MNIST_interp.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<style>.container { width:100% !important; }</style>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "from IPython.core.display import display, HTML\n",
    "display(HTML(\"<style>.container { width:100% !important; }</style>\"))\n",
    "\n",
    "from pathlib import Path\n",
    "from matplotlib import pyplot as plt\n",
    "from skimage import feature\n",
    "\n",
    "import torch\n",
    "from torch import nn\n",
    "from torch import optim\n",
    "from torch.utils.data import TensorDataset, DataLoader\n",
    "\n",
    "import numpy as np\n",
    "import requests\n",
    "import sklearn\n",
    "import sklearn.linear_model\n",
    "import sklearn.tree\n",
    "import sklearn.ensemble\n",
    "import pickle\n",
    "import gzip\n",
    "import graphviz\n",
    "import math"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# mnist loader code from:\n",
    "# https://pytorch.org/tutorials/beginner/nn_tutorial.html\n",
    "\n",
    "DATA_PATH = Path(\"data\")\n",
    "PATH = DATA_PATH / \"mnist\"\n",
    "\n",
    "PATH.mkdir(parents=True, exist_ok=True)\n",
    "\n",
    "URL = \"http://deeplearning.net/data/mnist/\"\n",
    "FILENAME = \"mnist.pkl.gz\"\n",
    "\n",
    "if not (PATH / FILENAME).exists():\n",
    "        content = requests.get(URL + FILENAME).content\n",
    "        (PATH / FILENAME).open(\"wb\").write(content)\n",
    "\n",
    "with gzip.open((PATH / FILENAME).as_posix(), \"rb\") as f:\n",
    "        ((x_train, y_train), (x_valid, y_valid), _) = pickle.load(f, encoding=\"latin-1\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [],
   "source": [
    "def show_mnist(x):\n",
    "    plt.imshow(x.reshape((28, 28)),cmap='Greys_r')\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Linear Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.9296"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# parameters for logistic regression from :\n",
    "# https://scikit-learn.org/stable/auto_examples/linear_model/plot_sparse_logistic_regression_mnist.html\n",
    "clf = sklearn.linear_model.LogisticRegression(solver='saga', \n",
    "                                              penalty='l1',\n",
    "                                              multi_class='multinomial', \n",
    "                                              tol=0.1).fit(x_train, y_train)\n",
    "clf.score(x_valid, y_valid)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 250,
       "width": 253
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 250,
       "width": 253
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "scale = 0.4\n",
    "plt.imshow(clf.coef_[0].reshape((28, 28)),\n",
    "                   cmap=plt.cm.RdBu, vmin=-scale, vmax=scale)\n",
    "plt.show()\n",
    "plt.imshow(clf.coef_[3].reshape((28, 28)),\n",
    "                   cmap=plt.cm.RdBu, vmin=-scale, vmax=scale)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        4.81118557e-03,  0.00000000e+00, -1.56610724e-04,  0.00000000e+00,\n",
       "        6.49496071e-03,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  3.90594519e-05,  7.10306823e-05,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00, -1.47542413e-03,\n",
       "       -1.67811041e-04, -3.83280468e-02, -8.10846867e-02, -5.01943218e-02,\n",
       "       -2.90314621e-02, -2.65494116e-02, -8.29385683e-03,  0.00000000e+00,\n",
       "        0.00000000e+00,  1.67390785e-04,  3.92789141e-04,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00, -7.71080841e-03, -2.00411429e-02, -2.88612280e-02,\n",
       "       -1.10008667e-03, -4.71999042e-03, -4.86830067e-03,  0.00000000e+00,\n",
       "       -3.06605308e-02,  2.59211361e-02,  1.07632647e-02, -2.28323153e-02,\n",
       "       -3.89948625e-02, -7.60822453e-02, -2.57722320e-01, -1.83172555e-01,\n",
       "       -1.16660411e-01, -1.17033806e-01, -5.96779470e-02, -4.12856074e-03,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00, -1.84694045e-02, -5.06460147e-02, -5.04724608e-03,\n",
       "        5.75169771e-03,  2.88787505e-02,  8.01685406e-03,  1.98964852e-01,\n",
       "       -3.51600030e-03,  4.04520699e-02, -5.54117933e-02,  3.38421178e-02,\n",
       "       -9.15697805e-02,  1.19186824e-01, -1.24064467e-03, -1.76550163e-01,\n",
       "       -1.96846275e-01, -1.21343368e-01, -7.93434054e-02, -1.21957144e-01,\n",
       "       -8.45574657e-02, -5.80837910e-03,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00, -9.91131282e-02, -3.12495506e-02,  5.76637729e-03,\n",
       "        6.21767255e-02,  4.56615839e-02,  1.64715609e-01,  8.51403677e-02,\n",
       "        5.28749999e-02,  1.96265338e-01,  3.40161703e-01,  1.82310818e-01,\n",
       "        1.77246746e-01,  2.06247221e-01,  2.52435487e-01,  5.88717193e-02,\n",
       "       -6.87739576e-02, -4.88348964e-02, -2.05339729e-02, -2.24217777e-01,\n",
       "       -2.08711393e-01, -5.31713262e-02,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "       -3.10942940e-02, -1.18934721e-02,  6.09848813e-03,  4.57354014e-02,\n",
       "        1.38676190e-02, -8.74488095e-02,  3.14658653e-02,  2.71251015e-02,\n",
       "       -2.92451050e-02,  1.03588048e-01,  6.51802733e-02, -4.11401370e-02,\n",
       "        1.79558939e-01,  1.46797575e-01,  3.48896886e-01,  2.21468067e-01,\n",
       "        8.96825390e-02,  2.98534157e-01,  3.17015259e-01, -4.13629648e-02,\n",
       "       -2.58113104e-01, -1.62125211e-01, -3.39625873e-02,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "       -2.51209593e-02,  1.11719315e-03,  4.85573395e-02,  2.27482685e-02,\n",
       "       -1.75682228e-01,  1.66274508e-02,  4.75295360e-02,  1.01894661e-01,\n",
       "        3.84475433e-02,  9.41392885e-02,  2.22483202e-02,  4.19964663e-02,\n",
       "        6.45305878e-02,  2.41351324e-01,  1.67940376e-01,  1.77964706e-01,\n",
       "        1.02514565e-02, -5.30421616e-04,  1.64423792e-01,  1.33931439e-01,\n",
