model-research.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "from matplotlib import pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3.1.0\n"
     ]
    }
   ],
   "source": [
    "import glob\n",
    "import sys\n",
    "import os\n",
    "import cv2\n",
    "import numpy as np\n",
    "print(cv2.__version__)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---\n",
    "\n",
    "## Load Traffic Lights data\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from data_utils import load_tl_extracts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "desired_dim = (32,32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "data_dirs = ['data/tl-extract-train', 'data/tl-extract-test', 'data/tl-extract-additional']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# OpenCV uses H, W, C. and C is ordered BGR\n",
    "x, y = load_tl_extracts(data_dirs, desired_dim)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(24504, 32, 32, 3)"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(24504,)"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "y_set = set(y)\n",
    "y_dict = {y: i for i,y in enumerate(y_set)}\n",
    "y_n = [y_dict[el] for el in y]\n",
    "n_classes = len(y_set)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---\n",
    "\n",
    "## Filter only classes relevant for us\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'green': 8,\n",
       " 'greenleft': 12,\n",
       " 'greenright': 9,\n",
       " 'greenstraight': 7,\n",
       " 'greenstraightleft': 0,\n",
       " 'greenstraightright': 5,\n",
       " 'off': 1,\n",
       " 'red': 4,\n",
       " 'redleft': 11,\n",
       " 'redright': 6,\n",
       " 'redstraight': 10,\n",
       " 'redstraightleft': 3,\n",
       " 'yellow': 2}"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZUAAAEKCAYAAADaa8itAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAGPpJREFUeJzt3XvUXXV95/H3Ry7eigQkZWgCJo7RDlJRmiJKx7FSBcQS\nxqUWbSUqq6mr1OJlVcCquKRYWNYbg6JUkNiyQIoXUkEwg6h1jSBBHK5SIqKEAolyVQYw8p0/zu+B\nk/gkOTzPPs/hJO/XWmedvX9nX75bYj757f3be6eqkCSpC08YdQGSpM2HoSJJ6oyhIknqjKEiSeqM\noSJJ6oyhIknqjKEiSeqMoSJJ6oyhIknqzNajLmCm7bTTTjVv3rxRlyFJY+WKK674WVXN3tRyW1yo\nzJs3jxUrVoy6DEkaK0l+Mshynv6SJHXGUJEkdcZQkSR1xlCRJHXGUJEkdcZQkSR1xlCRJHXGUJEk\ndcZQkSR1Zou7o17Sps07+vyhbv/mEw4a6vY1OvZUJEmdMVQkSZ0xVCRJnTFUJEmdMVQkSZ0xVCRJ\nnTFUJEmdMVQkSZ0xVCRJnTFUJEmdGVqoJDk9yeok1/S1fTjJD5NcleTLSWb1/XZMkpVJbkiyf1/7\nAa1tZZKj+9rnJ7mstX8hybbDOhZJ0mCG2VM5AzhgvbblwB5V9TzgP4BjAJLsDhwKPLet86kkWyXZ\nCvgkcCCwO/D6tizAicDHqupZwF3A4UM8FknSAIYWKlX1beDO9dq+XlVr2+ylwNw2vQg4u6oerKof\nAyuBvdtnZVXdVFUPAWcDi5IEeBlwblt/KXDIsI5FkjSYUV5TeQvwtTY9B7il77dVrW1D7U8H7u4L\nqIn2SSVZkmRFkhVr1qzpqHxJ0vpGEipJ/g5YC5w5E/urqlOramFVLZw9e/ZM7FKStkgz/j6VJG8C\nXgXsV1XVmm8Fdu1bbG5rYwPtPwdmJdm69Vb6l5ckjciM9lSSHAC8Gzi4qu7v+2kZcGiSJyaZDywA\nvgdcDixoI722pXcxf1kLo0uA17T1FwPnzdRxSJImN8whxWcB3wWek2RVksOBk4HtgOVJfpDk0wBV\ndS1wDnAdcCFwRFX9uvVC/hq4CLgeOKctC3AU8M4kK+ldYzltWMciSRrM0E5/VdXrJ2ne4F/8VXU8\ncPwk7RcAF0zSfhO90WGSpMcJ76iXJHXGUJEkdcZQkSR1xlCRJHXGUJEkdcZQkSR1xlCRJHXGUJEk\ndcZQkSR1xlCRJHXGUJEkdcZQkSR1xlCRJHXGUJEkdcZQkSR1xlCRJHXGUJEkdcZQkSR1xlCRJHXG\nUJEkdcZQkSR1xlCRJHXGUJEkdWZooZLk9CSrk1zT17ZjkuVJbmzfO7T2JDkpycokVyXZq2+dxW35\nG5Ms7mv//SRXt3VOSpJhHYskaTDD7KmcARywXtvRwMVVtQC4uM0DHAgsaJ8lwCnQCyHgWOCFwN7A\nsRNB1Jb5i7711t+XJGmGDS1UqurbwJ3rNS8ClrbppcAhfe2fr55LgVlJdgH2B5ZX1Z1VdRewHDig\n/fa0qrq0qgr4fN+2JEkjMtPXVHauqtva9O3Azm16DnBL33KrWtvG2ldN0j6pJEuSrEiyYs2aNdM7\nAknSBo3sQn3rYdQM7evUqlpYVQtnz549E7uUpC3STIfKHe3UFe17dWu/Fdi1b7m5rW1j7XMnaZck\njdBMh8oyYGIE12LgvL72w9oosH2Ae9ppsouAVyTZoV2gfwVwUfvt3iT7tFFfh/VtS5I0IlsPa8NJ\nzgJeCuyUZBW9UVwnAOckORz4CfC6tvgFwCuBlcD9wJsBqurOJMcBl7flPlhVExf//4reCLMnA19r\nH0nSCG0yVJIcCXwOuA/4LPAC4Oiq+vrG1quq12/gp/0mWbaAIzawndOB0ydpXwHssdHiJUkzapDT\nX2+pqnvpnXraAXgjvR6HJEnrGCRUJu5UfyXwz1V1bV+bJEmPGCRUrkjydXqhclGS7YCHh1uWJGkc\nDXKh/nDg+cBNVXV/kqfTLqRLktRvkJ5KAbsDf9Pmnwo8aWgVSZLG1iCh8ingRcDEaK77gE8OrSJJ\n0tga5PTXC6tqryRXAlTVXUm2HXJdkqQxNEhP5VdJtqI9pyvJbLxQL0maxCChchLwZeC3kxwPfAf4\n0FCrkiSNpU2e/qqqM5NcQe9O+ACHVNX1Q69MkjR2Nhgq7a2LE1YDZ/X/1vcMLkmSgI33VK6gdx1l\nsrvnC3jmUCqSJI2tDYZKVc2fyUIkSeNvoEffJ3k18If0eij/XlVfGWpVkqSxtMnRX0k+BbwVuBq4\nBnhrEm9+lCT9hkF6Ki8D/lt75wlJlgLXDrUqSdJYGuQ+lZXAbn3zu7Y2SZLWMUhPZTvg+iTfa/N/\nAKxIsgygqg4eVnGSpPEySKi8f+hVSJI2C4PcUf8tgCRP61/emx81U+Ydff7Qtn3zCQcNbdvSlmiT\noZJkCfBB4AF6D5IM3vwoSZrEIKe//hbYo6p+NuxiJEnjbZDRXz8C7h92IZKk8TdIqBwD/J8kn0ly\n0sRnOjtN8o4k1ya5JslZSZ6UZH6Sy5KsTPKFiReBJXlim1/Zfp/Xt51jWvsNSfafTk2SpOkbJFQ+\nA3wDuJTeQyYnPlOSZA69990vrKo9gK2AQ4ETgY9V1bOAu4DD2yqHA3e19o+15Uiye1vvucABwKfa\ny8QkSSMyyDWVbarqnUPY75OT/Ap4CnAbvTv339B+Xwp8ADgFWNSmAc4FTk6S1n52VT0I/DjJSmBv\n4Lsd1ypJGtAgPZWvJVmSZJckO058prrDqroV+Efgp/TC5B56PZ+7q2ptW2wVMKdNzwFuaeuubcs/\nvb99knUkSSMwSE/l9e37mL62KQ8pTrIDvV7GfOBu4F/pnb4amjYsegnAbrvttomlJUlTNcjNj12/\nV+WPgR9X1RqAJF8C9gVmJdm69UbmAre25W+l97yxVUm2BrYHft7XPqF/nfWP4VTgVICFCxdWx8cj\nSWoGOf1Fkj2SvC7JYROfaezzp8A+SZ7Sro3sB1wHXAK8pi2zGDivTS9r87Tfv9GemLwMOLSNDpsP\nLAAmnk8mSRqBQe6oPxZ4KbA7cAFwIPAd4PNT2WFVXZbkXOD7wFrgSnq9iPOBs5P8fWs7ra1yGvDP\n7UL8nfRGfFFV1yY5h14grQWOqKpfT6UmSVI3Brmm8hpgT+DKqnpzkp2Bf5nOTqvqWODY9Zpvojd6\na/1lHwBeu4HtHA8cP51aJEndGeT01/+rqoeBte2hkqtZ91qGJEnAYD2VFUlmAf9Eb+jvL/BeEEnS\nJAYZ/fVXbfLTSS4EnlZVVw23LEnSONrk6a8k+yZ5apv9Q+BNSZ4x3LIkSeNokGsqpwD3J9kTeBe9\npxZPaeSXJGnzNkiorG33hSwCTq6qT9J7b70kSesY5EL9fUmOAf4ceEmSJwDbDLcsSdI4GqSn8qfA\ng8DhVXU7vcehfHioVUmSxtIgo79uBz7aN/9TvKYiSZrEQM/+kiRpEIaKJKkzGwyVJBe37xNnrhxJ\n0jjb2DWVXZK8GDg4ydlA+n+squ8PtTJJ0tjZWKi8H3gfvdFeH13vt6L3TnlJkh6xwVCpqnOBc5O8\nr6qOm8GaJEljapAhxcclORh4SWv6ZlV9dbhlSZLG0SAPlPwH4Eh6b1i8DjgyyYeGXZgkafwM8piW\ng4Dntxd1kWQpvdf9vmeYhUmSxs+g96nM6pvefhiFSJLG3yA9lX8ArkxyCb1hxS8Bjh5qVZKksTTI\nhfqzknwT+IPWdFR7HpgkSesYpKdCVd0GLBtyLZKkMeezvyRJnRlJqCSZleTcJD9Mcn2SFyXZMcny\nJDe27x3asklyUpKVSa5Kslffdha35W9MsngUxyJJetRGQyXJVkl+OIT9fgK4sKp+F9gTuJ7exf+L\nq2oBcDGPDgY4EFjQPkuAU1ptOwLHAi8E9gaOnQgiSdJobDRUqurXwA1Jdutqh0m2pzeC7LS2j4eq\n6m5gEbC0LbYUOKRNLwI+Xz2XArOS7ALsDyyvqjur6i5gOXBAV3VKkh67QS7U7wBcm+R7wC8nGqvq\n4Cnucz6wBvhckj2BK+jdsb9zGxAAcDuwc5ueA9zSt/6q1rahdknSiAwSKu8bwj73At5WVZcl+QTr\n3fdSVZWkutphkiX0Tp2x226ddbokSevZ5IX6qvoWcDOwTZu+HJjOu1RWAauq6rI2fy69kLmjndai\nfa9uv98K7Nq3/tzWtqH2yY7h1KpaWFULZ8+ePY3SJUkbM8gDJf+C3l/8n2lNc4CvTHWH7cbJW5I8\npzXtR+9BlcuAiRFci4Hz2vQy4LA2Cmwf4J52muwi4BVJdmgX6F/R2iRJIzLI6a8j6I2uugygqm5M\n8tvT3O/bgDOTbAvcBLyZXsCdk+Rw4CfA69qyFwCvBFYC97dlqao7kxxHr+cE8MGqunOadUmSpmGQ\nUHmwqh5Kem8TTrI1vTc/TllV/QBYOMlP+02ybNELtsm2czpw+nRqkSR1Z5CbH7+V5D3Ak5O8HPhX\n4N+GW5YkaRwNEipH0xsCfDXwl/ROR713mEVJksbTIE8pfri9mOsyeqe9bminpCRJWscmQyXJQcCn\ngR/Re5/K/CR/WVVfG3ZxkqTxMsiF+o8Af1RVKwGS/FfgfMBQkSStY5BrKvdNBEpzE3DfkOqRJI2x\nDfZUkry6Ta5IcgFwDr1rKq/l0XtDJEl6xMZOf/1J3/QdwP9o02uAJw+tIknS2NpgqFTVm2eyEEnS\n+Btk9Nd8eo9Vmde//DQefS9J2kwNMvrrK/ReqPVvwMPDLUeSNM4GCZUHquqkoVciSRp7g4TKJ5Ic\nC3wdeHCisaqm804VSdJmaJBQ+T3gjcDLePT0V7V5SZIeMUiovBZ4ZlU9NOxiJEnjbZA76q8BZg27\nEEnS+BukpzIL+GGSy1n3mopDiiVJ6xgkVI4dehWSpM3CIO9T+dZMFCJJGn+D3FF/H4++k35bYBvg\nl1X1tGEWJkkaP4P0VLabmE4SYBGwzzCLkiSNp0FGfz2ier4C7D+keiRJY2yQ01+v7pt9ArAQeGBo\nFUmSxtYgPZU/6fvsT++tj4umu+MkWyW5MslX2/z8JJclWZnkC0m2be1PbPMr2+/z+rZxTGu/IYm9\nJ0kasUGuqQzrvSpHAtcDExf8TwQ+VlVnJ/k0cDhwSvu+q6qeleTQttyfJtkdOBR4LvA7wP9O8uyq\n+vWQ6pUkbcLGXif8/o2sV1V13FR3mmQucBBwPPDONgDgZcAb2iJLgQ/QC5VFbRrgXODkvgEDZ1fV\ng8CPk6wE9ga+O9W6JEnTs7HTX7+c5AO9nsNR09zvx4F38+gDKp8O3F1Va9v8KmBOm54D3ALQfr+n\nLf9I+yTrSJJGYGOvE/7IxHSS7eidrnozcDbwkQ2ttylJXgWsrqorkrx0qtt5jPtcAiwB2G233WZi\nl5K0RdrohfokOyb5e+AqegG0V1UdVVWrp7HPfYGDk9xML6BeBnwCmJVkIuTmAre26VuBXVs9WwPb\nAz/vb59knXVU1alVtbCqFs6ePXsapUuSNmaDoZLkw8Dl9EZ7/V5VfaCq7pruDqvqmKqaW1Xz6F1o\n/0ZV/RlwCfCatthi4Lw2vazN037/RlVVaz+0jQ6bDywAvjfd+iRJU7exnsq76I2qei/wn0nubZ/7\nktw7hFqOonfRfiW9ayantfbTgKe39ncCRwNU1bXAOcB1wIXAEY78kqTR2tg1lcd0t/1UVNU3gW+2\n6Zvojd5af5kH6L0obLL1j6c3gkyS9Dgw9OCQJG05DBVJUmcMFUlSZwwVSVJnDBVJUmcMFUlSZwwV\nSVJnDBVJUmcMFUlSZwwVSVJnDBVJUmcMFUlSZwwVSVJnDBVJUmcMFUlSZwwVSVJnDBVJUmcMFUlS\nZwwVSVJnDBVJUmcMFUlSZwwVSVJnDBVJUmdmPFSS7JrkkiTXJbk2yZGtfccky5Pc2L53aO1JclKS\nlUmuSrJX37YWt+VvTLJ4po9FkrSuUfRU1gLvqqrdgX2AI5LsDhwNXFxVC4CL2zzAgcCC9lkCnAK9\nEAKOBV4I7A0cOxFEkqTRmPFQqarbqur7bfo+4HpgDrAIWNoWWwoc0qYXAZ+vnkuBWUl2AfYHllfV\nnVV1F7AcOGAGD0WStJ6RXlNJMg94AXAZsHNV3dZ+uh3YuU3PAW7pW21Va9tQuyRpREYWKkl+C/gi\n8Paqurf/t6oqoDrc15IkK5KsWLNmTVeblSStZyShkmQbeoFyZlV9qTXf0U5r0b5Xt/ZbgV37Vp/b\n2jbU/huq6tSqWlhVC2fPnt3dgUiS1jGK0V8BTgOur6qP9v20DJgYwbUYOK+v/bA2Cmwf4J52muwi\n4BVJdmgX6F/R2iRJI7L1CPa5L/BG4OokP2ht7wFOAM5JcjjwE+B17bcLgFcCK4H7gTcDVNWdSY4D\nLm/LfbCq7pyZQ5AkTWbGQ6WqvgNkAz/vN8nyBRyxgW2dDpzeXXWSpOnwjnpJUmcMFUlSZwwVSVJn\nDBVJUmcMFUlSZwwVSVJnDBVJUmcMFUlSZ0ZxR70mMe/o84e6/ZtPOGio25ckMFQkaWD+42/TPP0l\nSeqMoSJJ6oyhIknqjKEiSeqMoSJJ6oyhIknqjKEiSeqM96lsIYY5vn5zGFsvqRv2VCRJnTFUJEmd\nMVQkSZ0xVCRJnTFUJEmdGftQSXJAkhuSrExy9KjrkaQt2VgPKU6yFfBJ4OXAKuDyJMuq6rrRViZJ\nj93mMPR/3HsqewMrq+qmqnoIOBtYNOKaJGmLNdY9FWAOcEvf/CrghSOqRZpRw35hlDQV4x4qA0my\nBFjSZn+R5IYpbmon4GfdVDVynR1LTuxiK1M2reMYce3r22L+fD3O/nfflM3iv0tOnPZxPGOQhcY9\nVG4Fdu2bn9va1lFVpwKnTndnSVZU1cLpbufxYHM5ls3lOMBjebzaXI5lpo5j3K+pXA4sSDI/ybbA\nocCyEdckSVusse6pVNXaJH8NXARsBZxeVdeOuCxJ2mKNdagAVNUFwAUztLtpn0J7HNlcjmVzOQ7w\nWB6vNpdjmZHjSFXNxH4kSVuAcb+mIkl6HDFUBrC5PAomya5JLklyXZJrkxw56pqmK8lWSa5M8tVR\n1zIdSWYlOTfJD5Ncn+RFo65pKpK8o/3ZuibJWUmeNOqaBpXk9CSrk1zT17ZjkuVJbmzfO4yyxkFt\n4Fg+3P58XZXky0lmDWPfhsom9D0K5kBgd+D1SXYfbVVTthZ4V1XtDuwDHDHGxzLhSOD6URfRgU8A\nF1bV7wJ7MobHlGQO8DfAwqrag97gmUNHW9VjcgZwwHptRwMXV9UC4OI2Pw7O4DePZTmwR1U9D/gP\n4Jhh7NhQ2bTN5lEwVXVbVX2/Td9H7y+uOaOtauqSzAUOAj476lqmI8n2wEuA0wCq6qGqunu0VU3Z\n1sCTk2wNPAX4zxHXM7Cq+jZw53rNi4ClbXopcMiMFjVFkx1LVX29qta22Uvp3dfXOUNl0yZ7FMzY\n/kU8Ick84AXAZaOtZFo+DrwbeHjUhUzTfGAN8Ll2Ku+zSZ466qIeq6q6FfhH4KfAbcA9VfX10VY1\nbTtX1W1t+nZg51EW06G3AF8bxoYNlS1Qkt8Cvgi8varuHXU9U5HkVcDqqrpi1LV0YGtgL+CUqnoB\n8EvG5zTLI9r1hkX0QvJ3gKcm+fPRVtWd6g2VHfvhskn+jt6p8DOHsX1DZdMGehTMuEiyDb1AObOq\nvjTqeqZhX+DgJDfTOyX5siT/MtqSpmwVsKqqJnqN59ILmXHzx8CPq2pNVf0K+BLw4hHXNF13JNkF\noH2vHnE905LkTcCrgD+rId1PYqhs2mbzKJgkoXfe/vqq+uio65mOqjqmquZW1Tx6/02+UVVj+a/i\nqroduCXJc1rTfsA4vhPop8A+SZ7S/qztxxgOOFjPMmBxm14MnDfCWqYlyQH0ThcfXFX3D2s/hsom\ntAtbE4+CuR44Z4wfBbMv8EZ6/6r/Qfu8ctRFCYC3AWcmuQp4PvChEdfzmLWe1rnA94Gr6f39MjZ3\noyc5C/gu8Jwkq5IcDpwAvDzJjfR6YieMssZBbeBYTga2A5a3/+9/eij79o56SVJX7KlIkjpjqEiS\nOmOoSJI6Y6hIkjpjqEiSOmOoSEOS5L8kOTvJj5JckeSCJM/uf3KstLkZ+zc/So9H7ea/LwNLq+rQ\n1rYnm8+zo6RJ2VORhuOPgF9V1SM3mFXV/6Xv4aRJ5iX59yTfb58Xt/Zdkny73aB2TZL/3t4bc0ab\nvzrJO2b+kKRNs6ciDccewKYedrkaeHlVPZBkAXAWsBB4A3BRVR3f3ufzFHp32c9p7ylhWC9YkqbL\nUJFGZxvg5CTPB34NPLu1Xw6c3h7++ZWq+kGSm4BnJvlfwPnAuD9SXpspT39Jw3Et8PubWOYdwB30\n3vS4ENgWHnnB0kvoPQ37jCSHVdVdbblvAm9lzF9Mps2XoSINxzeAJyZZMtGQ5Hms+xqF7YHbquph\neg/63Kot9wzgjqr6J3rhsVeSnYAnVNUXgfcyno/G1xbA01/SEFRVJfmfwMeTHAU8ANwMvL1vsU8B\nX0xyGHAhvZdzAbwU+NskvwJ+ARxG722jn0sy8Q/BobxfXJoun1IsSeqMp78kSZ0xVCRJnTFUJEmd\nMVQkSZ0xVCRJnTFUJEmdMVQkSZ0xVCRJnfn/l9UWw3DexuIAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fe29efe5630>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#distribution of examples per class\n",
    "plt.hist(y_n, bins=n_classes);\n",
    "plt.xlabel('Class');\n",
    "plt.ylabel('Number of samples');\n",
    "y_dict"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "relevant = ['green','off','yellow','red']\n",
    "x2 = x[np.isin(y, relevant)]\n",
    "y2 = y[np.isin(y, relevant)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(23164, 32, 32, 3)"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x2.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "y2_set = set(y2)\n",
    "y2_dict = {y: i for i,y in enumerate(y2_set)}\n",
    "y2_n = [y_dict[el] for el in y2]\n",
    "n_classes = len(y2_set)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'green': 0, 'off': 1, 'red': 2, 'yellow': 3}"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZUAAAEKCAYAAADaa8itAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nI2gAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fe29efe5208>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#distribution of examples per class\n",
    "plt.hist(y2_n, bins=n_classes);\n",
    "plt.xlabel('Class');\n",
    "plt.ylabel('Number of samples');\n",
    "y2_dict"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---\n",
    "\n",
    "## Visualize Data\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def show_images(images, labels_true, rows, cols, cls=None, random=True, labels_predicted=None):\n",
    "    fig, axes = plt.subplots(rows, cols)\n",
    "    fig.subplots_adjust(hspace=0.37, wspace=0.3)\n",
    "    imidx = 0\n",
    "    for i, ax in enumerate(axes.flat):\n",
    "        while True:\n",
    "            if random:\n",
    "                example = np.random.randint(0, len(images))\n",
    "            else:\n",
    "                example = imidx\n",
    "                imidx += 1\n",
    "            if cls is None or labels_true[example]==cls:\n",
    "                break\n",
    "        if example >= len(images):\n",
    "            return\n",
    "        if len(images.shape)==4:\n",
    "            ax.imshow(images[example,:,:,:]);\n",
    "        else:\n",
    "            ax.imshow(images[example,:,:], cmap='Greys_r');\n",
