sklearn.metrics import precision_score, recall_score, f1_score\n\ny_train_pred = cross_val_predict(gnb, X_train, y_train_one, cv=5)\nprecision = precision_score(y_train_one, y_train_pred) # == 36 / (36 + 5)\nrecall = recall_score(y_train_one, y_train_pred) # == 36 / (36 + 4)\nf1 = f1_score(y_train_one, y_train_pred)\nprint(precision, recall, f1)\n\ny_train_pred = cross_val_predict(log_reg, X_train, y_train_one, cv=5)\nprecision = precision_score(y_train_one, y_train_pred) # == 15 / (15 + 9)\nrecall = recall_score(y_train_one, y_train_pred) # == 15 / (15 + 25)\nf1 = f1_score(y_train_one, y_train_pred)\nprint(precision, recall, f1)", "0.20454545454545456 0.9863013698630136 0.3388235294117647\n0.832258064516129 0.8835616438356164 0.8571428571428571\n" ] ], [ [ "Oh no, poor gnb", "_____no_output_____" ], [ "- Precision-recall treade-off", "_____no_output_____" ], [ "Confidence score", "_____no_output_____" ] ], [ [ "log_reg.fit(X_train, y_train_one)\n\ny_scores = log_reg.decision_function([one_example])\ny_scores", "_____no_output_____" ], [ "threshold = 0\ny_one_pred = (y_scores > threshold)\ny_one_pred", "_____no_output_____" ], [ "threshold = -2\ny_one_pred = (y_scores > threshold)\ny_one_pred", "_____no_output_____" ] ], [ [ "Confidence scores", "_____no_output_____" ] ], [ [ "y_scores = cross_val_predict(log_reg, X_train, y_train_one, cv=5, method=\"decision_function\")\ny_scores", "_____no_output_____" ] ], [ [ "Plot precision vs recall", "_____no_output_____" ] ], [ [ "from sklearn.metrics import precision_recall_curve\n \nprecisions, recalls, thresholds = precision_recall_curve(y_train_one, y_scores)\n\ndef plot_pr_vs_threshold(precisions, recalls, thresholds):\n plt.plot(thresholds, precisions[:-1], \"b--\", label=\"Precision\")\n plt.plot(thresholds, recalls[:-1], \"g--\", label=\"Recall\")\n plt.xlabel(\"Threshold\")\n plt.legend(loc=\"upper right\")\n plt.ylim([0, 1])\n \nplot_pr_vs_threshold(precisions, recalls, thresholds)\nplt.show()", "_____no_output_____" ], [ "def plot_precision_vs_recall(precisions, recalls):\n plt.plot(recalls, precisions, \"b-\", linewidth=2)\n plt.xlabel(\"Recall\")\n plt.ylabel(\"Precision\")\n plt.axis([0, 1, 0, 1])\n\nplot_precision_vs_recall(precisions, recalls)\nplt.show()", "_____no_output_____" ] ], [ [ "- The Receiver Operating Characteristic (ROC)", "_____no_output_____" ] ], [ [ "from sklearn.metrics import roc_curve\n\nfpr, tpr, thresholds = roc_curve(y_train_one, y_scores)\n\ndef plot_roc_curve(fpr, tpr, label=None):\n plt.plot(fpr, tpr, linewidth=2, label=label)\n plt.plot([0, 1], [0, 1], \"k--\")\n plt.axis([0, 1, 0, 1.01])\n plt.xlabel(\"False positive rate (fpr)\")\n plt.ylabel(\"True positive rate (tpr)\")\n \nplot_roc_curve(fpr, tpr)\nplt.show()", "_____no_output_____" ], [ "# Area\nfrom sklearn.metrics import roc_auc_score\nroc_auc_score(y_train_one, y_scores)", "_____no_output_____" ], [ "# Now with GNB\ny_probas_gnb = cross_val_predict(gnb, X_train, y_train_one, cv=3, method=\"predict_proba\")\ny_scores_gnb = y_probas_gnb[:, 1] # score = proba of the positive class\nfpr_gnb, tpr_gnb, thresholds_gnb = roc_curve(y_train_one, y_scores_gnb)\n\nplt.plot(fpr, tpr, \"b:\", label=\"Logistic Regression\")\nplot_roc_curve(fpr_gnb, tpr_gnb, \"Gaussian Naive Bayes\")\nplt.legend(loc=\"lower right\")\nplt.show()", "_____no_output_____" ], [ "#Area\nroc_auc_score(y_train_one, y_scores_gnb)", "_____no_output_____" ] ], [ [ "Looks like Logistic Regression outperformed Gaussian Naive Bayes", "_____no_output_____" ], [ "# Step 4: Data transformations", "_____no_output_____" ], [ "## Kernel trick\n\n- with gamma = 1 we get EXTREMELY bad results\n- gamma = 0.001 solves that", "_____no_output_____" ] ], [ [ "from sklearn.kernel_approximation import RBFSampler\n\nrbf_features = RBFSampler(gamma=0.001, random_state=42)\nX_train_features = rbf_features.fit_transform(X_train)\nprint(X_train.shape, \"->\", X_train_features.shape)\nsgd_rbf = SGDClassifier(max_iter=100, random_state=42, loss=\"perceptron\", \n eta0=1, learning_rate=\"constant\", penalty=None)\nsgd_rbf.fit(X_train_features, y_train_one) \n\nsgd_rbf.score(X_train_features, y_train_one)\nprint(X_train_features[0])", "(1437, 64) -> (1437, 100)\n[-0.08948691 0.07840032 0.13012926 0.01843734 0.05243378 0.12363736\n -0.13138175 0.12487656 -0.06774296 -0.13116408 0.14131867 0.13014406\n -0.1412175 0.01781151 0.08980399 0.14045931 0.06150205 0.11652648\n 0.13544217 -0.11087697 0.13594337 -0.0800289 0.0574227 0.01216671\n 0.11133797 0.00604765 0.12907269 0.04008129 0.10124134 0.14130664\n 0.09733658 -0.14111269 0.11467299 -0.03910098 -0.05214749 -0.05723397\n -0.02252198 -0.1064269 0.00072984 -0.08188124 0.01504524 -0.1212134\n -0.0339027 0.10711778 0.01232271 -0.10386685 -0.08298496 0.13956306\n -0.03454778 0.14113989 -0.09677051 0.03187626 -0.07078854 -0.12390397\n 0.13693932 0.09349667 -0.12903172 0.0018465 -0.02683269 -0.062455\n 0.14121793 -0.01998847 0.13880371 0.13414756 -0.14132905 0.13276154\n -0.14141921 -0.05054704 0.12889829 0.13459871 -0.03282508 0.13367935\n -0.06263253 -0.11907552 0.14105804 0.13411986 0.06823374 0.08644726\n 0.09729963 0.14135676 -0.04737141 0.0218788 0.09904029 -0.12565361\n 0.1260095 0.04542973 0.08625159 -0.06465836 0.09918457 0.13192078\n 0.10236442 0.13360416 -0.081419 -0.09102759 0.13254435 -0.05242659\n 0.04783216 -0.14066595 -0.02853276 -0.11711412]\n" ] ], [ [ "- Precision, recall and F1 : non-kernel vs kernel GNB", "_____no_output_____" ] ], [ [ "y_train_pred = cross_val_predict(sgd, X_train, y_train_one, cv=5)\nprecision = precision_score(y_train_one, y_train_pred)\nrecall = recall_score(y_train_one, y_train_pred)\nf1 = f1_score(y_train_one, y_train_pred)\nprint(precision, recall, f1)\n\ny_train_pred = cross_val_predict(sgd_rbf, X_train_features, y_train_one, cv=5)\nprecision = precision_score(y_train_one, y_train_pred)\nrecall = recall_score(y_train_one, y_train_pred)\nf1 = f1_score(y_train_one, y_train_pred)\nprint(precision, recall, f1)", "0.8270676691729323 0.7534246575342466 0.7885304659498208\n0.9154929577464789 0.8904109589041096 0.9027777777777778\n" ] ], [ [ "## Case 2: Multi-class classification", "_____no_output_____" ] ], [ [ "sgd.fit(X_train, y_train) # i.e., all instances, not just one class\nprint(sgd.predict([zero_example]))\nprint(sgd.predict([one_example]))", "[0]\n[1]\n" ] ], [ [ "half good", "_____no_output_____" ] ], [ [ "sgd_rbf.fit(X_train_features, y_train) # i.e., all instances, not just one class\nX_test_features = rbf_features.transform(X_test)\nzero_rbf_example = X_test_features[10] # note that you need to transform the test data in the same way, too\none_rbf_example = X_test_features[3]\n\nprint(sgd_rbf.predict([zero_rbf_example]))\nprint(sgd_rbf.predict([one_rbf_example]))", "[0]\n[6]\n" ] ], [ [ "half good", "_____no_output_____" ] ], [ [ "zero_scores = sgd_rbf.decision_function([zero_rbf_example])\nprint(zero_scores)\n\n# check which class gets the maximum score\nprediction = np.argmax(zero_scores)\nprint(prediction)\n# check which class this corresponds to in the classifier\nprint(sgd_rbf.classes_[prediction])\nprint(digits.target_names[sgd_rbf.classes_[prediction]])\n", "[[ 1.32752637 -8.43047571 -0.56672488 -3.41607774 -2.43529497 -2.63031545\n -3.12302686 -2.30546944 -3.48919124 -6.71624512]]\n0\n0\n0\n" ] ], [ [ "good", "_____no_output_____" ] ], [ [ "# with the kernel\none_scores = sgd_rbf.decision_function([one_rbf_example])\nprint(one_scores)\nprediction = np.argmax(one_scores)\nprint(prediction)\nprint(digits.target_names[sgd_rbf.classes_[prediction]])", "[[-1.43787351 -1.87790689 -2.6351228 -3.70534368 -2.11141745 -3.57570642\n -0.83998359 -3.2773025 -2.55029636 -3.2336616 ]]\n6\n6\n" ] ], [ [ "):", "_____no_output_____" ] ], [ [ "# without the kernel\none_scores = sgd.decision_function([one_example])\nprint(one_scores)\nprediction = np.argmax(one_scores)\nprint(prediction)\nprint(digits.target_names[sgd.classes_[prediction]])", "[[-10026. -333. -5977. -2605. -5370. -6327. -7540. -2234. -1181.\n -6917.]]\n1\n1\n" ] ], [ [ "### One VS One", "_____no_output_____" ] ], [ [ "from sklearn.multiclass import OneVsOneClassifier\n\novo_clf = OneVsOneClassifier(SGDClassifier(max_iter=100, random_state=42, loss=\"perceptron\", \n eta0=1, learning_rate=\"constant\", penalty=None))\novo_clf.fit(X_train_features, y_train)\novo_clf.predict([one_rbf_example])", "_____no_output_____" ], [ "len(ovo_clf.estimators_)", "_____no_output_____" ] ], [ [ "10-way Naive Bayes ", "_____no_output_____" ] ], [ [ "gnb.fit(X_train, y_train)\ngnb.predict([one_example])", "_____no_output_____" ] ], [ [ "wow\n\nIt correctly classifies the *one* example, so let's check how confident it is about this prediction (use `predict_proba` with `NaiveBayes` and `decision_function` with the `SGDClassifier`):\n", "_____no_output_____" ] ], [ [ "gnb.predict_proba([one_example])", "_____no_output_____" ] ], [ [ "Cross-validation performance", "_____no_output_____" ] ], [ [ "print(cross_val_score(sgd_rbf, X_train_features, y_train, cv=5, scoring=\"accuracy\"))\nprint(cross_val_score(ovo_clf, X_train_features, y_train, cv=5, scoring=\"accuracy\"))\nprint(cross_val_score(gnb, X_train, y_train, cv=5, scoring=\"accuracy\"))", "[0.9375 0.90625 0.91637631 0.91986063 0.90592334]\n[0.93402778 0.92013889 0.92334495 0.91986063 0.91986063]\n[0.85416667 0.83333333 0.81881533 0.85365854 0.77700348]\n" ] ], [ [ "### Scaling\nLet's apply scaling", "_____no_output_____" ] ], [ [ "from sklearn.preprocessing import StandardScaler, MinMaxScaler\n\n#scaler = StandardScaler()\nscalar = MinMaxScaler()\nX_train_scaled = scaler.fit_transform(X_train.astype(np.float64))\nX_train_features_scaled = scaler.fit_transform(X_train_features.astype(np.float64))\n\nprint(cross_val_score(sgd_rbf, X_train_features_scaled, y_train, cv=5, scoring=\"accuracy\"))\nprint(cross_val_score(ovo_clf, X_train_features_scaled, y_train, cv=5, scoring=\"accuracy\"))\nprint(cross_val_score(gnb, X_train_scaled, y_train, cv=5, scoring=\"accuracy\"))", "[0.93402778 0.90625 0.8989547 0.87108014 0.87456446]\n[0.94444444 0.9375 0.90940767 0.92682927 0.89198606]\n[0.79166667 0.78472222 0.76655052 0.80836237 0.72473868]\n" ] ], [ [ "- StandardScaler() only made things worse\n\n[0.93402778 0.90625 0.8989547 0.87108014 0.87456446]\n\n[0.94444444 0.9375 0.90940767 0.92682927 0.89198606]\n\n[0.79166667 0.78472222 0.76655052 0.80836237 0.72473868]\n\n- MinMaxScaler() gives exact same values\n\n\n", "_____no_output_____" ], [ "# Step 5: Error Analysis", "_____no_output_____" ] ], [ [ "y_train_pred = cross_val_predict(sgd_rbf, X_train_features_scaled, y_train, cv=3)\nconf_mx = confusion_matrix(y_train, y_train_pred)\nconf_mx", "_____no_output_____" ], [ "plt.imshow(conf_mx, cmap = \"jet\")\nplt.show()", "_____no_output_____" ], [ "row_sums = conf_mx.sum(axis=1, keepdims=True)\nnorm_conf_mx = conf_mx / row_sums\nnp.fill_diagonal(norm_conf_mx, 0)\nplt.imshow(norm_conf_mx, cmap = \"jet\")\nplt.show()", "_____no_output_____" ], [ "y_train_pred = cross_val_predict(sgd, X_train_features_scaled, y_train, cv=3)\nconf_mx = confusion_matrix(y_train, y_train_pred)\nconf_mx", "_____no_output_____" ], [ "plt.imshow(conf_mx, cmap = \"jet\")\nplt.show()", "_____no_output_____" ], [ "row_sums = conf_mx.sum(axis=1, keepdims=True)\nnorm_conf_mx = conf_mx / row_sums\nnp.fill_diagonal(norm_conf_mx, 0)\nplt.imshow(norm_conf_mx, cmap = \"jet\")\nplt.show()", "_____no_output_____" ], [ "y_train_pred = cross_val_predict(ovo_clf, X_train_features_scaled, y_train, cv=3)\nconf_mx = confusion_matrix(y_train, y_train_pred)\nconf_mx", "_____no_output_____" ], [ "plt.imshow(conf_mx, cmap = \"jet\")\nplt.show()", "_____no_output_____" ], [ "row_sums = conf_mx.sum(axis=1, keepdims=True)\nnorm_conf_mx = conf_mx / row_sums\nnp.fill_diagonal(norm_conf_mx, 0)\nplt.imshow(norm_conf_mx, cmap = \"jet\")\nplt.show()", "_____no_output_____" ] ], [ [ "## Final step – evaluating on the test set", "_____no_output_____" ] ], [ [ "# non-kernel perceptron\nfrom sklearn.metrics import accuracy_score\n\ny_pred = sgd.predict(X_test)\naccuracy_score(y_test, y_pred)", "_____no_output_____" ], [ "precision = precision_score(y_test, y_pred, average='weighted')\nrecall = recall_score(y_test, y_pred, average='weighted')\nf1 = f1_score(y_test, y_pred, average='weighted')\nprint(precision, recall, f1)", "0.9506739556477815 0.9472222222222222 0.9472461128819086\n" ], [ "# kernel perceptron\nfrom sklearn.metrics import accuracy_score\n\nX_test_features_scaled = scaler.transform(X_test_features.astype(np.float64))\ny_pred = sgd_rbf.predict(X_test_features_scaled)\naccuracy_score(y_test, y_pred)", "_____no_output_____" ], [ "precision = precision_score(y_test, y_pred, average='weighted')\nrecall = recall_score(y_test, y_pred, average='weighted')\nf1 = f1_score(y_test, y_pred, average='weighted')\nprint(precision, recall, f1)", "0.9330274822584538 0.9305555555555556 0.9299190345492117\n" ] ], [ [ "The OvO SGD classifier:", "_____no_output_____" ] ], [ [ "# One vs one\nfrom sklearn.metrics import accuracy_score\n\nX_test_features_scaled = scaler.transform(X_test_features.astype(np.float64))\ny_pred = ovo_clf.predict(X_test_features_scaled)\naccuracy_score(y_test, y_pred)", "_____no_output_____" ], [ "precision = precision_score(y_test, y_pred, average='weighted')\nrecall = recall_score(y_test, y_pred, average='weighted')\nf1 = f1_score(y_test, y_pred, average='weighted')\nprint(precision, recall, f1)", "0.9038423919737274 0.8944444444444445 0.8906734076041858\n" ] ], [ [ "Naive Bayes", "_____no_output_____" ] ], [ [ "#Naive Bayes\ngnb.fit(X_train, y_train)\ny_pred = gnb.predict(X_test)\naccuracy_score(y_test, y_pred)", "_____no_output_____" ], [ "precision = precision_score(y_test, y_pred, average='weighted')\nrecall = recall_score(y_test, y_pred, average='weighted')\nf1 = f1_score(y_test, y_pred, average='weighted')\nprint(precision, recall, f1)", "0.8479871939298477 0.8111111111111111 0.8150828576150382\n" ] ], [ [ "Seems like non-kernel perceptron's doing the best", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Collision Avoidance - Data Collection\n\nIf you ran through the basic motion notebook, hopefully you're enjoying how easy it can be to make your Jetbot move around! Thats very cool! But what's even cooler, is making JetBot move around all by itself! \n\nThis is a super hard task, that has many different approaches but the whole problem is usually broken down into easier sub-problems. It could be argued that one of the most\nimportant sub-problems to solve, is the problem of preventing the robot from entering dangerous situations! We're calling this *collision avoidance*. \n\nIn this set of notebooks, we're going to attempt to solve the problem using deep learning and a single, very versatile, sensor: the camera. You'll see how with a neural network, camera, and the NVIDIA Jetson Nano, we can teach the robot a very useful behavior!\n\nThe approach we take to avoiding collisions is to create a virtual \"safety bubble\" around the robot. Within this safety bubble, the robot is able to spin in a circle without hitting any objects (or other dangerous situations like falling off a ledge). \n\n\nOf course, the robot is limited by what's in it's field of vision, and we can't prevent objects from being placed behind the robot, etc. But we can prevent the robot from entering these scenarios itself.\n\nThe way we'll do this is super simple: \n\nFirst, we'll manually place the robot in scenarios where it's \"safety bubble\" is violated, and label these scenarios ``blocked``. We save a snapshot of what the robot sees along with this label.\n\nSecond, we'll manually place the robot in scenarios where it's safe to move forward a bit, and label these scenarios ``free``. Likewise, we save a snapshot along with this label.\n\nThat's all that we'll do in this notebook; data collection. Once we have lots of images and labels, we'll upload this data to a GPU enabled machine where we'll *train* a neural network to predict whether the robot's safety bubble is being violated based off of the image it sees. We'll use this to implement a simple collision avoidance behavior in the end :)\n\n> IMPORTANT NOTE: When JetBot spins in place, it actually spins about the center between the two wheels, not the center of the robot chassis itself. This is an important detail to remember when you're trying to estimate whether the robot's safety bubble is violated or not. But don't worry, you don't have to be exact. If in doubt it's better to lean on the cautious side (a big safety bubble). We want to make sure JetBot doesn't enter a scenario that it couldn't get out of by turning in place.", "_____no_output_____" ], [ "### Display live camera feed\n\nSo let's get started. First, let's initialize and display our camera like we did in the *teleoperation* notebook. \n\n> Our neural network takes a 224x224 pixel image as input. We'll set our camera to that size to minimize the filesize of our dataset (we've tested that it works for this task).\n> In some scenarios it may be better to collect data in a larger image size and downscale to the desired size later.", "_____no_output_____" ] ], [ [ "import traitlets\nimport ipywidgets.widgets as widgets\nfrom IPython.display import display\nfrom jetbot import Camera, bgr8_to_jpeg\n#\ncamera = Camera.instance(width=224, height=224)\nimage = widgets.Image(format='jpeg', width=224, height=224) # this width and height doesn't necessarily have to match the camera\ncamera_link = traitlets.dlink((camera, 'value'), (image, 'value'), transform=bgr8_to_jpeg)\n", "_____no_output_____" ] ], [ [ "Awesome, next let's create a few directories where we'll store all our data. We'll create a folder ``dataset`` that will contain two sub-folders ``free`` and ``blocked``, \nwhere we'll place the images for each scenario.", "_____no_output_____" ] ], [ [ "import os\n\nblocked_dir = 'dataset/blocked'\nfree_dir = 'dataset/free'\n\n# we have this \"try/except\" statement because these next functions can throw an error if the directories exist already\ntry:\n os.makedirs(free_dir)\n os.makedirs(blocked_dir)\nexcept FileExistsError:\n print('Directories are not created because they already exist')", "Directories are not created because they already exist\n" ] ], [ [ "If you refresh the Jupyter file browser on the left, you should now see those directories appear. Next, let's create and display some buttons that we'll use to save snapshots\nfor each class label. We'll also add some text boxes that will display how many images of each category that we've collected so far. This is useful because we want to make\nsure we collect about as many ``free`` images as ``blocked`` images. It also helps to know how many images we've collected overall.", "_____no_output_____" ] ], [ [ "button_layout = widgets.Layout(width='128px', height='64px')\nfree_button = widgets.Button(description='add free', button_style='success', layout=button_layout)\nblocked_button = widgets.Button(description='add blocked', button_style='danger', layout=button_layout)\nfree_count = widgets.IntText(layout=button_layout, value=len(os.listdir(free_dir)))\nblocked_count = widgets.IntText(layout=button_layout, value=len(os.listdir(blocked_dir)))\n\nx=0 #from here TB\ncontroller = widgets.Controller(index=0) # replace with index of your controller\nbutton_layout = widgets.Layout(width='200px', height='64px') #TB\nfree_left = widgets.FloatText(layout=button_layout, value=x, description='forward') #TB\nfree_right = widgets.FloatText(layout=button_layout, value=x, description='turning') #TB\nmotorleft = widgets.FloatText(layout=button_layout, value=x, description='Motor Left') #TB\nmotorright = widgets.FloatText(layout=button_layout, value=x, description='Motor Right') #TB\n\nspeed_widget = widgets.FloatSlider(value=0.5, min=0.05, max=1.0, step=0.001, description='speed')\n#TB higher speed requires smaller turn_gain values: 2.5 for speed 0.22, around 2 for speed 0.4\nturn_gain_widget = widgets.FloatSlider(value=1, min=0.05, step=0.001, max=4.0, description='turn sensitivity')\n#TB value different for different forward speed, but very small differences\nmotoradjustment_widget = widgets.FloatSlider(value=0.02, min=0.00, max=0.2, step=0.0001, description='motoradjustment')\n", "_____no_output_____" ], [ "from jetbot import Robot\nimport traitlets\nimport math\n\nrobot = Robot()\n\n#TB to show the controller values\n\nleft_link = traitlets.dlink((controller.axes[1], 'value'), (free_left, 'value'), transform=lambda x: -x)\nright_link = traitlets.dlink((controller.axes[0], 'value'), (free_right, 'value'), transform=lambda x: -x)\n\ndef on_value_change(change):\n x= free_right.value\n y= free_left.value\n leftnew, rightnew = steering(x, y)\n motorright.value= round(float(leftnew),2) \n motorleft.value= round(float(rightnew + motoradjustment_widget.value),2) # adjust the motor that lags behind\n #motoradjustment value important to keep bot driving straight, small offset-values like 0.05\n robot.right_motor.value=motorright.value\n robot.left_motor.value=motorleft.value\n \ndef steering(x, y): \n #script from stackexchange of user Pedro Werneck \n #https://electronics.stackexchange.com/questions/19669/algorithm-for-mixing-2-axis-analog-input-to-control-a-differential-motor-drive\n # convert to polar\n r = math.hypot(x, y)\n t = math.atan2(y, x)\n\n # rotate by 45 degrees\n t += math.pi / -4.0\n\n # back to cartesian\n left = r * math.cos(t)\n right = r * math.sin(t)\n\n # rescale the new coords\n left = left * math.sqrt(2)\n right = right * math.sqrt(2)\n\n # clamp to -1/+1\n scalefactor= speed_widget.value\n left = max(scalefactor*-1.0, min(left, scalefactor))\n right = max(scalefactor*-1.0, min(right, scalefactor))\n \n #gamma correction for response sensitivity of joystick while turning : TB\n gamma=turn_gain_widget.value #using slider for joystick 1-4, for object recognition 2-40 \n if left <0 :\n left= -1* (((abs(left)/scalefactor)**(1/gamma))*scalefactor)\n else:\n left= ((abs(left)/scalefactor)**(1/gamma))*scalefactor\n \n if right <0:\n right= -1*(((abs(right)/scalefactor)**(1/gamma))*scalefactor)\n else:\n right= ((abs(right)/scalefactor)**(1/gamma))*scalefactor\n \n return left, right\n\n\nfree_left.observe(on_value_change, names='value')\nfree_right.observe(on_value_change, names='value')\n\n#left_link = traitlets.dlink((motorleft, 'value'), (robot.left_motor, 'value'))\n#right_link = traitlets.dlink((motorright, 'value'), (robot.right_motor, 'value'))", "_____no_output_____" ], [ "from jetbot import Heartbeat\n\n\ndef handle_heartbeat_status(change):\n if change['new'] == Heartbeat.Status.dead:\n camera_link.unlink()\n left_link.unlink()\n right_link.unlink()\n robot.stop()\n\nheartbeat = Heartbeat(period=0.5)\n\n# attach the callback function to heartbeat status\nheartbeat.observe(handle_heartbeat_status, names='status')", "_____no_output_____" ] ], [ [ "Right now, these buttons wont do anything. We have to attach functions to save images for each category to the buttons' ``on_click`` event. We'll save the value\nof the ``Image`` widget (rather than the camera), because it's already in compressed JPEG format!\n\nTo make sure we don't repeat any file names (even across different machines!) we'll use the ``uuid`` package in python, which defines the ``uuid1`` method to generate\na unique identifier. This unique identifier is generated from information like the current time and the machine address.", "_____no_output_____" ] ], [ [ "from uuid import uuid1\n\nsnapshot_image = widgets.Image(format='jpeg', width=224, height=224)\n\ndef save_snapshot(directory):\n image_path = os.path.join(directory, str(uuid1()) + '.jpg')\n with open(image_path, 'wb') as f:\n f.write(image.value)\n # display snapshot that was saved\n snapshot_image.value = image.value\n\ndef save_free(change):\n global free_dir, free_count\n if change['new']:\n save_snapshot(free_dir)\n free_count.value = len(os.listdir(free_dir))\n \ndef save_blocked(change):\n global blocked_dir, blocked_count\n if change['new']:\n save_snapshot(blocked_dir)\n blocked_count.value = len(os.listdir(blocked_dir))\n \ndef save_free_button():\n global free_dir, free_count\n save_snapshot(free_dir)\n free_count.value = len(os.listdir(free_dir))\n \ndef save_blocked_button():\n global blocked_dir, blocked_count\n save_snapshot(blocked_dir)\n blocked_count.value = len(os.listdir(blocked_dir))\n \n# attach the callbacks, we use a 'lambda' function to ignore the\n# parameter that the on_click event would provide to our function\n# because we don't need it.\n\ncontroller.buttons[5].observe(save_free, names='value') #TB gamepad button number 5\ncontroller.buttons[7].observe(save_blocked, names='value') #TB gamepad button numer 7\n\nfree_button.on_click(lambda x: save_free_button())\nblocked_button.on_click(lambda x: save_blocked_button())\n\n#display(image)\ndisplay(widgets.HBox([image, snapshot_image]))\ndisplay(controller)\n\ndisplay(widgets.VBox([\n speed_widget,\n turn_gain_widget,\n motoradjustment_widget,\n]))\n\ndisplay(widgets.HBox([free_left, free_right, motorleft, motorright]))\n\ndisplay(widgets.HBox([free_count, free_button]))\ndisplay(widgets.HBox([blocked_count, blocked_button]))", "_____no_output_____" ], [ "import time\n\ncamera.unobserve_all()\ntime.sleep(1.0)\nrobot.stop()", "_____no_output_____" ] ], [ [ "Great! Now the buttons above should save images to the ``free`` and ``blocked`` directories. You can use the Jupyter Lab file browser to view these files!\n\nNow go ahead and collect some data \n\n1. Place the robot in a scenario where it's blocked and press ``add blocked``\n2. Place the robot in a scenario where it's free and press ``add free``\n3. Repeat 1, 2\n\n> REMINDER: You can move the widgets to new windows by right clicking the cell and clicking ``Create New View for Output``. Or, you can just re-display them\n> together as we will below\n\nHere are some tips for labeling data\n\n1. Try different orientations\n2. Try different lighting\n3. Try varied object / collision types; walls, ledges, objects\n4. Try different textured floors / objects; patterned, smooth, glass, etc.\n\nUltimately, the more data we have of scenarios the robot will encounter in the real world, the better our collision avoidance behavior will be. It's important\nto get *varied* data (as described by the above tips) and not just a lot of data, but you'll probably need at least 100 images of each class (that's not a science, just a helpful tip here). But don't worry, it goes pretty fast once you get going :)", "_____no_output_____" ], [ "## Next\n\nOnce you've collected enough data, we'll need to copy that data to our GPU desktop or cloud machine for training. First, we can call the following *terminal* command to compress\nour dataset folder into a single *zip* file.\n\n> The ! prefix indicates that we want to run the cell as a *shell* (or *terminal*) command.\n\n> The -r flag in the zip command below indicates *recursive* so that we include all nested files, the -q flag indicates *quiet* so that the zip command doesn't print any output", "_____no_output_____" ] ], [ [ "!zip -r -q dataset.zip dataset", "_____no_output_____" ] ], [ [ "You should see a file named ``dataset.zip`` in the Jupyter Lab file browser. You should download the zip file using the Jupyter Lab file browser by right clicking and selecting ``Download``.\n\nNext, we'll need to upload this data to our GPU desktop or cloud machine (we refer to this as the *host*) to train the collision avoidance neural network. We'll assume that you've set up your training\nmachine as described in the JetBot WiKi. If you have, you can navigate to ``http://<host_ip_address>:8888`` to open up the Jupyter Lab environment running on the host. The notebook you'll need to open there is called ``collision_avoidance/train_model.ipynb``.\n\nSo head on over to your training machine and follow the instructions there! Once your model is trained, we'll return to the robot Jupyter Lab enivornment to use the model for a live demo!", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "empty" ] ] ]