       "       -5.98730146e-02, -2.30706557e-01, -7.20795756e-02,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "       -3.87277440e-02, -5.30896498e-02,  2.56906660e-03, -1.40424984e-01,\n",
       "       -1.49655157e-02,  1.38718250e-02,  1.28428749e-02, -2.55551956e-02,\n",
       "        1.15491452e-01,  1.33333451e-01,  2.67656382e-01,  1.62623109e-01,\n",
       "        4.19702053e-01,  2.21314952e-01,  1.62260430e-01,  3.13854736e-02,\n",
       "        8.49117336e-02,  2.66073574e-02,  2.37061108e-01,  2.92896837e-01,\n",
       "        4.84042272e-02, -2.81955472e-01, -8.12952777e-02,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "       -4.28894319e-02, -6.85778294e-02, -9.54232857e-02, -1.10535008e-01,\n",
       "       -1.29425790e-01, -2.77022139e-02,  8.15456045e-02, -1.41269209e-02,\n",
       "        6.35987545e-02,  3.13052318e-01,  7.23411106e-02, -1.81696811e-01,\n",
       "        1.97148326e-01,  2.19547613e-01,  3.49014135e-01,  2.94872569e-01,\n",
       "        1.52109459e-01, -6.02582695e-02, -2.71108406e-02,  3.44395832e-01,\n",
       "        9.19613876e-02, -3.31372994e-01, -8.01352425e-02,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "       -2.39345908e-02,  2.53768107e-03, -1.37613903e-01, -3.78978786e-02,\n",
       "       -9.61892600e-02, -1.08148032e-02, -1.32000088e-01,  3.02410731e-02,\n",
       "       -5.66188734e-03,  5.51205269e-02,  7.65865281e-02, -1.07685523e-01,\n",
       "        1.13308378e-02,  3.17631624e-01,  2.35484971e-01,  4.49761504e-04,\n",
       "        2.00286789e-02,  1.36662967e-01,  1.87868223e-01,  4.91620290e-01,\n",
       "        1.75579426e-01, -3.58357351e-01, -8.70351172e-02,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "       -9.58194435e-02, -1.98744761e-01,  7.68249173e-03, -4.08938630e-03,\n",
       "       -3.24970372e-02, -9.73336480e-02, -8.32186912e-02, -1.07853375e-02,\n",
       "       -2.30955584e-01, -8.35120942e-02, -3.37064891e-01, -6.61349602e-01,\n",
       "       -5.73121176e-01, -6.21281001e-03, -1.00135265e-01,  1.96945354e-02,\n",
       "        2.22344155e-01,  3.61462554e-01,  3.47262616e-01,  2.93186937e-01,\n",
       "        3.26893720e-01, -2.34562400e-01, -4.79955670e-02,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "       -1.04968764e-01, -5.65662349e-02,  2.44680966e-02,  1.24191424e-02,\n",
       "        4.70223463e-03, -8.20982440e-02, -4.78112355e-02, -4.16002528e-03,\n",
       "        9.07568262e-03,  1.38278222e-02, -5.12187793e-01, -8.24707454e-01,\n",
       "       -6.07372767e-01, -8.94049801e-02,  1.38901445e-02, -5.56902217e-03,\n",
       "        6.81150358e-02,  1.52615739e-01,  2.79126333e-01,  4.36927312e-01,\n",
       "        5.54557593e-01,  2.19759806e-03, -7.44892183e-03,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        2.97041473e-03,  4.42758814e-03,  2.65880096e-01,  3.19189031e-01,\n",
       "        1.44880458e-01,  1.31041158e-01,  1.11571960e-01,  4.92751980e-02,\n",
       "        1.48678258e-01, -2.28305647e-01, -6.45293939e-01, -8.72937066e-01,\n",
       "       -5.82334036e-01, -1.46910212e-01, -1.24929217e-03, -1.77023736e-01,\n",
       "       -2.68775445e-01,  7.18823352e-03,  7.92137160e-02,  4.37538431e-01,\n",
       "        3.94221207e-01,  4.02622804e-02,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00, -3.52966718e-03,\n",
       "       -3.87104314e-02,  2.66675417e-01,  2.23413823e-01,  4.38656063e-01,\n",
       "        1.93427086e-01,  2.23704723e-01,  7.55613319e-02,  2.24379238e-01,\n",
       "        1.11582600e-01, -1.58384330e-01, -7.80170057e-01, -7.69340736e-01,\n",
       "       -7.72877917e-01, -1.24404767e-01, -3.05758494e-03, -2.36273328e-01,\n",
       "        8.49112304e-02,  8.89689866e-02,  4.30898669e-02,  4.87891183e-01,\n",
       "        3.97798107e-01,  1.22042836e-02,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00, -2.24354420e-02,\n",
       "        2.71044713e-02,  4.03804157e-01,  2.24759597e-02,  3.62391390e-01,\n",
       "        3.02383284e-01,  1.85015501e-01,  2.06849856e-01,  1.71969116e-02,\n",
       "        2.78333798e-02, -4.00139463e-01, -9.41518072e-01, -6.12218615e-01,\n",
       "       -5.41997373e-01, -3.68514829e-01,  2.17724822e-04, -6.06236072e-02,\n",
       "        1.99454259e-01,  3.83515219e-02,  1.60614052e-01,  4.41824802e-01,\n",
       "        3.11325773e-01, -2.18420303e-03,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00, -5.13284037e-02,\n",
       "        5.06400221e-02,  3.70834186e-01,  2.33669706e-01,  1.28826116e-01,\n",
       "        3.92719118e-01,  2.96421997e-01,  2.85096201e-01,  8.44368526e-02,\n",
       "       -2.21760246e-01, -8.04276521e-01, -6.49682452e-01, -3.09732313e-01,\n",
       "       -2.94811948e-01, -9.09968826e-02,  1.39973549e-01,  1.47940426e-01,\n",
       "        2.21846417e-01,  8.00807709e-02,  1.61523119e-01,  2.63635944e-01,\n",
       "        7.26094396e-03, -8.18704327e-02,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00, -2.60143665e-02,\n",
       "        4.95086660e-02,  4.35272051e-01,  2.71086602e-01,  1.40519343e-01,\n",
       "        1.23206689e-01,  3.87122047e-01,  4.59665246e-01,  6.16310642e-02,\n",
       "       -1.02485558e-01, -6.28064601e-01, -6.09164608e-01, -3.91931200e-01,\n",
       "       -2.80067589e-02,  8.62552049e-02,  4.02459256e-02,  5.97353717e-02,\n",
       "        1.75887698e-01, -5.30968947e-03,  2.50235254e-01,  9.63958939e-02,\n",
       "        0.00000000e+00, -6.33991262e-02,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00, -7.84350254e-02,\n",
       "       -2.29175807e-02,  2.58179993e-01,  2.46648544e-01,  2.79827405e-02,\n",
       "        2.56176029e-01,  1.93640249e-01,  3.20238622e-01,  2.10186227e-01,\n",
       "       -1.83745526e-01, -6.05131116e-01, -5.85251677e-01, -1.82057739e-01,\n",
       "       -8.73455431e-03, -6.53488213e-02, -2.40069398e-02, -3.92991397e-02,\n",