    "        ax.set_xticks([])\n",
    "        ax.set_yticks([])\n",
    "        if labels_predicted is not None:\n",
    "            str = '{}: true {} pred {}'.format(example, labels_true[example], labels_predicted[example])\n",
    "        else:\n",
    "            str = '{}: true {}'.format(example, labels_true[example])\n",
    "        ax.set_xlabel(str)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX4AAAD5CAYAAAAgGF4oAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRaHIAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fe29edb6a58>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# randomly show grid 5x5 examples from training data set and their true class\n",
    "rows = 5\n",
    "cols = 5\n",
    "cls = None\n",
    "show_images(x2, y2, rows, cols, cls)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----\n",
    "\n",
    "## Preprocess and split train/test\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.preprocessing import LabelBinarizer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "x = x2\n",
    "y = y2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# apply One-Hot Encoding for class labels. at the moment they are integers 0..42\n",
    "encoder = LabelBinarizer()\n",
    "encoder.fit(y)\n",
    "y_onehot = encoder.transform(y)\n",
    "# Change to float32, so it can be multiplied against the features in \n",
    "# TensorFlow, which are float32\n",
    "y_onehot = y_onehot.astype(np.float32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pct_train = 85.\n",
    "pct_valid = 15.\n",
    "random_state = 123\n",
    "\n",
    "train_features, val_features, train_labels, val_labels = train_test_split(\n",
    "                                                            x, y_onehot, \n",
    "                                                            train_size = pct_train/100., \n",
    "                                                            test_size = pct_valid/100.,\n",
    "                                                            random_state = random_state)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(19689, 32, 32, 3)"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_features.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "dtype('uint8')"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_features.dtype"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(19689, 4)"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_labels.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----\n",
    "\n",
    "## Design and Test a Model Architecture\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Below is Tensorboard visualisation of the main part of my network (before softmax).\n",
    "\n",
    "Here is the description of it:\n",
    "\n",
    "1. input features. shape: (?, 32, 32, 3)\n",
    "2. convolution. kernel: [5,5] 3 input channels, 64 output channels. strides [1,1,1,1]. SAME padding.\n",
    "3. add bias\n",
    "4. relu\n",
    "5. max pool. strides [1,2,2,1]. kernel [1,2,2,1]. SAME padding\n",
    "6. convolution. kernel: [5,5] 64 input channels, 32 output channels. strides [1,1,1,1]. SAME padding.\n",
    "7. add bias\n",
    "8. relu\n",
    "9. max pool. strides [1,2,2,1]. kernel [1,2,2,1]. SAME padding\n",
    "10. reshape to flatten features. 2048 features\n",
    "11. dropout with keep_probability=0.5\n",
    "12. fully connected W*x+bias. output: number of classes\n",
    "13. softmax to get label probabilities\n",
    "\n",
    "."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "#from IPython.display import Image\n",
    "#Image('CNN-graph.png')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from tl_classifier_cnn import TLClassifierCNN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# define shapes\n",
    "features_shape = ((None,) + train_features.shape[1:])\n",
    "labels_shape = (None,train_labels.shape[1],)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "# create classifier\n",
    "tsc = TLClassifierCNN()\n",
    "\n",
    "# define model\n",
    "tsc.define_model(features_shape=features_shape, labels_shape=labels_shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# where we save trained variables\n",
    "model_param_file = 'ckpt/model.ckpt'\n",
    "\n",
