image_classification.ipynb

{
 "metadata": {
  "name": "",
  "signature": "sha256:384c80069ffc85cbc26f550163ed454cde7ae0b160b378bcaa97f7cea660c6a7"
 },
 "nbformat": 3,
 "nbformat_minor": 0,
 "worksheets": [
  {
   "cells": [
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "# Bag of Visual Words Image Classification\n",
      "\n",
      "This notebook walks through a simple bag of words based image classifier.  In this technique, small patches are extracted from an image and the number of each patch \"type\" is used as a feature to a classifier such as an SVM.  The patch types are automatically discovered via clustering of all patches found in the training set.\n",
      "\n",
      "Take a look [here](http://www.stanford.edu/class/cs221/assignments/images/index.html) for a more thorough exposition\n",
      "\n",
      "This notebook requires sklearn.  You can install something like [Enthought Canopy](https://www.enthought.com/downloads/) to get all of the packages without any hassle.\n",
      "\n",
      "<b>This approach is fairly naive (and extremely slow) as is</b>, but it gives an idea of how to implement your own classifiers for use in a pipeline."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "%pylab inline"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "Populating the interactive namespace from numpy and matplotlib\n"
       ]
      }
     ],
     "prompt_number": 1
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# Display given faces in a grid\n",
      "def show_patches(ims, grid_size=(16, 6)):\n",
      "    fig = plt.figure(figsize=grid_size)\n",
      "    for i in range(min(len(ims), 30)):\n",
      "        ax = fig.add_subplot(3, 10, i + 1, xticks=[], yticks=[])\n",
      "        ax.imshow(ims[i].reshape((8, 8)), cmap=plt.cm.bone, interpolation='nearest')"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 2
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "import sklearn\n",
      "from sklearn.pipeline import Pipeline\n",
      "from sklearn import svm"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 3
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "We are using the [CIFAR 10](http://www.cs.toronto.edu/~kriz/cifar.html) dataset - a dataset of 32x32 images from 10 classes"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# Load data\n",
      "import cPickle\n",
      "\n",
      "def load_data(file):\n",
      "    with open(file, 'rb') as f:\n",
      "        data = cPickle.load(f)\n",
      "    return data\n",
      "\n",
      "data = load_data(\"cifar-10-batches-py/data_batch_1\")"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 4
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "images = data['data']\n",
      "# Reshape to go from length 3072 vector to 32x32 rgb images\n",
      "# order='F' deals with specifics of how the data is laid out\n",
      "images = images.reshape((-1, 32, 32, 3), order='F')\n",
      "labels = np.array(data['labels'])\n",
      "\n",
      "def get_classes(classes=[0, 1, 2], per_class=100):\n",
      "    # Array of indices i where labels[i] is in classes\n",
      "    indices = np.concatenate([np.where(labels == c)[0][:per_class] for c in classes])\n",
      "    return images[indices], labels[indices]\n",
      "\n",
      "# For speed, let's consider only 2 classes, 100 images per class for now\n",
      "classes = [0,1,2,3]\n",
      "X, Y = get_classes(classes, 100)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 5
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# That's cars, cats, and birds in this case\n",
      "for c in classes:\n",
      "    plt.imshow(images[labels == c][0], interpolation='nearest')\n",
      "    plt.show()"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "display_data",