       "       -1.84857016e-02, -4.36532750e-03,  3.31042336e-01,  5.33133168e-03,\n",
       "       -7.51023994e-02, -3.70457431e-02,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00, -9.02205887e-02,\n",
       "       -1.37140590e-01,  8.95849936e-02,  2.84897702e-01,  2.12344318e-01,\n",
       "        3.04085467e-01, -1.48531813e-02,  1.97698815e-01,  2.59276781e-01,\n",
       "        2.08222298e-01, -1.35381441e-01, -1.50198567e-01, -2.21344397e-01,\n",
       "       -2.15526560e-01,  2.74216886e-03,  1.66723030e-02, -1.33903729e-01,\n",
       "       -2.34892359e-03,  1.76323969e-01,  2.82549395e-01, -1.34862589e-02,\n",
       "       -5.07013978e-02, -1.23157706e-02,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00, -8.18274427e-02,\n",
       "       -1.38088011e-02,  1.16085549e-01,  9.38903632e-02,  1.11317200e-01,\n",
       "        7.95591868e-02, -1.13269979e-02,  3.23483957e-01,  3.67424378e-01,\n",
       "        5.37367777e-01,  3.17360868e-01, -5.19006510e-02, -1.00664851e-01,\n",
       "       -1.07755557e-01, -6.72083692e-02, -5.83738514e-02, -2.34767069e-01,\n",
       "       -1.11305867e-01, -2.79007421e-02,  8.74275112e-02, -4.53538479e-02,\n",
       "       -6.28101144e-02,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00, -5.14104615e-02,\n",
       "       -7.18188182e-02,  8.96886079e-02,  1.63567400e-01,  1.32194105e-01,\n",
       "       -5.83653988e-02,  1.22372934e-02,  2.32616461e-01,  3.50575039e-01,\n",
       "        3.45412101e-01,  5.57699451e-02, -2.41708117e-02,  2.60300356e-02,\n",
       "        2.26921040e-03, -3.15186409e-01, -2.62168008e-01, -2.63493080e-01,\n",
       "       -1.42785774e-01, -3.67686134e-02,  1.59916332e-02, -5.10481022e-02,\n",
       "       -4.69881955e-02,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00, -1.51783495e-02,\n",
       "       -1.82779639e-01,  6.65250468e-04,  1.13989366e-01,  1.32254424e-02,\n",
       "        1.92024564e-01,  1.92102884e-01,  2.71429669e-01,  1.51049857e-01,\n",
       "        7.81186944e-02,  4.50102266e-01,  1.26853621e-01,  1.66276060e-01,\n",
       "        1.84328222e-02,  6.36102966e-02, -7.79614767e-02, -1.30364975e-01,\n",
       "       -2.24485880e-01, -1.00306439e-01,  1.06267158e-02,  2.14810210e-04,\n",
       "       -2.29631484e-02,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "       -1.03583975e-01, -7.66896177e-02, -1.81108255e-04,  1.49170346e-01,\n",
       "        1.41753690e-01,  2.75361494e-01,  3.86882222e-01,  3.58236516e-01,\n",
       "        1.39657288e-01,  2.47826085e-01,  2.90713154e-01,  1.56077835e-01,\n",
       "        1.07029569e-02,  2.58507121e-02, -2.16006338e-01, -3.26473460e-01,\n",
       "       -3.06821293e-01, -1.63387813e-01,  1.37169933e-05, -5.69511220e-05,\n",
       "       -8.59565418e-03,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "       -1.77563794e-03, -9.44401570e-04, -4.88125256e-02, -2.00470409e-01,\n",
       "       -2.25126220e-01, -8.49395308e-02, -1.45915244e-02, -1.04762558e-03,\n",
       "        2.29448842e-01,  1.80021594e-01,  1.08976496e-01,  3.92847995e-02,\n",
       "       -9.08724284e-02, -9.84975941e-02, -2.65683671e-01, -3.12607844e-01,\n",
       "       -2.80362413e-01, -1.73924281e-01, -4.70805374e-02,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00, -3.06901534e-02, -1.14143321e-01,\n",
       "       -2.26841769e-01, -3.25521240e-01, -4.16385539e-01, -2.88727828e-01,\n",
       "       -3.25762339e-01, -2.21544811e-01, -3.20775978e-01, -2.79264599e-01,\n",
       "       -2.65842529e-01, -3.12494942e-01, -2.78081150e-01, -1.21050904e-01,\n",
       "       -1.14539972e-01, -7.89729864e-02, -9.66171646e-03,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "       -2.09982108e-02, -8.58239334e-02, -6.41085388e-02, -2.35863804e-02,\n",
       "       -9.25264540e-03, -4.13561478e-02, -5.90876445e-02, -4.39728189e-02,\n",
       "       -2.54249768e-02, -2.98632249e-02, -2.93046180e-02,  0.00000000e+00,\n",
       "       -1.23354659e-02, -1.53128454e-02, -1.38382673e-03,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00, -3.99183685e-03,  0.00000000e+00,\n",
       "        0.00000000e+00,  0.00000000e+00,  0.00000000e+00,  0.00000000e+00])"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "clf.coef_[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Interpretable Features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7fd9198775f8>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 250,
       "width": 253
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(feature.canny(x_train[10].reshape((28, 28))), cmap='Greys_r')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 5, 11],\n",
       "       [13, 11],\n",
       "       [20,  7]])"
      ]
     },
     "execution_count": 75,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "feature.corner_peaks(feature.corner_harris(x_train[10].reshape((28, 28))), \n",
    "                     min_distance=5, threshold_rel=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7f123ee29ba8>"
      ]
     },
     "execution_count": 76,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 250,
       "width": 253
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(feature.corner_harris(x_train[10].reshape((28, 28))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 1min 53s, sys: 132 ms, total: 1min 53s\n",
      "Wall time: 1min 53s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "# hand-engineering features\n",
    "# 1] average luminosity\n",
    "# 2] proportion of canny edge detected image that is filled\n",
    "# 3] number of corner using corner_peak and corner_harris and some reasonable min (maybe 5)\n",
    "# 4] max width of pixels\n",
    "\n",
    "interp_x_train = []\n",
    "interp_x_valid = []\n",
    "\n",
    "for x in x_train:\n",
    "    avg_lumin = np.sum(x)/(28*28)\n",