[ "empty" ]

[ [ "empty" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# The Performance Of Models Trained On The MNIST Dataset On Custom-Drawn Images", "_____no_output_____" ] ], [ [ "import numpy as np\nimport tensorflow as tf\nimport sklearn, sklearn.linear_model, sklearn.multiclass, sklearn.naive_bayes\nimport matplotlib.pyplot as plt\nimport pandas as pd", "_____no_output_____" ], [ "plt.rcParams[\"figure.figsize\"] = (10, 10)\nplt.rcParams.update({'font.size': 12})", "_____no_output_____" ] ], [ [ "## Defining the data", "_____no_output_____" ] ], [ [ "(train_images, train_labels), (test_images, test_labels) = tf.keras.datasets.mnist.load_data()", "_____no_output_____" ] ], [ [ "### Making 1D versions of the MNIST images for the one-vs-rest classifier", "_____no_output_____" ] ], [ [ "train_images_flat = train_images.reshape((train_images.shape[0], train_images.shape[1] * train_images.shape[2])) / 255.0\n\ntest_images_flat = test_images.reshape((test_images.shape[0], test_images.shape[1] * test_images.shape[2])) / 255.0", "_____no_output_____" ] ], [ [ "### Making a 4D dataset and categorical labels for the neural net", "_____no_output_____" ] ], [ [ "train_images = np.expand_dims(train_images, axis=-1) / 255.0\ntest_images = np.expand_dims(test_images, axis=-1) / 255.0\n\n#train_images = train_images.reshape(60000, 28, 28, 1)\n#test_images = test_images.reshape(10000, 28, 28, 1)\n\ntrain_labels_cat = tf.keras.utils.to_categorical(train_labels)\ntest_labels_cat = tf.keras.utils.to_categorical(test_labels)", "_____no_output_____" ], [ "def plot_images(images, labels, rows=5, cols=5, label='Label'): \n fig, axes = plt.subplots(rows, cols)\n fig.figsize=(15, 15) \n\n indices = np.random.choice(len(images), rows * cols)\n counter = 0\n\n for i in range(rows):\n for j in range(cols):\n axes[i, j].imshow(images[indices[counter]])\n axes[i, j].set_title(f\"{label}: {labels[indices[counter]]}\")\n axes[i, j].set_xticks([])\n axes[i, j].set_yticks([])\n counter += 1\n \n plt.tight_layout()\n plt.show()", "_____no_output_____" ], [ "plot_images(train_images, train_labels)", "_____no_output_____" ] ], [ [ "## Training", "_____no_output_____" ], [ "### Defining and training the one-vs-rest classifier", "_____no_output_____" ] ], [ [ "log_reg = sklearn.linear_model.SGDClassifier(loss='log', max_iter=1000, penalty='l2')\n\nclassifier = sklearn.multiclass.OneVsRestClassifier(log_reg)\n\nclassifier.fit(train_images_flat, train_labels)", "_____no_output_____" ] ], [ [ "### Defining and training the neural net", "_____no_output_____" ] ], [ [ "from tensorflow.keras import Sequential\nfrom tensorflow.keras import layers", "_____no_output_____" ], [ "def create_model():\n model = Sequential([\n layers.Conv2D(64, 5, activation='relu', input_shape=(28, 28, 1)),\n layers.MaxPool2D(2),\n layers.Conv2D(128, 5, activation='relu'),\n layers.MaxPool2D(2),\n layers.GlobalAveragePooling2D(),\n layers.Dense(10, activation='softmax')\n ])\n \n model.compile(optimizer='Adam', loss='categorical_crossentropy', metrics=['accuracy'])\n \n return model", "_____no_output_____" ], [ "model = create_model()", "_____no_output_____" ], [ "model.summary()", "Model: \"sequential\"\n_________________________________________________________________\nLayer (type) Output Shape Param # \n=================================================================\nconv2d (Conv2D) (None, 24, 24, 64) 1664 \n_________________________________________________________________\nmax_pooling2d (MaxPooling2D) (None, 12, 12, 64) 0 \n_________________________________________________________________\nconv2d_1 (Conv2D) (None, 8, 8, 128) 204928 \n_________________________________________________________________\nmax_pooling2d_1 (MaxPooling2 (None, 4, 4, 128) 0 \n_________________________________________________________________\nglobal_average_pooling2d (Gl (None, 128) 0 \n_________________________________________________________________\ndense (Dense) (None, 10) 1290 \n=================================================================\nTotal params: 207,882\nTrainable params: 207,882\nNon-trainable params: 0\n_________________________________________________________________\n" ], [ "train_gen = tf.keras.preprocessing.image.ImageDataGenerator(zoom_range=0.3,\n height_shift_range=0.10,\n width_shift_range=0.10,\n rotation_range=10)\n\ntrain_datagen = train_gen.flow(train_images, train_labels_cat, batch_size=256)\n\n'''def scheduler(epoch): \n initial_lr = 0.001\n lr = initial_lr * np.exp(-0.1 * epoch)\n return lr\n\nfrom tensorflow.keras.callbacks import LearningRateScheduler\nlr_scheduler = LearningRateScheduler(scheduler, verbose=1)'''", "_____no_output_____" ], [ "history = model.fit(train_datagen, initial_epoch=0, epochs=30, batch_size=256,\n validation_data=(test_images, test_labels_cat))", "Epoch 1/30\n235/235 [==============================] - 15s 55ms/step - loss: 1.0590 - accuracy: 0.6757 - val_loss: 0.2701 - val_accuracy: 0.9319\nEpoch 2/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.4189 - accuracy: 0.8813 - val_loss: 0.1617 - val_accuracy: 0.9566\nEpoch 3/30\n235/235 [==============================] - 13s 53ms/step - loss: 0.3105 - accuracy: 0.9123 - val_loss: 0.1374 - val_accuracy: 0.9592\nEpoch 4/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.2602 - accuracy: 0.9252 - val_loss: 0.0984 - val_accuracy: 0.9728\nEpoch 5/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.2215 - accuracy: 0.9368 - val_loss: 0.0929 - val_accuracy: 0.9718\nEpoch 6/30\n235/235 [==============================] - 13s 53ms/step - loss: 0.1958 - accuracy: 0.9436 - val_loss: 0.0849 - val_accuracy: 0.9737\nEpoch 7/30\n235/235 [==============================] - 13s 53ms/step - loss: 0.1724 - accuracy: 0.9498 - val_loss: 0.0593 - val_accuracy: 0.9819\nEpoch 8/30\n235/235 [==============================] - 13s 53ms/step - loss: 0.1539 - accuracy: 0.9554 - val_loss: 0.0618 - val_accuracy: 0.9821\nEpoch 9/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.1443 - accuracy: 0.9575 - val_loss: 0.0533 - val_accuracy: 0.9832\nEpoch 10/30\n235/235 [==============================] - 13s 53ms/step - loss: 0.1306 - accuracy: 0.9614 - val_loss: 0.0499 - val_accuracy: 0.9842\nEpoch 11/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.1251 - accuracy: 0.9637 - val_loss: 0.0482 - val_accuracy: 0.9856\nEpoch 12/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.1145 - accuracy: 0.9662 - val_loss: 0.0411 - val_accuracy: 0.9872\nEpoch 13/30\n235/235 [==============================] - 13s 55ms/step - loss: 0.1080 - accuracy: 0.9681 - val_loss: 0.0412 - val_accuracy: 0.9872\nEpoch 14/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.1036 - accuracy: 0.9694 - val_loss: 0.0400 - val_accuracy: 0.9877\nEpoch 15/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.0974 - accuracy: 0.9701 - val_loss: 0.0495 - val_accuracy: 0.9846\nEpoch 16/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.0967 - accuracy: 0.9710 - val_loss: 0.0386 - val_accuracy: 0.9881\nEpoch 17/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.0932 - accuracy: 0.9720 - val_loss: 0.0403 - val_accuracy: 0.9866\nEpoch 18/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.0857 - accuracy: 0.9747 - val_loss: 0.0417 - val_accuracy: 0.9866\nEpoch 19/30\n235/235 [==============================] - 13s 53ms/step - loss: 0.0843 - accuracy: 0.9755 - val_loss: 0.0332 - val_accuracy: 0.9894\nEpoch 20/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.0822 - accuracy: 0.9749 - val_loss: 0.0420 - val_accuracy: 0.9863\nEpoch 21/30\n235/235 [==============================] - 13s 56ms/step - loss: 0.0767 - accuracy: 0.9771 - val_loss: 0.0354 - val_accuracy: 0.9896\nEpoch 22/30\n235/235 [==============================] - 13s 55ms/step - loss: 0.0731 - accuracy: 0.9779 - val_loss: 0.0356 - val_accuracy: 0.9876\nEpoch 23/30\n235/235 [==============================] - 13s 55ms/step - loss: 0.0752 - accuracy: 0.9773 - val_loss: 0.0315 - val_accuracy: 0.9898\nEpoch 24/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.0749 - accuracy: 0.9776 - val_loss: 0.0280 - val_accuracy: 0.9902\nEpoch 25/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.0697 - accuracy: 0.9791 - val_loss: 0.0389 - val_accuracy: 0.9877\nEpoch 26/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.0718 - accuracy: 0.9780 - val_loss: 0.0345 - val_accuracy: 0.9891\nEpoch 27/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.0653 - accuracy: 0.9799 - val_loss: 0.0292 - val_accuracy: 0.9902\nEpoch 28/30\n235/235 [==============================] - 13s 53ms/step - loss: 0.0682 - accuracy: 0.9796 - val_loss: 0.0321 - val_accuracy: 0.9892\nEpoch 29/30\n235/235 [==============================] - 13s 54ms/step - loss: 0.0646 - accuracy: 0.9805 - val_loss: 0.0297 - val_accuracy: 0.9905\nEpoch 30/30\n235/235 [==============================] - 13s 55ms/step - loss: 0.0641 - accuracy: 0.9805 - val_loss: 0.0364 - val_accuracy: 0.9887\n" ], [ "model.save('cnn-64-128-5-aug')\n#model.load_weights('cnn-64-128-5-aug')", "INFO:tensorflow:Assets written to: cnn-64-128-5-aug\\assets\n" ] ], [ [ "## Assessing model performance", "_____no_output_____" ], [ "### Loading drawn images", "_____no_output_____" ] ], [ [ "def read_images(filepaths, reverse=False):\n images = []\n images_flat = []\n \n for filepath in filepaths:\n image = tf.io.read_file(filepath)\n image = tf.image.decode_image(image, channels=1)\n image = tf.image.resize(image, (28, 28))\n \n if reverse: \n image = np.where(image == 255, 0, 255)\n else:\n image = image.numpy()\n \n image = image / 255.0 \n images.append(image)\n \n images_flat.append(image.reshape(28 * 28)) \n \n return np.array(images), np.array(images_flat)", "_____no_output_____" ], [ "filepaths = tf.io.gfile.glob('images/*.png')\n\nlist.sort(filepaths, key=lambda x: int(x[12:-4]))\n\nimages, images_flat = read_images(filepaths, True)", "_____no_output_____" ], [ "images.shape", "_____no_output_____" ] ], [ [ "### Creating labels for the one-vs-rest classifier and the neural net", "_____no_output_____" ] ], [ [ "labels = 100 * [0] + 98 * [1] + 100 * [2] + 101 * [3] + 99 * [4] + 111 * [5] + 89 * [6] + 110 * [7] + 93 * [8] + 112 * [9]\n\nlabels = np.array(labels)", "_____no_output_____" ], [ "labels.shape", "_____no_output_____" ], [ "labels_cat = tf.keras.utils.to_categorical(labels)\n\nlabels_cat.shape", "_____no_output_____" ], [ "labels_cat[0]", "_____no_output_____" ] ], [ [ "### Plotting the drawn images and their corresponding labels", "_____no_output_____" ] ], [ [ "plot_images(images, labels)", "_____no_output_____" ] ], [ [ "### Evaluating model performance", "_____no_output_____" ], [ "#### Neural net on MNIST test dataset", "_____no_output_____" ] ], [ [ "model.evaluate(test_images, test_labels_cat)", "313/313 [==============================] - 1s 3ms/step - loss: 0.0364 - accuracy: 0.9887\n" ], [ "from sklearn.metrics import classification_report, confusion_matrix\n\npredictions = np.argmax(model.predict(test_images), axis=-1)\n\nconf_mat = confusion_matrix(test_labels, predictions) ", "_____no_output_____" ], [ "conf_mat", "_____no_output_____" ], [ "class_report = classification_report(test_labels, predictions, output_dict=True)\n\nclass_report", "_____no_output_____" ] ], [ [ "#### Neural net on drawn images", "_____no_output_____" ] ], [ [ "model.evaluate(images, labels_cat)", "32/32 [==============================] - 0s 4ms/step - loss: 0.1527 - accuracy: 0.9467\n" ], [ "predictions = np.argmax(model.predict(images), axis=-1)", "_____no_output_____" ], [ "rows, cols = 5, 5\n\nfig, axes = plt.subplots(rows, cols)\nfig.figsize=(15, 15) \n\nindices = np.random.choice(len(images), rows * cols)\ncounter = 0\n \nfor i in range(rows):\n for j in range(cols):\n axes[i, j].imshow(images[indices[counter]])\n axes[i, j].set_title(f\"Prediction: {predictions[indices[counter]]}\\n\"\n f\"True label: {labels[indices[counter]]}\")\n axes[i, j].set_xticks([])\n axes[i, j].set_yticks([])\n counter += 1\n \nplt.tight_layout()\nplt.show()", "_____no_output_____" ] ], [ [ "##### Plotting wrong predictions", "_____no_output_____" ] ], [ [ "wrong_predictions = list(filter(lambda x: x[1][0] != x[1][1], list(enumerate(zip(predictions, labels)))))", "_____no_output_____" ], [ "len(wrong_predictions)", "_____no_output_____" ], [ "cols, rows = 5, 5\n\nfig, axes = plt.subplots(rows, cols)\nfig.figsize=(15, 15) \n\ncounter = 0\n\nfor i in range(rows):\n for j in range(cols):\n axes[i, j].imshow(images[wrong_predictions[counter][0]])\n axes[i, j].set_title(f\"Prediction: {wrong_predictions[counter][1][0]}\\n\"\n f\"True label: {wrong_predictions[counter][1][1]}\")\n axes[i, j].set_xticks([])\n axes[i, j].set_yticks([])\n counter += 1\n \nplt.tight_layout()\nplt.show()", "_____no_output_____" ], [ "from sklearn.metrics import classification_report, confusion_matrix\n\nconf_mat = confusion_matrix(labels, predictions) ", "_____no_output_____" ], [ "conf_mat", "_____no_output_____" ], [ "class_report = classification_report(labels, predictions, output_dict=True)\n\nclass_report", "_____no_output_____" ] ], [ [ "#### One-vs-rest classifier on MNIST test dataset", "_____no_output_____" ] ], [ [ "classifier.score(test_images_flat, test_labels)", "_____no_output_____" ], [ "predictions = classifier.predict(test_images_flat)\n\nconf_mat = confusion_matrix(test_labels, predictions) ", "_____no_output_____" ], [ "conf_mat", "_____no_output_____" ], [ "class_report = classification_report(test_labels, predictions, output_dict=True)", "_____no_output_____" ], [ "class_report", "_____no_output_____" ] ], [ [ "#### One-vs-rest classifier on drawn images", "_____no_output_____" ] ], [ [ "classifier.score(images_flat, labels)", "_____no_output_____" ], [ "predictions = classifier.predict(images_flat)\n\nconf_mat = confusion_matrix(labels, predictions) ", "_____no_output_____" ], [ "conf_mat", "_____no_output_____" ], [ "class_report = classification_report(labels, predictions, output_dict=True)", "_____no_output_____" ], [ "class_report", "_____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" ], [ "code", "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code", "code" ], [ "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code", "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "'''\n Pandas时刻数据：Timestamp 时间戳\n\n 时刻数据代表时间点，是pandas的数据类型，是将值与时间点相关联的最基本类型的时间序列数据\n\n pandas.Timestamp()\n'''\nimport datetime\nimport pandas as pd\n\ndate1 = datetime.datetime.today()\ndate2 = '2019-11-13'\nt1 = pd.Timestamp(date1)\nt2 = pd.Timestamp(date2)\nprint('t1:',t1)\nprint('type(t1):',type(t1)) # 数据类型为 pandas的Timestamp\nprint('t2:',t2)", "t1: 2019-12-15 07:53:53.091608\ntype(t1): <class 'pandas._libs.tslibs.timestamps.Timestamp'>\nt2: 2019-11-13 00:00:00\n" ], [ "# pd.to_datetime 多个时间数据转换时间戳索引\n\ndate1 = datetime.datetime(2019,11,13,20,22,22)\ndate2 = '2020-1-1'\n\n# pd.to_datetime()：如果是单个时间数据，转换成pandas的时刻数据，数据类型为Timestamp\nt1 = pd.to_datetime(date1)\nt2 = pd.to_datetime(date2)\nprint('t1:',t1)\nprint('type(t1):',type(t1))\nprint('t1:',t1)\nprint('type(t2):',type(t2))\n\n# 多个时间数据，将会转换为pandas的DatetimeIndex\ndate_lst = ['2018-1-1','2019-11-13','2019-5-29']\nt3 = pd.to_datetime(date_lst)\nprint('时间列表t3:',t3)\nprint('type(t3):',type(t3))\n\n# 当一组时间序列中夹杂其他格式数据，可用errors参数返回\ndate_lst2 = ['2018-1-1','2019-11-13','2019-5-29','头号玩家']\nprint(date_lst2)\n\n# errors = 'ignore':不可解析时返回原始输入，这里就是直接生成一般数组\nt4 = pd.to_datetime(date_lst2,errors='ignore')\nprint('t4:',t4) \nprint('type(t4):',type(t4))\n\n# errors = 'coerce':不可扩展，缺失值返回NaT（Not a Time），结果认为DatetimeIndex\nt5 = pd.to_datetime(date_lst2, errors='coerce')\nprint('t5:',t5) # 结果中,不可解析为时间的值变为NaT\nprint('type(t5):',type(t5))\n\n'''\n 源码:\n errors : {'ignore', 'raise', 'coerce'}, default 'raise'\n - If 'raise', then invalid parsing will raise an exception\n - If 'coerce', then invalid parsing will be set as NaT\n - If 'ignore', then invalid parsing will return the input\n coerce:胁迫 迫使\n'''", "t1: 2019-11-13 20:22:22\ntype(t1): <class 'pandas._libs.tslibs.timestamps.Timestamp'>\nt1: 2019-11-13 20:22:22\ntype(t2): <class 'pandas._libs.tslibs.timestamps.Timestamp'>\n时间列表t3: DatetimeIndex(['2018-01-01', '2019-11-13', '2019-05-29'], dtype='datetime64[ns]', freq=None)\ntype(t3): <class 'pandas.core.indexes.datetimes.DatetimeIndex'>\n['2018-1-1', '2019-11-13', '2019-5-29', '头号玩家']\nt4: Index(['2018-1-1', '2019-11-13', '2019-5-29', '头号玩家'], dtype='object')\ntype(t4): <class 'pandas.core.indexes.base.Index'>\nt5: DatetimeIndex(['2018-01-01', '2019-11-13', '2019-05-29', 'NaT'], dtype='datetime64[ns]', freq=None)\ntype(t5): <class 'pandas.core.indexes.datetimes.DatetimeIndex'>\n" ] ] ]

[ "code" ]