       "png": "iVBORw0KGgoAAAANSUhEUgAAAPwAAAD6CAYAAACF8ip6AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAHYlJREFUeJztnX1wVfWdxp+Te28gIS8EITcxiQ0VwmtILgkJutBZVoLD\nOo0gjotbMSPBdWmlQ2FK6R86dHZGwa1VXuyWKtpYZ6hspyBOhUErQozWVEgqiC8oRK8xCS8JIa/k\n5Z79w+Vqlny/SU5yc+P+ns8MM0mee8753XPPwzn3POf7/Vm2bdsghBhBRLgHQAgZPmh4QgyChifE\nIGh4QgyChifEIGh4QkzCdsiBAwfsKVOm2JMmTbI3b97cQwPAf/zHf2H8J2E5yeG7u7sxZcoUvPba\na0hJScGcOXOwe/duTJs2DQBgWRbK3307+Pqnf/sM7v+3VcHf3W63uG5Ni4iQL0hcLpeieYI/b9+2\nHWt+vKZfy1mWJWoqymLWNy6qnnziSaz9ydqvNVu74HI4ln7y5JNPYO3anwR/t6EfFt2BblGz7YCo\nud0eWXO58Ktf/Qrr1q27Rgso6+zukrWOjk5Ru3z5sqi1t7fjuWd34b6VxddoV65cEZfTjqexYxNE\nLSY6RtQsy8K27U/ix2vWXqMFAte+9ylTb4Rka0eX9OXl5Zg0aRLS09Ph8XiwfPlyvPTSS05WRQgZ\nRuTTqUJ1dTXS0tKCv6empuKdd97p8Zqnf/tM8Ocvv6xxODxCSF+8U/5XlJe/0/cL4dDw/bnU/eYl\n/LFjx51sJiTk5eeFewhB5s6dG+4hBBkpY7npppvCPYQg2T5fuIcQJD9P/nzy8+b20Hfs2Ca+1tEl\nfUpKCvx+f/B3v9+P1NRU8fU5ObOdbCYk5Ofnh3sIQebeNDJMBgBz544Mo40kw/t8I+m4HZpjxZHh\nc3Nzcfr0aVRVVaGjowMvvvgiCgsLh2RAhJDQ4eiS3u12Y8eOHbj11lvR3d2N4uLi4B36waJ9XXCq\noY87zs7WqWxNCz60O/jq9kIwlkEhr7epSb77feHCBVGLiZHvVCcnJ8tDUXZNZ6d8l/7ixYui1tjY\nKGrt7e2OxqIlBu4k2YqjRo0StYF+vo4MDwCLFy/G4sWLnS5OCAkDfNKOEIOg4QkxCBqeEIOg4Qkx\nCBqeEINwfJe+L7RCFz1ek5eLiJALE7TtOR2L0+W0wpMI5f9YC/L7cxrL9VZccRWtyMUVoR8aH3zw\nsajt2bNH1D755BNRGz/+OlFbtWqVqGVmZooa0CYq9fX1ovbZZ5+JWmtrq6h1dXWJWktTi6hFR0WL\nWmJioqgNNDrmGZ4Qg6DhCTEIGp4Qg6DhCTEIGp4Qg6DhCTGIsMRyerymafJwtThPj/qGXtNQYxTb\nWbWcVi8VoWyvW25LB5dbiwiB2tpaUXv7Lbn7yuXLTaJ2/pxcSfe5/3NRm5U1U9SsCPn9a7Hchx98\nIGqtLXIs163EcpcuytuLiZVjOU2Ljh4jar3BMzwhBkHDE2IQNDwhBkHDE2IQNDwhBkHDE2IQYYrl\nnEVozivpRAkRSmyjT18la1pjQbW/pcOoTwvmupU4z1JjQOfbHB0VJWptbfI0TW1Kc8jPPjsrau3t\nckWcFlk2NzeLmhY7dimNMbVqSG2cY+LleG2CUi13Q9oNotYbPMMTYhA0PCEGQcMTYhA0PCEGQcMT\nYhA0PCEG4TiWS09PR1xcHFwuFzweD8rLy3voTueB02Iy580oZc3t9oja+++/L2r+L/yi9r353xO1\nuLg4UdOnpJPfu9Y0E9CaWMrLtV9R5k8DcPq03IyyuUmuiPN45P3d3CLPSfeBUr2mzREXFSXHXVq0\nqjX/7OqSywxdlrJcmxznfXL6tKilpaWJ2tj4BFHrDceGtywLb7zxBsaNG+d0FYSQYWZQl/Shm5mU\nEBIKBnWGX7hwIVwuFx544AHcf//9PfT/+vXO4M+5c3IwZ06u81ESQkSOHz+GiuPH+/Vax4YvKytD\ncnIyzp8/j4KCAkydOhXz588P6qt/+IDTVRNCBsDs2TmYPTsn+PuuXc+Ir3V8SZ+cnAwAmDBhApYu\nXXrNTTtCyMjDkeFbW1vR9L93YltaWnDo0KE+pvshhIwEHF3S19XVYenSpQC+mkvrBz/4ARYtWtTj\nNU5jOadzvemRnSipyzU2Noranhf/W9Q87khRW7x4sagFAvJNUO09aJGdhvbez5yRq9MA4OTJE6Km\n38yVYyutAeTnn8lNLD/3yxHpDWnfETUtltP2TbfS/dPWPyiRi0pDzZMn5Xj4+uQUeaW94MjwEydO\nRGVlpZNFCSFhhE/aEWIQNDwhBkHDE2IQNDwhBkHDE2IQIWxi6Wz+OL1Rpbw9y9KiIK2ppBwT5eff\nJGqHXy8VtT/98WVRS//OZFGbOu27oqbFWS6X/DG6lTnitLnlPv3kU2UserVcV1eHqGmNOl1uWTt3\nTq6I++T0GVEbGy8Xd2nxoRYPa8dMtyzp8/wpkawWSX70sVxl1+sYBvRqQsi3GhqeEIOg4QkxCBqe\nEIOg4QkxCBqeEIMIWSynzxHnrOGk0+U8kfLb1CqmYmLkOdKW3blU1DY/+p+i9sILvxe1NT/+d1FL\nS5Orohob5eaP9fWXRK29TY7PGhrk5QCgXZkHTmuA6fE4O+TaWuV52bSIMO0Gee61K1fkee60Y02r\narSVXE6tpHPJWpPy+Z7++GN5nb3AMzwhBkHDE2IQNDwhBkHDE2IQNDwhBkHDE2IQIayWc9ZwUkOL\nSrqVOOTk8VOi1tTULGpRUXIs51Yq1FLTrhe1w4f/ImqJXrmyy+tNFLWjR8tErba2TtQ6O+RyuaYm\nOQoCgHql6aI2f5wWg2qfb5fS4PLMGblaLj1dbmLZ2tIiahoBpVouEFBKELVKUFs5tpXt1VRXy9vr\nbTsDejUh5FsNDU+IQdDwhBgEDU+IQdDwhBiEaviVK1fC6/X2mDeuvr4eBQUFyMjIwKJFi3Dpkl5k\nQQgZOVi20smvtLQUMTExuPfee3HixFdziW3YsAHjx4/Hhg0bsGXLFjQ0NGDz5s09V2pZ+PDjD8SN\nut1ypKVFdqNHjxa18+cviNrDD/2HqJ14T47stHFGjpLnj9PmSLvUKMdZNuTlOjs7Ra29Ta76UhId\nNSbS54cDbCV+0pbUok4tstMq21JS5UrC7OwsUYtU4sMvlbjrslK9FlA6g0Yok8tpDS61z8Idee17\neOudt8Rl1DP8/PnzkZCQ0ONv+/fvR1FREQCgqKgI+/bt01ZBCBlBDPg7fF1dHbxeLwDA6/Wirk5+\nsIMQMrIY1JN2lmWJT0ft2L4j+HNeXh7y8vMGsylCiEDj5UZcvqw/GXmVARve6/WitrYWSUlJqKmp\nQWJi7498PrjmwYGumhDigPi4eMTHxQd/91f7xdcO+JK+sLAQJSUlAICSkhIsWbLEwRAJIeFANfzd\nd9+Nm2++GR999BHS0tLw3HPPYePGjXj11VeRkZGB119/HRs3bhyusRJCBokayzle6SBiOS2aGTVq\nlKhdaZdjm9f+clTUjhyWK81OnJDfw/lz50WtOyDHa1HRcpx3qVGOFrs65XV6lP3ZcUVuVAltbjUl\nQgKAgC3HT1rFmHawuZQ5B91K88v4sfGi9t3vyvP1RUdHi5r2nViLCLsdxnLa/raVvdZbjP23Y+XO\nYjlCyP8vaHhCDIKGJ8QgaHhCDIKGJ8QgaHhCDCJkTSydpn36nF5y6VdUtFyFtXTpbaK28JZ/FLWq\nKrli6q9v/1XUyt56S9ROfyLPBRbTHStqlxoaRE2L7JyiTYMGQA3ttIpH9fNVIq2OK7LWojQibW1t\nFTUt5tWOX+39aceo5gjb4XLWAH3GMzwhBkHDE2IQNDwhBkHDE2IQNDwhBkHDE2IQIYvltPhFC3Wc\nRjoatlLZFT9Wrpjy5UwVtdk500XtloLvidqjjzwmau+dOClqWoTU3tomahpaZaIWE/WFFmlpn2GE\nMh6tiacWvWna2LFj5bE4nBvRUjRbmf9QDdeGsKCVZ3hCDIKGJ8QgaHhCDIKGJ8QgaHhCDIKGJ8Qg\nQhbLaVjW0FfS6XGeHPdoRETI23O75F133XXjRS0mNk7UYmPlarkOpXFiW4scPalVX6ICRCjzzn21\nXmexndbkUdO09xFQxtLa0iIvp0SP2rGmaS7lOOzSok7FEtr2BlqVyjM8IQZBwxNiEDQ8IQZBwxNi\nEDQ8IQahGn7lypXwer3IzMwM/m3Tpk1ITU2Fz+eDz+fDwYMHQz5IQsjQoMZy9913H9asWYN77703\n+DfLsrBu3TqsW7euj1XLcYEeJWiVVvJSWtTncskLdnXJUdC77x4TtTNnzoraF/4vRa2utkbUtGhR\n22cutxw7ag0uu5QYbJRHjzIjlFhSW6/20evFkNrxJC/V1t4ualoFnhq9KVV9WrSoxYBaDBrhsEq0\n13Vp4vz585GQkHDN30Mw/yQhZBhw9B1++/btyMrKQnFxMS5dujTUYyKEhIgBG3716tU4e/YsKisr\nkZycjPXr1/f6uh3bnwr+K3+nfNADJYT0TlPTZXxZUx38pzHgR2sTExODP69atQrf//73e33dg2t+\nNNBVE0IcEBsbh9hvPLL9ZY18D2nAZ/iamq9vOu3du7fHHXxCyMhGPcPffffdOHLkCC5cuIC0tDT8\n4he/wBtvvIHKykpYloWJEydi586dwzVWQsggUQ2/e/fua/62cuXKfq3YacTiNJbTmw7KMcrFi+dF\n7Te/+a2onXhPbjjp8cgNJ91uj6hFx4wRNa/XK2paJV1Tx2VR09IWLbIC9IaT+sxzThMeba43JSbr\nkmNJ7T1OmDBB1K4o+7tdiQH1ijhZ047tgcIn7QgxCBqeEIOg4QkxCBqeEIOg4QkxCBqeEIMIUxNL\nZw0Cnc5X54qQ3+a5uguidvGCXCcQFRUjapGRcizXpcREWuWeGs04bICo7s8+0jPtfWhVYU5jOa1e\nS9s1WvVaZGSkqMXFyc1GGxsbHa1TP3qdHvcDg2d4QgyChifEIGh4QgyChifEIGh4QgyChifEIEIW\nyzmOEhz2y9MTO1mMHBUlalFR0aJ26ZJWhaZ125T/j21raxO1dkXT9pjbI1fndXZ0yOvs42NwPi+b\ntlat4tHZvIJd3XJ82NTcLGqjRsnR6nXXXSdq2n5pUea561aajeox58DgGZ4Qg6DhCTEIGp4Qg6Dh\nCTEIGp4Qg6DhCTGI0FXLKbmO1jxRT4OURoZK3NVty7HG5KmTRW3t+rWi9oc/7BE1rcFlR4fcONEz\narSsKf83e8fEilpjQ72oXWltlTWlUWNfujovm3KKsZTPSZs7UIsQO5X9rUWd2vxxWkVcb1OzXSVq\ntPz5digRaYfSbFNr0tkbPMMTYhA0PCEGQcMTYhA0PCEGQcMTYhCq4f1+PxYsWIAZM2Zg5syZ2LZt\nGwCgvr4eBQUFyMjIwKJFizhHPCHfEixbychqa2tRW1uL7OxsNDc3IycnB/v27cNzzz2H8ePHY8OG\nDdiyZQsaGhqwefPmr1dqWfjwIzma0qqbtCqlqCi5sk2LSiLciqbMSxbhklPLulp5TrrDh98QtZdf\nflnUznz2uahFR8uVey3NchXWxfNyk04tz+qrQutKu7PKPijRGwJyw0lLiWS1CE1eI3B9SoqoFRYW\nitpoJV7TKuK0KFPT1MiuF23Pnj+I0bd6hk9KSkJ2djYAICYmBtOmTUN1dTX279+PoqIiAEBRURH2\n7dunrYYQMkLo93f4qqoqVFRUID8/H3V1dcHZTL1eL+rq6kI2QELI0NGvJ+2am5uxbNkybN26FbGx\nPZ/qsiyr1yerdmz/dfDnvLw5yMufM8ihEkJ649y5czh//ly/Xtun4Ts7O7Fs2TKsWLECS5YsAfDV\nWb22thZJSUmoqalBYmLiNcs9uOaHAxw2IcQJiYmJPTz4/vvK/TNtRbZto7i4GNOnT8fatV8/V15Y\nWIiSkhIAQElJSfA/AkLIyEY9w5eVleGFF17ArFmz4PP5AACPPvooNm7ciLvuugu7du1Ceno69uyR\nC0kIISMH1fDz5s0T45nXXntNXbFWMaXN9+V8zjJlLFCqsJRKK7tLFpO940XtX5ffJWrz/+FmUXvl\n4AFR+/OfZa36c7+oWUrsGB0tz4/ncuvf9tpampRtynOvtSmxlaWU0mmxXIRbfo+2EgM2NcnvQYt5\nU5Q47/JlubmpFr1pDUw1ra+qxv8Ln7QjxCBoeEIMgoYnxCBoeEIMgoYnxCBoeEIMIixzy3mU+c40\ntDhPI0IZi1K4p2JrlV3K/Gnp30kTtfuL7xM17/gJovZsye9FrUap6rNccvQUGSlXLQJ6I1KXW/58\nxypNHpuVSKuzQ46fupX542wlytXirk6lcWRycrKoaVWN7e3tjsbSqjQb1dbZGzzDE2IQNDwhBkHD\nE2IQNDwhBkHDE2IQNDwhBhGyWK6rS4411IaTSnWXbct5l1ZIZ0FezlLmpNOiRR15MAFbjvPcbnl7\nty/5Z1GbNWumqL134pSo/a38uKi9fqRU1ACgvVOOg8YmjBU1p3GmFtl1dcqfoVYR2H5Ffg/NzXJM\nljDuOlHTGly2KRGa1vhTi+W0OK83eIYnxCBoeEIMgoYnxCBoeEIMgoYnxCBoeEIMImSxnNY8UKt6\nsyw5llP6GMIOyKKtLOg8etNQBqqgDcWtNGqcMmWSqGVkZIjawn9aIGqzc7LlwQDY9fvnRa36yy9F\nLS42TtQi4uNFLUqpQmuobxC1znZ5XjYoce3lJjkG1CrixoyRNbUZpRbZKRqbWBJCRGh4QgyChifE\nIGh4QgyChifEIFTD+/1+LFiwADNmzMDMmTOxbds2AMCmTZuQmpoKn88Hn8+HgwcPDstgCSGDQ43l\nPB4PnnjiCWRnZ6O5uRk5OTkoKCiAZVlYt24d1q1bJy6rNTnU5ohzqmnb0+KX0MRyTnE2lm6lAg1K\ndV782DGi9i/L71C3GRklVzw+uXWbqDU1yXPLRY5W4q44ufmlDTmyrP1CnndPa1TZ2CjHchoTJsjN\nRluUefU6O+T4cChjOdXwSUlJSEpKAgDExMRg2rRpqK6uBtCXwQghI5F+P3hTVVWFiooKzJ07F2Vl\nZdi+fTuef/555Obm4vHHH8fYsT1roH/91M7gz3Pm5GBOXu7QjZoQEqSiogKVlRX9em2/DN/c3Iw7\n77wTW7duRUxMDFavXo2HH34YAPDQQw9h/fr12LVrV49lfvijBwY4bEKIE67eS7vK7373nPjaPu/S\nd3Z2YtmyZbjnnnuwZMkSAEBiYiIsy4JlWVi1ahXKy8uHYNiEkFCjGt62bRQXF2P69OlYu3Zt8O81\nNTXBn/fu3YvMzMzQjZAQMmSol/RlZWV44YUXMGvWrOAlwyOPPILdu3ejsrISlmVh4sSJ2Llzp7Ya\nQsgIQTX8vHnzeo3CFi9e3OeKtbv4Tu/whyayk7GUiee0OcuGGy1Z1CoFbaXZpt2tv7/Fty4UteuV\nudf++KeXRO3kqQ9F7WJ9o6i5PHJEGBevNNRUKjrPn5fn5GtqahK1aVOniJrbJceHXV3y/Hha9MZq\nOUKICA1PiEHQ8IQYBA1PiEHQ8IQYBA1PiEGErImlFpNFRGiRnbPoTWuMqRfSaXPLKfPcqVV2TguL\nhr4gSS8GdF4pOCpSPnTm5s0WtZnTp4qa318tahWVfxe1vxw+Kmqf+WtF7Qu/XEk3OkquJPS4PfJy\nHlnriooSNS0FjYyU56sbFclYjhAiQMMTYhA0PCEGQcMTYhA0PCEGQcMTYhAhnFvOWbVcQMknApYS\ny6mRnVZJJ0p9MLxz0mnNNp028BwMlqU1KZUrv8ZEyxHT9GlypVnG5BtFbf68eaL295Mfi9qhQ4dE\nzaPsNm9ioqhp0bEW50XY8gYjIuR42KVova5rQK8mhHyroeEJMQganhCDoOEJMQganhCDoOEJMYgQ\nxnIO549Toje1Wk5pAtjVLWvaOl1K00G1caSSkjmfoisUkV1oUMejVIVpjRy1t5Fyvdw0M+X6FFHL\nzZohag0NDaKWqMwf1x1QBqo0RdViOY2Bzo3IMzwhBkHDE2IQNDwhBkHDE2IQNDwhBqEavr29Hfn5\n+cjOzsb06dPx85//HABQX1+PgoICZGRkYNGiRbh06dKwDJYQMjgsu4/cprW1FdHR0ejq6sK8efPw\ny1/+Evv378f48eOxYcMGbNmyBQ0NDdi8efPXK7UsHDv+trhOl0tOA91KRZHbJWseZX6xMbExohYX\nFydqkZGjRE2LHZ1WClpq9CZKYYnl9DhInexORJ+uz1kOqqS8fe1UUdKqLwNKLKd9EraynFYJ2lus\nnDAuVvzc+7ykj46OBgB0dHSgu7sbCQkJ2L9/P4qKigAARUVF2LdvX1+rIYSMAPp88CYQCGD27Nn4\n9NNPsXr1asyYMQN1dXXwer0AAK/Xi7q6umuW27nzmeDPOTmzkZsrty4mhDjnzTdL8WZZab9e2+cl\n/VUaGxtx66234tFHH8Udd9zR40mkcePGob6+/uuV8pLekcZLenWLyjp5Sf9NBnVJf5X4+Hjcdttt\nOHbsGLxeL2prv2rwX1NTg0SlAwghZOSgGv7ChQvBO/BtbW149dVX4fP5UFhYiJKSEgBASUkJlixZ\nEvqREkIGjfodvqamBkVFRQgEAggEAlixYgVuueUW+Hw+3HXXXdi1axfS09OxZ8+e4RovIWQQ9Ps7\n/IBWaln427tvibpWhaZ9h9eW077DR4+JFbX4+HhRi4qKFjWn39Mdf4cXlfCgfYe3lPn6nO83ZTAO\nj2D1+7a2QbUa0GHzVu3ehtaEtZfv8GOH4js8IeTbz7AY/tix48OxmX7x9ltyejDcHDlyJNxDCDJS\nxjJSxgEAR4/Ks9ION6VvDs1YhsnwFcOxmX7x9tsjx/BHj4ycA+rICBnLSBkHABwtHTljefPN/uXs\nfcFLekIMgoYnxCBCdpeeEBI+JFuHpIllOBonEkL6hpf0hBgEDU+IQYTc8AcPHsTUqVMxefJkbNmy\nJdSbU0lPT8esWbPg8/mQl5c3rNteuXIlvF4vMjMzg38LV+eg3sayadMmpKamwufzwefz4eDBgyEf\nh9/vx4IFCzBjxgzMnDkT27ZtAxCe/SKNJRz7JaSdpuwQ0tXVZd9444322bNn7Y6ODjsrK8s+depU\nKDepkp6ebl+8eDEs2z569Kh9/Phxe+bMmcG//fSnP7W3bNli27Ztb9682f7Zz34WtrFs2rTJfvzx\nx4dl+1epqamxKyoqbNu27aamJjsjI8M+depUWPaLNJZw7Bfbtu2Wlhbbtm27s7PTzs/Pt0tLS4dk\nv4T0DF9eXo5JkyYhPT0dHo8Hy5cvx0svvRTKTfaJHaYbivPnz0dCQkKPv4Wrc1BvYwGGf98kJSUh\nOzsbABATE4Np06ahuro6LPtFGgsQnmMmVJ2mQmr46upqpKWlBX9PTU0N7sRwYFkWFi5ciNzcXDz9\n9NNhG8dV+tM5aDjZvn07srKyUFxcPOyNSauqqlBRUYH8/Pyw75erY5k7dy6A8OyXQCCA7OxseL3e\n4FeNodgvITX8SMvjy8rKUFFRgQMHDuCpp55CaenQPK44FFiWFdb9tXr1apw9exaVlZVITk7G+vXr\nh23bzc3NWLZsGbZu3YrY2J6VjcO9X5qbm3HnnXdi69atiImJCdt+iYiIQGVlJb744gscPXoUhw8f\n7qE73S8hNXxKSgr8fn/wd7/fj9TU1FBuUiU5+atJBydMmIClS5eivLw8bGMBMKI6ByUmJgYPolWr\nVg3bvuns7MSyZcuwYsWKYCOVcO2Xq2O55557gmMJ1365ylB3mgqp4XNzc3H69GlUVVWho6MDL774\nIgoLC0O5SZHW1lY0NTUBAFpaWnDo0KEed6nDwUjqHFRTUxP8ee/evcOyb2zbRnFxMaZPn461a9cG\n/x6O/SKNJRz7JaSdpobqrqLEK6+8YmdkZNg33nij/cgjj4R6cyJnzpyxs7Ky7KysLHvGjBnDPpbl\ny5fbycnJtsfjsVNTU+1nn33Wvnjxon3LLbfYkydPtgsKCuyGhoawjGXXrl32ihUr7MzMTHvWrFn2\n7bffbtfW1oZ8HKWlpbZlWXZWVpadnZ1tZ2dn2wcOHAjLfultLK+88kpY9st7771n+3w+Oysry87M\nzLQfe+wx27btIdkvIXmWnhAyMuGTdoQYBA1PiEHQ8IQYBA1PiEHQ8IQYBA1PiEHQ8IQYxP8A+cYi\nBNQ6gWkAAAAASUVORK5CYII=\n",
       "text": [
        "<matplotlib.figure.Figure at 0x10aba8810>"
       ]
      },
      {
       "metadata": {},
       "output_type": "display_data",
       "png": "iVBORw0KGgoAAAANSUhEUgAAAPwAAAD6CAYAAACF8ip6AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nY3AAAAAASUVORK5CYII=\n",
       "text": [
        "<matplotlib.figure.Figure at 0x10f744290>"
       ]
      },
      {
       "metadata": {},
       "output_type": "display_data",
       "png": "iVBORw0KGgoAAAANSUhEUgAAAPwAAAD6CAYAAACF8ip6AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nHy4DvYQzgVSQAAAAAElFTkSuQmCC\n",
       "text": [
        "<matplotlib.figure.Figure at 0x108329110>"
       ]
      },
      {
       "metadata": {},
       "output_type": "display_data",
       "png": "iVBORw0KGgoAAAANSUhEUgAAAPwAAAD6CAYAAACF8ip6AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAHv1JREFUeJztnWtsVHea5p9TF18K34ldNraT4hITMMau4IRMD2iGTkyE\nWk2DiLJkO8QKJsoy2kgItmn6Q1r0l7RpKYqA7tmNsnTkVrQokVYQPgQmYRIIcffECeDJ9pIhJNjE\nNrbbxvdr3c58yFAJjd/Xdtnlcub//CRLVfWcy7/+dV6fU/Wc930t27ZtEEKMwJHoARBC5g4GPCEG\nwYAnxCAY8IQYBAOeEINgwBNiEnaMnD592l6+fLm9bNkyu7a29g4NAP/4x78E/klYsfjw4XAYy5cv\nx9mzZ1FYWIiHHnoIx48fx4oVKwAAlmXhxRf/R3T58+f/iL/7ux9En2fnLhK3XbJ8hahlp6eL2uf/\n719FraevK/r4vffOoarq76PPMzIzxfXgShalsVBE1LLuyRO14mJf9HHda6+h+rnnos89zhR5f4Mj\nojY+NipqIyO9otbwyfno448++hesW/dI9PlXV/8srgcA3a1tojYwOiZqo4GgqIVG+3GrP4CFmUl3\naffe4xHXW1p8j6gtyFwgapn35IvaeAi48PFVrF+7/C5tdFj+LMaGBkUtEpLnxYI8L7BtfPp5NypX\n3P0+LcfdF+n/8/9egRTWMV3SNzQ0YNmyZfD5fHC73di+fTvefvvtWDZFCJlDXLGs1NbWhuLi4ujz\noqIifPzxx3csc/78H6OP+/r6YxweIWQy2rqGcbNLvur4LjEFvGVZky7z3Uv45uaWWHYTF5Ys8SV6\nCFHKH3ww0UOIcu+9RYkeAgAgNdmZ6CFEubdwYaKHEGWR8pWmMHcBCnO//ery6edd4rIxXdIXFhai\npeXbIG5paUFRkXzA+HzFojbXLF3qS/QQolSsWZPoIUSZLwHvSZk/AX9fkfzbwFyzKFcO+OkQU8BX\nVlbi2rVraG5uRiAQwJtvvonNmzfPyoAIIfEjpkt6l8uF3/72t3j88ccRDodRU1MT/YX+Nv0d7eL6\nuQu9onbt6jVRW7LsflHLWCj/Mv7nxo9FrWlkSNSysnNELSlF/o/bee1zUfvikz+KWsF3fsH/awoX\nyVdJLqf8Md5oluez7esmUevp7hY1AIiE5a91HrfsNjgRkrepnH8W5sjz7UlPEzVHcpaofd0u/6Le\nevMvohYMak6D7JjkZLhFLTfnbmfiNi6XPC9O5/SuiGIKeADYtGkTNm3aFOvqhJAEwDvtCDEIBjwh\nBsGAJ8QgGPCEGAQDnhCDiPlX+skYG5YTBe4tuk/UuvsVq6TlhqilKtaFW0l06e7oFLXO682i5nTJ\nU+dSrJJb/fJtxsFUOVknLUu2CLMzskUtNUW2e9wuOW9qyRL5MwIAROT3P9DXJ2o93TflbYblzzAr\nW7Zy3Qtk621E2eaNdvmOtNYO5W41S96mFZaPNZdL1nIXypadZsslJcnrTQTP8IQYBAOeEINgwBNi\nEAx4QgyCAU+IQTDgCTGIuNlyrmS5llh2jpxn7EmXa8z19MgWWu8tObspf7FsMY1G5MynlFG5ikgo\nGBC1MWW90d5xeZvj8lhGO+UsrJBS0y4rQ64DODwk22fJqXLGGwCkemQrzKWsm+uV68hZtpxJ5/Zk\niNqNm3JmX8Qpn9PCSiGXQES2LMPKZ++EvN64/PYQiYRFza1kH6ak0JYjhAgw4AkxCAY8IQbBgCfE\nIBjwhBgEA54Qg4ibLYcJWuDcJmDJVkJyqqxlKJbd6KhslQSUbLl0pS3UQOvXoma7lOKBSfK0Fvnu\nFbXxgGzp9PQqzTyUZmHjSsFFV0qqqI0FFA8JwF+6WkXNo9hyhYvk+dayycbC8ud7a3BA3qYlZwsG\nRjWfTDsXyuMMh+Vt2hF5LA7ImZJup6y5nLTlCCECDHhCDIIBT4hBMOAJMQgGPCEGwYAnxCBituV8\nPh8yMjLgdDrhdrvR0NBwh7502VJxXcsl2xNjY3Kvt2TFRsrLLxQ1T5qcuae1vu7slLPzOpTeY26X\nvE2Hsr++Abnw5/CInGU3ovQzc8gJcUhLk3uyBQLyWAAgLVVe160cVSnJ8mefXyBn0nV1y/M9Mixb\ndlr76THFyg0ptqRLycCzLHl/bsVCczq0UJSPGUvRJiLmgLcsC+fOnUNOjlxNlRAyv5jRJb1tK3d8\nEELmHTM6wz/22GNwOp14/vnn8dxzz92h/9PZ96OPly5ZjGVLFsc+SkKISHNbH5pvKndifoeYA76+\nvh4FBQXo6upCVVUVHnjgAaxfvz6qP/7YD2PdNCFkGvgKs+Ar/LYC0blP5VvCY76kLygoAADk5uZi\n69atd/1oRwiZf8QU8CMjIxgc/KYl1PDwMN59912UlZXN6sAIIbNPTJf0nZ2d2Lp1KwAgFArhpz/9\nKTZu3