    "    edge_prop = np.sum(feature.canny(x.reshape((28, 28))))/(28*28)\n",
    "    num_corners = len(feature.corner_peaks(feature.corner_harris(x.reshape((28, 28))), \n",
    "                     min_distance=5, threshold_rel=0))\n",
    "    max_width = max(np.sum(x.reshape((28,28)) > 0, axis=1))\n",
    "    max_height = max(np.sum(x.reshape((28,28)) > 0, axis=0))\n",
    "    \n",
    "    interp_x_train.append([\n",
    "        avg_lumin,\n",
    "        edge_prop,\n",
    "        num_corners,\n",
    "        max_width,\n",
    "        max_height\n",
    "    ])\n",
    "    \n",
    "interp_x_train = np.array(interp_x_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.33354\n"
     ]
    }
   ],
   "source": [
    "clf = sklearn.tree.DecisionTreeClassifier(max_depth = 3, criterion=\"gini\")\n",
    "clf.fit(interp_x_train, y_train)\n",
    "print(clf.score(interp_x_train, y_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/svg+xml": [
       "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"no\"?>\n",
       "<!DOCTYPE svg PUBLIC \"-//W3C//DTD SVG 1.1//EN\"\n",
       " \"http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd\">\n",
       "<!-- Generated by graphviz version 2.40.1 (20161225.0304)\n",
       " -->\n",
       "<!-- Title: Tree Pages: 1 -->\n",
       "<svg width=\"2456pt\" height=\"493pt\"\n",
       " viewBox=\"0.00 0.00 2455.50 493.00\" xmlns=\"http://www.w3.org/2000/svg\" xmlns:xlink=\"http://www.w3.org/1999/xlink\">\n",
       "<g id=\"graph0\" class=\"graph\" transform=\"scale(1 1) rotate(0) translate(4 489)\">\n",
       "<title>Tree</title>\n",
       "<polygon fill=\"#ffffff\" stroke=\"transparent\" points=\"-4,4 -4,-489 2451.5,-489 2451.5,4 -4,4\"/>\n",
       "<!-- 0 -->\n",
       "<g id=\"node1\" class=\"node\">\n",
       "<title>0</title>\n",
       "<polygon fill=\"#fffffd\" stroke=\"#000000\" points=\"1184,-485 815,-485 815,-387 1184,-387 1184,-485\"/>\n",
       "<text text-anchor=\"middle\" x=\"999.5\" y=\"-469.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">max_width &lt;= 7.5</text>\n",
       "<text text-anchor=\"middle\" x=\"999.5\" y=\"-454.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">gini = 0.9</text>\n",
       "<text text-anchor=\"middle\" x=\"999.5\" y=\"-439.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">samples = 50000</text>\n",
       "<text text-anchor=\"middle\" x=\"999.5\" y=\"-424.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">value = [4932, 5678, 4968, 5101, 4859, 4506, 4951, 5175, 4842</text>\n",
       "<text text-anchor=\"middle\" x=\"999.5\" y=\"-409.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">4988]</text>\n",
       "<text text-anchor=\"middle\" x=\"999.5\" y=\"-394.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">class = 1</text>\n",
       "</g>\n",
       "<!-- 1 -->\n",
       "<g id=\"node2\" class=\"node\">\n",
       "<title>1</title>\n",
       "<polygon fill=\"#e3e640\" stroke=\"#000000\" points=\"873,-343.5 600,-343.5 600,-260.5 873,-260.5 873,-343.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"736.5\" y=\"-328.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">edge_prop &lt;= 0.091</text>\n",
       "<text text-anchor=\"middle\" x=\"736.5\" y=\"-313.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">gini = 0.072</text>\n",
       "<text text-anchor=\"middle\" x=\"736.5\" y=\"-298.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">samples = 5325</text>\n",
       "<text text-anchor=\"middle\" x=\"736.5\" y=\"-283.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">value = [7, 5130, 2, 32, 10, 16, 32, 19, 46, 31]</text>\n",
       "<text text-anchor=\"middle\" x=\"736.5\" y=\"-268.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">class = 1</text>\n",
       "</g>\n",
       "<!-- 0&#45;&gt;1 -->\n",
       "<g id=\"edge1\" class=\"edge\">\n",
       "<title>0&#45;&gt;1</title>\n",
       "<path fill=\"none\" stroke=\"#000000\" d=\"M903.0619,-386.8642C878.2541,-374.2245 851.6504,-360.6698 827.1477,-348.1855\"/>\n",
       "<polygon fill=\"#000000\" stroke=\"#000000\" points=\"828.7217,-345.0594 818.2226,-343.6381 825.5438,-351.2965 828.7217,-345.0594\"/>\n",
       "<text text-anchor=\"middle\" x=\"825.9484\" y=\"-363.7144\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">True</text>\n",
       "</g>\n",
       "<!-- 8 -->\n",
       "<g id=\"node9\" class=\"node\">\n",
       "<title>8</title>\n",
       "<polygon fill=\"#ffffff\" stroke=\"#000000\" points=\"1526.5,-351 1164.5,-351 1164.5,-253 1526.5,-253 1526.5,-351\"/>\n",
       "<text text-anchor=\"middle\" x=\"1345.5\" y=\"-335.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">edge_prop &lt;= 0.119</text>\n",
       "<text text-anchor=\"middle\" x=\"1345.5\" y=\"-320.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">gini = 0.891</text>\n",
       "<text text-anchor=\"middle\" x=\"1345.5\" y=\"-305.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">samples = 44675</text>\n",
       "<text text-anchor=\"middle\" x=\"1345.5\" y=\"-290.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">value = [4925, 548, 4966, 5069, 4849, 4490, 4919, 5156, 4796</text>\n",
       "<text text-anchor=\"middle\" x=\"1345.5\" y=\"-275.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">4957]</text>\n",
       "<text text-anchor=\"middle\" x=\"1345.5\" y=\"-260.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">class = 7</text>\n",
       "</g>\n",
       "<!-- 0&#45;&gt;8 -->\n",
       "<g id=\"edge8\" class=\"edge\">\n",
       "<title>0&#45;&gt;8</title>\n",
       "<path fill=\"none\" stroke=\"#000000\" d=\"M1126.3729,-386.8642C1153.3733,-376.4075 1181.9908,-365.3244 1209.3225,-354.7393\"/>\n",
       "<polygon fill=\"#000000\" stroke=\"#000000\" points=\"1210.8705,-357.8931 1218.9316,-351.0178 1208.3425,-351.3655 1210.8705,-357.8931\"/>\n",
       "<text text-anchor=\"middle\" x=\"1208.8312\" y=\"-370.1866\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">False</text>\n",