[ [ "code", "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Plot Comparison Between Algorithms", "_____no_output_____" ], [ "Expects the input data to contain CSV files containing rewards per timestep", "_____no_output_____" ] ], [ [ "import os\nimport numpy as np\n\n%reload_ext autoreload\n%autoreload 2", "_____no_output_____" ] ], [ [ "We need to read the CSV files (from a function in another file) to get the reward at each timestep for each run of each algorithm. Only the `dataPath` directories will be loaded.\n\n`load_data` loads the CSV files containing rewards as a python list of Pandas DataFrames.\n\n`dataPath` contains the exact path of the directories containing the CSV files. This path is relative to the `data` directory. It assumes every element will be path for a different algorithm. It will overwrite if two paths are for different parameter settings of the same algorithm.\n\nExpects there to be more than 1 input CSV file.", "_____no_output_____" ] ], [ [ "dataPath = ['esarsa/alpha-0.015625_driftProb--1,-1,-1,-1_driftScale-1000_enable-debug-0_epsilon-0.05_gamma-0.95_lambda-0.8_sensorLife-1,1,1,1_tiles-4_tilings-32',\n 'dqn/alpha-0.015625_driftProb--1,-1,-1,-1_driftScale-100_enable-debug-0_epsilon-0.05_gamma-0.95_lambda-0.8_sensorLife-1,1,1,1_tiles-4_tilings-32/']\n\nbasePath = '../data/'\n\nalgorithms = [dataPath[i].split('/')[0] for i in range(len(dataPath))]\n\nData = {}\n\nfrom loadFromRewards import load_data\n\nfor i in range(len(dataPath)):\n if os.path.isdir(basePath + dataPath[i]) == True:\n Data[algorithms[i]] = load_data(basePath+dataPath[i])\n\nprint('Data will be stored for', ', '.join([k for k in Data.keys()]))\nprint('Loaded the rewards data from the csv files')", "Data will be stored for esarsa\nLoaded the rewards data from the csv files\n" ] ], [ [ "The rewards can be transformed into the following values of transformation =\n1. 'Returns'\n2. 'Failures'\n3. 'Average-Rewards'\n4. 'Rewards' (no change)\n\n----------------------------------------------------------------------------------------------\n\nThere is an additional parameter of window which can be any non-negative integer. It is used for the 'Average-Rewards' transformation to maintain a moving average over a sliding window. By default window is 0.\n\n- If window is 500 and timesteps are 10000, then the first element is the average of the performances of timesteps from 1 - 500. The second element is the average of the performances of timesteps from 2 - 501. The last element is the average of the performances of timesteps from 9501 - 10000.\n\n----------------------------------------------------------------------------------------------\n\n`transform_data` transforms the absolute failure timesteps (python list of Pandas DataFrames) into the respective `transformation` (a numpy array of numpy arrays) for plotting", "_____no_output_____" ] ], [ [ "plottingData = {}\n\nfrom loadFromRewards import transform_data\n\ntransformation = 'Returns'\nwindow = 2500\n\nfor alg, data in Data.items():\n plottingData[alg] = transform_data(alg, data, transformation, window)\n\nprint('Data will be plotted for', ', '.join([k for k in plottingData.keys()]))\nprint('The stored rewards are transformed to: ', transformation)", "0 esarsa\nData will be plotted for esarsa\nThe stored rewards are transformed to: Returns\n" ] ], [ [ "Here, we can plot the following statistics:\n\n1. Mean of all the runs\n\n2. Median run\n\n3. Run with the best performance (highest return, or equivalently least failures)\n\n4. Run with the worst performance (lowest return, or equivalently most failures)\n\n5. Mean along with the confidence interval (Currently, plots the mean along with 95% confidence interval, but should be changed to make it adaptive to any confidence interval)\n\n6. Mean along with percentile regions (Plots the mean and shades the region between the run with the lower percentile and the run with the upper percentile)\n\n----------------------------------------------------------------------------------------------\n\nDetails:\n\nplotBest, plotWorst, plotMeanAndPercentileRegions sort the performances based on their final performance\n\n ----------------------------------------------------\n\nMean, Median, MeanAndConfidenceInterval are all symmetric plots so 'Failures' does not affect their plots\n \nBest, Worst, MeanAndPercentileRegions are all asymmetric plots so 'Failures' affects their plots, and has to be treated in the following way: \n\n ----------------------------------------------------\n\n1. plotBest for Returns will plot the run with the highest return (least failures)\n plotBest for Failures will plot the run with the least failures and not the highest failures\n\n2. plotWorst for Returns will plot the run with the lowest return (most failures)\n plotWorst for Failures will plot the run with the most failures and not the least failures\n\n3. plotMeanAndPercentileRegions for Returns uses the lower variable to select the run with the 'lower' percentile and uses the upper variable to select the run with the 'upper' percentile\n plotMeanAndPercentileRegions for Failures uses the lower variable along with some calculations to select the run with 'upper' percentile and uses the upper variable along with some calculations to select the run with the 'lower' percentile \n \n----------------------------------------------------------------------------------------------\n\nCaution:\n- Jupyter notebooks (mostly) or matplotlib gives an error when displaying very dense plots. For example: plotting best and worst case for transformation of 'Rewards' for 'example' algorithm, or when trying to zoom into dense plots. Most of the plots for 'Rewards' and 'example' fail.", "_____no_output_____" ] ], [ [ "from stats import getMean, getMedian, getBest, getWorst, getConfidenceIntervalOfMean, getRegion\n\n# Add color, linestyles as needed\n\ndef plotMean(xAxis, data, color):\n mean = getMean(data)\n plt.plot(xAxis, mean, label=alg+'-mean', color=color)\n\ndef plotMedian(xAxis, data, color):\n median = getMedian(data)\n plt.plot(xAxis, median, label=alg+'-median', color=color)\n\ndef plotBest(xAxis, data, transformation, color):\n best = getBest(data, transformation)\n plt.plot(xAxis, best, label=alg+'-best', color=color)\n\ndef plotWorst(xAxis, data, transformation, color):\n worst = getWorst(data, transformation)\n plt.plot(xAxis, worst, label=alg+'-worst', color=color)\n\ndef plotMeanAndConfidenceInterval(xAxis, data, confidence, color):\n plotMean(xAxis, data, color=color)\n lowerBound, upperBound = getConfidenceIntervalOfMean(data, confidence)\n plt.fill_between(xAxis, lowerBound, upperBound, alpha=0.25, color=color)\n\ndef plotMeanAndPercentileRegions(xAxis, data, lower, upper, transformation, color):\n plotMean(xAxis, data, color)\n lowerRun, upperRun = getRegion(data, lower, upper, transformation)\n plt.fill_between(xAxis, lowerRun, upperRun, alpha=0.25, color=color)", "_____no_output_____" ] ], [ [ "Details:\n\n- X axis for 'Average-Rewards' will start from 'window' timesteps and end with the final timesteps\n\n- Need to add color (shades), linestyle as per requirements\n\n- Currently plot one at a time by commenting out the others otherwise, it displays different colors for all.\n", "_____no_output_____" ] ], [ [ "# For saving figures\n#%matplotlib inline\n\n# For plotting in the jupyter notebook\n%matplotlib notebook \n\nimport matplotlib.pyplot as plt\n\ncolors = plt.rcParams['axes.prop_cycle'].by_key()['color']\n\nfor alg, data in plottingData.items():\n lenRun = len(data[0])\n xAxis = np.array([i for i in range(1,lenRun+1)])\n \n if transformation == 'Average-Rewards':\n xAxis += (window-1)\n \n if alg == 'esarsa':\n color = colors[0]\n if alg == 'hand':\n color = colors[1] \n \n plotMean(xAxis, data, color=color)\n\n #plotMedian(xAxis, data, color=color)\n \n #plotBest(xAxis, data, transformation=transformation, color=color)\n \n #plotWorst(xAxis, data, transformation=transformation, color=color)\n \n #plotMeanAndConfidenceInterval(xAxis, data, confidence=0.95, color=color)\n \n #plotMeanAndPercentileRegions(xAxis, data, lower=0.025, upper=0.975, transformation=transformation, color=color)\n\n \n#plt.title('Rewards averaged with sliding window of 1000 timesteps across 100 runs', pad=25, fontsize=10)\nplt.xlabel('Timesteps', labelpad=35)\nplt.ylabel(transformation, rotation=0, labelpad=45)\nplt.rcParams['figure.figsize'] = [8, 5.33]\nplt.legend(loc=0)\nplt.yticks()\nplt.xticks()\nplt.tight_layout()\n\n#plt.savefig('../img/'+transformation+'.png',dpi=500, bbox_inches='tight')", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "### Classification with MNIST Dataset and ResNet network\nThis script sets up a ResNet-style network to classify digits from the MNIST dataset.", "_____no_output_____" ] ], [ [ "import keras\nimport keras.backend as K\nfrom keras.datasets import mnist\nfrom keras.models import Model\nfrom keras.layers import Input, Conv2D, Dense, Flatten, MaxPooling2D, Add, Activation, Dropout\nfrom keras.optimizers import SGD\nfrom matplotlib import pyplot as plt\nimport numpy as np", "Using TensorFlow backend.\n" ] ], [ [ "Use a Keras utility function to load the MNIST dataset. We select only zeros and ones to do binary classification.", "_____no_output_____" ] ], [ [ "(x_train, y_train), (x_test, y_test) = mnist.load_data()\ny_train = keras.utils.to_categorical(y_train,10)\ny_test = keras.utils.to_categorical(y_test,10)", "_____no_output_____" ] ], [ [ "Resize the images to vectors and convert their datatype and range.", "_____no_output_____" ] ], [ [ "x_train = np.expand_dims(x_train,axis=-1)\nx_test = np.expand_dims(x_test,axis=-1)\nx_train = x_train.astype('float32')\nx_test = x_test.astype('float32')\nx_train /= 255\nx_test /= 255\nx_train = x_train*2.-1.\nx_test = x_test*2.-1.", "_____no_output_____" ] ], [ [ "Build a multi-class classifier model.", "_____no_output_____" ] ], [ [ "def res_block(x,c,s=1):\n if K.shape(x)[3] <> c or s <> 1:\n x_save = Conv2D(c,1,strides=s,activation=None)(x)\n else:\n x_save = x\n x = Conv2D(c,3,strides=s,padding='same',activation='relu',kernel_initializer='he_normal')(x)\n x = Conv2D(c,3,padding='same',activation=None,kernel_initializer='he_normal')(x)\n x = Add()([x,x_save])\n x = Activation('relu')(x)\n return x\n\nx_in = Input((28,28,1))\n\nx = res_block(x_in,64,2)\nx = res_block(x,64)\n\nx = res_block(x,128,2)\nx = res_block(x,128)\n\nx = res_block(x,256,2)\nx = res_block(x,256)\n\nx = Flatten()(x)\nx = Dense(200,kernel_initializer='he_normal')(x)\nx = Dropout(0.5)(x)\nx = Dense(10,activation='softmax',kernel_initializer='he_normal')(x)\nmodel = Model(inputs=x_in,outputs=x)\nmodel.summary()", "____________________________________________________________________________________________________\nLayer (type) Output Shape Param # Connected to \n====================================================================================================\ninput_1 (InputLayer) (None, 28, 28, 1) 0 \n____________________________________________________________________________________________________\nconv2d_2 (Conv2D) (None, 14, 14, 64) 640 input_1[0][0] \n____________________________________________________________________________________________________\nconv2d_3 (Conv2D) (None, 14, 14, 64) 36928 conv2d_2[0][0] \n____________________________________________________________________________________________________\nconv2d_1 (Conv2D) (None, 14, 14, 64) 128 input_1[0][0] \n____________________________________________________________________________________________________\nadd_1 (Add) (None, 14, 14, 64) 0 conv2d_3[0][0] \n conv2d_1[0][0] \n____________________________________________________________________________________________________\nactivation_1 (Activation) (None, 14, 14, 64) 0 add_1[0][0] \n____________________________________________________________________________________________________\nconv2d_5 (Conv2D) (None, 14, 14, 64) 36928 activation_1[0][0] \n____________________________________________________________________________________________________\nconv2d_6 (Conv2D) (None, 14, 14, 64) 36928 conv2d_5[0][0] \n____________________________________________________________________________________________________\nconv2d_4 (Conv2D) (None, 14, 14, 64) 4160 activation_1[0][0] \n____________________________________________________________________________________________________\nadd_2 (Add) (None, 14, 14, 64) 0 conv2d_6[0][0] \n conv2d_4[0][0] \n____________________________________________________________________________________________________\nactivation_2 (Activation) (None, 14, 14, 64) 0 add_2[0][0] \n____________________________________________________________________________________________________\nconv2d_8 (Conv2D) (None, 7, 7, 128) 73856 activation_2[0][0] \n____________________________________________________________________________________________________\nconv2d_9 (Conv2D) (None, 7, 7, 128) 147584 conv2d_8[0][0] \n____________________________________________________________________________________________________\nconv2d_7 (Conv2D) (None, 7, 7, 128) 8320 activation_2[0][0] \n____________________________________________________________________________________________________\nadd_3 (Add) (None, 7, 7, 128) 0 conv2d_9[0][0] \n conv2d_7[0][0] \n____________________________________________________________________________________________________\nactivation_3 (Activation) (None, 7, 7, 128) 0 add_3[0][0] \n____________________________________________________________________________________________________\nconv2d_11 (Conv2D) (None, 7, 7, 128) 147584 activation_3[0][0] \n____________________________________________________________________________________________________\nconv2d_12 (Conv2D) (None, 7, 7, 128) 147584 conv2d_11[0][0] \n____________________________________________________________________________________________________\nconv2d_10 (Conv2D) (None, 7, 7, 128) 16512 activation_3[0][0] \n____________________________________________________________________________________________________\nadd_4 (Add) (None, 7, 7, 128) 0 conv2d_12[0][0] \n conv2d_10[0][0] \n____________________________________________________________________________________________________\nactivation_4 (Activation) (None, 7, 7, 128) 0 add_4[0][0] \n____________________________________________________________________________________________________\nconv2d_14 (Conv2D) (None, 4, 4, 256) 295168 activation_4[0][0] \n____________________________________________________________________________________________________\nconv2d_15 (Conv2D) (None, 4, 4, 256) 590080 conv2d_14[0][0] \n____________________________________________________________________________________________________\nconv2d_13 (Conv2D) (None, 4, 4, 256) 33024 activation_4[0][0] \n____________________________________________________________________________________________________\nadd_5 (Add) (None, 4, 4, 256) 0 conv2d_15[0][0] \n conv2d_13[0][0] \n____________________________________________________________________________________________________\nactivation_5 (Activation) (None, 4, 4, 256) 0 add_5[0][0] \n____________________________________________________________________________________________________\nconv2d_17 (Conv2D) (None, 4, 4, 256) 590080 activation_5[0][0] \n____________________________________________________________________________________________________\nconv2d_18 (Conv2D) (None, 4, 4, 256) 590080 conv2d_17[0][0] \n____________________________________________________________________________________________________\nconv2d_16 (Conv2D) (None, 4, 4, 256) 65792 activation_5[0][0] \n____________________________________________________________________________________________________\nadd_6 (Add) (None, 4, 4, 256) 0 conv2d_18[0][0] \n conv2d_16[0][0] \n____________________________________________________________________________________________________\nactivation_6 (Activation) (None, 4, 4, 256) 0 add_6[0][0] \n____________________________________________________________________________________________________\nflatten_1 (Flatten) (None, 4096) 0 activation_6[0][0] \n____________________________________________________________________________________________________\ndense_1 (Dense) (None, 200) 819400 flatten_1[0][0] \n____________________________________________________________________________________________________\ndropout_1 (Dropout) (None, 200) 0 dense_1[0][0] \n____________________________________________________________________________________________________\ndense_2 (Dense) (None, 10) 2010 dropout_1[0][0] \n====================================================================================================\nTotal params: 3,642,786\nTrainable params: 3,642,786\nNon-trainable params: 0\n____________________________________________________________________________________________________\n" ] ], [ [ "Set up the model to optimize the categorical crossentropy loss using stochastic gradient descent.", "_____no_output_____" ] ], [ [ "model.compile(loss='categorical_crossentropy',\n optimizer='sgd',\n metrics=['accuracy'])", "_____no_output_____" ] ], [ [ "Optimize the model over the training data.", "_____no_output_____" ] ], [ [ "history = model.fit(x_train, y_train,\n batch_size=100,\n epochs=20,\n verbose=1,\n validation_data=(x_test, y_test))", "Train on 60000 samples, validate on 10000 samples\nEpoch 1/20\n60000/60000 [==============================] - 44s - loss: 0.3635 - acc: 0.8831 - val_loss: 0.0851 - val_acc: 0.9725\nEpoch 2/20\n60000/60000 [==============================] - 42s - loss: 0.1030 - acc: 0.9686 - val_loss: 0.0563 - val_acc: 0.9805\nEpoch 3/20\n60000/60000 [==============================] - 42s - loss: 0.0728 - acc: 0.9771 - val_loss: 0.0444 - val_acc: 0.9848\nEpoch 4/20\n60000/60000 [==============================] - 42s - loss: 0.0545 - acc: 0.9833 - val_loss: 0.0408 - val_acc: 0.9858\nEpoch 5/20\n60000/60000 [==============================] - 42s - loss: 0.0436 - acc: 0.9864 - val_loss: 0.0367 - val_acc: 0.9872\nEpoch 6/20\n60000/60000 [==============================] - 42s - loss: 0.0358 - acc: 0.9888 - val_loss: 0.0359 - val_acc: 0.9881\nEpoch 7/20\n60000/60000 [==============================] - 42s - loss: 0.0291 - acc: 0.9908 - val_loss: 0.0344 - val_acc: 0.9881\nEpoch 8/20\n60000/60000 [==============================] - 42s - loss: 0.0251 - acc: 0.9926 - val_loss: 0.0291 - val_acc: 0.9894\nEpoch 9/20\n60000/60000 [==============================] - 42s - loss: 0.0210 - acc: 0.9931 - val_loss: 0.0281 - val_acc: 0.9906\nEpoch 10/20\n60000/60000 [==============================] - 42s - loss: 0.0179 - acc: 0.9943 - val_loss: 0.0304 - val_acc: 0.9893\nEpoch 11/20\n60000/60000 [==============================] - 42s - loss: 0.0147 - acc: 0.9952 - val_loss: 0.0282 - val_acc: 0.9912\nEpoch 12/20\n60000/60000 [==============================] - 42s - loss: 0.0137 - acc: 0.9953 - val_loss: 0.0259 - val_acc: 0.9914\nEpoch 13/20\n60000/60000 [==============================] - 42s - loss: 0.0121 - acc: 0.9960 - val_loss: 0.0294 - val_acc: 0.9906\nEpoch 14/20\n60000/60000 [==============================] - 42s - loss: 0.0100 - acc: 0.9968 - val_loss: 0.0317 - val_acc: 0.9895\nEpoch 15/20\n60000/60000 [==============================] - 42s - loss: 0.0088 - acc: 0.9976 - val_loss: 0.0337 - val_acc: 0.9900\nEpoch 16/20\n60000/60000 [==============================] - 42s - loss: 0.0074 - acc: 0.9976 - val_loss: 0.0275 - val_acc: 0.9918\nEpoch 17/20\n60000/60000 [==============================] - 42s - loss: 0.0063 - acc: 0.9980 - val_loss: 0.0304 - val_acc: 0.9917\nEpoch 18/20\n60000/60000 [==============================] - 42s - loss: 0.0062 - acc: 0.9981 - val_loss: 0.0282 - val_acc: 0.9916\nEpoch 19/20\n60000/60000 [==============================] - 42s - loss: 0.0057 - acc: 0.9983 - val_loss: 0.0297 - val_acc: 0.9912\nEpoch 20/20\n60000/60000 [==============================] - 42s - loss: 0.0047 - acc: 0.9986 - val_loss: 0.0291 - val_acc: 0.9914\n" ], [ "plt.plot(history.history['loss'])\nplt.plot(history.history['val_loss'])\nplt.legend(['Training Loss','Testing Loss'])\nplt.xlabel('Epoch')\nplt.ylabel('Loss')\nplt.show()", "_____no_output_____" ], [ "plt.plot(history.history['acc'])\nplt.plot(history.history['val_acc'])\nplt.legend(['Training Accuracy','Testing Accuracy'])\nplt.xlabel('Epoch')\nplt.ylabel('Accuracy')\nplt.show()", "_____no_output_____" ] ] ]