HjHMkuWyrac9mNfUpJs2yQ5lR8JHXIRQI+nQNRSU+TMrouXLonaV00t8jaT5Wl1KFmEQVu2\nWEIh2SZKUjL3chfKLkpRgTwvbqWfGQAMDvWK2viYbCH23JILTubdI/ePu69QPp667u0RtcFh+btt\nn9L/0LNAtnKdyvFrB+X3nuSaXmZbdJvK/iIROXNvImIK+MWLF6OxsTGWVQkhCYR32hFiEAx4QgyC\nAU+IQTDgCTEIBjwhBhG/IpaKlTCm9EKLBGRtQZJmacmarfxfW5i7SNRWriwXtT/96V9EbUjJbFOS\n5eDW+rk55BVHx2R7KRCQM8L6e+RUOguyzQkA7iR5PGke2dLq7rolaqe/ekfUigvl4p/5hbK9OByU\n+/xZSlqfSyl86lLcYZfyOSU5ZPtUs9fCYfmz0LSJ4BmeEINgwBNiEAx4QgyCAU+IQTDgCTEIBjwh\nBhE3W66tVc4mW57vE7XhcdlGGR3T+sfJXonisMClZDB5FmSImkMpSBhS+rnZtjwYKyxrkbC8zYiy\nv9EReT57FUvHqfmHAMYUK1C1XZX3sahAtkhz78kSta6uDlG73twkauNBxe4KKhmdWuFIRQpH5GNG\ns+UCSkFNe5rZcjzDE2IQDHhCDIIBT4hBMOAJMQgGPCEGwYAnxCDiZstd+eQTUVv6wGpRy82TM5/s\nsNJjKyLbKCOjg6I2PirbVoFx2XpyOeWpiyi98yK2bLGEleKPkbDW60yel4Bi2aWmJova4IhsrQGA\n0pYN9ri8T49DGY9S4zGk2HnXm2ULuKdnQNSUQwa2IkaS5Xmzk+WMx36lEGnGiDyhWt85j1vutzgR\nPMMTYhAMeEIMggFPiEEw4AkxCAY8IQahBvzOnTvh9Xrv6BvX09ODqqoqlJSUYOPGjejrk+uiEULm\nF6ot9+yzz+KFF17AM888E32ttrYWVVVV2L9/Pw4dOoTa2lrU1tbetW7z9evidq9/8YWo/U2hT9QU\nVwMOxUZRagciNVW20JR6hMjNkbO3Bvrk/mmhgJyhFXLIvlQA8vvTEtsCAdnO6lX+WdvapAFI9ciZ\nhEFbtpi0Qo4jSjbkzf//uai1dXSJWkg5LlwupQegWz4XBkLKZ2jL1mqKUjTTnSwX/kzP8MjrKZbs\nRKhn+PXr1yM7O/uO106dOoXq6moAQHV1NU6ePDmtHRJCEse0v8N3dnbC6/2my6fX60VnZ+esD4oQ\nEh9mdKedZVmwhOvJa23t0cc56WlYmJE+k10RQgRudPTjRod8R+F3mXbAe71edHR0ID8/H+3t7cjL\ny5twufuV5gCEkNnjvvxM3JefGX3+YaN8q/G0L+k3b96Muro6AEBdXR22bNkSwxAJIYlADfinnnoK\nP/jBD3D16lUUFxfj9ddfx4EDB/Dee++hpKQE77//Pg4cODBXYyWEzBD1kv748eMTvn727NlJNxxW\n/pX09/aIWtvXzaKWmZktam7F8ggGZKvk2hey3fPZpQZRQ1DOpAuPDYvaAqV/nO2Us7B6B+TvaA6H\nZj3JNphD+YySU/QsrP7+XlGzQ7K9lrcoX9SGlaKS3X1yxiMszUKUbauIkn2p1BpVtxlQvONQirye\nQ/GAI4rNGWJvOUKIBAOeEINgwBNiEAx4QgyCAU+IQTDgCTGIuBWxdCg2WV+PbMuN9MsZXIFxubBi\ndracvfVv/3pR1P700QVR6x+Qx+JS/lVmZimZdIOyveRKlrPlkpJkbWxMLsTpdMoDDSm97ELDig0G\nwA5p/fPk9Tq6++V9KpltAcUmsxVbztJsOaUvm6ZByVy0NctOeRPjY7K9ZkF5fw72liOECDDgCTEI\nBjwhBsGAJ8QgGPCEGAQDnhCDiJstl5wqF977+kaTqP3N364TtVBYzhqqr/9A1C6de1fUHEq2UUgp\nVjg4INtWAcVeGhyWM/dc44o1o2RTOSz5/3YwKGdvjY7KWX32JFlYnlS56GJY8eV6B4ZELaKcfywl\ntU/rA2crY9GcN8VZnESUCQUV21E+LADFllNc1wnhGZ4Qg2DAE2IQDHhCDIIBT4hBMOAJMQgGPCEG\nETdbLitdzl5r+Vq25do720SteOlSUUtPlQtALlggW0h93XIfuIiSTbYwPU3UxsZljyVVsZfGbVkb\nHZctSa25nGYtOiH31QtElEZ+AAZGFTtTy8JTNK0Yp2VpmtZfTTmnWfJYLM0iVHYXVjPwYlvPcshz\nrRW/nHD5aS1NCPlew4AnxCAY8IQYBAOeEINgwBNiEGrA79y5E16vF2VlZdHXDh48iKKiIvj9fvj9\nfpw5cybugySEzA6qLffss8/ihRdewDPPPBN9zbIs7N27F3v37lU3nJGZKWouxUlIcsv2WmqKXBxy\nZenDolaQK7euvtkqt9bVDI+kZNnSGhqSM8IuXrokan++dk3e5rjcyy4QkC07ze7Rqk0qCWjf6CFt\nAW3mlH5ucUhR07LlNByKfar3gZP3p30W2jhVS1JrEDgB6tLr169HdvbdDRxjnURCSGKJ6Tv80aNH\nUV5ejpqaGvT1yaWcCSHzi2kH/O7du9HU1ITGxkYUFBRg3759Ey738eXPon+t7Z0zHighZGKabw7g\n3MXW6J/GtG+tzcvLiz7etWsXfvzjH0+43Fr/6ulumhASA75FGfAt+vZW9vOX5NvTp32Gb29vjz4+\nceLEHb/gE0LmN+oZ/qmnnsL58+fR3d2N4uJi/OpXv8K5c+fQ2NgIy7KwePFivPrqq3M1VkLIDFED\n/vjx43e9tnPnziltWLPlUpU+aVlp6aKW5pFtuXGnXDTTW5wqasnpC0VNtUoUO8TllKe1+S9yb7Vb\nFxtFbTyoVjmUidlR0bOwHErKmKUU1dQz22Yf7TOM1W2K0emcwViUuZ7kc/preKcdIQbBgCfEIBjw\nhBgEA54Qg2DAE2IQDHhCDCJuRSyzc+5OurlNODAian/+7BNRu6d4sagFlFZoSW75bSZ55GKUwUBQ\n1MIReYdDY7KFlp4pz4snVR6LVoxStXtibYQ2KVrWW2z7jIdlp21T7zs3+/OmWWgOh9w/Tp2XaU4Z\nz/CEGAQDnhCDYMATYhAMeEIMggFPiEEw4AkxiLjZcpmZOaLmdskWxPl/fkfURhQbJSM3X9TKVpaL\nWpZikw3Zw6IWUXzADKWv3gKP3OcuySV/HE7lf3PElseiZXZpbs9kplSsdp9mMcVqocWK6nYpmlbE\n0rZlzemU30OSWymaOZMP6q+3Nb3FCSHfZxjwhBgEA54Qg2DAE2IQDHhCDIIBT4hBxM2W86TJdldK\nipJJ95frovbuiTdFLeiSC1W6dvyDqK3/+x+Kmkeuiwm3W+4tFwnKXliSJRfw1CwdRYKt+kuKbxOj\nRQZMO0lrytuNBdWyU96/y6kU21T8Lgvy5xtRZibJJdunqUlyZqZ2Vp6uW8kzPCEGwYAnxCAY8IQY\nBAOeEINgwBNiEGrAt7S0YMOGDSgtLcWqVatw5MgRAEBPTw+qqqpQUlKCjRs3skc8Id8TVFvO7Xbj\nlVdeQUVFBYaGhrBmzRpUVVXh9ddfR1VVFfbv349Dhw6htrYWtbW1d6ybni73lnMlp4ialonkDMke\nRG9vr6i1tdyIaX/Jbtnqy8ySrcXujk5R++KLfxO10VG5uKfWyy7mlLiZ2HIxZrbphSPl96HtT81e\nU95GxI6tSZwTsr2mnUEzF8jHfZonWdRizeqbCPUMn5+fj4qKim8GlJaGFStWoK2tDadOnUJ1dTUA\noLq6GidPnpzeXgkhCWHK3+Gbm5tx+fJlrF27Fp2dnfB6vQAAr9eLzk75jEYImT9M6U67oaEhbNu2\nDYcPH0Z6+p3tnC3LmvBy6/+89e1dcWWlpSgrXTXDoRJCJqL55gCa2wemtOykAR8MBrFt2zbs2LED\nW7ZsAfDNWb2jowP5+flob29HXl7eXev91yf/yzSHTQiJBd+iDPgWfVtl6dylVnFZ9ZLetm3U1NRg\n5cqV2LNnT/T1zZs3o66uDgBQV1cX/UdACJnfqGf4+vp6vPHGG1i9ejX8fj8A4Ne//jUOHDiAJ598\nEseOHYPP58Nbb701J4MlhMwMNeDXrVsnWiVnz55VN5yaJqeapaTLhRy1zK8kp5xp5nHJFkt/X7eo\njY7JhSpTF8jFKCNheX/Dw4Oi1tvXJW9TKUapFTJU2urFbJHFI6ttMvRClYr1pmS2uTTLTjn8w2HN\nepP3px2H3hzZqvakyvGiW5JyQdgJl5/W0oSQ7zUMeEIMggFPiEEw4AkxCAY8IQbBgCfEIOJWxDJF\nqQCZmp4mas5kuTikOxIQNY9i2Q0O