6_Generative_Models_number_9-FINAL.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h1 align=\"center\"> Worksheet 6 Generative models II # 9 </h1>\n",
    "<h2 align=\"center\"> Michael Galarnyk </h2> <p></p>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "from struct import unpack\n",
    "%matplotlib inline\n",
    "import numpy as np\n",
    "from scipy.stats import multivariate_normal\n",
    "from sklearn import datasets"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2 align=\"center\">Loading MNIST Dataset </h2> <p></p>\n",
    "<h5 align=\"center\">(very similar to loader.py) </h5> <p></p>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "#dataset = datasets.fetch_mldata(\"MNIST Original\") in case you have trouble with loadmnist function"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "def loadmnist(imagefile, labelfile):\n",
    "\n",
    "    # Open the images with gzip in read binary mode\n",
    "    images = open(imagefile, 'rb')\n",
    "    labels = open(labelfile, 'rb')\n",
    "\n",
    "    # Get metadata for images\n",
    "    images.read(4)  # skip the magic_number\n",
    "    number_of_images = images.read(4)\n",
    "    number_of_images = unpack('>I', number_of_images)[0]\n",
    "    rows = images.read(4)\n",
    "    rows = unpack('>I', rows)[0]\n",
    "    cols = images.read(4)\n",
    "    cols = unpack('>I', cols)[0]\n",
    "\n",
    "    # Get metadata for labels\n",
    "    labels.read(4)\n",
    "    N = labels.read(4)\n",
    "    N = unpack('>I', N)[0]\n",
    "\n",
    "    # Get data\n",
    "    x = np.zeros((N, rows*cols), dtype=np.uint8)  # Initialize numpy array\n",
    "    y = np.zeros(N, dtype=np.uint8)  # Initialize numpy array\n",
    "    for i in range(N):\n",
    "        for j in range(rows*cols):\n",
    "            tmp_pixel = images.read(1)  # Just a single byte\n",
    "            tmp_pixel = unpack('>B', tmp_pixel)[0]\n",
    "            x[i][j] = tmp_pixel\n",
    "        tmp_label = labels.read(1)\n",
    "        y[i] = unpack('>B', tmp_label)[0]\n",
    "\n",
    "    images.close()\n",
    "    labels.close()\n",
    "    return (x, y)\n",
    "\n",
    "def displaychar(image):\n",
    "    plt.imshow(np.reshape(image, (28,28)), cmap=plt.cm.gray)\n",
    "    plt.axis('off')\n",
    "    #plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "path = '/Users/mgalarny/Documents/youtube/DSE210_Probability_Statistics_Python/'\n",
    "\n",
    "train_img_path = path + 'train-images-idx3-ubyte' \n",
    "train_lbl_path = path + 'train-labels-idx1-ubyte' \n",
    "test_img_path = path + 't10k-images-idx3-ubyte' \n",
    "test_lbl_path = path + 't10k-labels-idx1-ubyte'\n",
    "\n",
    "X_train, Y_train = loadmnist(train_img_path, train_lbl_path)\n",
    "X_test, Y_test = loadmnist(test_img_path, test_lbl_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "class GaussianClassify:\n",
    "    \n",
    "    def __init__(self):\n",
    "        self.mean = np.zeros((10, 784))\n",
    "        self.cov = np.zeros((10, 784, 784))\n",
    "        \n",
    "    def sort_by_number(self,train_img, train_label):\n",