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

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "# export\nimport time,os,shutil\nfrom sys import stdout\nfrom warnings import warn\nfrom fastprogress.core import *", "_____no_output_____" ], [ "#default_exp fastprogress", "_____no_output_____" ] ], [ [ "## Base class", "_____no_output_____" ] ], [ [ "#export\nclass ProgressBar():\n update_every,first_its = 0.2,5\n\n def __init__(self, gen, total=None, display=True, leave=True, parent=None, master=None, comment=''):\n self.gen,self.parent,self.master,self.comment = gen,parent,master,comment\n self.total = len(gen) if total is None else total\n self.last_v = 0\n if parent is None: self.leave,self.display = leave,display\n else:\n self.leave,self.display=False,False\n parent.add_child(self)\n self.last_v = None\n\n def on_iter_begin(self):\n if self.master is not None: self.master.on_iter_begin()\n \n def on_interrupt(self):\n if self.master is not None: self.master.on_interrupt()\n \n def on_iter_end(self):\n if self.master is not None: self.master.on_iter_end()\n \n def on_update(self, val, text): pass\n\n def __iter__(self):\n if self.total != 0: self.update(0)\n try:\n for i,o in enumerate(self.gen):\n if i >= self.total: break\n yield o\n self.update(i+1)\n except Exception as e:\n self.on_interrupt()\n raise e\n\n def update(self, val):\n if self.last_v is None: \n self.on_iter_begin()\n self.last_v = 0\n if val == 0:\n self.start_t = self.last_t = time.time()\n self.pred_t,self.last_v,self.wait_for = 0,0,1\n self.update_bar(0)\n elif val <= self.first_its or val >= self.last_v + self.wait_for or val >= self.total:\n cur_t = time.time()\n avg_t = (cur_t - self.start_t) / val\n self.wait_for = max(int(self.update_every / (avg_t+1e-8)),1)\n self.pred_t = avg_t * self.total\n self.last_v,self.last_t = val,cur_t\n self.update_bar(val)\n if val >= self.total: \n self.on_iter_end()\n self.last_v = None\n\n def update_bar(self, val):\n elapsed_t = self.last_t - self.start_t\n remaining_t = format_time(self.pred_t - elapsed_t)\n elapsed_t = format_time(elapsed_t)\n end = '' if len(self.comment) == 0 else f' {self.comment}'\n if self.total == 0:\n warn(\"Your generator is empty.\")\n self.on_update(0, '100% [0/0]')\n else: self.on_update(val, f'{100 * val/self.total:.2f}% [{val}/{self.total} {elapsed_t}<{remaining_t}{end}]')", "_____no_output_____" ], [ "class VerboseProgressBar(ProgressBar):\n def on_iter_begin(self): super().on_iter_begin(); print(\"on_iter_begin\")\n def on_interrupt(self): print(\"on_interrupt\")\n def on_iter_end(self): print(\"on_iter_end\"); super().on_iter_end()\n def on_update(self, val, text): print(f\"on_update {val}\")", "_____no_output_____" ], [ "from contextlib import redirect_stdout\nimport io", "_____no_output_____" ], [ "tst_pb = VerboseProgressBar(range(6))\ns = io.StringIO()\nwith redirect_stdout(s): \n for i in tst_pb: time.sleep(0.1)\n\nassert s.getvalue() == '\\n'.join(['on_iter_begin'] + [f'on_update {i}' for i in range(7)] + ['on_iter_end']) + '\\n'", "_____no_output_____" ], [ "tst_pb = VerboseProgressBar(range(6))\ns = io.StringIO()\nwith redirect_stdout(s): \n for i in range(7): \n tst_pb.update(i)\n time.sleep(0.1)\n\nassert s.getvalue() == '\\n'.join(['on_iter_begin'] + [f'on_update {i}' for i in range(7)] + ['on_iter_end']) + '\\n'", "_____no_output_____" ], [ "#export\nclass MasterBar(ProgressBar):\n def __init__(self, gen, cls, total=None): \n self.main_bar = cls(gen, total=total, display=False, master=self)\n \n def on_iter_begin(self): pass\n def on_interrupt(self): pass\n def on_iter_end(self): pass\n def add_child(self, child): pass\n def write(self, line): pass\n def update_graph(self, graphs, x_bounds, y_bounds): pass\n \n def __iter__(self):\n for o in self.main_bar:\n yield o\n \n def update(self, val): self.main_bar.update(val)", "_____no_output_____" ], [ "class VerboseMasterBar(MasterBar):\n def __init__(self, gen, total=None): super().__init__(gen, VerboseProgressBar, total=total)\n def on_iter_begin(self): print(\"master_on_iter_begin\")\n def on_interrupt(self): print(\"master_on_interrupt\")\n def on_iter_end(self): print(\"master_on_iter_end\")\n #def on_update(self, val, text): print(f\"master_on_update {val}\")", "_____no_output_____" ], [ "tst_mb = VerboseMasterBar(range(6))\nfor i in tst_mb: time.sleep(0.1)", "master_on_iter_begin\non_iter_begin\non_update 0\non_update 1\non_update 2\non_update 3\non_update 4\non_update 5\non_update 6\non_iter_end\nmaster_on_iter_end\n" ], [ "#hide\n#Test an empty progress bar doesn't crash\nfor i in ProgressBar([]): pass", "_____no_output_____" ] ], [ [ "## Notebook progress bars", "_____no_output_____" ] ], [ [ "#export\nif IN_NOTEBOOK:\n try:\n from IPython.display import clear_output, display, HTML\n import matplotlib.pyplot as plt\n import ipywidgets as widgets\n except:\n warn(\"Couldn't import ipywidgets properly, progress bar will use console behavior\")\n IN_NOTEBOOK = False", "_____no_output_____" ], [ "#export\nclass NBOutput():\n def __init__(self, to_display):\n self.out = widgets.Output()\n display(self.out)\n with self.out:\n display(to_display)\n\n def update(self, to_update):\n with self.out:\n clear_output(wait=True)\n display(to_update)", "_____no_output_____" ], [ "#export\nclass NBProgressBar(ProgressBar):\n def on_iter_begin(self):\n super().on_iter_begin()\n self.progress = html_progress_bar(0, self.total, \"\")\n if self.display:\n display(HTML(html_progress_bar_styles))\n self.out = NBOutput(HTML(self.progress))\n self.is_active=True\n\n def on_interrupt(self):\n self.on_update(0, 'Interrupted', interrupted=True)\n super().on_interrupt()\n self.on_iter_end()\n\n def on_iter_end(self):\n if not self.leave and self.display: self.out.update(HTML(''))\n self.is_active=False\n super().on_iter_end()\n \n def on_update(self, val, text, interrupted=False):\n self.progress = html_progress_bar(val, self.total, text, interrupted)\n if self.display: self.out.update(HTML(self.progress))\n elif self.parent is not None: self.parent.show()", "_____no_output_____" ], [ "tst = NBProgressBar(range(100))\nfor i in tst: time.sleep(0.05)", "_____no_output_____" ], [ "tst = NBProgressBar(range(100))\nfor i in range(50): \n time.sleep(0.05)\n tst.update(i)\ntst.on_interrupt()", "_____no_output_____" ], [ "#hide\nfor i in NBProgressBar([]): pass", "_____no_output_____" ], [ "#export\nclass NBMasterBar(MasterBar):\n names = ['train', 'valid']\n def __init__(self, gen, total=None, hide_graph=False, order=None, clean_on_interrupt=False, total_time=False):\n super().__init__(gen, NBProgressBar, total)\n if order is None: order = ['pb1', 'text', 'pb2']\n self.hide_graph,self.order = hide_graph,order\n self.report,self.clean_on_interrupt,self.total_time = [],clean_on_interrupt,total_time\n self.inner_dict = {'pb1':self.main_bar, 'text':\"\"}\n self.text,self.lines = \"\",[]\n \n def on_iter_begin(self):\n self.html_code = '\\n'.join([html_progress_bar(0, self.main_bar.total, \"\"), \"\"])\n display(HTML(html_progress_bar_styles))\n self.out = NBOutput(HTML(self.html_code))\n\n def on_interrupt(self):\n if self.clean_on_interrupt: self.out.update(HTML(''))\n\n def on_iter_end(self):\n if hasattr(self, 'imgs_fig'):\n plt.close()\n self.imgs_out.update(self.imgs_fig)\n if hasattr(self, 'graph_fig'):\n plt.close()\n self.graph_out.update(self.graph_fig)\n if self.text.endswith('<p>'): self.text = self.text[:-3]\n if self.total_time: \n total_time = format_time(time.time() - self.main_bar.start_t)\n self.text = f'Total time: {total_time} <p>' + self.text\n if hasattr(self, 'out'): self.out.update(HTML(self.text))\n\n def add_child(self, child):\n self.child = child\n self.inner_dict['pb2'] = self.child\n #self.show()\n\n def show(self):\n self.inner_dict['text'] = self.text\n to_show = [name for name in self.order if name in self.inner_dict.keys()]\n self.html_code = '\\n'.join([getattr(self.inner_dict[n], 'progress', self.inner_dict[n]) for n in to_show])\n self.out.update(HTML(self.html_code))\n\n def write(self, line, table=False):\n if not table: self.text += line + \"<p>\"\n else:\n self.lines.append(line)\n self.text = text2html_table(self.lines)\n\n def show_imgs(self, imgs, titles=None, cols=4, imgsize=4, figsize=None):\n if self.hide_graph: return\n rows = len(imgs)//cols if len(imgs)%cols == 0 else len(imgs)//cols + 1\n plt.close()\n if figsize is None: figsize = (imgsize*cols, imgsize*rows)\n self.imgs_fig, imgs_axs = plt.subplots(rows, cols, figsize=figsize)\n if titles is None: titles = [None] * len(imgs)\n for img, ax, title in zip(imgs, imgs_axs.flatten(), titles): img.show(ax=ax, title=title)\n for ax in imgs_axs.flatten()[len(imgs):]: ax.axis('off')\n if not hasattr(self, 'imgs_out'): self.imgs_out = NBOutput(self.imgs_fig)\n else: self.imgs_out.update(self.imgs_fig)\n\n def update_graph(self, graphs, x_bounds=None, y_bounds=None, figsize=(6,4)):\n if self.hide_graph: return\n if not hasattr(self, 'graph_fig'):\n self.graph_fig, self.graph_ax = plt.subplots(1, figsize=figsize)\n self.graph_out = NBOutput(self.graph_ax.figure)\n self.graph_ax.clear()\n if len(self.names) < len(graphs): self.names += [''] * (len(graphs) - len(self.names))\n for g,n in zip(graphs,self.names): self.graph_ax.plot(*g, label=n)\n self.graph_ax.legend(loc='upper right')\n if x_bounds is not None: self.graph_ax.set_xlim(*x_bounds)\n if y_bounds is not None: self.graph_ax.set_ylim(*y_bounds)\n self.graph_out.update(self.graph_ax.figure)", "_____no_output_____" ], [ "mb = NBMasterBar(range(5))\nfor i in mb:\n for j in NBProgressBar(range(10), parent=mb, comment=f'first bar stat'):\n time.sleep(0.01)\n #mb.child.comment = f'second bar stat'\n mb.write(f'Finished loop {i}.')", "_____no_output_____" ], [ "mb = NBMasterBar(range(5))\nmb.update(0) \nfor i in range(5):\n for j in NBProgressBar(range(10), parent=mb):\n time.sleep(0.01)\n #mb.child.comment = f'second bar stat'\n mb.main_bar.comment = f'first bar stat'\n mb.write(f'Finished loop {i}.')\n mb.update(i+1)", "_____no_output_____" ] ], [ [ "## Console progress bars", "_____no_output_____" ] ], [ [ "#export\nNO_BAR = False\nWRITER_FN = print\nFLUSH = True\nSAVE_PATH = None\nSAVE_APPEND = False\nMAX_COLS = 160", "_____no_output_____" ], [ "#export \ndef printing():\n return False if NO_BAR else (stdout.isatty() or IN_NOTEBOOK)", "_____no_output_____" ], [ "#export\nclass ConsoleProgressBar(ProgressBar):\n fill:str='█'\n end:str='\\r'\n\n def __init__(self, gen, total=None, display=True, leave=True, parent=None, master=None, txt_len=60):\n self.cols,_ = shutil.get_terminal_size((100, 40))\n if self.cols > MAX_COLS: self.cols=MAX_COLS\n self.length = self.cols-txt_len\n self.max_len,self.prefix = 0,''\n #In case the filling char returns an encoding error\n try: print(self.fill, end='\\r', flush=FLUSH)\n except: self.fill = 'X'\n super().__init__(gen, total, display, leave, parent, master)\n\n def on_interrupt(self):\n super().on_interrupt()\n self.on_iter_end()\n\n def on_iter_end(self):\n if not self.leave and printing():\n print(f'\\r{self.prefix}' + ' ' * (self.max_len - len(f'\\r{self.prefix}')), end='\\r', flush=FLUSH)\n super().on_iter_end()\n\n def on_update(self, val, text):\n if self.display:\n if self.length > self.cols-len(text)-len(self.prefix)-4:\n self.length = self.cols-len(text)-len(self.prefix)-4\n filled_len = int(self.length * val // self.total) if self.total else 0\n bar = self.fill * filled_len + '-' * (self.length - filled_len)\n to_write = f'\\r{self.prefix} |{bar}| {text}'\n if val >= self.total: end = '\\r'\n else: end = self.end\n if len(to_write) > self.max_len: self.max_len=len(to_write)\n if printing(): WRITER_FN(to_write, end=end, flush=FLUSH)", "_____no_output_____" ], [ "tst = ConsoleProgressBar(range(100))\nfor i in tst: time.sleep(0.05)", " |█████████████████████████████████████████████████████████████████████████████████████████████| 100.00% [100/100 00:05<00:00]\r" ], [ "tst = ConsoleProgressBar(range(100))\nfor i in range(50): \n time.sleep(0.05)\n tst.update(i)\ntst.on_interrupt()", " |███████████████████████████████████████████--------------------------------------------------| 47.00% [47/100 00:02<00:02]\r" ], [ "#export\ndef print_and_maybe_save(line):\n WRITER_FN(line)\n if SAVE_PATH is not None:\n attr = \"a\" if os.path.exists(SAVE_PATH) else \"w\"\n with open(SAVE_PATH, attr) as f: f.write(line + '\\n')", "_____no_output_____" ], [ "#export\nclass ConsoleMasterBar(MasterBar):\n def __init__(self, gen, total=None, hide_graph=False, order=None, clean_on_interrupt=False, total_time=False):\n super().__init__(gen, ConsoleProgressBar, total)\n self.total_time = total_time\n\n def add_child(self, child):\n self.child = child\n v = 0 if self.main_bar.last_v is None else self.main_bar.last_v\n self.child.prefix = f'Epoch {v+1}/{self.main_bar.total} :'\n self.child.display = True\n\n def on_iter_begin(self):\n super().on_iter_begin()\n if SAVE_PATH is not None and os.path.exists(SAVE_PATH) and not SAVE_APPEND:\n with open(SAVE_PATH, 'w') as f: f.write('')\n\n def write(self, line, table=False):\n if table:\n text = ''\n if not hasattr(self, 'names'):\n self.names = [name + ' ' * (8-len(name)) if len(name) < 8 else name for name in line]\n text = ' '.join(self.names)\n else:\n for (t,name) in zip(line,self.names): text += t + ' ' * (2 + len(name)-len(t))\n print_and_maybe_save(text)\n else: print_and_maybe_save(line)\n if self.total_time:\n total_time = format_time(time() - self.start_t)\n print_and_maybe_save(f'Total time: {total_time}')\n\n def show_imgs(*args, **kwargs): pass\n def update_graph(*args, **kwargs): pass", "_____no_output_____" ], [ "mb = ConsoleMasterBar(range(5))\nfor i in mb:\n for j in ConsoleProgressBar(range(10), parent=mb):\n time.sleep(0.01)\n #mb.child.comment = f'second bar stat'\n mb.main_bar.comment = f'first bar stat'\n mb.write(f'Finished loop {i}.')", "Finished loop 0. \nFinished loop 1. \nFinished loop 2. \nFinished loop 3. \nFinished loop 4. \n" ], [ "mb = ConsoleMasterBar(range(5))\nmb.update(0) \nfor i in range(5):\n for j in ConsoleProgressBar(range(10), parent=mb):\n time.sleep(0.01)\n #mb.child.comment = f'second bar stat'\n mb.main_bar.comment = f'first bar stat'\n mb.write(f'Finished loop {i}.')\n mb.update(i+1)\n \n # confirming a kwarg can be passed to ConsoleMasterBar instance\n mb.update_graph([[1,2],[3,4]], figsize=(10,5,))\n mb.show_imgs(figsize=(10,5,))", "Finished loop 0. \nFinished loop 1. \nFinished loop 2. \nFinished loop 3. \nFinished loop 4. \n" ], [ "#export\nif IN_NOTEBOOK: master_bar, progress_bar = NBMasterBar, NBProgressBar\nelse: master_bar, progress_bar = ConsoleMasterBar, ConsoleProgressBar", "_____no_output_____" ], [ "#export\n_all_ = ['master_bar', 'progress_bar']", "_____no_output_____" ], [ "#export\ndef force_console_behavior():\n \"Return the console progress bars\"\n return ConsoleMasterBar, ConsoleProgressBar", "_____no_output_____" ], [ "#export\ndef workaround_empty_console_output():\n \"Change console output behaviour to correctly show progress in consoles not recognizing \\r at the end of line\"\n ConsoleProgressBar.end = ''", "_____no_output_____" ] ], [ [ "## Export -", "_____no_output_____" ] ], [ [ "from nbdev.export import notebook2script\nnotebook2script()", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "<a href=\"https://colab.research.google.com/github/bsyllaisidk/Architecture-L/blob/main/Prediction_diabet.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>", "_____no_output_____" ], [ "Impotations de dependances\n", "_____no_output_____" ] ], [ [ "import numpy as np\nimport pandas as pd\n#import matplotlib.pyplot as plt\n#import seaborn as sns\nfrom sklearn.preprocessing import StandardScaler\nfrom sklearn.model_selection import train_test_split\nfrom sklearn import svm\nfrom sklearn.metrics import accuracy_score\n#from sklearn.cluster import KMeans\n#from sklearn.model_selection import train_test_split\n#from sklearn.ensemble import RandomForestRegressor\n#from sklearn import metrics\n", "_____no_output_____" ] ], [ [ "\nchargement des données du fichier csv dans une trame de données pendas", "_____no_output_____" ] ], [ [ "\ndiabet_data = pd.read_csv('/content/diabete.csv')", "_____no_output_____" ] ], [ [ "lecture des donnes du tableau ", "_____no_output_____" ] ], [ [ "pd.read_csv\n", "_____no_output_____" ] ], [ [ "les 5 premiers lignes du tableau", "_____no_output_____" ] ], [ [ "diabet_data.head()", "_____no_output_____" ], [ "# le nombre de ligne et de colonnes dans notre base de donnes ", "_____no_output_____" ], [ "diabet_data.shape", "_____no_output_____" ], [ "#Pour obtenir les mesures statistique de cette donnee.", "_____no_output_____" ], [ "diabet_data.describe()", "_____no_output_____" ] ], [ [ "le nombre de diabetique et le non diabetique \n0--> non Diabetique\n1--> Diabetique", "_____no_output_____" ] ], [ [ "diabet_data['diabete'].value_counts()", "_____no_output_____" ], [ "diabet_data.groupby('diabete').mean()", "_____no_output_____" ], [ "#Sepearation de données et les etiquettes\nX=diabet_data.drop(columns='diabete', axis=1)\nY=diabet_data['diabete']", "_____no_output_____" ], [ "#lecture de X\nprint(X)", " n_pregnant glucose tension thickness insulin bmi pedigree age\n0 6 148 72 35 0 33.6 0.627 50\n1 1 85 66 29 0 26.6 0.351 31\n2 8 183 64 0 0 23.3 0.672 32\n3 1 89 66 23 94 28.1 0.167 21\n4 0 137 40 35 168 43.1 2.288 33\n.. ... ... ... ... ... ... ... ...\n763 10 101 76 48 180 32.9 0.171 63\n764 2 122 70 27 0 36.8 0.340 27\n765 5 121 72 23 112 26.2 0.245 30\n766 1 126 60 0 0 30.1 0.349 47\n767 1 93 70 31 0 30.4 0.315 23\n\n[768 rows x 8 columns]\n" ], [ "print(Y)", "0 1\n1 0\n2 1\n3 0\n4 1\n ..\n763 0\n764 0\n765 0\n766 1\n767 0\nName: diabete, Length: 768, dtype: int64\n" ] ], [ [ "Normalisation des donnees.", "_____no_output_____" ] ], [ [ "scaler = StandardScaler()", "_____no_output_____" ], [ "scaler.fit(X)", "_____no_output_____" ], [ "standardized_date = scaler.transform(X)\n", "_____no_output_____" ], [ "print(standardized_date)", "[[ 0.63994726 0.84832379 0.14964075 ... 0.20401277 0.46849198\n 1.4259954 ]\n [-0.84488505 -1.12339636 -0.16054575 ... -0.68442195 -0.36506078\n -0.19067191]\n [ 1.23388019 1.94372388 -0.26394125 ... -1.10325546 0.60439732\n -0.10558415]\n ...\n [ 0.3429808 0.00330087 0.14964075 ... -0.73518964 -0.68519336\n -0.27575966]\n [-0.84488505 0.1597866 -0.47073225 ... -0.24020459 -0.37110101\n 1.17073215]\n [-0.84488505 -0.8730192 0.04624525 ... -0.20212881 -0.47378505\n -0.87137393]]\n" ], [ "X = standardized_date\nY = diabet_data['diabete']", "_____no_output_____" ], [ "print(X)\nprint(Y)", "[[ 0.63994726 0.84832379 0.14964075 ... 0.20401277 0.46849198\n 1.4259954 ]\n [-0.84488505 -1.12339636 -0.16054575 ... -0.68442195 -0.36506078\n -0.19067191]\n [ 1.23388019 1.94372388 -0.26394125 ... -1.10325546 0.60439732\n -0.10558415]\n ...\n [ 0.3429808 0.00330087 0.14964075 ... -0.73518964 -0.68519336\n -0.27575966]\n [-0.84488505 0.1597866 -0.47073225 ... -0.24020459 -0.37110101\n 1.17073215]\n [-0.84488505 -0.8730192 0.04624525 ... -0.20212881 -0.47378505\n -0.87137393]]\n0 1\n1 0\n2 1\n3 0\n4 1\n ..\n763 0\n764 0\n765 0\n766 1\n767 0\nName: diabete, Length: 768, dtype: int64\n" ], [ "#Essaye de train fractionné", "_____no_output_____" ], [ "X_train, X_test, Y_train, Y_test = train_test_split(X,Y,test_size=0.2, stratify=Y, random_state=2)", "_____no_output_____" ], [ "print(X.shape, X_train.shape, X_test.shape)", "(768, 8) (614, 8) (154, 8)\n" ] ], [ [ "Model de formation", "_____no_output_____" ] ], [ [ "classifier = svm.SVC(kernel='linear')", "_____no_output_____" ] ], [ [ "Formation de la machine a vecteur classée", "_____no_output_____" ] ], [ [ "classifier.fit(X_train, Y_train)", "_____no_output_____" ] ], [ [ "Evaluation de modele", "_____no_output_____" ], [ "Note de precision", "_____no_output_____" ] ], [ [ "# Score de precision sur les données de formation\nX_train_prediction = classifier.predict(X_train)\ntraining_data_accuracy = accuracy_score(X_train_prediction, Y_train)", "_____no_output_____" ], [ "print('Score de precision sur les données de formation : ', training_data_accuracy)", "Score de precision sur les données de formation : 0.7866449511400652\n" ], [ "# Score de precision sur les données de formation\nX_test_prediction = classifier.predict(X_test)\ntest_data_accuracy = accuracy_score(X_test_prediction, Y_test)", "_____no_output_____" ], [ "print('Score de precision sur les données de teste : ', test_data_accuracy)", "Score de precision sur les données de teste : 0.7727272727272727\n" ] ], [ [ "Faaier un systeme predictif", "_____no_output_____" ] ], [ [ "input_data = (2,197,70,45,543,30.5,0.158,53)\n\n# Changement de inptu_data en tableau nympy\ninput_data_as_numpy_array = np.asarray(input_data)\n\n# Remodeler le tableau comme nous le predisons une instance \ninput_data_reshaped = input_data_as_numpy_array.reshape(1,-1)\n\n# normaliser l’entrée de donnees\nstd_date = scaler.transform(input_data_reshaped)\nprint(std_date)\n\nprediction = classifier.predict(std_date)\nprint(prediction)\n\nif (prediction[0] == 0) :\n print('La personne n\\'est pas diabetique')\nelse:\n print('La peronne est diabetique') ", "[[-0.54791859 2.38188392 0.04624525 1.53455054 4.02192191 -0.18943689\n -0.94794368 1.68125866]]\n[1]\nLa peronne est diabetique\n" ] ] ]

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

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

[ "Unlicense" ]

[ "Unlicense" ]

[ "Unlicense" ]

[ [ [ "sql(\n '''\n WITH daily_avg_sales AS\n (SELECT\n date_format(order_date, '%Y-%m-%d') order_day,\n avg(sales) avg_sales\n FROM \n superstore\n GROUP BY\n order_day\n )\n SELECT order_day, avg_sales\n FROM daily_avg_sales\n ORDER BY order_day;\n '''\n)", "_____no_output_____" ] ] ]

[ "code" ]