    "# where we save graph and events/summaries for tensorflow\n",
    "summary_dir = 'train_summaries'\n",
    "\n",
    "tsc.set_save_files(model_param_file, summary_dir)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# learning parameters\n",
    "epochs = 50\n",
    "batch_size = 250\n",
    "learning_rate = 0.001\n",
    "max_iterations_without_improvement = 10\n",
    "dropout_keep_probability=0.7"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create Tensorflow session\n",
    "tsc.create_session(learning_rate)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Restoring parameters from ckpt/model.ckpt-1501\n"
     ]
    }
   ],
   "source": [
    "# i have trained model several times, restarting from where I left off, sometimes changing parameters\n",
    "tsc.restore_variables()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Train Epoch 1502/50: 100%|██████████| 79/79 [00:49<00:00,  1.69batches/s]\n",
      "Train Accuracy Epoch 1502/50: 100%|██████████| 79/79 [00:01<00:00, 39.77batches/s]\n",
      "Val Accuracy Epoch 1502/50: 100%|██████████| 14/14 [00:00<00:00, 38.93batches/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1501, val accuracy: 0.9769784936801993\n",
      "validation accuracy improved\n",
      "checkpoint saved to ckpt/model.ckpt-1580\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Train Epoch 1503/50: 100%|██████████| 79/79 [00:49<00:00,  1.73batches/s]\n",
      "Train Accuracy Epoch 1503/50: 100%|██████████| 79/79 [00:02<00:00, 39.04batches/s]\n",
      "Val Accuracy Epoch 1503/50: 100%|██████████| 14/14 [00:00<00:00, 38.65batches/s]\n",
      "Train Epoch 1504/50:   0%|          | 0/79 [00:00<?, ?batches/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1502, val accuracy: 0.9738130140647614\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Train Epoch 1504/50: 100%|██████████| 79/79 [00:49<00:00,  1.67batches/s]\n",
      "Train Accuracy Epoch 1504/50: 100%|██████████| 79/79 [00:01<00:00, 40.28batches/s]\n",
      "Val Accuracy Epoch 1504/50: 100%|██████████| 14/14 [00:00<00:00, 39.34batches/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1503, val accuracy: 0.9749640861003519\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Train Epoch 1505/50: 100%|██████████| 79/79 [00:49<00:00,  1.71batches/s]\n",
      "Train Accuracy Epoch 1505/50: 100%|██████████| 79/79 [00:01<00:00, 40.26batches/s]\n",
      "Val Accuracy Epoch 1505/50: 100%|██████████| 14/14 [00:00<00:00, 40.42batches/s]\n",
      "Train Epoch 1506/50:   0%|          | 0/79 [00:00<?, ?batches/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1504, val accuracy: 0.9775540365589609\n",
      "validation accuracy improved\n",
      "checkpoint saved to ckpt/model.ckpt-1817\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Train Epoch 1506/50: 100%|██████████| 79/79 [00:49<00:00,  1.72batches/s]\n",
      "Train Accuracy Epoch 1506/50: 100%|██████████| 79/79 [00:02<00:00, 39.19batches/s]\n",
      "Val Accuracy Epoch 1506/50: 100%|██████████| 14/14 [00:00<00:00, 40.59batches/s]\n",
      "Train Epoch 1507/50:   0%|          | 0/79 [00:00<?, ?batches/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1505, val accuracy: 0.9761151744307374\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Train Epoch 1507/50: 100%|██████████| 79/79 [00:49<00:00,  1.71batches/s]\n",
      "Train Accuracy Epoch 1507/50: 100%|██████████| 79/79 [00:01<00:00, 39.81batches/s]\n",
      "Val Accuracy Epoch 1507/50: 100%|██████████| 14/14 [00:00<00:00, 39.30batches/s]\n",
      "Train Epoch 1508/50:   0%|          | 0/79 [00:00<?, ?batches/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1506, val accuracy: 0.9775540245522698\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Train Epoch 1508/50: 100%|██████████| 79/79 [00:49<00:00,  1.71batches/s]\n",
      "Train Accuracy Epoch 1508/50: 100%|██████████| 79/79 [00:02<00:00, 38.98batches/s]\n",
      "Val Accuracy Epoch 1508/50: 100%|██████████| 14/14 [00:00<00:00, 40.95batches/s]\n",
      "Train Epoch 1509/50:   0%|          | 0/79 [00:00<?, ?batches/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1507, val accuracy: 0.9772662430358448\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Train Epoch 1509/50: 100%|██████████| 79/79 [00:49<00:00,  1.70batches/s]\n",