       "</g>\n",
       "<!-- 2 -->\n",
       "<g id=\"node3\" class=\"node\">\n",
       "<title>2</title>\n",
       "<polygon fill=\"#e3e63e\" stroke=\"#000000\" points=\"510.5,-209.5 244.5,-209.5 244.5,-126.5 510.5,-126.5 510.5,-209.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"377.5\" y=\"-194.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">max_width &lt;= 6.5</text>\n",
       "<text text-anchor=\"middle\" x=\"377.5\" y=\"-179.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">gini = 0.05</text>\n",
       "<text text-anchor=\"middle\" x=\"377.5\" y=\"-164.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">samples = 5239</text>\n",
       "<text text-anchor=\"middle\" x=\"377.5\" y=\"-149.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">value = [0, 5106, 2, 27, 7, 10, 21, 16, 22, 28]</text>\n",
       "<text text-anchor=\"middle\" x=\"377.5\" y=\"-134.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">class = 1</text>\n",
       "</g>\n",
       "<!-- 1&#45;&gt;2 -->\n",
       "<g id=\"edge2\" class=\"edge\">\n",
       "<title>1&#45;&gt;2</title>\n",
       "<path fill=\"none\" stroke=\"#000000\" d=\"M624.8781,-260.3361C584.6818,-245.3325 539.1598,-228.341 498.2226,-213.0608\"/>\n",
       "<polygon fill=\"#000000\" stroke=\"#000000\" points=\"499.3369,-209.7409 488.7443,-209.523 496.889,-216.299 499.3369,-209.7409\"/>\n",
       "</g>\n",
       "<!-- 5 -->\n",
       "<g id=\"node6\" class=\"node\">\n",
       "<title>5</title>\n",
       "<polygon fill=\"#ffffff\" stroke=\"#000000\" points=\"849.5,-209.5 623.5,-209.5 623.5,-126.5 849.5,-126.5 849.5,-209.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"736.5\" y=\"-194.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">max_height &lt;= 10.5</text>\n",
       "<text text-anchor=\"middle\" x=\"736.5\" y=\"-179.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">gini = 0.809</text>\n",
       "<text text-anchor=\"middle\" x=\"736.5\" y=\"-164.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">samples = 86</text>\n",
       "<text text-anchor=\"middle\" x=\"736.5\" y=\"-149.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">value = [7, 24, 0, 5, 3, 6, 11, 3, 24, 3]</text>\n",
       "<text text-anchor=\"middle\" x=\"736.5\" y=\"-134.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">class = 1</text>\n",
       "</g>\n",
       "<!-- 1&#45;&gt;5 -->\n",
       "<g id=\"edge5\" class=\"edge\">\n",
       "<title>1&#45;&gt;5</title>\n",
       "<path fill=\"none\" stroke=\"#000000\" d=\"M736.5,-260.3361C736.5,-247.4794 736.5,-233.163 736.5,-219.7145\"/>\n",
       "<polygon fill=\"#000000\" stroke=\"#000000\" points=\"740.0001,-219.6381 736.5,-209.6381 733.0001,-219.6382 740.0001,-219.6381\"/>\n",
       "</g>\n",
       "<!-- 3 -->\n",
       "<g id=\"node4\" class=\"node\">\n",
       "<title>3</title>\n",
       "<polygon fill=\"#e2e53a\" stroke=\"#000000\" points=\"233,-75.5 0,-75.5 0,-7.5 233,-7.5 233,-75.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"116.5\" y=\"-60.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">gini = 0.012</text>\n",
       "<text text-anchor=\"middle\" x=\"116.5\" y=\"-45.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">samples = 4517</text>\n",
       "<text text-anchor=\"middle\" x=\"116.5\" y=\"-30.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">value = [0, 4489, 0, 10, 1, 1, 2, 2, 6, 6]</text>\n",
       "<text text-anchor=\"middle\" x=\"116.5\" y=\"-15.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">class = 1</text>\n",
       "</g>\n",
       "<!-- 2&#45;&gt;3 -->\n",
       "<g id=\"edge3\" class=\"edge\">\n",
       "<title>2&#45;&gt;3</title>\n",
       "<path fill=\"none\" stroke=\"#000000\" d=\"M291.6773,-126.404C260.8429,-111.4593 226.2241,-94.6805 195.9547,-80.0097\"/>\n",
       "<polygon fill=\"#000000\" stroke=\"#000000\" points=\"197.2347,-76.7406 186.7094,-75.5287 194.1816,-83.0398 197.2347,-76.7406\"/>\n",
       "</g>\n",
       "<!-- 4 -->\n",
       "<g id=\"node5\" class=\"node\">\n",
       "<title>4</title>\n",
       "<polygon fill=\"#e6e957\" stroke=\"#000000\" points=\"504,-75.5 251,-75.5 251,-7.5 504,-7.5 504,-75.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"377.5\" y=\"-60.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">gini = 0.266</text>\n",
       "<text text-anchor=\"middle\" x=\"377.5\" y=\"-45.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">samples = 722</text>\n",
       "<text text-anchor=\"middle\" x=\"377.5\" y=\"-30.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">value = [0, 617, 2, 17, 6, 9, 19, 14, 16, 22]</text>\n",
       "<text text-anchor=\"middle\" x=\"377.5\" y=\"-15.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">class = 1</text>\n",
       "</g>\n",
       "<!-- 2&#45;&gt;4 -->\n",
       "<g id=\"edge4\" class=\"edge\">\n",
       "<title>2&#45;&gt;4</title>\n",
       "<path fill=\"none\" stroke=\"#000000\" d=\"M377.5,-126.2286C377.5,-113.2814 377.5,-98.9718 377.5,-85.8557\"/>\n",
       "<polygon fill=\"#000000\" stroke=\"#000000\" points=\"381.0001,-85.6415 377.5,-75.6416 374.0001,-85.6416 381.0001,-85.6415\"/>\n",
       "</g>\n",
       "<!-- 6 -->\n",
       "<g id=\"node7\" class=\"node\">\n",
       "<title>6</title>\n",
       "<polygon fill=\"#e6e955\" stroke=\"#000000\" points=\"734.5,-75.5 522.5,-75.5 522.5,-7.5 734.5,-7.5 734.5,-75.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"628.5\" y=\"-60.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">gini = 0.234</text>\n",
       "<text text-anchor=\"middle\" x=\"628.5\" y=\"-45.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">samples = 23</text>\n",
       "<text text-anchor=\"middle\" x=\"628.5\" y=\"-30.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">value = [0, 20, 0, 0, 0, 1, 0, 2, 0, 0]</text>\n",
       "<text text-anchor=\"middle\" x=\"628.5\" y=\"-15.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">class = 1</text>\n",