[ [ "code" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "### Breakdown of lethality", "_____no_output_____" ] ], [ [ "import pickle\nimport pandas as pd\nimport os\nimport seaborn as sns\nimport matplotlib.pyplot as plt\nimport numpy as np\nfrom ast import literal_eval", "_____no_output_____" ] ], [ [ "#### Read in files", "_____no_output_____" ] ], [ [ "dir_in_res = '../out/20.0216 feat/reg_rf_boruta'\ndir_in_anlyz = os.path.join(dir_in_res, 'anlyz_filtered')\ndf_featSummary = pd.read_csv(os.path.join(dir_in_anlyz, 'feat_summary.csv')) #feature summary\ndf_featSummary['feat_sources'] = df_featSummary['feat_sources'].apply(literal_eval)\ndf_featSummary['feat_genes'] = df_featSummary['feat_genes'].apply(literal_eval)", "_____no_output_____" ], [ "feat_summary_annot_gene = pd.read_csv(os.path.join(dir_in_anlyz, 'onsamegene', 'feat_summary_annot.csv'), header=0, index_col=0)\n\ngs_name = 'paralog'\nfeat_summary_annot_paralog = pd.read_csv(os.path.join(dir_in_anlyz, f'insame{gs_name}', 'feat_summary_annot.csv'), header=0, index_col=0)\n\ngs_name = 'Panther'\nfeat_summary_annot_panther = pd.read_csv(os.path.join(dir_in_anlyz, f'insamegeneset{gs_name}', 'feat_summary_annot.csv'), header=0, index_col=0)\n", "_____no_output_____" ] ], [ [ "#### Breakdown - basic - top most important feature", "_____no_output_____" ] ], [ [ "df_counts = df_featSummary.groupby('feat_source1')['feat_source1'].count()\ndf_counts = df_counts.to_dict()\ndf_sl = pd.DataFrame([{'new_syn_lethal':df_counts['CERES'],\n 'classic_syn_lethal': sum([df_counts[k] for k in ['CN','Mut','RNA-seq']]) }])\ndf_sl = df_sl.T.squeeze()", "_____no_output_____" ], [ "df_sl", "_____no_output_____" ] ], [ [ "#### Breakdown of lethality, top most important feature", "_____no_output_____" ] ], [ [ "df_src1 = df_featSummary[['target','feat_source1']].set_index('target')\ndf = pd.DataFrame({'isNotCERES': df_src1.feat_source1.isin(['RNA-seq', 'CN', 'Mut']),\n 'sameGene': feat_summary_annot_gene.inSame_1,\n 'sameParalog': feat_summary_annot_paralog.inSame_1,\n 'sameGS': feat_summary_annot_panther.inSame_1,\n 'isCERES': df_src1.feat_source1 == 'CERES'\n })\n\nlethal_dict = {'sameGene': 'Same gene',\n 'sameParalog': 'Paralog',\n 'sameGS': 'Gene set',\n 'isCERES': 'Functional',\n 'isNotCERES': 'Classic synthetic'}", "_____no_output_____" ], [ "df_counts = pd.DataFrame({'sum':df.sum(axis=0)})\ndf_counts['lethality'] = [lethal_dict[n] for n in df_counts.index]", "_____no_output_____" ], [ "df_counts", "_____no_output_____" ], [ "plt.figure()\nax = sns.barplot(df_counts['lethality'], df_counts['sum'], color='steelblue')\nax.set(xlabel='Lethality types', ylabel='Number of genes')\nplt.tight_layout()", "_____no_output_____" ] ], [ [ "#### Breakdown of lethality, top 10 most important feature", "_____no_output_____" ] ], [ [ "df_src = df_featSummary.set_index('target').feat_sources\ndf = pd.DataFrame({'hasNoCERES': df_src.apply(lambda x: any([n in x for n in ['CN','Mut','RNA-seq','Lineage']])),\n 'sameGene': feat_summary_annot_gene.inSame_top10,\n 'sameParalog': feat_summary_annot_paralog.inSame_top10,\n 'sameGS': feat_summary_annot_panther.inSame_top10,\n 'hasCERES': df_src.apply(lambda x: 'CERES' in x)\n })\n\nlethal_dict = {'sameGene': 'Same gene',\n 'sameParalog': 'Paralog',\n 'sameGS': 'Gene set',\n 'hasCERES': 'Functional',\n 'hasNoCERES': 'Classic synthetic'}", "_____no_output_____" ], [ "df_counts = pd.DataFrame({'sum':df.sum(axis=0)})\ndf_counts['lethality'] = [lethal_dict[n] for n in df_counts.index]", "_____no_output_____" ], [ "df_counts", "_____no_output_____" ], [ "plt.figure()\nax = sns.barplot(df_counts['lethality'], df_counts['sum'], color='steelblue')\nax.set(xlabel='Lethality types', ylabel='Number of genes', ylim=[0,500])\nplt.tight_layout()", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "# Module 2: Inversion", "_____no_output_____" ], [ "In the previous module we started with a continuous distribution of a physical property and discretized it into many cells, then we performed a forward simulation that created data from known model parameters. Inversion, of course, is exactly the opposite process. Imagine each model parameter that we had represents a layer in a 1D layered earth. At the surface of the earth we measure the data, and when we invert we do so for the model parameters. Our goal is to take the observed data and recover models that emulate the real Earth as closely as possible. \n\nYou may have noticed that the act of discretizing our problem created more cells than data values. In our latter example we produced 20 data points from 1000 model parameters, which is only a few data points and many model parameters. While this was not much of a problem in the forward simulation, when we want to do the inverse process, that is, obtain the model parameters from the data, it is clear that we have many more unknowns than knowns. In short, we have an underdetermined problem, and therefore infinite possible solutions. In mathematical terms, geophysical surveys represent what are called \"ill-posed\" problems. \n\nAn \"ill-posed\" problem is any problem that does not satisfy the requirements for the definition of \"well-posed\" problem. A *well-posed* problem is a problem in mathematics that must satisfy all three of the following criteria:\n\n<ol>\n<li> A solution exists.\n<li> The solution is unique.\n<li> The solution's behaviors change continuously with continuously changing initial conditions.\n</ol>\n\nAny mathematical formulation that does not satisfy all three of the above is, by definition, an ill-posed problem. Since we are dealing with an underdetermined system, I hope that it is clear that we are dealing with an ill-posed problem (i.e., we have no unique solution), and we are going to have to come up with a method (or methods) that can help us choose from the available solutions. We need to devise an algorithm that can choose the \"best\" model from the infinitely many that are available to us. \n\nIn short, we are going to have to find an optimum model. More specifically, in the context of most geophysics problems, we are going to use gradient-based optimization. This process involves building an *objective function*, which is a function that casts our inverse problem as an optimization problem. We will build an objective function consisting of two parts:\n\n$$\n\\phi = \\phi_d + \\beta \\phi_m\n$$\n\nWhere the terms on the right hand side are (1) a data misfit (denoted as $\\phi_d$) and (2) a model regularization (denoted as $\\phi_m$). These two parts will be elaborated in detail below.\n\nOnce we have formulated the objective function, we will take derivatives and obtain a recovered model. This module will flesh out the details of the model objective function, and then take first and second derivatives to derive an expression that gives us a solution for our model parameters.\n\n", "_____no_output_____" ], [ "## The Data Misfit, $\\phi_d$", "_____no_output_____" ], [ "A *misfit* describes how close synthetic data matches measurements that are made in the field. Traditionally this term refers to the difference between the measured data and the predicted data. If these two quantities are sufficiently close, then we consider the model to be a viable candidate for the solution to our problem. Because the data are inaccurate, a model that reproduces those data exactly is not feasible. A realistic goal, rather, is to find a model whose predicted data are consistent with the errors in the observations, and this requires incorporating knowledge about the noise and uncertainties. The concept of fitting the data means that some estimate of the “noise” be available. Unfortunately “noise” within the context of inversion is everything that cannot be accounted for by a compatible relationship between the model and the data. More specifically, noise refers to (1) noise from data aquisition in the field, (2) uncertainty in source and receiver locations, (3) numerical error, (4) physical assumptions about our model that do not capture all of the physics. \n\nA standard approach is to assume that each datum, $d_i$, contains errors that can be described as Gaussian with a standard deviation $\\epsilon_i$. It is important to give a significant amount of thought towards assigning standard deviations in the data, but a reasonable starting point is to assign each $\\epsilon_i$ as $\\epsilon_i= floor +\\%|d_i|$. \n\nIncorporating both the differences between predicted and measured data and a measure of the uncertainties in the data yields our misfit function, $\\phi_d$:\n\n$$\n\\phi_d (m) = \\frac{1}{2} \\sum_{i=1}^N \\left( \\frac{F[m] -d_i^{obs} }{\\epsilon_i}\\right)^2 = \\frac{1}{2} \\|W_d(F[m] - d^{obs}) \\|_2^2\n$$ \n\nNote that the right hand size of the equation is written as a matrix-vector product, with each $\\epsilon_i$ in the denominator placed as elements on a diagonal matrix $W_d$, as follows:\n\n\\begin{equation}\n\\begin{split}\nW_d = \n\\begin{bmatrix}\n \\frac{1}{\\epsilon_1} & 0 & 0 & \\cdots & 0\\\\\n 0 & \\frac{1}{\\epsilon_2} & 0 & \\cdots & 0\\\\\n 0 & 0 & \\frac{1}{\\epsilon_3} & \\cdots & \\vdots\\\\\n 0 & 0 & 0 & \\ddots & \\frac{1}{\\epsilon_M}\\\\ \n\\end{bmatrix}\n\\end{split}\n\\end{equation}\n\nIf we return to linear problem from the previous section where our forward operator was simply a matrix of kernel functions, we can substitute $F[m]$ with $G$ and obtain\n$$\n\\phi_d (m) = \\frac{1}{2} \\sum_{i=1}^N \\left( \\frac{(Gm)_i -d_i^{obs} }{\\epsilon_i}\\right)^2 = \\frac{1}{2} \\|W_d(Gm - d^{obs}) \\|_2^2\n$$ \n\nNow that we have defined a measure of misfit, the next task is to determine a tolerance value, such that if the misfit is about equal to that value, then we have an acceptable fit. Suppose that the standard deviations are known and that errors are Gaussian, then $\\phi_d$ becomes a $\\chi_N^2$ variable with $N$ degrees of freedom. This is a well-known quantity with an expected value $E[\\chi_N^2]=N$ and a standard deviation of $\\sqrt{2N}$. Basically, what this means is that computing $\\phi_d$ should give us a value that is close to the number of data, $N$.\n\n\n\n", "_____no_output_____" ], [ "## The Model Regularization, $\\phi_m$", "_____no_output_____" ], [ "There are many options for choosing a model regularization, but the goal in determining a model regularization is the same: given that we have no unique solution, we must make assumptions in order to recast the problem in such a way that a solution exists. A general function used in 1D is as follows:\n\n$$\n\\phi_m = \\alpha_s \\int (m)^2 dx + \\alpha_x \\int \\left( \\frac{dm}{dx} \\right)^2 dx\n$$\n\nEach term in the above expression is a norm that measures characteristics about our model. The first term is a representation of the square of the Euclidean length for a continuous function, and therefore measures the length of the model, while the second term uses derivative information to measure the model's smoothness. Usually the model regularization is defined with respect to a reference model. In the above, the reference model would simply be zero, but choosing a non-zero reference model $m_{ref}$, yields the following:\n$$\n\\phi_m = \\alpha_s \\int (m-m_{ref})^2 dx + \\alpha_x \\int \\left( \\frac{d}{dx} (m-m_{ref}) \\right)^2 dx\n$$\n\nAs before, we will discretize this expression. It is easiest to break up each term and treat them separately, at first.\nWe will denote each term of $\\phi_m$ as $\\phi_s$ and $\\phi_x$, respectively. Consider the first term. Translating the integral to a sum yields:\n\n$$\n\\phi_s = \\alpha_s \\int (m)^2 dx \\rightarrow \\sum_{i=1}^N \\int_{x_{i-1}}^{x_i} (m_i)^2 dx = \\sum_{i=1}^N m_i^2 (x_i - x_{i-1})\n$$\n\nEach spatial \"cell\" is $x_i - x_{i-1}$, which is the distance between nodes, as you may recall from the previous module. To simplify notation, we will use $\\Delta x_{n_i}$ to denote the *ith* distance between nodes:\n\n<img src=\"figures/1D_domain_dx.png\" width=\"40%\" height=\"40%\"> <br>\n\n\nWe can then write $\\phi_s$ as:\n\n$$\n\\phi_s = \\alpha_s \\sum_{i=1}^N m_i^2 \\Delta x_{n_i} = \\alpha_s m^T W_s^T W_s m = \\alpha_s \\|W_s m\\|_2^2\n$$\n\nwith:\n\n\\begin{equation}\n\\begin{split}\nW_s = \n\\begin{bmatrix}\n {\\sqrt{\\Delta x_{n_1}}} & 0 & 0 & \\cdots & 0\\\\\n 0 & {\\sqrt{\\Delta x_{n_2}}} & 0 & \\cdots & 0\\\\\n 0 & 0 & {\\sqrt{\\Delta x_{n_3}}} & \\cdots & \\vdots\\\\\n 0 & 0 & 0 & \\ddots & {\\sqrt{\\Delta x_{n_N}}}\\\\ \n\\end{bmatrix}\n\\end{split}\n\\end{equation}\n\n\nFor the second term, we will do a similar process. First, we will delineate $\\Delta x_{c_i}$ as the distance between cell centers:\n\n<img src=\"figures/1D_h_lengths_dx.png\" width=\"40%\" height=\"40%\"> <br>\n\nA discrete approximation to the integral can be made by evaluating the derivative of the model based on how much it changes between the cell-centers, that is, we will take the average gradient between the *ith* and *i+1th* cells:\n\n$$\n\\phi_x = \\alpha_x \\int \\left( \\frac{dm}{dx} \\right)^2 dx \\rightarrow \\sum_{i=1}^{N-1} \\left( \\frac{m_{i+1}-m_i}{h_k}\\right) \\Delta x_{c_i} = m^T W_x^T W_x m = \\|W_x m\\|_2^2\n$$\n\nThe matrix $W_x$ is a finite difference matrix constructed thus:\n\n\n\\begin{equation}\n\\begin{split}\nD_x = \n\\begin{bmatrix}\n -\\frac{1}{{\\Delta x_{c_1}}} & \\frac{1}{{\\Delta x_{c_1}}} & 0 & \\cdots & 0\\\\\n 0 & -\\frac{1}{{\\Delta x_{c_2}}} & \\frac{1}{{\\Delta x_{c_2}}} & \\cdots & 0\\\\\n 0 & 0 & \\ddots & \\ddots & \\vdots\\\\\n 0 & 0 & 0 & -\\frac{1}{{\\Delta x_{c_{M-1}}}} & \\frac{1}{{\\Delta x_{c_{M-1}}}}\\\\ \n\\end{bmatrix}\n\\end{split}\n\\end{equation}\n\nand then we need to account for the integration, so we multiply by a diagonal matrix $\\rm diag(\\sqrt{v})$\n\n\\begin{equation}\nW_x = D_x \\rm diag(\\sqrt{v})\n\\end{equation}\n\nSo to summarize, we have $\\phi_m = \\phi_s + \\phi_x$ with \n\n\\begin{equation}\n\\begin{split}\n \\phi_m & = \\phi_s + \\phi_x \\\\[0.4em]\n & = \\alpha_s \\|W_s (m - m_{ref})\\|_2^2 + \\alpha_x \\|W_x (m - m_{ref})\\|_2^2 \\\\[0.4em] \n\\end{split}\n\\end{equation}\n\nNext, we will write this more compactly by stacking $W_s$ and $W_x$ into a matrix $W_m$ as follows\n\n\\begin{equation}\n\\begin{split}\nW_m = \n\\begin{bmatrix}\n \\sqrt{\\alpha_s} W_s\\\\\n \\sqrt{\\alpha_x} W_x\n\\end{bmatrix}\n\\end{split}\n\\end{equation}", "_____no_output_____" ], [ "## Model Objective Function", "_____no_output_____" ], [ "If we go back and recall what was discussed in the introduction, the model objective function casts the inverse problem as an optimization problem, and as mentioned, we will be using gradient-based optimization, so we will need to take derivatives. The complete model objective function that we are dealing with will contain both the data misfit and the model regularization. This means that we can write it as $\\phi$ as the sum of the two and then differentiate:\n\n$$\n\\phi = \\phi_d + \\beta \\phi_m\n$$\nFor the linear problem we are considering\n$$\n\\phi_d = \\frac{1}{2}\\| W_d (Gm-d^{obs})\\|_2^2 = \\frac{1}{2}(Gm-d^{obs})^T W_d^T W_d (Gm-d^{obs})\n$$\nand\n$$\n\\phi_m = \\frac{1}{2} \\|W_m (m-m_{ref}) \\|^2_2 = \\frac{1}{2}(m-m_{ref})^T W_m^T W_m (m-m_{ref})\n$$\n\nTo simplify the terms and see the math a little more clearly, let's note that $W_d(Gm-d^{obs})$, and $\\beta W_m(m-m_{ref})$ are simply vectors. And since we are taking the square of the 2-norm, all that we are really doing is taking the dot product of each vector with itself. So let $z=W_d(Gm-d^{obs})$, and let $y=W_m(m-m_{ref})$ where both $z$ and $y$ vectors are functions of $m$. So then:\n\n$$\n\\phi_d = \\frac{1}{2}\\|z\\|_2^2 = \\frac{1}{2}z^T z \n$$<br>\n$$\n\\phi_m = \\frac{1}{2}\\|y\\|_2^2 =\\frac{1}{2}y^T y \n$$\n\n\nTo minimize this, we want to look at $\\nabla \\phi$. Using our compact expressions:\n$$\n\\phi = \\phi_d + \\beta \\phi_m = \\frac{1}{2}z^Tz + \\beta \\frac{1}{2}y^Ty \\\\ \n$$\n\nTaking the derivative with respect to $m$ yields:\n\\begin{equation}\n\\begin{split}\n\\frac{d \\phi}{dm}& = \\frac{1}{2} \\left(z^T \\frac{dz}{dm} + z^T \\frac{dz}{dm} + \\beta y^T \\frac{dy}{dm} + \\beta y^T \\frac{dy}{dm}\\right)\\\\\\\\[0.6em]\n& = z^T \\frac{dz}{dm} + \\beta y^T \\frac{dy}{dm}\n\\end{split}\n\\end{equation}\n\nNote that \n$$\\frac{dz}{dm} = \\frac{d}{dm}(W_d(Gm-d^{obs})) = W_d G $$ \n\nand \n\n$$ \\frac{dy}{dm} = \\frac{d}{dm}(W_m (m-m_{ref})) = W_m $$\n\nNext, let's substitute both derivatives, our expressions for $z$ and $y$, apply the transposes, and rearrange:<br>\n\n\\begin{equation}\n\\begin{split}\n\\frac{d \\phi}{dm} & = z^T \\frac{dz}{dm} + \\beta y^T \\frac{dy}{dm} \\\\[0.6em]\n & = (W_d(Gm-d^{obs}))^T W_d G + \\beta (W_m (m-m_{ref}))^T W_m\\\\[0.6em]\n & = (Gm-d^{obs})^T W_d^T W_d G + \\beta (m-m_{ref})^T W_m^T W_m \\\\[0.6em]\n & = ((Gm)^T - d^T) W_d^T W_d G + \\beta (m^T-m_{ref}^T)W_m^T W_m \\\\[0.6em]\n & = (m^T G^T - d^T) W_d^T W_d G + \\beta m^T W_m^T W_m - \\beta m_{ref}^T W_m^T W_m \\\\[0.6em]\n & = m^T G^T W_d^T W_d G - d^T W_d^T W_d G + \\beta m^T W_m^T W_m - \\beta m_{ref}^T W_m^T W_m\\\\[0.6em]\n & = m^T G^T W_d^T W_d G + \\beta m^T W_m^T W_m - d^T W_d^T W_d G - \\beta m_{ref}^T W_m^T W_m\n \\end{split}\n\\end{equation}\n\nNow we have an expression for the derivative of our equation that we can work with. Setting the gradient to zero and gathering like terms gives:<br>\n\n\\begin{equation} \n\\begin{split}\n m^T G^T W_d^T W_d G + \\beta m^T W_m^T W_m = d^T W_d^T W_d G + \\beta m_{ref}^T W_m^T W_m\\\\[0.6em]\n (G^T W_d^T W_d G + \\beta W_m^T W_m)m = G^T W_d^T W_d d + \\beta W_m^T W_m m_{ref}\\\\[0.6em]\n\\end{split}\n\\end{equation}\n\nFrom here we can do two things. First, we can solve for $m$, our recovered model:\n\n\\begin{equation}\n\\begin{split}\n m = (G^T W_d^T W_d G + \\beta W_m^T W_m)^{-1} (G^T W_d^T W_d d + \\beta W_m^T W_m m_{ref})\\\\[0.6em]\n\\end{split}\n\\end{equation}\n\nSecond, we can get the second derivative simply from the bracketed terms on the left hand side of the equation above:\n\\begin{equation} \n\\frac{d^2 \\phi}{dm^2} = G^T W_d^T W_d G + \\beta W_m^T W_m\n\\end{equation}\n\nIn the model problem that we are solving, second derivative information is not required to obtain a solution, however, in non-linear problems or situations when higher order information is required, it is useful to have this available when we need it. \n\n\n\n", "_____no_output_____" ], [ "## Solving for $m$ in Python", "_____no_output_____" ], [ "Before we solve for $m$, we will recreate what we had in the first module. First, install the appropriate packages:", "_____no_output_____" ] ], [ [ "# Import Packages\nimport numpy as np\nimport matplotlib.pyplot as plt", "_____no_output_____" ] ], [ [ "Here is the model that we had previously:", "_____no_output_____" ] ], [ [ "# Begin by creating a ficticious set of model data\n\nn_cells = 1000 # Set number of model parameters \nn_nodes = n_cells + 1\nxn = np.linspace(0, 1, n_nodes) # Define 1D domain on nodes\nxc = 0.5*(xn[1:] + xn[:-1]) # Define 1D domain on cell centers", "_____no_output_____" ], [ "# Define Gaussian function:\ndef gauss(x, amplitude, mean, std):\n \"\"\"Define a gaussian function\"\"\"\n return amplitude * np.exp(-((x-mean)/std)**2 / 2) \n\n# Choose parameters for Gaussian, evaluate, and store in an array, f.\nstd = 6e-2 \nmean = 0.7\namplitude_gaussian = 1 \ngaussian = gauss(xc, amplitude_gaussian, mean, std)\n\nfig, ax = plt.subplots(1, 1)\nax.plot(xc, gaussian)\nax.set_title(\"Gaussian\")", "_____no_output_____" ], [ "# Define a boxcar function:\nx_boxcar = np.r_[0.2, 0.35]\namplitude_boxcar = 1\nboxcar = np.zeros(n_cells) # initialize an array of all zeros\nboxcar_inds = (xc >= x_boxcar.min()) & (xc <= x_boxcar.max()) # find the indices of the boxcar\nboxcar[boxcar_inds] = amplitude_boxcar \n\n# construct the model\nmtrue = gaussian + boxcar\n \n# Plot \nfig, ax = plt.subplots(1, 1)\nax.plot(xc, mtrue)\nax.set_xlabel('x')\nax.set_ylabel('m(x)')\nax.set_title('Model, $m(x)$')", "_____no_output_____" ] ], [ [ "Again, we define out kernel functions and averaging and volume matrices as before:", "_____no_output_____" ] ], [ [ "# Make the set of kernel functions\n\ndef kernel_functions(x, j, p, q): \n return np.exp(-p*j*x) * np.cos(2*np.pi*q*j*x) \n\np = 0.01 # Set values for p, q\nq = 0.15\nn_data = 20 # specify number of output data\n\nj_min = 0\nj_max = n_data\nj_values = np.linspace(j_min, j_max, n_data)\n\nGn = np.zeros((n_nodes, n_data))\n\nfor i, j in enumerate(j_values):\n Gn[:, i] = kernel_functions(xn, j, p, q)\n \n# Plot \nfig, ax = plt.subplots(1, 1)\nax.plot(xn, Gn)\nax.set_xlabel('x')\nax.set_ylabel('g(x)')\nax.set_title('Kernel functions, $g(x)$')", "_____no_output_____" ], [ "# Make Averaging Matrix\nAv = np.zeros((n_cells, n_nodes)) # Create a matrix of zeros of the correct dimensions \n\n# and fill in with elements usin the loop below (note the 1/2 is included in here). \nfor i in range(n_cells):\n Av[i, i] = 0.5 \n Av[i, i+1] = 0.5 \n\nprint(Av.shape)", "(1000, 1001)\n" ], [ "# make the Volume, \"delta x\" array\ndelta_x = np.diff(xn) # set x-spacings\nV = np.diag(delta_x) # create diagonal matrix \n\nprint(V.shape)", "(1000, 1000)\n" ] ], [ [ "Last, we produce our data:", "_____no_output_____" ] ], [ [ "G = Gn.T @ Av.T @ V", "_____no_output_____" ], [ "d = G @ mtrue", "_____no_output_____" ], [ "# Plot\nfig, ax = plt.subplots(1, 1)\nax.plot(d, '-o')\nax.set_title('Synthetic Data $d$')", "_____no_output_____" ] ], [ [ "## Introducing noise to the data", "_____no_output_____" ], [ "This is where we stood at the end of the last module. Next, to simulate taking data in the field, we are going to add a noise to the data before we perform our inversion. We will do this by defining a lambda function that assigns a floor value and percent scaling factor. Also, we will assume that the noise is Gaussian. We then add the noise to the original data to make a simulated vector of observed data. The superposition of our noise and original data is plotted below.", "_____no_output_____" ] ], [ [ "# Add noise to our synthetic data\n\nadd_noise = False # set to true if you want to add noise to the data\n\nif add_noise is True:\n relative_noise = 0.04\n noise_floor = 1e-2\n noise = (\n relative_noise * np.random.randn(n_data) * np.abs(d) + # percent of data\n noise_floor * np.random.randn(n_data)\n )\n dobs = d + noise\nelse: \n dobs = d\n\nfig, ax = plt.subplots(1, 1)\nax.plot(d, '-o', label=\"d clean\")\nax.plot(dobs, '-s', label=\"dobs\")\nax.set_title(\"synthetic data\")\nax.legend()", "_____no_output_____" ] ], [ [ "# Setting up the inverse problem\n\nNow we will assemble the pieces for constructing an objective function to be minimized in the inversion. \n\nThroughout we use L2 norms, so the first thing we will do is define a simple function for computing a weighted L2 norm. ", "_____no_output_____" ] ], [ [ "def weighted_l2_norm(W, v):\n \"\"\"\n A function that returns a weighted L2 norm. The parameter W is a weighting matrix \n and v is a vector. \n \"\"\"\n Wv = W @ v\n return Wv.T @ Wv", "_____no_output_____" ] ], [ [ "## Calculating $\\phi_d$", "_____no_output_____" ], [ "We are now in a position to build up the data misfit term, $\\phi_d$. We will need a function to compute the 2-norm, so constructing a function to do this is useful. Next we will make the matrix $W_d$, which is a diagonal matrix that contains the inverses of the uncertainty in our data. Again, we will define a floor and percent error for each datum. Last, we calculate $\\phi_d$ using our 2-norm function that we created. It is insightful to see what values have been assigned to our floor and misfit, so they are printed below.", "_____no_output_____" ] ], [ [ "# Calculate the data misfit, phi_d\n\nnoise_floor = 1e-3\nrelative_error = 0.05\nstandard_deviation = noise_floor + relative_error * np.abs(dobs)\n\n# construct Wd\nWd = np.diag(1/standard_deviation)\n\nfig, ax = plt.subplots(1, 1)\nimg = ax.imshow(Wd, \"Greys\")\nplt.colorbar(img, ax=ax)\nax.set_title(\"Wd\")", "_____no_output_____" ] ], [ [ "## Calculating $\\phi_m$\n\nAs discussed above, we are going to first need to make our $W_m$ matrix, which is a partitioned matrix from two other matrices, $W_s$ and $W_x$, each scaled by a separate parameter $\\alpha_s$ and $\\alpha_x$. We are going to discuss the manner in which $\\alpha_s$ and $\\alpha_x$ are selected in more detail during the next module. But for the moment, we will set them as they are defined below. Once this matrix is built up, calculating $\\phi_m$ is a simple matter, given that we have made a function to compute the 2-norm already. For the sake of illustration, I compute and print $\\phi_m$ from the residual of our reference model $m_{ref}$ and our true model $m$. However, of interest to us will be the residual of the model that we recover $m_{rec}$ and our reference model.", "_____no_output_____" ] ], [ [ "# Start with Ws \nsqrt_vol = np.sqrt(delta_x) # in 1D - the \"Volume\" = length of each cell (delta_x)\nWs = np.diag(sqrt_vol)\n\n# and now Wx\nDx = np.zeros((n_cells-1, n_cells)) # differencing matrix\n\nfor i, dx in enumerate(delta_x[:-1]):\n Dx[i, i] = -1/dx\n Dx[i, i+1] = 1/dx\n\nWx = Dx @ np.diag(sqrt_vol)\n\nprint(Ws.shape, Wx.shape)", "(1000, 1000) (999, 1000)\n" ], [ "# plot both\nfig, ax = plt.subplots(1, 2, figsize=(10, 4))\nplot_up_to = 10 # plot 10 entries\n\n# plot Ws\nimg = ax[0].imshow(Ws[:plot_up_to, :plot_up_to], \"Greys\")\nplt.colorbar(img, ax=ax[0])\nax[0].set_title(\"Ws\")\n\n# plot Wx\nimg = ax[1].imshow(Wx[:plot_up_to, :plot_up_to+1], \"bwr\")\nplt.colorbar(img, ax=ax[1])\nax[1].set_title(\"Wx\")\n\nplt.tight_layout()", "_____no_output_____" ] ], [ [ "### Stack Ws, Wx to make a single regularization matrix Wm", "_____no_output_____" ] ], [ [ "alpha_s = 1e-6\nalpha_x = 1\n\nWm = np.vstack([\n np.sqrt(alpha_s)*Ws, \n np.sqrt(alpha_x)*Wx\n])\n\nprint(Wm.shape)", "(1999, 1000)\n" ] ], [ [ "## Inverting for our recovered model", "_____no_output_____" ], [ "At last we can invert to find our recovered model and see how it compares with the true model. First we will assign a value for $\\beta$. As with the $\\alpha$ parameters from before, we will assign a value, but the choice of beta will be a topic that we explore more fully in the next module. Once our $\\beta$ value is assigned, we will define yet another lambda function to obtain the recovered model, plot it against our true model, and then output our results for $\\phi_d$ and $\\phi_m$. ", "_____no_output_____" ] ], [ [ "beta = 1e-1 # Set beta value\nmref = 0.5 * np.ones(n_cells) # choose a reference model\n\nWdG = Wd @ G\nmrec = (\n np.linalg.inv(WdG.T @ WdG + beta * Wm.T @ Wm) @ \n (WdG.T @ Wd @ dobs + beta * Wm.T @ Wm @ mref)\n)", "_____no_output_____" ], [ "fig, ax = plt.subplots(1, 1)\n\nax.plot(xc, mtrue, label=\"true\")\nax.plot(xc, mrec, label=\"recovered\")\nax.legend()", "_____no_output_____" ], [ "dpred = G @ mrec\n\nfig, ax = plt.subplots(1, 1)\nax.plot(dobs, '-o', label=\"observed\")\nax.plot(dpred, '-s', label=\"predicted\")\nax.legend()", "_____no_output_____" ] ], [ [ "This concludes the current module. For the next module, we will examine the constraints on our choice for $\\alpha_s$ and $\\alpha_x$, and then introduce the Tikhonov curve and a method for choosing $\\beta$.", "_____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", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown" ], [ "code", "code", "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code" ], [ "markdown" ], [ "code", "code" ], [ "markdown" ], [ "code" ], [ "markdown", "markdown" ], [ "code", "code", "code" ], [ "markdown" ] ]