      "Train Accuracy Epoch 1509/50: 100%|██████████| 79/79 [00:01<00:00, 39.77batches/s]\n",
      "Val Accuracy Epoch 1509/50: 100%|██████████| 14/14 [00:00<00:00, 39.17batches/s]\n",
      "Train Epoch 1510/50:   0%|          | 0/79 [00:00<?, ?batches/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1508, val accuracy: 0.9752518564677067\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Train Epoch 1510/50: 100%|██████████| 79/79 [00:49<00:00,  1.73batches/s]\n",
      "Train Accuracy Epoch 1510/50: 100%|██████████| 79/79 [00:01<00:00, 39.65batches/s]\n",
      "Val Accuracy Epoch 1510/50: 100%|██████████| 14/14 [00:00<00:00, 38.64batches/s]\n",
      "Train Epoch 1511/50:   0%|          | 0/79 [00:00<?, ?batches/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1509, val accuracy: 0.9761151915831532\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Train Epoch 1511/50: 100%|██████████| 79/79 [00:49<00:00,  1.72batches/s]\n",
      "Train Accuracy Epoch 1511/50: 100%|██████████| 79/79 [00:01<00:00, 39.99batches/s]\n",
      "Val Accuracy Epoch 1511/50: 100%|██████████| 14/14 [00:00<00:00, 41.91batches/s]\n",
      "Train Epoch 1512/50:   0%|          | 0/79 [00:00<?, ?batches/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1510, val accuracy: 0.9772662468951383\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Train Epoch 1512/50: 100%|██████████| 79/79 [00:49<00:00,  1.72batches/s]\n",
      "Train Accuracy Epoch 1512/50: 100%|██████████| 79/79 [00:02<00:00, 39.45batches/s]\n",
      "Val Accuracy Epoch 1512/50: 100%|██████████| 14/14 [00:00<00:00, 40.48batches/s]\n",
      "Train Epoch 1513/50:   0%|          | 0/79 [00:00<?, ?batches/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1511, val accuracy: 0.9764029636657495\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Train Epoch 1513/50: 100%|██████████| 79/79 [00:49<00:00,  1.73batches/s]\n",
      "Train Accuracy Epoch 1513/50: 100%|██████████| 79/79 [00:02<00:00, 38.77batches/s]\n",
      "Val Accuracy Epoch 1513/50: 100%|██████████| 14/14 [00:00<00:00, 38.97batches/s]\n",
      "Train Epoch 1514/50:   0%|          | 0/79 [00:00<?, ?batches/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1512, val accuracy: 0.9766907233128445\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Train Epoch 1514/50: 100%|██████████| 79/79 [00:49<00:00,  1.70batches/s]\n",
      "Train Accuracy Epoch 1514/50: 100%|██████████| 79/79 [00:01<00:00, 39.84batches/s]\n",
      "Val Accuracy Epoch 1514/50: 100%|██████████| 14/14 [00:00<00:00, 39.50batches/s]\n",
      "Train Epoch 1515/50:   0%|          | 0/79 [00:00<?, ?batches/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1513, val accuracy: 0.9764029473709546\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Train Epoch 1515/50: 100%|██████████| 79/79 [00:50<00:00,  1.73batches/s]\n",
      "Train Accuracy Epoch 1515/50: 100%|██████████| 79/79 [00:01<00:00, 39.61batches/s]\n",
      "Val Accuracy Epoch 1515/50: 100%|██████████| 14/14 [00:00<00:00, 38.70batches/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 1514, val accuracy: 0.9758274036345722\n",
      "no validation accuracy improvement over 10 epochs. stop\n",
      "runtime: 0:12:21.434049\n",
      "best val accuracy: 0.9775540365589609\n",
      "epochs: 1514\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "# main training\n",
    "loss_epoch, train_acc_epoch, valid_acc_epoch, best_validation_accuracy = \\\n",
    "    tsc.train(train_images      = train_features, \n",
    "              train_labels      = train_labels, \n",
    "              validation_images = val_features, \n",
    "              validation_labels = val_labels, \n",
    "              dropout_keep_probability = dropout_keep_probability,\n",
    "              batch_size        = batch_size, \n",
    "              epochs            = epochs, \n",
    "              max_iterations_without_improvement = max_iterations_without_improvement)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "STOP"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# accuracy statistics for last epoch (train set and validation set). and also best seen validation set accuracy \n",