9IjayIhsoWVm330H4W3GxsZE7eoXn4tac/M1UbOV7C0tk0wj\nEfZarOh955QVlbmJaMUo3XKGmlOxuxxh+bP3uOWBZqYqvQO1gppWjH3nJlp+WksTQr7XMOAJMQgG\nPCEGwYAnxCAY8IQYBAOeEIOIny2XKmfEZS3MFbWkJKVn26hsyyW75LcyNCBX1R0e7Be1lCTZ6vvi\nqmy9ffrJx6LW23dL1JyKtRhREsnUjDjFsppmW7Ip73MGW1W02Cw79f0rGY9a0VCnkoGXoRSjTEmS\nrT6nsj/VWlUsu4ngGZ4Qg2DAE2IQDHhCDIIBT4hBMOAJMQgGPCEGETdbzpkk92XLylkkrxdRCvaF\ng6JmuWQ7ZHhYLlTZ2Sr3nUtyyDZZ/bn3Re369auiFlYKVQJKhpal9GTTMsIUCylexFo4UzflYuuf\nF1HsPKdWMFU5F6YqFrDHo/RNdClZb05lnIpdCxaxJIRIMOAJMQgGPCEGwYAnxCAY8IQYhBrwLS0t\n2LBhA0pLS7Fq1SocOXIEAHDw4EEUFRXB7/fD7/fjzJkzczJYQsjMUG05t9uNV155BRUVFRgaGsKa\nNWtQVVUFy7Kwd+9e7N27V1xXs0Mys7LlfSrZcmNqnzCZ0dFRUWtuahK1gf4Reb3m66I2PDQkaraW\n9uaILQPNkQDrTetppmV+hbWCk2qhTsV6U6xOrQBkROkfZys2mcul2KdOWdOy3pwOpcClVlBzmrac\nGvD5+fnIz88HAKSlpWHFihVoa2sDEK/0SEJIPJnyjTfNzc24fPkyHnnkEdTX1+Po0aP4wx/+gMrK\nSrz88svIysq6Y/lX/9dvo4/XVD6MysqHZ2/UhJAo11t70NTWO6VlpxTwQ0NDeOKJJ3D48GGkpaVh\n9+7d+OUvfwkAePHFF7Fv3z4cO3bsjnWe/2//fZrDJoTEwpKiHCwpyok+/+eGr8RlJ/2VPhgMYtu2\nbXj66aexZcsWAEBeXh4sy4JlWdi1axcaGhpmYdiEkHijBrxt26ipqcHKlSuxZ8+e6Ovt7e3RxydO\nnEBZWVn8RkgImTXUS/r6+nq88cYbWL16Nfx+PwDgpZdewvHjx9HY2AjLsrB48WK8+uqrczJYQsjM\nUAN+3bp1E1olmzZtmnTDmsGSlpkhDyhZznqLaFlRimkQDMpZdte+/FLUfvjDpaK2aFGhqDXdkL9D\nKe6S6nyEtYKLMfaPm0nfOW2sEUVzKJadpmljtW15vWA4JGqhcGxZfZYlW2EuJZPO6ZQ1h5KZqWrK\nWCZcflpLE0K+1zDgCTEIBjwhBsGAJ8QgGPCEGAQDnhCDiFsRS8UpQbpiy3kWyMUvuxSzz9bsLiWX\nbsECuQee1+sVte5euUec2y1Pa1jJ0HIotk08mEkClG4hatlr8cjsU8ZiK8eMctBomXshxebVrEWX\nSylGqZx7LUuz82jLEUIEGPCEGAQDnhCDYMATYhAMeEIMggFPiEHMrQ/0H6SmpYta5kLZCrO/vCZr\niv2iFTIsKioWtbR02T601aKKSmaXlkaoFE7Uy3TGRqw94GaCakvGmEmnpSBq9T0txTuOKNmJIUXT\nj0OtCKuWYai8P3VeJlh+WksTQr7XMOAJMQgGPCEGwYAnxCAY8IQYBAOeEIOIny2nFAh0JntELT2n\nSN6mU14PgUFRSnLJGXgLFRvQlSLvz5UiZ9k5nHJ/PDuiVrGUNYVE2GuxEqv15lR6tmmFQR3KIa5l\noUUC4/JGIWfLacmAltY7UDn1Woot51St3GnthhDynw0GPCEGwYAnxCAY8IQYBAOeEINQA35sbAxr\n165FRUUFVq5ciV/84hcAgJ6eHlRVVaGkpAQbN25EX1/fnAyWEDIzVFsuJSUFH3zwATweD0KhENat\nW4ePPvoIp06dQlVVFfbv349Dhw6htrYWtbW1d6yrFQF0Kv23srNzRM2pFOzT7Kckt1w8MDVVtuy0\n9+BStqlZSLFaaKq9pq0nrzUj4lGMUnuPoZDcI04josyAZcmfr0NNa1SOC2eM1qJy7rW187JWLXZa\ne/kPPJ5vvOhAIIBwOIzs7GycOnUK1dXVAIDq6mqcPHlyWjslhCSGSW+8iUQiePDBB/HVV19h9+7d\nKC0tRWdnZ7SEs9frRWdn513r/e9X/zH6+ME1D+HByodmcdiEkNt82dKFL1u6prTspAHvcDjQ2NiI\n/v5+PP744/jggw/u0C3LmvDybtfz/zDF4RJCZsKy4lwsK86NPv+nP14Rl53yF4DMzEz86Ec/wsWL\nF+H1etHR0QEAaG9vR15e3gyGSwiZK9SA7+7ujv4CPzo6ivfeew9+vx+bN29GXV0dAKCurg5btmyJ\n/0gJITNGvaRvb29HdXU1IpEIIpEIduzYgUcffRR+vx9PPvkkjh07Bp/Ph7feemuuxksImQFqwJeV\nleHSpUt3vZ6Tk4OzZ8+qG3Yo9oRLsScWLlwoasnJyaI2OCLbL7edhonIyZFtwJSUFFHT+otpFlKs\nxRg1F0y13mK0+iYrjqjZcpqdqRZyjDWzT3UIlW1aWp8/JdtTyVBzJ2m2nGK9RTTbVTkupnnvHO+0\nI8Qg5iTgL37aMBe7mRK9g0OJHkKUQDC2m0niwXwZSyAgXznNNePj82csX7VOzXabjDkJ+EuffjIX\nu5kSfYPDiR5ClPkSZAAQnCdjmU9zMh6YP2O53to9K9vhJT0hBsGAJ8QgLDsORc/ikVhBCJk6UljH\npYjlfCucSAj5Bl7SE2IQDHhCDCLuAX/mzBk88MADuP/++3Ho0KF4707F5/Nh9erV8Pv9ePjhh+d0\n3zt37oTX60VZWVn0tURVDppoLAcPHkRRURH8fj/8fj/OnDkT93G0tLRgw4YNKC0txapVq3DkyBEA\niZkXaSyJmJe4Vpqy40goFLKXLl1qNzU12YFAwC4vL7evXLkSz12q+Hw++9atWwnZ94cffmhfunTJ\nXrVqVfS1n/3sZ/ahQ4ds27bt2tpa++c//3nCxnLw4EH75ZdfnpP936a9vd2+fPmybdu2PTg4aJeU\nlNhXrlxJyLxIY0nEvNi2bQ8PD9u2bdvBYNBeu3atfeHChVmZl7ie4RsaGrBs2TL4fD643W5s374d\nb7/9djx3OSl2gn5QXL9+PbKzs+94LVGVgyYaCzD3c5Ofn4+KigoAQFpaGlasWIG2traEzIs0FiAx\nx0y8Kk3FNeDb2tpQXFwcfV5UVBSdxERgWRYee+wxVFZW4rXXXkvYOG4zlcpBc8nRo0dRXl6Ompqa\nOS9M2tzcjMuXL2Pt2rUJn5fbY3nkkUcAJGZeIpEIKioq4PV6o181ZmNe4hrw882Pr6+vx+XLl3H6\n9Gn87ne/w4ULFxI9pChS5aC5Yvfu3WhqakJjYyMKCgqwb9++Odv30NAQtm3bhsOHDyM9Pf0Oba7n\nZWhoCE888QQOHz6MtLS0hM3L7UpTra2t+PDDD6dcaWrS7c7WACeisLAQLS0t0ectLS0oKlKaRcaZ\ngoICAEBubi62bt2KhobEJvXMp8pBeXl50YNo165dczY3wWAQ27Ztw44dO6KFVBI1L7fH8vTTT0fH\nkqh5uc1sV5qKa8BXVlbi2rVraG5uRiAQwJtvvonNmzfHc5ciIyMjGBz8psPs8PAw3n333Tt+pU4E\n86lyUHt7e/TxiRMn5mRubNtGTU0NVq5ciT179kRfT8S8SGNJxLzEtdLUbP2qKPHOO+/YJSUl9tKl\nS+2XXnop3rsTuX79ul1eXm6Xl5fbpaWlcz6W7du32wUFBbbb7baLiors3//+9/atW7fsRx991L7/\n/vvtqqoqu7e3NyFjOXbsmL1jxw67rKzMXr16tf2Tn/zE7ujoiPs4Lly4YFuWZZeXl9sVFRV2RUWF\nffr06YTMy0RjeeeddxIyL5999pnt9/vt8vJyu6yszP7Nb35j27Y9K/MSl3vpCSHzE95pR4hBMOAJ\nMQgGPCEGwYAnxCAY8IQYBAOeEINgwBNiEP8Olfalq6oHmAwAAAAASUVORK5CYII=\n",
       "text": [
        "<matplotlib.figure.Figure at 0x10f734390>"
       ]
      }
     ],
     "prompt_number": 6
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "from sklearn.cross_validation import KFold\n",
      "from sklearn.metrics import accuracy_score\n",
      "\n",