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "import tensorflow as tf\nimport os\nimport pickle\nimport numpy as np\n\nCIFAR_DIR = \"./../../cifar-10-batches-py\"\nprint(os.listdir(CIFAR_DIR))", "_____no_output_____" ], [ "def load_data(filename):\n \"\"\"read data from data file.\"\"\"\n with open(filename, 'rb') as f:\n data = pickle.load(f, encoding='bytes')\n return data[b'data'], data[b'labels']\n\n# tensorflow.Dataset.\nclass CifarData:\n def __init__(self, filenames, need_shuffle):\n all_data = []\n all_labels = []\n for filename in filenames:\n data, labels = load_data(filename)\n all_data.append(data)\n all_labels.append(labels)\n self._data = np.vstack(all_data)\n self._data = self._data / 127.5 - 1\n self._labels = np.hstack(all_labels)\n print(self._data.shape)\n print(self._labels.shape)\n \n self._num_examples = self._data.shape[0]\n self._need_shuffle = need_shuffle\n self._indicator = 0\n if self._need_shuffle:\n self._shuffle_data()\n \n def _shuffle_data(self):\n # [0,1,2,3,4,5] -> [5,3,2,4,0,1]\n p = np.random.permutation(self._num_examples)\n self._data = self._data[p]\n self._labels = self._labels[p]\n \n def next_batch(self, batch_size):\n \"\"\"return batch_size examples as a batch.\"\"\"\n end_indicator = self._indicator + batch_size\n if end_indicator > self._num_examples:\n if self._need_shuffle:\n self._shuffle_data()\n self._indicator = 0\n end_indicator = batch_size\n else:\n raise Exception(\"have no more examples\")\n if end_indicator > self._num_examples:\n raise Exception(\"batch size is larger than all examples\")\n batch_data = self._data[self._indicator: end_indicator]\n batch_labels = self._labels[self._indicator: end_indicator]\n self._indicator = end_indicator\n return batch_data, batch_labels\n\ntrain_filenames = [os.path.join(CIFAR_DIR, 'data_batch_%d' % i) for i in range(1, 6)]\ntest_filenames = [os.path.join(CIFAR_DIR, 'test_batch')]\n\ntrain_data = CifarData(train_filenames, True)\ntest_data = CifarData(test_filenames, False)", "_____no_output_____" ], [ "x = tf.placeholder(tf.float32, [None, 3072])\ny = tf.placeholder(tf.int64, [None])\n# [None], eg: [0,5,6,3]\nx_image = tf.reshape(x, [-1, 3, 32, 32])\n# 32*32\nx_image = tf.transpose(x_image, perm=[0, 2, 3, 1])\n\n# conv1: 神经元图， feature_map, 输出图像\nconv1_1 = tf.layers.conv2d(x_image,\n 32, # output channel number\n (3,3), # kernel size\n padding = 'same',\n activation = tf.nn.relu,\n name = 'conv1_1')\nconv1_2 = tf.layers.conv2d(conv1_1,\n 32, # output channel number\n (3,3), # kernel size\n padding = 'same',\n activation = tf.nn.relu,\n name = 'conv1_2')\n\n# 16 * 16\npooling1 = tf.layers.max_pooling2d(conv1_2,\n (2, 2), # kernel size\n (2, 2), # stride\n name = 'pool1')\n\n\nconv2_1 = tf.layers.conv2d(pooling1,\n 32, # output channel number\n (3,3), # kernel size\n padding = 'same',\n activation = tf.nn.relu,\n name = 'conv2_1')\nconv2_2 = tf.layers.conv2d(conv2_1,\n 32, # output channel number\n (3,3), # kernel size\n padding = 'same',\n activation = tf.nn.relu,\n name = 'conv2_2')\n# 8 * 8\npooling2 = tf.layers.max_pooling2d(conv2_2,\n (2, 2), # kernel size\n (2, 2), # stride\n name = 'pool2')\n\nconv3_1 = tf.layers.conv2d(pooling2,\n 32, # output channel number\n (3,3), # kernel size\n padding = 'same',\n activation = tf.nn.relu,\n name = 'conv3_1')\nconv3_2 = tf.layers.conv2d(conv3_1,\n 32, # output channel number\n (3,3), # kernel size\n padding = 'same',\n activation = tf.nn.relu,\n name = 'conv3_2')\n# 4 * 4 * 32\npooling3 = tf.layers.max_pooling2d(conv3_2,\n (2, 2), # kernel size\n (2, 2), # stride\n name = 'pool3')\n# [None, 4 * 4 * 32]\nflatten = tf.layers.flatten(pooling3)\ny_ = tf.layers.dense(flatten, 10)\n\nloss = tf.losses.sparse_softmax_cross_entropy(labels=y, logits=y_)\n# y_ -> sofmax\n# y -> one_hot\n# loss = ylogy_\n\n# indices\npredict = tf.argmax(y_, 1)\n# [1,0,1,1,1,0,0,0]\ncorrect_prediction = tf.equal(predict, y)\naccuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float64))\n\nwith tf.name_scope('train_op'):\n train_op = tf.train.AdamOptimizer(1e-3).minimize(loss)", "_____no_output_____" ], [ "init = tf.global_variables_initializer()\nbatch_size = 20\ntrain_steps = 10000\ntest_steps = 100\n\n# train 10k: 73.4%\nwith tf.Session() as sess:\n sess.run(init)\n for i in range(train_steps):\n batch_data, batch_labels = train_data.next_batch(batch_size)\n loss_val, acc_val, _ = sess.run(\n [loss, accuracy, train_op],\n feed_dict={\n x: batch_data,\n y: batch_labels})\n if (i+1) % 100 == 0:\n print('[Train] Step: %d, loss: %4.5f, acc: %4.5f' \n % (i+1, loss_val, acc_val))\n if (i+1) % 1000 == 0:\n test_data = CifarData(test_filenames, False)\n all_test_acc_val = []\n for j in range(test_steps):\n test_batch_data, test_batch_labels \\\n = test_data.next_batch(batch_size)\n test_acc_val = sess.run(\n [accuracy],\n feed_dict = {\n x: test_batch_data, \n y: test_batch_labels\n })\n all_test_acc_val.append(test_acc_val)\n test_acc = np.mean(all_test_acc_val)\n print('[Test ] Step: %d, acc: %4.5f' % (i+1, test_acc))\n \n \n ", "_____no_output_____" ] ] ]

[ "code" ]

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "%matplotlib inline\n\"\"\"\ndata source:\nhttps://zola.planning.nyc.gov/data#12.31/40.73327/-73.92447\nhttps://www1.nyc.gov/site/planning/data-maps/open-data/dwn-pluto-mappluto.page\nhttps://sfplanninggis.org/PIM/help.html\n\"\"\"\n\nimport os\nimport time\nimport pickle\n\nimport numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nfrom scipy import ndimage\n\nimport torch\nimport torch.nn as nn\nimport torch.optim as optim\nimport torch.nn.functional as F\n\nfrom IPython.display import clear_output\nfrom datetime import datetime\n\nfrom lib.utils import SamplePool, make_seed, make_circle_masks, get_rand_avail\nfrom lib.utils import get_sobel, softmax\nfrom lib.NCCAModel2 import NCCAModel2", "_____no_output_____" ], [ "with open('anchor_loc.pickle', 'rb') as handle:\n anchor_loc = pickle.load(handle)\n\nroot = \"_maps/\"\nfull_size = (100,100)\nmap_size = (80,80)\ncolor_map = [(0.5,0.5,0.5),\n (0.5,1.0,0.5),\n (1.0,1.0,0.5),\n (1.0,0.7,0.2),\n (1.0,0.5,0.5),\n (1.0,0.5,1.0)]\n\n################################################################\n\nd_trains = []\nd_tests = []\nalive_maps = []\n\nfor d_i, obj_name in enumerate(list(anchor_loc.keys())[:10]):\n\n filenames = []\n common_index = {}\n\n for filename in os.listdir(root):\n if filename[:len(obj_name)]==obj_name:\n filenames.append(root+filename)\n\n for filename in filenames:\n with open(filename, 'rb') as handle:\n map_dict = pickle.load(handle)\n for index in map_dict:\n try:\n tmp = int(map_dict[index]['status'])\n if index in common_index:\n common_index[index]+= 1\n else:\n common_index[index] = 1\n except (TypeError, KeyError):\n continue\n\n common_index = [x for x in common_index.keys() if common_index[x]==len(filenames)]\n\n d_train = np.zeros([64, full_size[0], full_size[1], 4])\n d_test = np.zeros([len(filenames)-d_train.shape[0], full_size[0], full_size[1], d_train.shape[-1]])\n\n for i,filename in enumerate(filenames[:d_train.shape[0]]):\n with open(filename, 'rb') as handle:\n map_dict = pickle.load(handle)\n for index in common_index:\n try:\n status = min(int(map_dict[index]['status'])-1, 3)\n d_train[i, index[0], index[1]] = np.zeros(d_train.shape[-1])\n d_train[i, index[0], index[1], status] = 1\n except (TypeError, KeyError):\n continue\n\n for i,filename in enumerate(filenames[d_train.shape[0]:]):\n with open(filename, 'rb') as handle:\n map_dict = pickle.load(handle)\n for index in common_index:\n try:\n status = min(int(map_dict[index]['status'])-1, 3)\n d_test[i, index[0], index[1]] = np.zeros(d_test.shape[-1])\n d_test[i, index[0], index[1], status] = 1\n except (TypeError, KeyError):\n continue\n\n alive_map = np.expand_dims(np.expand_dims(np.sum(d_train[0, ...], -1)>0.001, 0), -1)\n\n cut_off = ((full_size[0]-map_size[0])//2, (full_size[1]-map_size[1])//2)\n d_train = d_train[:, cut_off[0]:(cut_off[0]+map_size[0]),\n cut_off[1]:(cut_off[1]+map_size[1]), :]\n d_test = d_test[:, cut_off[0]:(cut_off[0]+map_size[0]),\n cut_off[1]:(cut_off[1]+map_size[1]), :]\n alive_map = alive_map[:, cut_off[0]:(cut_off[0]+map_size[0]),\n cut_off[1]:(cut_off[1]+map_size[1]), :]\n\n print(d_train.shape, d_test.shape, alive_map.shape)\n \n d_trains.append(d_train)\n d_tests.append(d_test)\n alive_maps.append(alive_map)", "(64, 80, 80, 4) (28, 80, 80, 4) (1, 80, 80, 1)\n(64, 80, 80, 4) (206, 80, 80, 4) (1, 80, 80, 1)\n(64, 80, 80, 4) (30, 80, 80, 4) (1, 80, 80, 1)\n(64, 80, 80, 4) (44, 80, 80, 4) (1, 80, 80, 1)\n(64, 80, 80, 4) (32, 80, 80, 4) (1, 80, 80, 1)\n(64, 80, 80, 4) (32, 80, 80, 4) (1, 80, 80, 1)\n(64, 80, 80, 4) (45, 80, 80, 4) (1, 80, 80, 1)\n(64, 80, 80, 4) (12, 80, 80, 4) (1, 80, 80, 1)\n(64, 80, 80, 4) (39, 80, 80, 4) (1, 80, 80, 1)\n(64, 80, 80, 4) (18, 80, 80, 4) (1, 80, 80, 1)\n" ], [ "DEVICE = torch.device(\"cuda:0\")\nmodel_path = \"models/ncca_softmax_multi_traffic.pth\"\n\nCHANNEL_N = 16\nALPHA_CHANNEL = 4\n\nlr = 1e-3\nlr_gamma = 0.99995\nbetas = (0.5, 0.5)\nn_epoch = 100000\n\nBATCH_SIZE = 10\nN_STEPS = 128\nPOOL_SIZE = 256\nCELL_FIRE_RATE = 0.5\nCALIBRATION = 1.0\neps = 1e-3\n\nUSE_PATTERN_POOL = 1\nDAMAGE_N = 3\nTRANS_N = 3", "_____no_output_____" ], [ "valid_masks = []\nfor alive_map in alive_maps:\n valid_masks.append(alive_map.astype(bool))\nvalid_masks = np.concatenate(valid_masks, 0)\n\npools_list = []\nfor d_i, d_train in enumerate(d_trains):\n pools = []\n for _ in range(d_train.shape[0]):\n init_coord = get_rand_avail(valid_masks[d_i:(d_i+1)])\n seed = make_seed(map_size, CHANNEL_N, np.arange(CHANNEL_N-ALPHA_CHANNEL)+ALPHA_CHANNEL, init_coord)\n pools.append(SamplePool(x=np.repeat(seed[None, ...], POOL_SIZE, 0)))\n pools_list.append(pools)\n \nmy_model = NCCAModel2(CHANNEL_N, ALPHA_CHANNEL, CELL_FIRE_RATE, DEVICE).to(DEVICE)\nmy_model.load_state_dict(torch.load(model_path))\nfor param in my_model.parameters():\n param.requires_grad = False", "_____no_output_____" ], [ "def test(x, target, valid_mask_t, calibration_map, steps):\n history = [x.detach().cpu().numpy(),]\n for _ in range(steps):\n x = my_model(x, valid_mask_t, 1)\n h = torch.softmax(x[..., :ALPHA_CHANNEL], -1)\n t = target[..., :ALPHA_CHANNEL]\n _delta = t*(h-1)\n delta = _delta * calibration_map * CALIBRATION\n y1 = x[..., :ALPHA_CHANNEL]-delta\n \n alpha_h = x[..., ALPHA_CHANNEL:(ALPHA_CHANNEL+1)]\n y2 = alpha_h - 2 * (alpha_h-valid_mask_t) * calibration_map * CALIBRATION\n x = torch.cat((y1,y2,x[..., (ALPHA_CHANNEL+1):]), -1)\n history.append(x.detach().cpu().numpy())\n return x, history\n \n# 重新选择target\ncalibration_map = make_circle_masks(BATCH_SIZE, map_size[0], map_size[1], rmin=0.5, rmax=0.5)[..., None]\ncalibration_map = torch.from_numpy(calibration_map.astype(np.float32)).to(DEVICE)\n\ntargets = []\n# pre_target_i = [-1]*10\npre_target_i = [1, 122, 7, 6, 19, 27, 19, 11, 22, 6]\ntarget_is = []\nfor d_i in range(10):\n if pre_target_i[d_i]<0:\n target_i = np.random.randint(d_tests[d_i].shape[0])\n else:\n target_i = pre_target_i[d_i]\n print(target_i)\n target_is.append((d_i, target_i))\n target = np.concatenate((d_tests[d_i][target_i:target_i+1], valid_masks[d_i:(d_i+1)]), -1)\n targets.append(target)\ntargets = np.concatenate(targets, 0).astype(np.float32)\ntargets[..., :-1] += eps\ntargets[..., :-1] /= np.sum(targets[..., :-1], axis=-1, keepdims=True)\n_target = torch.from_numpy(targets).to(DEVICE)\n\nx0 = np.repeat(seed[None, ...], BATCH_SIZE, 0)*0\nx0 = torch.from_numpy(x0.astype(np.float32)).to(DEVICE)\n\nvalid_mask_t = valid_masks[[tmp[0] for tmp in target_is]]\nvalid_mask_t = torch.from_numpy(valid_mask_t.astype(np.float32)).to(DEVICE)\n\nx, history = test(x0, _target, valid_mask_t, calibration_map, N_STEPS)\n# x.backward()\nhistory = np.array(history)\ncali_map_numpy = calibration_map.detach().cpu().numpy()\nprint(\"history generated\", history.shape)", "1\n122\n7\n6\n19\n27\n19\n11\n22\n6\nhistory generated (129, 10, 80, 80, 16)\n" ], [ "color_map = [(0.0,0.0,0.0),\n (0.5,1.0,0.5),\n (1.0,1.0,0.5),\n (1.0,0.7,0.2),\n (1.0,0.5,0.5)]\n\nfor history_i in range(10):\n history_t = history[:,history_i,...,:(ALPHA_CHANNEL+1)]\n targets_t = targets[history_i,...]\n\n map_dict = np.argmax(targets_t[..., :-1], -1)\n _map = np.zeros([map_dict.shape[0], map_dict.shape[1], 3])\n for i in range(_map.shape[0]):\n for j in range(_map.shape[1]):\n if targets_t[i,j,-1]>0.1:\n _map[i,j] = color_map[map_dict[i,j]+1]\n\n plt.figure(figsize=(14,6))\n plt.subplot(1,9,1)\n rotated_img = ndimage.rotate(_map*0.999+0.00001, 90)\n plt.imshow(rotated_img)\n if history_i==0: plt.gca().set_title(\"Target\")\n plt.axis('off')\n plt.subplot(1,9,2)\n rotated_img = ndimage.rotate(cali_map_numpy[history_i, ..., 0], 90)\n plt.imshow(rotated_img, cmap=plt.cm.gray, vmin=0, vmax=1)\n if history_i==0: plt.gca().set_title(\"Pre-explored\")\n plt.axis('off')\n \n shown_steps = [2,4,8,16,32,64,128]\n for index, i_map in enumerate(shown_steps):\n plt.subplot(1,9,index+3)\n i_map-=1\n\n map_dict = np.argmax(history_t[i_map, ..., :-1], -1)\n _map = np.zeros([map_dict.shape[0], map_dict.shape[1], 3])\n for i in range(_map.shape[0]):\n for j in range(_map.shape[1]):\n if history_t[i_map,i,j,-1]>0.1:\n _map[i,j] = color_map[map_dict[i,j]+1]\n\n rotated_img = ndimage.rotate(_map*0.999+0.00001, 90)\n plt.imshow(rotated_img)\n if history_i==0: plt.gca().set_title('Step #'+str(i_map+1))\n plt.axis('off')\n plt.show()", "_____no_output_____" ], [ "percentages = []\nfor _ in range(10):\n for d_i in range(10):\n targets = []\n target_is = []\n for target_i in range(d_tests[d_i].shape[0]):\n target_is.append((d_i, target_i))\n target = np.concatenate((d_tests[d_i][target_i:target_i+1], valid_masks[d_i:(d_i+1)]), -1)\n targets.append(target)\n targets = np.concatenate(targets, 0).astype(np.float32)\n _target = torch.from_numpy(targets).to(DEVICE)\n \n calibration_map = make_circle_masks(_target.size(0), map_size[0], map_size[1],\n rmin=0.5, rmax=0.5)[..., None]\n calibration_map = torch.from_numpy(calibration_map.astype(np.float32)).to(DEVICE)\n x0 = np.repeat(seed[None, ...], _target.size(0), 0)*0\n x0 = torch.from_numpy(x0.astype(np.float32)).to(DEVICE)\n\n valid_mask_t = valid_masks[[tmp[0] for tmp in target_is]]\n valid_mask_t = torch.from_numpy(valid_mask_t.astype(np.float32)).to(DEVICE)\n\n x, history = test(x0, _target, valid_mask_t, calibration_map, N_STEPS)\n \n hyp = x.detach().cpu().numpy()\n hyp = np.argmax(hyp[..., :(ALPHA_CHANNEL+1)], -1)\n y = np.argmax(targets, -1)\n percentage = np.sum((hyp==y)*alive_maps[d_i][...,0])/(np.sum(alive_maps[d_i])*hyp.shape[0])\n percentages.append(percentage)\n print(percentage)\nprint(\"---------\")\nprint(np.mean(percentages))", "0.6557621502209131\n0.7531524208975867\n0.7594786015672091\n0.7239869964007895\n0.6608549288617886\n0.803313364624506\n0.7698341994555803\n0.8309084126095878\n0.8490995998221432\n0.8345033486645687\n0.637711708394698\n0.7593831673297827\n0.7592224231464738\n0.7202542668059909\n0.6776549796747967\n0.7906913908102767\n0.7610987379361545\n0.8126748740906548\n0.8429116644434563\n0.8495521665456306\n0.6487297496318115\n0.7521038592457541\n0.7658981314044605\n0.7331185417392314\n0.6732405995934959\n0.7840059906126482\n0.7914625092798813\n0.8107162842753217\n0.8525270490588409\n0.8260308238521746\n0.6477724594992637\n0.7522655218558868\n0.757157926461724\n0.7086845466155811\n0.6781631097560976\n0.8037302371541502\n0.7849913387775304\n0.8409811602313001\n0.8408366681488069\n0.8326474622770919\n0.6371686303387334\n0.7546500453553434\n0.776130198915009\n0.7314582607686057\n0.667603531504065\n0.7841681077075099\n0.7678173719376392\n0.8002704719268793\n0.8513598636431007\n0.8260509965302993\n0.6548969072164949\n0.746346874073808\n0.7728149487643159\n0.7198827353999768\n0.6712874745934959\n0.7975543478260869\n0.8025736203909923\n0.80325498974072\n0.8520082999851786\n0.8442669248769467\n0.6425810014727541\n0.7474201341799664\n0.7632459312839059\n0.7100371531406015\n0.6613471798780488\n0.8129245923913043\n0.7700569166048008\n0.8149132624510352\n0.8533792796798577\n0.8254458161865569\n0.6292893961708395\n0.749337632361262\n0.7579716696805304\n0.7195402298850575\n0.6765275660569106\n0.8148005187747036\n0.8039594159861421\n0.8467636634956165\n0.8622165406847487\n0.8255265068990559\n0.6346649484536082\n0.7490547227935299\n0.7816455696202531\n0.7287008011145942\n0.6670954014227642\n0.7853415266798419\n0.7579683246721108\n0.8168718522663683\n0.8548984733955832\n0.8229444040990882\n0.6347938144329897\n0.7503861940130947\n0.7637733574442436\n0.7351793800069663\n0.668286331300813\n0.7906836709486166\n0.7706632021776788\n0.8242865137101287\n0.8497480361642211\n0.8360566448801743\n---------\n0.763090055501292\n" ], [ "percentages = []\nfor _ in range(10):\n for d_i in range(10):\n targets = []\n target_is = []\n for target_i in range(d_trains[d_i].shape[0]):\n target_is.append((d_i, target_i))\n target = np.concatenate((d_trains[d_i][target_i:target_i+1], valid_masks[d_i:(d_i+1)]), -1)\n targets.append(target)\n targets = np.concatenate(targets, 0).astype(np.float32)\n _target = torch.from_numpy(targets).to(DEVICE)\n \n calibration_map = make_circle_masks(_target.size(0), map_size[0], map_size[1],\n rmin=0.5, rmax=0.5)[..., None]\n calibration_map = torch.from_numpy(calibration_map.astype(np.float32)).to(DEVICE)\n\n x0 = np.repeat(seed[None, ...], _target.size(0), 0)*0\n x0 = torch.from_numpy(x0.astype(np.float32)).to(DEVICE)\n\n valid_mask_t = valid_masks[[tmp[0] for tmp in target_is]]\n valid_mask_t = torch.from_numpy(valid_mask_t.astype(np.float32)).to(DEVICE)\n\n x, history = test(x0, _target, valid_mask_t, calibration_map, N_STEPS)\n \n hyp = x.detach().cpu().numpy()\n hyp = np.argmax(hyp[..., :6], -1)\n y = np.argmax(targets, -1)\n percentage = np.sum((hyp==y)*alive_maps[d_i][...,0])/(np.sum(alive_maps[d_i])*hyp.shape[0])\n percentages.append(percentage)\n