       "</g>\n",
       "<!-- 5&#45;&gt;6 -->\n",
       "<g id=\"edge6\" class=\"edge\">\n",
       "<title>5&#45;&gt;6</title>\n",
       "<path fill=\"none\" stroke=\"#000000\" d=\"M700.8374,-126.2286C689.1271,-112.5123 676.1112,-97.2668 664.3845,-83.5313\"/>\n",
       "<polygon fill=\"#000000\" stroke=\"#000000\" points=\"666.8035,-80.9743 657.6485,-75.6416 661.4798,-85.5194 666.8035,-80.9743\"/>\n",
       "</g>\n",
       "<!-- 7 -->\n",
       "<g id=\"node8\" class=\"node\">\n",
       "<title>7</title>\n",
       "<polygon fill=\"#f8ceef\" stroke=\"#000000\" points=\"972,-75.5 753,-75.5 753,-7.5 972,-7.5 972,-75.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"862.5\" y=\"-60.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">gini = 0.791</text>\n",
       "<text text-anchor=\"middle\" x=\"862.5\" y=\"-45.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">samples = 63</text>\n",
       "<text text-anchor=\"middle\" x=\"862.5\" y=\"-30.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">value = [7, 4, 0, 5, 3, 5, 11, 1, 24, 3]</text>\n",
       "<text text-anchor=\"middle\" x=\"862.5\" y=\"-15.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">class = 8</text>\n",
       "</g>\n",
       "<!-- 5&#45;&gt;7 -->\n",
       "<g id=\"edge7\" class=\"edge\">\n",
       "<title>5&#45;&gt;7</title>\n",
       "<path fill=\"none\" stroke=\"#000000\" d=\"M778.1063,-126.2286C792.0238,-112.2559 807.5218,-96.6964 821.4004,-82.7627\"/>\n",
       "<polygon fill=\"#000000\" stroke=\"#000000\" points=\"823.9161,-85.1966 828.4934,-75.6416 818.9565,-80.2567 823.9161,-85.1966\"/>\n",
       "</g>\n",
       "<!-- 9 -->\n",
       "<g id=\"node10\" class=\"node\">\n",
       "<title>9</title>\n",
       "<polygon fill=\"#fdfafe\" stroke=\"#000000\" points=\"1523.5,-217 1167.5,-217 1167.5,-119 1523.5,-119 1523.5,-217\"/>\n",
       "<text text-anchor=\"middle\" x=\"1345.5\" y=\"-201.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">num_corners &lt;= 2.5</text>\n",
       "<text text-anchor=\"middle\" x=\"1345.5\" y=\"-186.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">gini = 0.874</text>\n",
       "<text text-anchor=\"middle\" x=\"1345.5\" y=\"-171.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">samples = 27146</text>\n",
       "<text text-anchor=\"middle\" x=\"1345.5\" y=\"-156.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">value = [564, 542, 2724, 2444, 4340, 2547, 3132, 4923, 1945</text>\n",
       "<text text-anchor=\"middle\" x=\"1345.5\" y=\"-141.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">3985]</text>\n",
       "<text text-anchor=\"middle\" x=\"1345.5\" y=\"-126.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">class = 7</text>\n",
       "</g>\n",
       "<!-- 8&#45;&gt;9 -->\n",
       "<g id=\"edge9\" class=\"edge\">\n",
       "<title>8&#45;&gt;9</title>\n",
       "<path fill=\"none\" stroke=\"#000000\" d=\"M1345.5,-252.8642C1345.5,-244.5802 1345.5,-235.903 1345.5,-227.3918\"/>\n",
       "<polygon fill=\"#000000\" stroke=\"#000000\" points=\"1349.0001,-227.1298 1345.5,-217.1298 1342.0001,-227.1299 1349.0001,-227.1298\"/>\n",
       "</g>\n",
       "<!-- 12 -->\n",
       "<g id=\"node13\" class=\"node\">\n",
       "<title>12</title>\n",
       "<polygon fill=\"#fcf2eb\" stroke=\"#000000\" points=\"2086.5,-209.5 1718.5,-209.5 1718.5,-126.5 2086.5,-126.5 2086.5,-209.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"1902.5\" y=\"-194.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">num_corners &lt;= 2.5</text>\n",
       "<text text-anchor=\"middle\" x=\"1902.5\" y=\"-179.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">gini = 0.846</text>\n",
       "<text text-anchor=\"middle\" x=\"1902.5\" y=\"-164.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">samples = 17529</text>\n",
       "<text text-anchor=\"middle\" x=\"1902.5\" y=\"-149.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">value = [4361, 6, 2242, 2625, 509, 1943, 1787, 233, 2851, 972]</text>\n",
       "<text text-anchor=\"middle\" x=\"1902.5\" y=\"-134.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">class = 0</text>\n",
       "</g>\n",
       "<!-- 8&#45;&gt;12 -->\n",
       "<g id=\"edge12\" class=\"edge\">\n",
       "<title>8&#45;&gt;12</title>\n",
       "<path fill=\"none\" stroke=\"#000000\" d=\"M1526.7268,-258.4014C1588.672,-243.499 1657.9602,-226.83 1720.0049,-211.9037\"/>\n",
       "<polygon fill=\"#000000\" stroke=\"#000000\" points=\"1720.9943,-215.2656 1729.8983,-209.5236 1719.357,-208.4598 1720.9943,-215.2656\"/>\n",
       "</g>\n",
       "<!-- 10 -->\n",
       "<g id=\"node11\" class=\"node\">\n",
       "<title>10</title>\n",
       "<polygon fill=\"#fcf9fe\" stroke=\"#000000\" points=\"1346.5,-83 990.5,-83 990.5,0 1346.5,0 1346.5,-83\"/>\n",
       "<text text-anchor=\"middle\" x=\"1168.5\" y=\"-67.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">gini = 0.858</text>\n",
       "<text text-anchor=\"middle\" x=\"1168.5\" y=\"-52.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">samples = 18568</text>\n",
       "<text text-anchor=\"middle\" x=\"1168.5\" y=\"-37.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">value = [492, 468, 1090, 1019, 1875, 1559, 2815, 4109, 1520</text>\n",
       "<text text-anchor=\"middle\" x=\"1168.5\" y=\"-22.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">3621]</text>\n",
       "<text text-anchor=\"middle\" x=\"1168.5\" y=\"-7.8\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">class = 7</text>\n",
       "</g>\n",
       "<!-- 9&#45;&gt;10 -->\n",
       "<g id=\"edge10\" class=\"edge\">\n",
       "<title>9&#45;&gt;10</title>\n",
       "<path fill=\"none\" stroke=\"#000000\" d=\"M1276.8138,-118.9107C1263.1335,-109.1335 1248.7872,-98.8804 1235.1756,-89.1524\"/>\n",
       "<polygon fill=\"#000000\" stroke=\"#000000\" points=\"1236.9789,-86.1391 1226.8079,-83.1721 1232.9086,-91.8342 1236.9789,-86.1391\"/>\n",
       "</g>\n",