    "        list_by_class = {i:[] for i in xrange(10)}\n",
    "        for i in xrange(len(train_label)):\n",
    "            class_img = train_label[i]\n",
    "            list_by_class[class_img].append(train_img[i])\n",
    "        return list_by_class\n",
    "\n",
    "    def find_mean_cov(self,list_by_class, i, c):\n",
    "        images = np.matrix(list_by_class[i])\n",
    "        total_pixels = images.shape[1] # should always be 784; total number of pixels \n",
    "        mean = np.mean(images, axis = 0)\n",
    "        self.mean[i] = mean\n",
    "        cov = np.cov(np.transpose(images)) + c * np.identity(total_pixels)\n",
    "        self.cov[i] = cov\n",
    "        return mean, cov\n",
    "\n",
    "    def test(self, test_image, test_label, mean, cov):\n",
    "        list_by_class = self.sort_by_number(test_image, test_label)\n",
    "        images = np.matrix(list_by_class)\n",
    "        test_img_array = np.array(test_image)\n",
    "        prob = []\n",
    "        for k in xrange(10):\n",
    "            images = np.matrix(list_by_class[k])\n",
    "            number_images = images.shape[0]\n",
    "            pi = float(number_images) / float(len(test_label)) \n",
    "            prob.append(np.log(pi) + multivariate_normal.logpdf(test_img_array,\n",
    "                                             mean=self.mean[k],\n",
    "                                             cov=self.cov[k]))\n",
    "        max_prob = np.argmax(prob, axis = 0)\n",
    "        self.boolean_label = [i!=j for i,j in zip(max_prob, test_label)]\n",
    "        self.max_prob = max_prob\n",
    "        self.test_label = test_label\n",
    "        self.prob = prob\n",
    "        error = 100 * np.sum([i!=j for i,j in zip(max_prob, test_label)])/float(len(test_label))\n",
    "        return error, max_prob, test_label, prob\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2 align=\"center\"> Smoothing through optimizing c value </h2>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 110,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "c: 1 Error rate:  15.88  %\n",
      "c: 10 Error rate:  12.26  %\n",
      "c: 100 Error rate:  8.05  %\n",
      "c: 1000 Error rate:  4.89  %\n",
      "c: 2000 Error rate:  4.37  %\n",
      "c: 2100 Error rate:  4.35  %\n",
      "c: 2200 Error rate:  4.36  %\n",
      "c: 2300 Error rate:  4.36  %\n",
      "c: 2400 Error rate:  4.35  %\n",
      "c: 2500 Error rate:  4.34  %\n",
      "c: 2600 Error rate:  4.31  %\n",
      "c: 2700 Error rate:  4.3  %\n",
      "c: 2800 Error rate:  4.29  %\n",
      "c: 2900 Error rate:  4.31  %\n",
      "c: 3000 Error rate:  4.35  %\n",
      "c: 3100 Error rate:  4.35  %\n",
      "c: 3200 Error rate:  4.36  %\n",
      "c: 3300 Error rate:  4.38  %\n",
      "c: 3400 Error rate:  4.35  %\n",
      "c: 3500 Error rate:  4.38  %\n",
      "c: 3600 Error rate:  4.36  %\n",
      "c: 3700 Error rate:  4.35  %\n",
      "c: 3800 Error rate:  4.34  %\n",
      "c: 3900 Error rate:  4.3  %\n",
      "c: 4000 Error rate:  4.31  %\n"
     ]
    }
   ],
   "source": [
    "error = []\n",
    "c_value = []\n",