print(percentage)\nprint(\"---------\")\nprint(np.mean(percentages))", "0.6239449097938145\n0.7395206984273821\n0.7449635510849909\n0.6936981162196679\n0.6618711890243902\n0.7784322504940712\n0.705517469376392\n0.7859103945159485\n0.8766821712427746\n0.7884837962962963\n0.6338635631443299\n0.7393472479185939\n0.7378503616636528\n0.6970266602809706\n0.6633161839430894\n0.7776718441205533\n0.71118109688196\n0.7701804700615557\n0.8788046423410405\n0.7903560729847494\n0.62277706185567\n0.7383137719703978\n0.7385637997287523\n0.6968590357598978\n0.6614662728658537\n0.7817788105237155\n0.7356973830734966\n0.7790203553441523\n0.8808819544797688\n0.7855108478576616\n0.6176183956185567\n0.7480559088806661\n0.7359643422242315\n0.70242656449553\n0.6624587144308943\n0.7794396924407114\n0.7270409938752784\n0.7764234750979295\n0.8741306900289018\n0.7857888525780683\n0.6394249355670103\n0.7480197733580018\n0.7491735420433996\n0.6860791826309067\n0.6590447154471545\n0.7800997406126482\n0.7274063891982183\n0.7749720201454953\n0.880915823699422\n0.7866455610021786\n0.6187822164948453\n0.739751965772433\n0.7371581148282098\n0.7043702107279693\n0.6641577743902439\n0.7760545331027668\n0.7059437639198218\n0.7816522104085059\n0.8817851336705202\n0.7867930737109659\n0.6273236146907216\n0.7434233348751156\n0.7333154385171791\n0.6878591954022989\n0.6558292047764228\n0.781940927618577\n0.7243701280623608\n0.785429490766648\n0.8803061777456648\n0.7885121641249092\n0.6314030283505154\n0.7412118408880666\n0.738938178119349\n0.6993095466155811\n0.6622205284552846\n0.7795053112648221\n0.7059089643652561\n0.7699531337437046\n0.8806110007225434\n0.7922340232389252\n0.6231636597938144\n0.740069958371878\n0.7362257007233273\n0.6951149425287356\n0.6629906631097561\n0.7738080533596838\n0.7533842566815144\n0.7854906966983771\n0.880227149566474\n0.7922737381989833\n0.630114368556701\n0.7404385407030527\n0.7395315325497287\n0.6917784163473819\n0.663427337398374\n0.7745838994565217\n0.7389598413140311\n0.7876329043088975\n0.8832189306358381\n0.7880752995642701\n---------\n0.7414718541588834\n" ] ], [ [ "## Speed Test", "_____no_output_____" ] ], [ [ "times = []\nvalid_mask_t = torch.from_numpy(np.ones([1,80,80,1]).astype(np.float32)).to(DEVICE)\nfor d_i in range(10):\n _target = torch.from_numpy(d_trains[d_i].astype(np.float32)).to(DEVICE)\n\n calibration_map = make_circle_masks(_target.size(0), map_size[0], map_size[1],\n rmin=0.5, rmax=0.5)[..., None]\n calibration_map = torch.from_numpy(calibration_map.astype(np.float32)).to(DEVICE)\n\n x0 = np.repeat(seed[None, ...], _target.size(0), 0)*0\n x0 = torch.from_numpy(x0.astype(np.float32)).to(DEVICE)\n\n start_time = time.time()\n x, history = test(x0, _target, valid_mask_t, calibration_map, N_STEPS)\n times.append((time.time()-start_time)/_target.size(0))\n print(times[-1])\nprint(\"---------\")\nprint(np.mean(times))", "0.03891327977180481\n0.04005876183509827\n0.04132132604718208\n0.04402513429522514\n0.04346586391329765\n0.04147135838866234\n0.03921307995915413\n0.038483794778585434\n0.04098214581608772\n0.044003926217556\n---------\n0.04119386710226536\n" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "Fit interpretable models to the training set and test on validation sets.", "_____no_output_____" ] ], [ [ "#%matplotlib inline\n#%load_ext autoreload\n#%autoreload 2\n\nimport os\nimport pickle as pkl\nfrom os.path import join as oj\nimport numpy as np\n\nimport matplotlib.pyplot as plt\nfrom sklearn import metrics\nfrom sklearn.tree import DecisionTreeClassifier, plot_tree\nfrom sklearn.feature_selection import RFE\nfrom sklearn.linear_model import LogisticRegression\nfrom sklearn.ensemble import AdaBoostClassifier\n\nimport imodels\nfrom rulevetting.api import validation\nfrom rulevetting.projects.csi_pecarn.dataset import Dataset\n\nMODELS_DIR = './models'\nos.makedirs(MODELS_DIR, exist_ok=True)\n\noutcome_def = 'outcome' # output", "/Users/seunghoonpaik/Desktop/SH/Berkeley/Coursework/215A/Lab/final-proj/andy-github/rule-env/lib/python3.8/site-packages/redis/connection.py:72: UserWarning: redis-py works best with hiredis. Please consider installing\n warnings.warn(msg)\n" ], [ "def var_selection(df,method=['rfe',10]): ## input: a dataframe with outcome as the last column, method: ['rfe',number of\n ## features to choose] or ['lasso',penalty] output: a dataframe containing the columns we select and the outcome column\n algo=method[0]\n param=method[1]\n X=df.drop(columns=['outcome'])\n y=df.outcome\n if algo=='rfe':\n mymodel=LogisticRegression()\n myrfe = RFE(mymodel,n_features_to_select=param)\n myfit = myrfe.fit(X, y)\n index=np.append(myfit.support_,True)\n \n elif algo=='lasso':\n mylasso = LogisticRegression(penalty='l1', solver='liblinear',C=param) ## for example C=0.1\n myfit=mylasso.fit(X, y)\n \n index=np.append(myfit.coef_[0]!=0,True)\n \n \n return index\n ", "_____no_output_____" ], [ "df_train, df_tune, _ = Dataset().get_data(load_csvs=True)", "_____no_output_____" ], [ "def predict_and_save(model, model_name='decision_tree'):\n '''Plots cv and returns cv, saves all stats\n '''\n results = {'model': model}\n for x, y, suffix in zip([X_train, X_tune],\n [y_train, y_tune],\n ['_train', '_tune']):\n \n stats, threshes = validation.all_stats_curve(y, model.predict_proba(x)[:, 1],\n plot=suffix == '_tune')\n for stat in stats.keys():\n results[stat + suffix] = stats[stat]\n results['threshes' + suffix] = threshes\n pkl.dump(results, open(oj(MODELS_DIR, model_name + '.pkl'), 'wb'))\n return stats, threshes\n\n\n\ndef model_valid(max_num=20,model_name='decision_tree'):\n '''use validation set to select # of features'''\n \n record=np.zeros(max_num)\n sensitivity=np.zeros(max_num)\n for num in range(1,max_num+1):\n index=var_selection(df_train,method=['rfe',num])\n loc_train=df_train.loc[:,index]\n loc_tune=df_tune.loc[:,index]\n loc_=_.loc[:,index]\n X_train = loc_train.drop(columns=outcome_def)\n y_train = loc_train[outcome_def].values\n X_tune = loc_tune.drop(columns=outcome_def)\n y_tune = loc_tune[outcome_def].values\n \n if model_name=='decision_tree':\n model = DecisionTreeClassifier(max_depth=4, class_weight={0: 1, 1: 1e3})\n model.fit(X_train, y_train)\n elif model_name=='logistic':\n model= LogisticRegression()\n model.fit(X_train, y_train)\n elif model_name=='adaboost':\n model= AdaBoostClassifier(n_estimators=50, learning_rate=1)\n model.fit(X_train, y_train)\n \n \n \n stats, threshes = validation.all_stats_curve(y_tune, model.predict_proba(X_tune)[:, 1],\n plot=False)\n sens=stats['sens']\n spec=stats['spec']\n if sens[0]<0.98:\n record[num-1]=0.\n sensitivity[num-1]=sens[0]\n continue\n j=0\n while sens[j]>0.98:\n #print([j, sens[j]], spec[j])\n #print(sens[j])\n cur_pec=spec[j]\n \n j+=1\n record[num-1]=cur_pec\n sensitivity[num-1]=sens[j]\n \n print(record)\n print(sensitivity)\n return np.argmax(record)+1 ## output the optimal number of features via validation", "_____no_output_____" ], [ "# print(model_valid(20,model_name='adaboost')) ## output zero when sens<.98, otherwise output spec (adaboost,decision_tree,logistic)", "_____no_output_____" ], [ "# print(model_valid(20,model_name='decision_tree')) ## output zero when sens<.98, otherwise output spec (adaboost,decision_tree,logistic)", "_____no_output_____" ], [ "# print(model_valid(30,model_name='logistic')) ## output zero when sens<.98, otherwise output spec (adaboost,decision_tree,logistic)", "_____no_output_____" ], [ "index=var_selection(df_train,method=['rfe',9])\n\nprint(df_train.columns[index])\n\ndf_train=df_train.loc[:,index]\ndf_tune=df_tune.loc[:,index]\n_=_.loc[:,index]\n\nX_train = df_train.drop(columns=outcome_def)\ny_train = df_train[outcome_def].values\nX_tune = df_tune.drop(columns=outcome_def)\ny_tune = df_tune[outcome_def].values\n\nprocessed_feats = df_train.keys().values.tolist()\nfeature_names=processed_feats", "Index(['ArrPtIntub', 'DxCspineInjury', 'FocalNeuroFindings', 'HighriskDiving',\n 'IntervForCervicalStab', 'PtExtremityWeakness', 'PtSensoryLoss',\n 'PtTenderExt', 'SubInj_TorsoTrunk', 'outcome'],\n dtype='object')\n" ] ], [ [ "# fit simple models", "_____no_output_____" ], [ "**decision tree**", "_____no_output_____" ] ], [ [ "# fit decision tree\ndt = DecisionTreeClassifier(max_depth=4, class_weight={0: 1, 1: 1e3})\ndt.fit(X_train, y_train)\n\nstats, threshes = predict_and_save(dt, model_name='decision_tree')\nprint(stats,threshes)\n\nplt.show()\nplt.savefig(\"tree-roc.png\", dpi='figure', format=None, metadata=None,\n bbox_inches=None, pad_inches=0,\n facecolor='auto', edgecolor='auto',\n backend=None)", "100%|███████████████████████████████████████████████████| 14/14 [00:00<00:00, 1871.38it/s]\n100%|███████████████████████████████████████████████████| 14/14 [00:00<00:00, 2638.52it/s]" ], [ "fig = plt.figure(figsize=(50, 40))\nplot_tree(dt, feature_names=feature_names, filled=True)\nplt.show()", "_____no_output_____" ], [ "# fit logitstic\ndt= LogisticRegression()\ndt.fit(X_train, y_train)\n\nstats_lr, threshes_lr = predict_and_save(dt, model_name='logistic')\nprint(stats_lr, \"\\n\")\nprint(threshes_lr)\n\nplt.show()\nfig = plt.figure(figsize=(50, 40))\nplt.show()", "100%|███████████████████████████████████████████████████| 74/74 [00:00<00:00, 2963.78it/s]\n100%|███████████████████████████████████████████████████| 48/48 [00:00<00:00, 3448.55it/s]" ], [ "# fit adaboost\ndt= AdaBoostClassifier(n_estimators=100, learning_rate=1)\ndt.fit(X_train, y_train)\n\nstats_ab, threshes_ab = predict_and_save(dt, model_name='adaboost')\nprint(stats_ab, \"\\n\")\nprint(threshes_ab)\n\nplt.show()\nfig = plt.figure(figsize=(50, 40))\nplt.show()", "100%|███████████████████████████████████████████████████| 74/74 [00:00<00:00, 3015.37it/s]\n100%|███████████████████████████████████████████████████| 48/48 [00:00<00:00, 3513.92it/s]" ], [ "(np.asarray(stats_lr[\"sens\"]) - np.asarray(stats_ab[\"sens\"])) * 1000", "_____no_output_____" ] ], [ [ "**bayesian rule list (this one is slow)**", "_____no_output_____" ] ], [ [ "np.random.seed(13)\n# train classifier (allow more iterations for better accuracy; use BigDataRuleListClassifier for large datasets)\nprint('training bayesian_rule_list...')\nbrl = imodels.BayesianRuleListClassifier(listlengthprior=2, max_iter=10000, class1label=\"IwI\", verbose=False)\nbrl.fit(X_train, y_train, feature_names=feature_names)\nstats, threshes = predict_and_save(brl, model_name='bayesian_rule_list')\nprint(brl)", "training bayesian_rule_list...\n" ], [ "print(brl)", "Trained RuleListClassifier for detecting IwI\n=============================================\nIF IntervForCervicalStab > 0.5 THEN probability of IwI: 59.3% (54.8%-63.8%)\nELSE IF FocalNeuroFindings > 0.5 THEN probability of IwI: 15.8% (9.7%-23.0%)\nELSE IF DxCspineInjury > 0.5 THEN probability of IwI: 10.0% (5.1%-16.2%)\nELSE probability of IwI: 1.3% (0.8%-2.0%)\n============================================\n\n" ] ], [ [ "**rulefit**", "_____no_output_____" ] ], [ [ "# fit a rulefit model\nnp.random.seed(13)\nrulefit = imodels.RuleFitRegressor(max_rules=4)\nrulefit.fit(X_train, y_train, feature_names=feature_names)\n\n# preds = rulefit.predict(X_test)\nstats, threshes = predict_and_save(rulefit, model_name='rulefit')\n'''\ndef print_best(sens, spec):\n idxs = np.array(sens) > 0.9\n print(np.array(sens)[idxs], np.array(spec)[idxs])\nprint_best(sens, spec)\n'''", "100%|█████████████████████████████████████████████████████| 7/7 [00:00<00:00, 2259.51it/s]\n100%|█████████████████████████████████████████████████████| 7/7 [00:00<00:00, 1988.09it/s]\n" ], [ "# pd.reset_option('display.max_colwidth')\nrulefit.visualize()", "_____no_output_____" ] ], [ [ "**greedy (CART) rule list**", "_____no_output_____" ] ], [ [ "class_weight = {0: 1, 1: 100}\nd = imodels.GreedyRuleListClassifier(max_depth=9, class_weight=class_weight, criterion='neg_corr')\nd.fit(X_train, y_train, feature_names=feature_names, verbose=False)\nstats, threshes = predict_and_save(d, model_name='grl')\n# d.print_list()\nprint(d)", "/Users/seunghoonpaik/Desktop/SH/Berkeley/Coursework/215A/Lab/final-proj/andy-github/rule-env/lib/python3.8/site-packages/numpy/lib/function_base.py:2691: RuntimeWarning: invalid value encountered in true_divide\n c /= stddev[:, None]\n/Users/seunghoonpaik/Desktop/SH/Berkeley/Coursework/215A/Lab/final-proj/andy-github/rule-env/lib/python3.8/site-packages/numpy/lib/function_base.py:2692: RuntimeWarning: invalid value encountered in true_divide\n c /= stddev[None, :]\n100%|█████████████████████████████████████████████████████| 7/7 [00:00<00:00, 2065.43it/s]\n100%|█████████████████████████████████████████████████████| 6/6 [00:00<00:00, 1816.90it/s]" ] ], [ [ "**rf**", "_____no_output_____" ], [ "# look at all the results", "_____no_output_____" ] ], [ [ "def plot_metrics(suffix, title=None, fs=15):\n for fname in sorted(os.listdir(MODELS_DIR)):\n if 'pkl' in fname:\n if not fname[:-4] == 'rf':\n r = pkl.load(open(oj(MODELS_DIR, fname), 'rb'))\n # print(r)\n # print(r.keys())\n\n threshes = np.array(r['threshes' + suffix])\n sens = np.array(r['sens' + suffix])\n spec = np.array(r['spec' + suffix])\n plt.plot(100 * sens, 100 * spec, 'o-', label=fname[:-4], alpha=0.6, markersize=3)\n plt.xlabel('Sensitivity (%)', fontsize=fs)\n plt.ylabel('Specificity (%)', fontsize=fs)\n s = suffix[1:]\n if title is None:\n plt.title(f'{s}\\n{data_sizes[s][0]} IAI-I / {data_sizes[s][1]}')\n else:\n plt.title(title, fontsize=fs)\n\n # print best results\n if suffix == '_test2':\n idxs = (sens > 0.95) & (spec > 0.43)\n if np.sum(idxs) > 0:\n idx_max = np.argmax(spec[idxs])\n print(fname, f'{100 * sens[idxs][idx_max]:0.2f} {100 * spec[idxs][idx_max]:0.2f}')\n\n if suffix == '_test2':\n plt.plot(96.77, 43.98, 'o', color='black', label='Original CDR', ms=4)\n else:\n plt.plot(97.0, 42.5, 'o', color='black', label='Original CDR', ms=4)\n plt.grid()\n\n\nsuffixes = ['_train', '_tune'] # _train, _test1, _test2, _cv\ntitles = ['Train (PECARN)', 'Tune (PECARN)']\nR, C = 1, len(suffixes)\nplt.figure(dpi=200, figsize=(C * 2.5, R * 3), facecolor='w')\nfs = 10\nfor i, suffix in enumerate(suffixes):\n ax = plt.subplot(R, C, i + 1)\n plot_metrics(suffix, title=titles[i], fs=fs)\n if i > 0:\n plt.ylabel('')\n plt.yticks([0, 25, 50, 75, 100], labels=[''] * 5)\n # ax.yaxis.set_visible(False)\n plt.xlim((50, 101))\n plt.ylim((0, 101))\nplt.tight_layout()\n# plt.subplot(R, C, 1)\n# plt.legend(fontsize=20)\nplt.legend(bbox_to_anchor=(1.1, 1), fontsize=fs, frameon=False)\n# plt.savefig('figs/metrics_3_splits')\nplt.show()", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "**This notebook is an exercise in the [Introduction to Machine Learning](https://www.kaggle.com/learn/intro-to-machine-learning) course. You can reference the tutorial at [this link](https://www.kaggle.com/dansbecker/underfitting-and-overfitting).**\n\n---\n", "_____no_output_____" ], [ "## Recap\nYou've built your first model, and now it's time to optimize the size of the tree to make better predictions. Run this cell to set up your coding environment where the previous step left off.", "_____no_output_____" ] ], [ [ "# Code you have previously used to load data\nimport pandas as pd\nfrom sklearn.metrics import mean_absolute_error\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.tree import DecisionTreeRegressor\n\n\n# Path of the file to read\niowa_file_path = '../input/home-data-for-ml-course/train.csv'\n\nhome_data = pd.read_csv(iowa_file_path)\n# Create target object and call it y\ny = home_data.SalePrice\n# Create X\nfeatures = ['LotArea', 'YearBuilt', '1stFlrSF', '2ndFlrSF', 'FullBath', 'BedroomAbvGr', 'TotRmsAbvGrd']\nX = home_data[features]\n\n# Split into validation and training data\ntrain_X, val_X, train_y, val_y = train_test_split(X, y, random_state=1)\n\n# Specify Model\niowa_model = DecisionTreeRegressor(random_state=1)\n# Fit Model\niowa_model.fit(train_X, train_y)\n\n# Make validation predictions and calculate mean absolute error\nval_predictions = iowa_model.predict(val_X)\nval_mae = mean_absolute_error(val_predictions, val_y)\nprint(\"Validation MAE: {:,.0f}\".format(val_mae))\n\n# Set up code checking\nfrom learntools.core import binder\nbinder.bind(globals())\nfrom learntools.machine_learning.ex5 import *\nprint(\"\\nSetup complete\")", "_____no_output_____" ] ], [ [ "# Exercises\nYou could write the function `get_mae` yourself. For now, we'll supply it. This is the same function you read about in the previous lesson. Just run the cell below.", "_____no_output_____" ] ], [ [ "def get_mae(max_leaf_nodes, train_X, val_X, train_y, val_y):\n model = DecisionTreeRegressor(max_leaf_nodes=max_leaf_nodes, random_state=0)\n model.fit(train_X, train_y)\n preds_val = model.predict(val_X)\n mae = mean_absolute_error(val_y, preds_val)\n return(mae)", "_____no_output_____" ] ], [ [ "## Step 1: Compare Different Tree Sizes\nWrite a loop that tries the following values for *max_leaf_nodes* from a set of possible values.\n\nCall the *get_mae* function on each value of max_leaf_nodes. Store the output in some way that allows you to select the value of `max_leaf_nodes` that gives the most accurate model on your data.", "_____no_output_____" ] ], [ [ "candidate_max_leaf_nodes = [5, 25, 50, 100, 250, 500]\n# Write loop to find the ideal tree size from candidate_max_leaf_nodes\nresults = []\nfor max_leaf_nodes in candidate_max_leaf_nodes:\n my_mae = get_mae(max_leaf_nodes, train_X, val_X, train_y, val_y)\n print(\"Max leaf nodes: %d \\t\\t Mean Absolute Error: %d\" %(max_leaf_nodes, my_mae))\n results.append(my_mae)\nbest_value = min(results) \n\n# Store the best value of max_leaf_nodes (it will be either 5, 25, 50, 100, 250 or 500)\nbest_tree_size = candidate_max_leaf_nodes[results.index(best_value)]\n\n# Check your answer\nstep_1.check()", "_____no_output_____" ], [ "# The lines below will show you a hint or the solution.\n# step_1.hint() \n# step_1.solution()", "_____no_output_____" ] ], [ [ "## Step 2: Fit Model Using All Data\nYou know the best tree size. If you were going to deploy this model in practice, you would make it even more accurate by using all of the data and keeping that tree size. That is, you don't need to hold out the validation data now that you've made all your modeling decisions.", "_____no_output_____" ] ], [ [ "# Fill in argument to make optimal size and uncomment\nfinal_model = DecisionTreeRegressor(max_leaf_nodes=best_tree_size, random_state=1)\n\n# fit the final model and uncomment the next two lines\nfinal_model.fit(X, y)\n\n# Check your answer\nstep_2.check()", "_____no_output_____" ], [ "step_2.hint()\nstep_2.solution()", "_____no_output_____" ] ], [ [ "You've tuned this model and improved your results. But we are still using Decision Tree models, which are not very sophisticated by modern machine learning standards. In the next step you will learn to use Random Forests to improve your models even more.\n\n# Keep Going\n\nYou are ready for **[Random Forests](https://www.kaggle.com/dansbecker/random-forests).**\n", "_____no_output_____" ], [ "---\n\n\n\n\n*Have questions or comments? Visit the [course discussion forum](https://www.kaggle.com/learn/intro-to-machine-learning/discussion) to chat with other learners.*", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]