      "def score(clf, X, Y, folds=2, verbose=False):\n",
      "    predictions = np.zeros(len(Y))\n",
      "    for i, (train, test) in enumerate(KFold(len(X), n_folds=folds, shuffle=True)):\n",
      "        clf.fit(X[train], Y[train])\n",
      "        predictions[test] = clf.predict(X[test])\n",
      "        if verbose:\n",
      "            print(\"Fold {}: {}\".format(i + 1, accuracy_score(Y[test], predictions[test])))\n",
      "    return accuracy_score(Y, predictions)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 7
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "### The pipeline begins with the patch extractor.  This extracts small (8x8) patches from the image."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "from sklearn.feature_extraction import image\n",
      "def rgb2gray(rgb):\n",
      "    return np.dot(rgb[...,:3], [0.299, 0.587, 0.144])\n",
      "\n",
      "# Used to whiten patches - reduces variance, helps with classification\n",
      "from sklearn.decomposition import RandomizedPCA\n",
      "\n",
      "class PatchExtractor(sklearn.base.BaseEstimator):\n",
      "    def __init__(self, patch_size=(8, 8)):\n",
      "        self.patch_size = patch_size\n",
      "    \n",
      "    def _extract_patches(self, x):\n",
      "        \"\"\" Extracts patches from given H x W image \"\"\"\n",
      "        # Extract patches in a grid, reshape to proper size\n",
      "        # Details not really important - written this way for speed\n",
      "        return image.extract_patches(x, self.patch_size,\n",
      "                                     self.patch_size[0]).reshape((-1,\n",
      "                                                                  self.patch_size[0] * self.patch_size[1]))\n",
      "    \n",
      "    def fit(self, X, Y=None):\n",
      "        patches = np.concatenate([self._extract_patches(rgb2gray(x)) for x in X])\n",
      "        print(patches.shape)\n",
      "        #self.pca = RandomizedPCA(whiten=True)\n",
      "        #self.pca.fit(patches)\n",
      "        return self\n",
      "    \n",
      "    def transform(self, X, Y=None):\n",
      "        return np.array([self._extract_patches(rgb2gray(x)) for x in X])\n",
      "        #return np.array([self.pca.transform(self._extract_patches(rgb2gray(x))) for x in X])"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 8
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "### The next step is taking a collection of patches and creating a histogram of patch types.\n",
      "\n",
      "Patch types are automatically discovered during training via KMeans clustering.  For feature transformation of a single image, each patch is assigned to its nearest cluster and the feature vector of an image is the number of patches from each type (a histogram)."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "from sklearn.cluster import KMeans\n",
      "class Codebook(sklearn.base.BaseEstimator):\n",
      "    def __init__(self, size=10):\n",
      "        self.size = size\n",
      "        self.clusterer = KMeans(n_clusters=size)\n",
      "    \n",
      "    def _get_histogram(self, x):\n",
      "        \"\"\" Returns histogram of codewords for given features \"\"\"\n",
      "        # Alternative method: return distance of each patch to cluster centers\n",
      "        # return self.clusterer.transform(x).ravel()\n",
      "        \n",
      "        # Assign each patch to a cluster\n",
      "        clusters = self.clusterer.predict(x)\n",
      "        \n",
      "        # Get the number of each patch type\n",
      "        return np.bincount(clusters, minlength=self.size)\n",
      "    \n",
      "    def fit(self, X, Y=None):\n",
      "        # print(\"Fitting clusterer\")\n",
      "        self.clusterer.fit(np.concatenate(X))\n",
      "        return self\n",
      "        \n",
      "    def transform(self, X, Y=None):\n",
      "        return np.array([self._get_histogram(x) for x in X])"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 9
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Let's put those together in a Pipeline"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "X, Y = get_classes([1, 3, 5, 7, 9], 100)\n",
      "patcher = PatchExtractor(patch_size=(8, 8))\n",
      "codebook = Codebook(size=50)\n",
      "clf = svm.SVC(kernel='linear')\n",