       "<!-- 11 -->\n",
       "<g id=\"node12\" class=\"node\">\n",
       "<title>11</title>\n",
       "<polygon fill=\"#e7fcf8\" stroke=\"#000000\" points=\"1711,-75.5 1364,-75.5 1364,-7.5 1711,-7.5 1711,-75.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"1537.5\" y=\"-60.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">gini = 0.826</text>\n",
       "<text text-anchor=\"middle\" x=\"1537.5\" y=\"-45.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">samples = 8578</text>\n",
       "<text text-anchor=\"middle\" x=\"1537.5\" y=\"-30.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">value = [72, 74, 1634, 1425, 2465, 988, 317, 814, 425, 364]</text>\n",
       "<text text-anchor=\"middle\" x=\"1537.5\" y=\"-15.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">class = 4</text>\n",
       "</g>\n",
       "<!-- 9&#45;&gt;11 -->\n",
       "<g id=\"edge11\" class=\"edge\">\n",
       "<title>9&#45;&gt;11</title>\n",
       "<path fill=\"none\" stroke=\"#000000\" d=\"M1420.007,-118.9107C1438.849,-106.4966 1458.8555,-93.3153 1476.9167,-81.4156\"/>\n",
       "<polygon fill=\"#000000\" stroke=\"#000000\" points=\"1479.1883,-84.1103 1485.6132,-75.6858 1475.3371,-78.265 1479.1883,-84.1103\"/>\n",
       "</g>\n",
       "<!-- 13 -->\n",
       "<g id=\"node14\" class=\"node\">\n",
       "<title>13</title>\n",
       "<polygon fill=\"#fae8db\" stroke=\"#000000\" points=\"2076,-75.5 1729,-75.5 1729,-7.5 2076,-7.5 2076,-75.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"1902.5\" y=\"-60.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">gini = 0.804</text>\n",
       "<text text-anchor=\"middle\" x=\"1902.5\" y=\"-45.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">samples = 8399</text>\n",
       "<text text-anchor=\"middle\" x=\"1902.5\" y=\"-30.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">value = [2807, 1, 447, 479, 200, 538, 1551, 188, 1523, 665]</text>\n",
       "<text text-anchor=\"middle\" x=\"1902.5\" y=\"-15.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">class = 0</text>\n",
       "</g>\n",
       "<!-- 12&#45;&gt;13 -->\n",
       "<g id=\"edge13\" class=\"edge\">\n",
       "<title>12&#45;&gt;13</title>\n",
       "<path fill=\"none\" stroke=\"#000000\" d=\"M1902.5,-126.2286C1902.5,-113.2814 1902.5,-98.9718 1902.5,-85.8557\"/>\n",
       "<polygon fill=\"#000000\" stroke=\"#000000\" points=\"1906.0001,-85.6415 1902.5,-75.6416 1899.0001,-85.6416 1906.0001,-85.6415\"/>\n",
       "</g>\n",
       "<!-- 14 -->\n",
       "<g id=\"node15\" class=\"node\">\n",
       "<title>14</title>\n",
       "<polygon fill=\"#f6fef7\" stroke=\"#000000\" points=\"2447.5,-75.5 2093.5,-75.5 2093.5,-7.5 2447.5,-7.5 2447.5,-75.5\"/>\n",
       "<text text-anchor=\"middle\" x=\"2270.5\" y=\"-60.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">gini = 0.829</text>\n",
       "<text text-anchor=\"middle\" x=\"2270.5\" y=\"-45.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">samples = 9130</text>\n",
       "<text text-anchor=\"middle\" x=\"2270.5\" y=\"-30.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">value = [1554, 5, 1795, 2146, 309, 1405, 236, 45, 1328, 307]</text>\n",
       "<text text-anchor=\"middle\" x=\"2270.5\" y=\"-15.3\" font-family=\"Times,serif\" font-size=\"14.00\" fill=\"#000000\">class = 3</text>\n",
       "</g>\n",
       "<!-- 12&#45;&gt;14 -->\n",
       "<g id=\"edge14\" class=\"edge\">\n",
       "<title>12&#45;&gt;14</title>\n",
       "<path fill=\"none\" stroke=\"#000000\" d=\"M2023.2518,-126.4916C2067.959,-111.1235 2118.3381,-93.8057 2161.8996,-78.8314\"/>\n",
       "<polygon fill=\"#000000\" stroke=\"#000000\" points=\"2163.0545,-82.1355 2171.3736,-75.5747 2160.7789,-75.5157 2163.0545,-82.1355\"/>\n",
       "</g>\n",
       "</g>\n",
       "</svg>\n"
      ],
      "text/plain": [
       "<graphviz.files.Source at 0x7fa723eb9c18>"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dot_data = sklearn.tree.export_graphviz(clf, out_file = None, \n",
    "                                feature_names = [\n",
    "                                    \"avg_lumin\", \"edge_prop\", \"num_corners\", \"max_width\", \"max_height\"\n",
    "                                ],\n",
    "                                class_names = [str(i) for i in range(0, 10)],\n",
    "                                filled=True)\n",
    "\n",
    "# Draw graph\n",
    "graph = graphviz.Source(dot_data, format=\"png\") \n",
    "graph"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/owen/program_files/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/_sag.py:330: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge\n",
      "  \"the coef_ did not converge\", ConvergenceWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.46714\n",
      "[32.66 28.24 -0.21 -0.2  -0.2 ]\n",
      "[ 5.15 -5.61 -0.98 -1.25 -0.01]\n",
      "[-7.46 -6.79  0.75  0.49 -0.04]\n",
      "[ 2.59  4.77  1.04 -0.02  0.04]\n",
      "[-22.35 -19.43   0.66   0.39   0.09]\n",
      "[-2.89  0.31  0.52  0.16 -0.26]\n",
      "[ 4.49  2.87 -0.94  0.1  -0.04]\n",
      "[-21.59 -28.91  -0.31   0.47  -0.06]\n",
      "[16.7  23.96  0.04 -0.3   0.31]\n",
      "[-7.31  0.53 -0.57  0.04  0.2 ]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/owen/program_files/anaconda3/lib/python3.6/site-packages/sklearn/linear_model/_sag.py:330: ConvergenceWarning: The max_iter was reached which means the coef_ did not converge\n",
      "  \"the coef_ did not converge\", ConvergenceWarning)\n"
     ]
    }
   ],
   "source": [
    "clf = sklearn.linear_model.LogisticRegression(solver='saga', \n",
    "                                              penalty='l1',\n",
    "                                              C=10,\n",
    "                                              max_iter=200,\n",
    "                                              multi_class='multinomial').fit(interp_x_train, y_train)\n",
    "clf.fit(interp_x_train, y_train)\n",