[ "MIT" ]

[ [ [ "%load_ext autoreload\n%autoreload 2", "_____no_output_____" ], [ "import os\nos.chdir(\"..\")", "_____no_output_____" ], [ "from deepsvg.svglib.geom import Point\nfrom deepsvg.svglib.svg import SVG\nfrom deepsvg.svglib.svg_path import SVGPath\nfrom deepsvg.svglib.utils import to_gif\n\nfrom deepsvg.difflib.tensor import SVGTensor\nfrom deepsvg.difflib.utils import *\nfrom deepsvg.difflib.loss import *", "_____no_output_____" ], [ "import torch.optim as optim\nimport IPython.display as ipd\nfrom moviepy.editor import ImageClip, concatenate_videoclips, ipython_display", "_____no_output_____" ], [ "import pickle\nimport csv", "_____no_output_____" ] ], [ [ "# Differentiable SVGTensor optimization", "_____no_output_____" ], [ "Load a target SVG and apply the standard pre-processing.", "_____no_output_____" ] ], [ [ "svg = SVG.load_svg(\"docs/imgs/dolphin.svg\").normalize().zoom(0.9).canonicalize().simplify_heuristic()", "simplify\n" ] ], [ [ "Convert the SVG to the differentiable SVGTensor data-structure.", "_____no_output_____" ] ], [ [ "svg_target = SVGTensor.from_data(svg.to_tensor())", "_____no_output_____" ], [ "p_target = svg_target.sample_points()\nplot_points(p_target, show_color=True)", "_____no_output_____" ] ], [ [ "Create an arbitrary SVG whose Bézier parameters will be optimized to match the target shape.", "_____no_output_____" ] ], [ [ "circle = SVG.unit_circle().normalize().zoom(0.9).split(8) # split: 1/2/4/8\nsvg_pred = SVGTensor.from_data(circle.to_tensor())", "_____no_output_____" ] ], [ [ "SVGTensor enables to sample points in a differentiable way, so that the loss that will be backpropagated down to the SVG Bézier parameters.", "_____no_output_____" ] ], [ [ "p_pred = svg_pred.sample_points()\nplot_points(p_pred, show_color=True)", "_____no_output_____" ], [ "svg_pred.control1.requires_grad_(True)\nsvg_pred.control2.requires_grad_(True)\nsvg_pred.end_pos.requires_grad_(True);", "_____no_output_____" ], [ "optimizer = optim.Adam([svg_pred.control1, svg_pred.control2, svg_pred.end_pos], lr=0.1)", "_____no_output_____" ] ], [ [ "Write a standard gradient descent algorithm and observe the step-by-step optimization!", "_____no_output_____" ] ], [ [ "img_list = []\n\nfor i in range(150):\n optimizer.zero_grad()\n\n p_pred = svg_pred.sample_points()\n l = svg_emd_loss(p_pred, p_target)\n l.backward()\n optimizer.step()\n \n if i % 4 == 0:\n img = svg_pred.draw(with_points=True, do_display=False, return_png=True)\n img_list.append(img)\n \nto_gif(img_list)", "_____no_output_____" ], [ "svg = SVG.load_svg(\"docs/imgs/dolphin.svg\")\nprint(svg)\nsvg_tensor = svg.to_tensor()\nprint(svg_tensor)", "SVG[Bbox(0.0 0.0 294.8680114746094 294.8680114746094)](\n\tSVGPathGroup(SVGPath(M[P(0.0, 0.0), P(284.3949890136719, 115.12999725341797)] C[P(284.3949890136719, 115.12999725341797), P(280.5419921875, 119.21499633789062), P(274.864990234375, 119.21499633789062), P(272.9989929199219, 119.21499633789062)] C[P(272.9989929199219, 119.21499633789062), P(269.53900146484375, 119.21499633789062), P(265.260986328125, 118.64199829101562), P(259.1309814453125, 117.35599517822266)] C[P(259.1309814453125, 117.35599517822266), P(254.31597900390625, 116.34599304199219), P(250.16998291015625, 115.33899688720703), P(246.51397705078125, 114.45199584960938)] C[P(246.51397705078125, 114.45199584960938), P(239.3219757080078, 112.70499420166016), P(234.1239776611328, 111.4419937133789), P(229.23097229003906, 111.4419937133789)] C[P(229.23097229003906, 111.4419937133789), P(226.4729766845703, 111.4419937133789), P(221.44097900390625, 112.63699340820312), P(216.11196899414062, 113.90299224853516)] C[P(216.11196899414062, 113.90299224853516), P(207.81497192382812, 115.87399291992188), P(197.48997497558594, 118.32598876953125), P(187.5029754638672, 118.32598876953125)] C[P(187.5029754638672, 118.32598876953125), P(177.41998291015625, 118.32598876953125), P(166.80697631835938, 117.5419921875), P(160.11997985839844, 116.9369888305664)] C[P(160.11997985839844, 116.9369888305664), P(157.89898681640625, 125.26298522949219), P(152.61497497558594, 138.99298095703125), P(140.6209716796875, 148.82098388671875)] C[P(140.6209716796875, 148.82098388671875), P(124.31397247314453, 162.1809844970703), P(112.22396850585938, 168.1409912109375), P(111.71697235107422, 168.3879852294922)] C[P(111.71697235107422, 168.3879852294922), P(108.6199722290039, 169.89797973632812), P(104.8959732055664, 169.1069793701172), P(102.68296813964844, 166.4659881591797)] C[P(102.68296813964844, 166.4659881591797), P(100.469970703125, 163.82699584960938), P(100.33796691894531, 160.02098083496094), P(102.36396789550781, 157.23599243164062)] C[P(102.36396789550781, 157.23599243164062), P(102.43096923828125, 157.1409912109375), P(110.58097076416016, 145.6199951171875), P(110.2679672241211, 132.6719970703125)] C[P(110.2679672241211, 132.6719970703125), P(110.14396667480469, 127.52099609375), P(109.95697021484375, 123.48500061035156), P(109.7659683227539, 120.41099548339844)] C[P(109.7659683227539, 120.41099548339844), P(99.60897064208984, 123.40499877929688), P(82.31497192382812, 130.09799194335938), P(73.39396667480469, 142.69699096679688)] C[P(73.39396667480469, 142.69699096679688), P(59.880001068115234, 161.77999877929688), P(54.33599853515625, 191.16000366210938), P(55.82099914550781, 200.79800415039062)] C[P(55.82099914550781, 200.79800415039062), P(57.367000579833984, 210.83999633789062), P(57.551998138427734, 210.83999633789062), P(62.31699752807617, 210.83999633789062)] C[P(62.31699752807617, 210.83999633789062), P(72.6989974975586, 210.83999633789062), P(81.4209976196289, 211.26499938964844), P(91.2449951171875, 216.6129913330078)] C[P(91.2449951171875, 216.6129913330078), P(99.93499755859375, 221.3419952392578), P(104.72099304199219, 226.5159912109375), P(105.23699188232422, 227.0909881591797)] C[P(105.23699188232422, 227.0909881591797), P(107.18598937988281, 229.2559814453125), P(107.70499420166016, 232.35398864746094), P(106.5679931640625, 235.0349884033203)] C[P(106.5679931640625, 235.0349884033203), P(105.43099212646484, 237.71798706054688), P(102.843994140625, 239.49899291992188), P(99.93299102783203, 239.6029815673828)] C[P(99.93299102783203, 239.6029815673828), P(97.12499237060547, 239.70797729492188), P(88.62799072265625, 240.41598510742188), P(83.1669921875, 242.39797973632812)] C[P(83.1669921875, 242.39797973632812), P(80.2669906616211, 243.4509735107422), P(77.76898956298828, 244.6829833984375), P(75.35298919677734, 245.87498474121094)] C[P(75.35298919677734, 245.87498474121094), P(71.48799133300781, 247.781982421875), P(67.83699035644531, 249.58297729492188), P(63.57798767089844, 250.0859832763672)] C[P(63.57798767089844, 250.0859832763672), P(62.192989349365234, 250.24998474121094), P(60.77098846435547, 250.32998657226562), P(59.42498779296875, 250.3049774169922)] C[P(59.42498779296875, 250.3049774169922), P(57.85498809814453, 253.9519805908203), P(55.182987213134766, 258.2989807128906), P(50.688987731933594, 261.74798583984375)] C[P(50.688987731933594, 261.74798583984375), P(47.24898910522461, 264.38897705078125), P(43.74898910522461, 266.0179748535156), P(40.660987854003906, 267.4539794921875)] C[P(40.660987854003906, 267.4539794921875), P(36.868988037109375, 269.218994140625), P(33.87298583984375, 270.61297607421875), P(31.572986602783203, 273.4649658203125)] C[P(31.572986602783203, 273.4649658203125), P(25.544986724853516, 280.9399719238281), P(23.521987915039062, 289.0449523925781), P(23.502986907958984, 289.1259765625)] C[P(23.502986907958984, 289.1259765625), P(22.71498680114746, 292.36297607421875), P(19.8809871673584, 294.708984375), P(16.55298614501953, 294.8599853515625)] C[P(16.55298614501953, 294.8599853515625), P(16.437986373901367, 294.8659973144531), P(16.321985244750977, 294.86798095703125), P(16.206985473632812, 294.86798095703125)] C[P(16.206985473632812, 294.86798095703125), P(13.015985488891602, 294.86798095703125), P(10.152985572814941, 292.8559875488281), P(9.11198616027832, 289.8119812011719)] C[P(9.11198616027832, 289.8119812011719), P(8.642986297607422, 288.4449768066406), P(4.689986228942871, 275.9159851074219), P(9.9509859085083, 257.57696533203125)] C[P(9.9509859085083, 257.57696533203125), P(11.831985473632812, 251.02096557617188), P(14.745985984802246, 245.34596252441406), P(17.562986373901367, 239.85595703125)] C[P(17.562986373901367, 239.85595703125), P(22.14698600769043, 230.9269561767578), P(26.105987548828125, 223.21595764160156), P(24.44498634338379, 213.76695251464844)] C[P(24.44498634338379, 213.76695251464844), P(16.636985778808594, 169.35595703125), P(17.175987243652344, 133.24594116210938), P(26.04598617553711, 106.44395446777344)] C[P(26.04598617553711, 106.44395446777344), P(39.8380012512207, 64.76000213623047), P(73.22599792480469, 41.53499984741211), P(79.77899932861328, 37.30400085449219)] C[P(79.77899932861328, 37.30400085449219), P(83.02999877929688, 35.202999114990234), P(85.95600128173828, 33.35499954223633), P(88.4219970703125, 31.819000244140625)] C[P(88.4219970703125, 31.819000244140625), P(86.13499450683594, 29.996999740600586), P(83.22799682617188, 28.08300018310547), P(79.7719955444336, 26.61400032043457)] C[P(79.7719955444336, 26.61400032043457), P(71.90599822998047, 23.270000457763672), P(68.5999984741211, 22.356000900268555), P(67.68399810791016, 22.13800048828125)] C[P(67.68399810791016, 22.13800048828125), P(67.625, 22.14000129699707), P(67.56800079345703, 22.14000129699707), P(67.51100158691406, 22.14000129699707)] C[P(67.51100158691406, 22.14000129699707), P(64.24600219726562, 22.141000747680664), P(61.736000061035156, 19.985000610351562), P(60.781002044677734, 16.801000595092773)] C[P(60.781002044677734, 16.801000595092773), P(59.78900146484375, 13.495000839233398), P(61.63200378417969, 9.946001052856445), P(64.6030044555664, 8.191000938415527)] C[P(64.6030044555664, 8.191000938415527), P(65.16799926757812, 7.855999946594238), P(78.68099975585938, 0.0), P(96.48999786376953, 0.0)] C[P(96.48999786376953, 0.0), P(100.16500091552734, 0.0), P(103.81599426269531, 0.3370000123977661), P(107.34099578857422, 1.0019999742507935)] C[P(107.34099578857422, 1.0019999742507935), P(118.3239974975586, 3.0749998092651367), P(126.08599853515625, 6.171999931335449), P(133.5919952392578, 9.165999412536621)] C[P(133.5919952392578, 9.165999412536621), P(140.92098999023438, 12.089999198913574), P(147.843994140625, 14.851999282836914), P(158.17999267578125, 17.02899932861328)] C[P(158.17999267578125, 17.02899932861328), P(163.20098876953125, 18.086999893188477), P(167.8249969482422, 18.902999877929688), P(172.29598999023438, 19.6929988861084)] C[P(172.29598999023438, 19.6929988861084), P(187.67898559570312, 22.408998489379883), P(200.9639892578125, 24.7549991607666), P(220.82699584960938, 34.89799880981445)] C[P(220.82699584960938, 34.89799880981445), P(246.75099182128906, 48.13800048828125), P(261.3280029296875, 62.624000549316406), P(261.8739929199219, 75.68499755859375)] C[P(261.8739929199219, 75.68499755859375), P(261.9889831542969, 78.43799591064453), P(261.9289855957031, 80.89799499511719), P(261.6929931640625, 83.05999755859375)] C[P(261.6929931640625, 83.05999755859375), P(264.1189880371094, 84.27299499511719), P(267.0669860839844, 85.80599975585938), P(270.04498291015625, 87.5)] C[P(270.04498291015625, 87.5), P(282.16796875, 94.3949966430664), P(287.2519836425781, 99.55899810791016), P(287.5909729003906, 105.3219985961914)] C[P(287.5909729003906, 105.3219985961914), P(287.82598876953125, 109.33200073242188), P(286.7510070800781, 112.63099670410156), P(284.3949890136719, 115.12999725341797)] Z[P(284.3949890136719, 115.12999725341797), P(284.3949890136719, 115.12999725341797)]))\n)\ntensor([[ 0.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 0.0000,\n 0.0000, -1.0000, -1.0000, -1.0000, -1.0000, 284.3950, 115.1300],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 284.3950,\n 115.1300, 280.5420, 119.2150, 274.8650, 119.2150, 272.9990, 119.2150],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 272.9990,\n 119.2150, 269.5390, 119.2150, 265.2610, 118.6420, 259.1310, 117.3560],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 259.1310,\n 117.3560, 254.3160, 116.3460, 250.1700, 115.3390, 246.5140, 114.4520],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 246.5140,\n 114.4520, 239.3220, 112.7050, 234.1240, 111.4420, 229.2310, 111.4420],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 229.2310,\n 111.4420, 226.4730, 111.4420, 221.4410, 112.6370, 216.1120, 113.9030],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 216.1120,\n 113.9030, 207.8150, 115.8740, 197.4900, 118.3260, 187.5030, 118.3260],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 187.5030,\n 118.3260, 177.4200, 118.3260, 166.8070, 117.5420, 160.1200, 116.9370],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 160.1200,\n 116.9370, 157.8990, 125.2630, 152.6150, 138.9930, 140.6210, 148.8210],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 140.6210,\n 148.8210, 124.3140, 162.1810, 112.2240, 168.1410, 111.7170, 168.3880],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 111.7170,\n 168.3880, 108.6200, 169.8980, 104.8960, 169.1070, 102.6830, 166.4660],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 102.6830,\n 166.4660, 100.4700, 163.8270, 100.3380, 160.0210, 102.3640, 157.2360],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 102.3640,\n 157.2360, 102.4310, 157.1410, 110.5810, 145.6200, 110.2680, 132.6720],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 110.2680,\n 132.6720, 110.1440, 127.5210, 109.9570, 123.4850, 109.7660, 120.4110],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 109.7660,\n 120.4110, 99.6090, 123.4050, 82.3150, 130.0980, 73.3940, 142.6970],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 73.3940,\n 142.6970, 59.8800, 161.7800, 54.3360, 191.1600, 55.8210, 200.7980],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 55.8210,\n 200.7980, 57.3670, 210.8400, 57.5520, 210.8400, 62.3170, 210.8400],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 62.3170,\n 210.8400, 72.6990, 210.8400, 81.4210, 211.2650, 91.2450, 216.6130],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 91.2450,\n 216.6130, 99.9350, 221.3420, 104.7210, 226.5160, 105.2370, 227.0910],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 105.2370,\n 227.0910, 107.1860, 229.2560, 107.7050, 232.3540, 106.5680, 235.0350],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 106.5680,\n 235.0350, 105.4310, 237.7180, 102.8440, 239.4990, 99.9330, 239.6030],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 99.9330,\n 239.6030, 97.1250, 239.7080, 88.6280, 240.4160, 83.1670, 242.3980],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 83.1670,\n 242.3980, 80.2670, 243.4510, 77.7690, 244.6830, 75.3530, 245.8750],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 75.3530,\n 245.8750, 71.4880, 247.7820, 67.8370, 249.5830, 63.5780, 250.0860],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 63.5780,\n 250.0860, 62.1930, 250.2500, 60.7710, 250.3300, 59.4250, 250.3050],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 59.4250,\n 250.3050, 57.8550, 253.9520, 55.1830, 258.2990, 50.6890, 261.7480],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 50.6890,\n 261.7480, 47.2490, 264.3890, 43.7490, 266.0180, 40.6610, 267.4540],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 40.6610,\n 267.4540, 36.8690, 269.2190, 33.8730, 270.6130, 31.5730, 273.4650],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 31.5730,\n 273.4650, 25.5450, 280.9400, 23.5220, 289.0450, 23.5030, 289.1260],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 23.5030,\n 289.1260, 22.7150, 292.3630, 19.8810, 294.7090, 16.5530, 294.8600],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 16.5530,\n 294.8600, 16.4380, 294.8660, 16.3220, 294.8680, 16.2070, 294.8680],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 16.2070,\n 294.8680, 13.0160, 294.8680, 10.1530, 292.8560, 9.1120, 289.8120],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 9.1120,\n 289.8120, 8.6430, 288.4450, 4.6900, 275.9160, 9.9510, 257.5770],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 9.9510,\n 257.5770, 11.8320, 251.0210, 14.7460, 245.3460, 17.5630, 239.8560],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 17.5630,\n 239.8560, 22.1470, 230.9270, 26.1060, 223.2160, 24.4450, 213.7670],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 24.4450,\n 213.7670, 16.6370, 169.3560, 17.1760, 133.2459, 26.0460, 106.4440],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 26.0460,\n 106.4440, 39.8380, 64.7600, 73.2260, 41.5350, 79.7790, 37.3040],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 79.7790,\n 37.3040, 83.0300, 35.2030, 85.9560, 33.3550, 88.4220, 31.8190],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 88.4220,\n 31.8190, 86.1350, 29.9970, 83.2280, 28.0830, 79.7720, 26.6140],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 79.7720,\n 26.6140, 71.9060, 23.2700, 68.6000, 22.3560, 67.6840, 22.1380],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 67.6840,\n 22.1380, 67.6250, 22.1400, 67.5680, 22.1400, 67.5110, 22.1400],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 67.5110,\n 22.1400, 64.2460, 22.1410, 61.7360, 19.9850, 60.7810, 16.8010],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 60.7810,\n 16.8010, 59.7890, 13.4950, 61.6320, 9.9460, 64.6030, 8.1910],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 64.6030,\n 8.1910, 65.1680, 7.8560, 78.6810, 0.0000, 96.4900, 0.0000],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 96.4900,\n 0.0000, 100.1650, 0.0000, 103.8160, 0.3370, 107.3410, 1.0020],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 107.3410,\n 1.0020, 118.3240, 3.0750, 126.0860, 6.1720, 133.5920, 9.1660],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 133.5920,\n 9.1660, 140.9210, 12.0900, 147.8440, 14.8520, 158.1800, 17.0290],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 158.1800,\n 17.0290, 163.2010, 18.0870, 167.8250, 18.9030, 172.2960, 19.6930],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 172.2960,\n 19.6930, 187.6790, 22.4090, 200.9640, 24.7550, 220.8270, 34.8980],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 220.8270,\n 34.8980, 246.7510, 48.1380, 261.3280, 62.6240, 261.8740, 75.6850],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 261.8740,\n 75.6850, 261.9890, 78.4380, 261.9290, 80.8980, 261.6930, 83.0600],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 261.6930,\n 83.0600, 264.1190, 84.2730, 267.0670, 85.8060, 270.0450, 87.5000],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 270.0450,\n 87.5000, 282.1680, 94.3950, 287.2520, 99.5590, 287.5910, 105.3220],\n [ 2.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 287.5910,\n 105.3220, 287.8260, 109.3320, 286.7510, 112.6310, 284.3950, 115.1300],\n [ 6.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 284.3950,\n 115.1300, -1.0000, -1.0000, -1.0000, -1.0000, 284.3950, 115.1300]])\n" ] ], [ [ "# Save svg tensor in a .pkl file", "_____no_output_____" ] ] ]

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

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

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ "Apache-2.0" ]

[ [ [ "<h1>Table of Contents<span class=\"tocSkip\"></span></h1>\n<div class=\"toc\" style=\"margin-top: 1em;\"><ul class=\"toc-item\"><li><span><a href=\"#Intro\" data-toc-modified-id=\"Intro-1\"><span class=\"toc-item-num\">1&nbsp;&nbsp;</span>Intro</a></span></li><li><span><a href=\"#Load-Data\" data-toc-modified-id=\"Load-Data-2\"><span class=\"toc-item-num\">2&nbsp;&nbsp;</span>Load Data</a></span></li><li><span><a href=\"#Cyclical-Feeding\" data-toc-modified-id=\"Cyclical-Feeding-3\"><span class=\"toc-item-num\">3&nbsp;&nbsp;</span>Cyclical Feeding</a></span><ul class=\"toc-item\"><li><span><a href=\"#TODOs\" data-toc-modified-id=\"TODOs-3.1\"><span class=\"toc-item-num\">3.1&nbsp;&nbsp;</span>TODOs</a></span></li></ul></li><li><span><a href=\"#Image-Sharpening\" data-toc-modified-id=\"Image-Sharpening-4\"><span class=\"toc-item-num\">4&nbsp;&nbsp;</span>Image Sharpening</a></span></li><li><span><a href=\"#Source-Data-FaceSwap-and-Upscaling\" data-toc-modified-id=\"Source-Data-FaceSwap-and-Upscaling-5\"><span class=\"toc-item-num\">5&nbsp;&nbsp;</span>Source Data FaceSwap and Upscaling</a></span></li><li><span><a href=\"#Celeba-Test\" data-toc-modified-id=\"Celeba-Test-6\"><span class=\"toc-item-num\">6&nbsp;&nbsp;</span>Celeba Test</a></span></li></ul></div>", "_____no_output_____" ], [ "# Intro\nNotebook exploring random experiments around the use of the trained Faceswap generators.", "_____no_output_____" ] ], [ [ "import numpy as np\nimport pandas as pd\nimport seaborn as sns\nfrom PIL import Image\nimport matplotlib.pyplot as plt\n\nfrom pathlib import Path\nimport sys\nimport pickle\nimport yaml\nfrom numpy.random import shuffle\nfrom ast import literal_eval\nimport tensorflow as tf\n\nimport cv2\n\nfrom tqdm import tqdm\n\n# Plotting\n%matplotlib notebook\n#%matplotlib inline\n\nsns.set_context(\"paper\")\nsns.set_style(\"dark\")\n\nsys.path.append('../face_swap')\n\nfrom utils import image_processing\nfrom utils import super_resolution\n\nfrom face_swap.deep_swap import swap_faces, Swapper\nfrom face_swap import faceswap_utils as utils\nfrom face_swap.plot_utils import stack_images\nfrom face_swap import FaceGenerator, FaceDetector\nfrom face_swap.train import get_original_data\nfrom face_swap import gan, gan_utils\nfrom face_swap import CONFIG_PATH\nfrom face_swap.Face import Face\n\n%load_ext autoreload\n%autoreload 2", "_____no_output_____" ], [ "data_folder = Path.home() / \"Documents/datasets/\"\nmodels_folder = Path.home() / \"Documents/models/\"", "_____no_output_____" ] ], [ [ "# Load Data", "_____no_output_____" ] ], [ [ "# Load two random celeba faces\nfrom_face_img = cv2.cvtColor(cv2.imread(str(data_folder / \"img_align_celeba\" / \n \"000{}{}{}.jpg\".format(*np.random.randint(0, 9, 3)))),\n cv2.COLOR_BGR2RGB)\nto_face_img = cv2.cvtColor(cv2.imread(str(data_folder / \"img_align_celeba\" / \n \"000{}{}{}.jpg\".format(*np.random.randint(0, 9, 3)))),\n cv2.COLOR_BGR2RGB)", "_____no_output_____" ], [ "plt.imshow(from_face_img)\nplt.show()\nplt.imshow(to_face_img)\nplt.show()", "_____no_output_____" ] ], [ [ "# Cyclical Feeding\nCycling feeding own output to generator. Can start with actual face or random noise. \n\n## TODOs\n* Try apply text on image before feeding to generator", "_____no_output_____" ] ], [ [ "def crop(img, crop_factor=0.2):\n h, w = img.shape[:2]\n h_crop = int((h * crop_factor)//2)\n w_crop = int((w * crop_factor)//2)\n return img[h_crop:h-h_crop, w_crop:w-w_crop]", "_____no_output_____" ], [ "def zoom(img, zoom_factor=1.5):\n h, w = img.shape[:2]\n mat = cv2.getRotationMatrix2D((w//2, h//2), 0, zoom_factor)\n #mat[:, 2] -= (w//2, h//2)\n result = cv2.warpAffine(img, mat, (w, h), borderMode=cv2.BORDER_REPLICATE)\n return result", "_____no_output_____" ], [ "# load config\nwith open(CONFIG_PATH, 'r') as ymlfile:\n cfg = yaml.load(ymlfile)\nmodel_cfg = cfg['masked_gan']['v1']", "_____no_output_____" ], [ "# load generator and related functions\ngen_a, gen_b, _, _ = gan.get_gan(model_cfg, load_discriminators=False)\n_, _, _, fun_generate_a, fun_mask_a, fun_abgr_a = gan_utils.cycle_variables_masked(gen_a)\n_, _, _, fun_generate_b, fun_mask_b, fun_abgr_b = gan_utils.cycle_variables_masked(gen_b)", "_____no_output_____" ], [ "gen_fun_a = lambda x: fun_abgr_a([np.expand_dims(x, 0)])[0][0]\ngen_fun_b = lambda x: fun_abgr_b([np.expand_dims(x, 0)])[0][0]", "_____no_output_____" ], [ "generator_a = FaceGenerator.FaceGenerator(\n lambda face_img: FaceGenerator.gan_masked_generate_face(gen_fun_a, face_img),\n input_size=(64, 64), tanh_fix=True)\ngenerator_b = FaceGenerator.FaceGenerator(\n lambda face_img: FaceGenerator.gan_masked_generate_face(gen_fun_b, face_img),\n input_size=(64, 64), tanh_fix=True)", "_____no_output_____" ], [ "gen_input = Face(img, img)\nuse_a = True\ngenerator = generator_a if use_a else generator_b\nfor i in range(500):\n out = get_hr_version(sr_model, generator.generate(gen_input, (64, 64))[0])\n #out = generator.generate(gen_input, (128, 128))[0]\n gen_input.face_img = FaceGenerator.random_transform(out, **cfg['random_transform'])\n #gen_input.img = zoom(out)\n res_path = str(data_folder / 'faceswap_experiments/cycle_feed/02/_{:04d}.png'.format(i))\n #cv2.imwrite(res_path, zoom(out))\n cv2.imwrite(res_path, out)\n # swap generator randomly every epoch\n #generator = generator_a if np.random.rand() > 0.5 else generator_b\n # swap generator every N epoch\n if i%50 == 0:\n use_a = not use_a\n generator = generator_a if use_a else generator_b", "_____no_output_____" ] ], [ [ "# Image Sharpening", "_____no_output_____" ] ], [ [ "# adapted from https://github.com/AdityaPokharel/Sharpen-Image\nregular_kernel = np.array([[-1,-1,-1], [-1,9,-1], [-1,-1,-1]])\nedge_enhance_kernel = np.array([[-1,-1,-1,-1,-1],\n [-1,2,2,2,-1],\n [-1,2,8,2,-1],\n [-2,2,2,2,-1],\n [-1,-1,-1,-1,-1]])/8.0\ndef sharpen(img, kernel=regular_kernel):\n # apply kernel to input image\n res = cv2.filter2D(img, -1, kernel)\n return res\n\n# see also cv2.detailEnhance(src, sigma_s=10, sigma_r=0.15)", "_____no_output_____" ], [ "plt.imshow(sharpen(to_face_img))\nplt.show()", "_____no_output_____" ] ], [ [ "# Source Data FaceSwap and Upscaling\nTry to cherry pick some results of face-swapping on the training data, apply upscaling to a reasonable size (e.g. 128x128) and any possible post-processing that might help in improving image quality.\n", "_____no_output_____" ] ], [ [ "input_path = data_folder / \"facesets\" / \"cage\"\nout_path = data_folder / \"faceswap_experiments\" / \"source_faceswap\" / \"cage_trump\"\n\nout_size = (64, 64)", "_____no_output_____" ], [ "# collected all image paths\nimg_paths = image_processing.get_imgs_paths(input_path, as_str=False)\n\n# iterate over all collected image paths\nfor i, img_path in enumerate(img_paths):\n img = cv2.imread(str(img_path))\n gen_input = Face(img, img)\n gen_face = generator_b.generate(gen_input)[0]\n gen_face = sharpen(gen_face)\n gen_face = cv2.resize(gen_face, out_size)\n cv2.imwrite(str(out_path / \"out_{:04d}.jpg\".format(i)),\n gen_face)", "_____no_output_____" ] ], [ [ "# Celeba Test\nTest Celeba training and generation of artworks", "_____no_output_____" ] ], [ [ "def plot_sample(images: list, predict_fun,\n tanh_fix=False, save_to: str=None, \n nb_test_imgs=14, nb_columns=3, white_border=3):\n # need number of images divisible by number of columns\n nb_rows = nb_test_imgs//nb_columns\n assert nb_test_imgs % nb_columns == 0\n images = images[0:nb_test_imgs]\n\n figure = np.stack([\n images,\n predict_fun(images),\n ], axis=1)\n # we split images on two columns\n figure = figure.reshape((nb_columns, nb_rows) + figure.shape[1:])\n figure = stack_images(figure)\n img_width = images[0].shape[1]\n img_height = images[0].shape[0]\n for i in range(1, nb_columns):\n x = img_width*2*i\n figure[:, x-white_border:x+white_border, :] = 255.0\n for i in range(1, nb_rows):\n y = img_height*i\n figure[y-white_border:y+white_border, :, :] = 255.0\n\n if save_to:\n cv2.imwrite(save_to, figure)\n else:\n figure = cv2.cvtColor(figure, cv2.COLOR_BGR2RGB)\n #plt.imshow(figure)\n #plt.show()\n display(Image.fromarray(figure))\n # crashes in notebooks\n #cv2.imshow('', figure)\n #cv2.waitKey(0)", "_____no_output_____" ], [ "# load config\nwith open(CONFIG_PATH, 'r') as ymlfile:\n cfg = yaml.load(ymlfile)\nmodel_cfg = cfg['masked_gan']['v1']\nmodel_cfg['models_path'] = str(models_folder / \"face_recognition/deep_faceswap/masked_gan/cage_celeba/v4\")", "_____no_output_____" ], [ "#tf.reset_default_graph()\nface_detector = FaceDetector.FaceDetector(cfg)", "_____no_output_____" ], [ "# load generator and related functions\nnetGA, netGB, _, _ = gan.get_gan(model_cfg, load_discriminators=False)\n\n# define generation and plotting function\n# depending if using masked gan model or not\nif model_cfg['masked']:\n distorted_A, fake_A, mask_A, path_A, fun_mask_A, fun_abgr_A = gan_utils.cycle_variables_masked(netGA)\n distorted_B, fake_B, mask_B, path_B, fun_mask_B, fun_abgr_B = gan_utils.cycle_variables_masked(netGB)\n #gen_plot_a = lambda x: np.array(path_A([x])[0]) \n #gen_plot_b = lambda x: np.array(path_B([x])[0])\n gen_plot_a = lambda x: np.array(fun_abgr_A([x])[0][ :, :, :, 1:]) \n gen_plot_b = lambda x: np.array(fun_abgr_B([x])[0][ :, :, :, 1:])\n gen_plot_mask_a = lambda x: np.array(fun_mask_A([x])[0])*2-1\n gen_plot_mask_b = lambda x: np.array(fun_mask_B([x])[0])*2-1\nelse:\n gen_plot_a = lambda x: netGA.predict(x)\n gen_plot_b = lambda x: netGB.predict(x)", "_____no_output_____" ], [ "sr_model = super_resolution.get_SRResNet(cfg['super_resolution'])\nresize_fun = lambda img, size: FaceGenerator.super_resolution_resizing(sr_model, img, size)", "_____no_output_____" ], [ "gen_fun_a = lambda x: fun_abgr_A([np.expand_dims(x, 0)])[0][0]\ngen_fun_b = lambda x: fun_abgr_B([np.expand_dims(x, 0)])[0][0]\ngen_input_size = literal_eval(model_cfg['img_shape'])[:2]\nface_generator = FaceGenerator.FaceGenerator(\n lambda face_img: FaceGenerator.gan_masked_generate_face(gen_fun_a, face_img),\n input_size=gen_input_size, config=cfg['swap'], resize_fun=resize_fun)", "_____no_output_____" ], [ "swapper = Swapper(face_detector, face_generator, cfg['swap'], save_all=True)", "_____no_output_____" ], [ "def swap(img):\n face = Face(img.copy(), Face.Rectangle(0, 64, 64, 0))\n #return swap_faces(face, face_detector, cfg['swap'], face_generator)\n return face.get_face_img()\n#gen_plot_b = lambda x: [swap(img) for img in x]\ngen_plot = lambda x: [swapper.swap(img) for img in x]", "_____no_output_____" ], [ "img_dir_a = data_folder / 'facesets/cage'\nimg_dir_b = data_folder / 'celeba_tmp'\n#images_a, images_b = get_original_data(img_dir_a, img_dir_b, img_size=None, tanh_fix=False)\nimages = image_processing.load_data(image_processing.get_imgs_paths(img_dir_a), (128, 128))", "_____no_output_____" ], [ "dest_folder = str(data_folder / \"faceswap_experiments/source_faceswap/cage_celeba_masked/test_1/_{}.png\")\nswapper.config['mask_method'] = \"gen_mask\"\nface_generator.border_expand = (0.1, 0.1)\nface_generator.blur_size = 13\nface_generator.align = False\n#shuffle(images)\nfor i in range(20):\n print(i)\n images_subset = images[i*15:(i+1)*15]\n try:\n plot_sample(images_subset, gen_plot, nb_test_imgs=15, nb_columns=3, \n save_to=dest_folder.format(i), tanh_fix=False)\n except FaceDetector.FaceSwapException:\n pass", "_____no_output_____" ] ] ]

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

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

[ "MIT" ]

[ "MIT" ]