    "train_acc_epoch[-1], best_validation_accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAagAAAEYCAYAAAAJeGK1AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nEZGYpIASEZGY9P8B0JRcXOYvjRUAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fe2506bea20>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "loss_plot = plt.subplot(211)\n",
    "loss_plot.set_title('Loss')\n",
    "loss_plot.plot(loss_epoch, 'g')\n",
    "acc_plot = plt.subplot(212)\n",
    "acc_plot.set_title('Accuracy')\n",
    "acc_plot.plot(train_acc_epoch, 'r', label='Training Accuracy')\n",
    "acc_plot.plot(valid_acc_epoch, 'b', label='Validation Accuracy')\n",
    "acc_plot.set_ylim([0, 1.0])\n",
    "acc_plot.legend(loc=2)\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## investigating accuracy on out of sample data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "predicted_probabilities, predicted_classes, accuracy = \\\n",
    "  tsc.predict(images=val_features, \n",
    "              true_labels=val_labels, \n",
    "              batch_size=150)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# great, accuracy 99.74% on validation set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "predicted_probabilities.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "predicted_probabilities[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "predicted_classes[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "certainty = predicted_probabilities.max(axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "plt.hist(certainty, bins=100);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "len(certainty[certainty<.98])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "plt.plot(certainty[certainty<.98]);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# NOW test model predictions on test.p data which was not used in training or cross-validation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "predicted_probabilities, predicted_classes, accuracy = \\\n",
    "  tsc.predict(images = x_test_norm, \n",
    "              true_labels = y_test_norm, \n",
    "              batch_size = 150)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 95.64% accuracy. great! model generalises very well! lets analyse what is working and what is not."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.metrics import confusion_matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def print_test_accuracy(images, cls_pred, cls_true, \n",
    "                        show_example_errors=False,\n",
    "                        show_confusion_matrix=False):\n",
    "    num_test = len(images)\n",
    "    num_classes = len(set(cls_true))\n",
    "\n",
    "    correct = (cls_true == cls_pred)\n",
    "    correct_sum = correct.sum()\n",
    "    acc = float(correct_sum) / num_test\n",
    "\n",
    "    msg = \"Accuracy: {0:.1%} ({1} / {2})\"\n",
    "    print(msg.format(acc, correct_sum, num_test))\n",
    "\n",
    "    if show_example_errors:\n",
    "        print(\"Example errors:\")\n",
    "        incorrect = (correct == False)\n",
    "        im = images[incorrect]\n",
    "        cp = cls_pred[incorrect]\n",
    "        ct = cls_true[incorrect]\n",
    "        show_images(im, ct, 3, 4, labels_predicted=cp, random=False)\n",
    "                \n",
    "    if show_confusion_matrix:\n",
    "        print(\"Confusion Matrix (log):\")\n",
    "        # Get the confusion matrix using sklearn.\n",
    "        cm = confusion_matrix(y_true=cls_true,\n",
    "                              y_pred=cls_pred)\n",
    "        #print(cm)\n",
    "        plt.matshow(np.log(cm))\n",
    "        plt.colorbar()\n",
    "        tick_marks = np.arange(num_classes)\n",
    "        plt.xticks(tick_marks, range(num_classes))\n",
    "        plt.yticks(tick_marks, range(num_classes))\n",
    "        plt.xlabel('Predicted')\n",
    "        plt.ylabel('True')\n",
    "        plt.show()        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# lets look at first 26x26 cells of confusion matrix\n",
    "cm = confusion_matrix(y_true=y_test,\n",
    "                      y_pred=predicted_classes)\n",
    "np.set_printoptions(threshold=np.nan, linewidth=200)\n",
    "print(cm[:25,:25])\n",
    "np.set_printoptions(threshold=30, linewidth=75)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# visualize example error images and confusion matrix\n",