      "pipeline = Pipeline([(\"Patch_extractor\", patcher), (\"Codebook\", codebook), (\"svm\", clf)])\n",
      "score(pipeline, X, Y, 2, verbose=True)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "(4000, 64)\n",
        "Fold 1: 0.304"
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "\n",
        "(4000, 64)\n",
        "Fold 2: 0.328"
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "\n"
       ]
      },
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 10,
       "text": [
        "0.316"
       ]
      }
     ],
     "prompt_number": 10
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Heads up: This takes a really long time to run\n",
      "\n",
      "We end up getting about a 30% accuracy across 10 classes, which is significantly better than chance but can definitely be improved."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "X, Y = get_classes([0, 1, 2, 3, 4, 5, 6, 7, 8, 9], 900)\n",
      "patcher = PatchExtractor(patch_size=(8, 8))\n",
      "codebook = Codebook(size=50)\n",
      "clf = svm.SVC(kernel='linear')\n",
      "pipeline = Pipeline([(\"Patch_extractor\", patcher), (\"Codebook\", codebook), (\"svm\", clf)])\n",
      "score(pipeline, X, Y, 2, verbose=True)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "(72000, 64)\n",
        "Fold 1: 0.316444444444"
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "\n",
        "(72000, 64)"
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "\n",
        "Fold 2: 0.313777777778"
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "\n"
       ]
      },
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 14,
       "text": [
        "0.31511111111111112"
       ]
      }
     ],
     "prompt_number": 14
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "# You end up with edge like features\n",
      "show_patches(codebook.clusterer.cluster_centers_)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "display_data",
       "png": "iVBORw0KGgoAAAANSUhEUgAAA4sAAAFGCAYAAADD+gogAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nxf8LW6I5uWdZS9oAAAAASUVORK5CYII=\n",
       "text": [
        "<matplotlib.figure.Figure at 0x110b37250>"
       ]
      }
     ],
     "prompt_number": 15
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "If you wanted to try different codebook sizes, GridSearchCV is a good approach:\n",
      "\n",
      "    X, Y = get_classes([5, 7], 100)\n",
      "    from sklearn.grid_search import GridSearchCV\n",
      "    param_grid = [\n",
      "      {'Codebook__size': [3, 10]},\n",
      "     ]\n",
      "    grid_search = GridSearchCV(pipeline, param_grid, scoring=score)\n",
      "    grid_search.fit(X, Y)"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Now let's compare this to a simple RGB color based approach\n",
      "\n",
      "While this may be useful on some datasets with characteristic colors per category, that doesn't appear to be the case with CIFAR.  We end up doing worse than chance."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "class RGBFeature(sklearn.base.BaseEstimator):\n",
      "    \"\"\" Maps an image to its RGB color averages \"\"\"\n",
      "    def fit(self, X, Y=None):\n",
      "        return self\n",
      "    def transform(self, X, Y=None):\n",
      "        # Get average of each color channel\n",
      "        return np.array([[np.average(X[:, :, i]) for i in range(3)] for x in X])"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 16
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "X, Y = get_classes([1, 3, 5, 7, 9], 500)\n",
      "rgb = RGBFeature()\n",
      "clf = svm.SVC(kernel='linear')\n",
      "pipeline = Pipeline([(\"RGB Average\", rgb), (\"svm\", clf)])\n",
      "score(pipeline, X, Y, 5, verbose=True)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "Fold 1: 0.164\n",
        "Fold 2: 0.168"
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "\n",
        "Fold 3: 0.18"
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "\n",
        "Fold 4: 0.16"
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "\n",
        "Fold 5: 0.178"
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "\n"
       ]
      },
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 24,
       "text": [
        "0.17000000000000001"
       ]
      }
     ],
     "prompt_number": 24
    }
   ],
   "metadata": {}
  }
 ]
}