    "print(clf.score(interp_x_train, y_train))\n",
    "for c in clf.coef_:\n",
    "    print(np.round(c, decimals=2))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Neural Net Interpretability"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/owen/program_files/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:2: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).\n",
      "  \n"
     ]
    }
   ],
   "source": [
    "x_train, y_train, x_valid, y_valid = map(\n",
    "    torch.tensor, (x_train, y_train, x_valid, y_valid)\n",
    ")\n",
    "n, c = x_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "class MNIST_Net(nn.Module):\n",
    "    def __init__(self):\n",
    "        super().__init__()\n",
    "        self.conv1 = nn.Conv2d(1, 16, kernel_size=3, stride=2, padding=1)\n",
    "        self.conv2 = nn.Conv2d(16, 16, kernel_size=3, stride=2, padding=1)\n",
    "        self.conv3 = nn.Conv2d(16, 10, kernel_size=3, stride=2, padding=1)\n",
    "\n",
    "    def forward(self, xb):\n",
    "        xb = xb.view(-1, 1, 28, 28)\n",
    "        xb = torch.nn.functional.relu(self.conv1(xb))\n",
    "        xb = torch.nn.functional.relu(self.conv2(xb))\n",
    "        xb = torch.nn.functional.relu(self.conv3(xb))\n",
    "        xb = torch.nn.functional.avg_pool2d(xb, 4)\n",
    "        return xb.view(-1, xb.size(1))\n",
    "\n",
    "loss_func = torch.nn.functional.cross_entropy\n",
    "lr = 0.1\n",
    "epochs = 10\n",
    "bs = 64\n",
    "train_data = TensorDataset(x_train, y_train)\n",
    "valid_data = TensorDataset(x_valid, y_valid)\n",
    "\n",
    "def accuracy(out, yb):\n",
    "    preds = torch.argmax(out, dim=1)\n",
    "    return (preds == yb).float().mean()\n",
    "\n",
    "def get_model():\n",
    "    model = MNIST_Net()\n",
    "    opt = optim.SGD(model.parameters(), lr=lr, momentum=0.9)\n",
    "    return(model, opt)\n",
    "\n",
    "def get_data(train_ds, valid_ds, bs):\n",
    "    return (\n",
    "        DataLoader(train_ds, batch_size=bs, shuffle=True),\n",
    "        DataLoader(valid_ds, batch_size=bs * 2),\n",
    "    )\n",
    "\n",
    "def loss_batch(model, loss_func, xb, yb, opt=None):\n",
    "    loss = loss_func(model(xb), yb)\n",
    "\n",
    "    if opt is not None:\n",
    "        loss.backward()\n",
    "        opt.step()\n",
    "        opt.zero_grad()\n",
    "\n",
    "    return loss.item(), len(xb)\n",
    "\n",
    "def fit(epochs, model, loss_func, opt, train_dl, valid_dl):\n",
    "    for epoch in range(epochs):\n",
    "        model.train()\n",
    "        for xb, yb in train_dl:\n",
    "            loss_batch(model, loss_func, xb, yb, opt)\n",
    "\n",
    "        model.eval()\n",
    "        with torch.no_grad():\n",
    "            losses, nums = zip(\n",
    "                *[loss_batch(model, loss_func, xb, yb) for xb, yb in valid_dl]\n",
    "            )\n",
    "        val_loss = np.sum(np.multiply(losses, nums)) / np.sum(nums)\n",
    "\n",
    "        print(epoch, val_loss)\n",
    "    print(accuracy(model.forward(x_valid), y_valid))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 0.29766359733343123\n",
      "1 0.20563192915320397\n",
      "2 0.2096680709719658\n",
      "3 0.1848541338145733\n",
      "4 0.14094552401304244\n",
      "5 0.14855193548202514\n",
      "6 0.14791034851670265\n",
      "7 0.13577543746829032\n",
      "8 0.14206985997855665\n",
      "9 0.15643914679288864\n",
      "tensor(0.9521)\n",
      "CPU times: user 5min 23s, sys: 2min 48s, total: 8min 11s\n",
      "Wall time: 1min 12s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "train_dl, valid_dl = get_data(train_data, valid_data, bs)\n",
    "model, opt = get_model()\n",
    "fit(epochs, model, loss_func, opt, train_dl, valid_dl)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "MNIST_Net(\n",
       "  (conv1): Conv2d(1, 16, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))\n",
       "  (conv2): Conv2d(16, 16, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))\n",
       "  (conv3): Conv2d(16, 10, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))\n",
       ")"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.eval()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor(3)"
      ]
     },
     "execution_count": 92,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "torch.argmax(model.forward(x_valid[0]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 250,
       "width": 253
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "show_mnist(x_valid[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 228,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 250,
       "width": 253
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 250,
       "width": 253
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 250,
       "width": 253
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 250,
       "width": 253
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[ 0.5324,  2.1158, 13.9041, 29.6140,  4.2903, 14.5835,  7.6240, 10.4082,\n",
      "         11.9075,  7.7537]], grad_fn=<ViewBackward>)\n",
      "tensor(3)\n"
     ]
    }
   ],
   "source": [
    "img = x_valid[0].clone()\n",
    "img.requires_grad = True\n",
    "\n",
    "for i in range(3):\n",
    "    logits = model.forward(img)\n",
    "    loss = -(logits[0][3]**2).mean()\n",
    "    loss.backward()\n",
    "    with torch.no_grad():\n",
    "        g = img.grad\n",
    "        g = torch.clamp(g/g.abs().mean(), -1, 1)\n",
    "        show_mnist(-g)\n",
    "        img.data = img - 0.05*g\n",
    "        \n",
    "show_mnist(img.detach())\n",
    "print(model.forward(img))\n",
    "print(torch.argmax(model.forward(img)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.6.7 64-bit ('base': conda)",
   "language": "python",
   "name": "python36764bitbasecondab7a8f070add3487eadfe15e733184a9b"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}