    "print_test_accuracy(images=X_test, \n",
    "                    cls_pred=predicted_classes, \n",
    "                    cls_true=y_test, \n",
    "                    show_example_errors=True, \n",
    "                    show_confusion_matrix=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# read the sign names into a dict\n",
    "import csv\n",
    "reader = csv.DictReader(open('signnames.csv'))\n",
    "signnames = {}\n",
    "for row in reader:\n",
    "    signnames[int(row['ClassId'])] = row['SignName']    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# visualize instances where model is misclassifying big number of images"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def print_one_confusion(images, true_labels, predicted_labels, show_true_class, show_false_class):\n",
    "    idx = np.logical_and(true_labels==show_true_class, predicted_labels==show_false_class)\n",
    "    cnt = sum(idx)\n",
    "    print('true class \"{}\", predicted class \"{}\". found: {}'.format(signnames[show_true_class], signnames[show_false_class], cnt))\n",
    "    cols = int(np.ceil(cnt/10))\n",
    "    show_images(images[idx], true_labels[idx], cols, 10, random=False, labels_predicted=predicted_labels[idx])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "print_one_confusion(images=X_test, true_labels=y_test, predicted_labels=predicted_classes, \n",
    "                    show_true_class=21, show_false_class=12)\n",
    "# what we see here is the same test set image (may be with small transformations) that was misclassified.\n",
    "# this seems like a genuine defficiency of the model -- the predicted sign is very different visually than the image"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "print_one_confusion(images=X_test, true_labels=y_test, predicted_labels=predicted_classes, \n",
    "                    show_true_class=30, show_false_class=23)\n",
    "# with this one it looks like the same image, with slight transformations. the quality is quite poor, so its \n",
    "# understandable that model had hard time predicting it. the shape of the sign is actually right. and the picture\n",
    "# inside triangle is not very legible even for a human, so this is probably ok"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "print_one_confusion(images=X_test, true_labels=y_test, predicted_labels=predicted_classes, \n",
    "                    show_true_class=7, show_false_class=5)\n",
    "# we see in confusion matrix that all the speed limit signs have high misclassification errors between them.\n",
    "# looks like the quality of the picture need to increase (better resolution is required)\n",
    "# for model to be more discerning between different speed limits."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "print_one_confusion(images=X_test, true_labels=y_test, predicted_labels=predicted_classes, \n",
    "                    show_true_class=6, show_false_class=42)\n",
    "# looks like in this case the model is confused by the 'end' part -- the line across the sign. so it used the shape\n",
    "# and the cross line as main features. it also interpreted 80 as two vehicles passing each other (see below for\n",
    "# images of the 'end of passing' sign examples)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "rows = 3\n",
    "cols = 10\n",
    "cls = 42\n",
    "show_images(X_test, y_test, rows, cols, cls)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "x = normalize_data(im_array)\n",
    "y = encoder.transform(im_true_label)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "x.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "y.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "predicted_probabilities, predicted_classes, accuracy = \\\n",
    "  tsc.predict(images = x, \n",
    "              true_labels = y, \n",
    "              batch_size = 150)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "predicted_classes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "im_true_label"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "rows = 3\n",
    "cols = 10\n",
    "cls = 11\n",
    "show_images(X_test, y_test, rows, cols, cls)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "tsc.close_session()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}