    "c = 2000 # default value\n",
    "constant = [1,10,100,1000,2000,2100,2200,2300,2400,2500,2600,2700,2800,2900,3000,3100,3200,3300,3400,3500,3600,3700,3800,3900,4000]\n",
    "GC = GaussianClassify()\n",
    "list_by_class = GC.sort_by_number(X_train, Y_train)\n",
    "\n",
    "for c in constant:\n",
    "    for i in xrange(10):\n",
    "        GC.find_mean_cov(list_by_class, i, c)\n",
    "\n",
    "    err_rate, max_prob, test_label, prob = GC.test(X_test, Y_test, mean, cov)\n",
    "    print 'c:', c, \"Error rate: \", err_rate, \" %\" \n",
    "    error.append(err_rate)\n",
    "    c_value.append(c)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The best c value for me was c = 2800 with a error rate of: 4.29 %\n"
     ]
    }
   ],
   "source": [
    "print \"The best c value for me was c =\",c_value[np.argmin(error)], \"with a error rate of:\", error[np.argmin(error)], \"%\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2 align=\"center\">Best c value: 2800, error rate: 4.29 </h2> <p></p>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "boolean_list = [i!=j for i,j in zip(max_prob, test_label)]\n",
    "index_error = [index for index, tf in enumerate(boolean_list) if tf]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 136,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAlQAAACQCAYAAADOfvWiAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJztvXmQ5F12FXZe7vu+Vdba3dX9LTMjjSRkYJA0GrCxJDZh\nCHkJbEeIkMHhIMBGFoYwQsLGIgiCJbDDeJFs7AgMDgQoZJCQQazGsgEPaObj+7qnu7rW3LfKPasq\n8+c/ss7t98vO6qX2ynon4hfVnZmVmZUv33vn3XvuucqyLBgYGBgYGBgYGJwfjpt+AwYGBgYGBgYG\ndx2GUBkYGBgYGBgYXBCGUBkYGBgYGBgYXBCGUBkYGBgYGBgYXBCGUBkYGBgYGBgYXBCGUBkYGBgY\nGBgYXBCGUF0jlFI/ppSaKKW+66bfi8HFYMZysWDGc3FgxnKxcJfG89oIlVJqQyn155VSnyqlekqp\nklLqHyulfkgp5b6u93HDsE6vC0Ep5VJK/Val1E8qpb6mlDo8/Ux/WSn140qp0CW817Ne+7copf7S\n6Tg2lFJ9pdQzpdRfVEp921W97i3EpYwloZT6ilLqbyqlakqpoVLqG0qpn7jKsTzjfSSVUsXTBewf\nXOdr3zDu/Hgqpf7I6bjNu8ZKqV9/Va99y3Dn11ntPZh98/Lnpu90/D5TSg2UUmWl1F9WSn140ed2\nXcYbfBuUUt8O4O8C8AL4eQB/HUAEwG8C8N8B+DcAfO91vJcFwSMAPw2gi+nn+n8ACAH41wH8YQA/\noJT6NZZlNa7gtX8LgF8B4J8AKAA4ArAJ4PsB/JtKqR+yLOunruB1FxZKqd8N4L8BcAzgrwLYB/Bt\nAP4AgO9VSn2nZVmda3o7/z2AAC5xAbtvuOHxtAD8BQDbc+57fkWvuai4yXXW7JtXAKWUB8DfBvAl\nTPewPwNgFcAPAPgNSqmvWJb1T879ApZlXfkF4G8AGAP4HTO3+wF8/fS+77iO93KTF4A/cvq3ftcF\nnycP4HcD8M/c7gLws6ev8Wev6G/wnHH75wAMADQAuG76s75DY5kD0AcwAvBtM/f9ZwAmAP7MNf1N\n/97p3/S7Tl/3H9z052zG8/r/hrt+LcI6e/o6Zt+8xPE8fa4/eDoH/9LM7b/p9PavXeT5ryvl9+D0\n58/qN1qWNQDwd07/m77ICyilvnwa2v5RpdSvUkr9baVUSynVVkr9/Lx0lJ6bVUr9O0qpX1JKdZRS\nWzOP+5VKqb9ymgoZKaV2T8OwS2e8l287fc32aZj4/1RK/aqL/H06LMsqWJb1508/P/32EwD/FQAF\n4Lsv6/VmXuPojNs/AfApgCjMWL4PvheAD8Bftyzrn83c9ycwJag/qJTyXeJrvgal1BqAPwvgf8T0\nNHyZz23Gc4prG8+rwn0ay5tcZ09h9s3LnZvAlCBbmEaLBZZl/SyAfwjgY6XUl8/75NdFqD45/fkb\n9BuVUgEAvw7TE93/fUmv9asA/D1MoyX/NYC/CeDXAviHSqlfM/NY5mZ/GNONZAfAnwPwc9p7/EEA\n/wjTMO8vAvjTmIYKfyeAf6qUWpn5m74E4B+cvubfOH2+0el7+pXz3rD2Bf3Rc/7NOo5Pf55cwnO9\nM5RSTwB8AKBqWVbxkp72Poxl7vTn1uwdlmVNTv+O4Fmvd4n4nwG0APz+K3wNM55XP54KwHcqpX6/\nUupHlFI/oJRKXsHr3IexfBOuY501++YljqdS6hGm6b1nlmXtzHnIz2E6f37tuzzfXFxTyO4DTHUE\nRwB+BsBPYKox2Dm9/ddfwmt8GdOQ3RjAf3hGOO/pnFDiBEAHwDfNec7HmA7qUwC5mfu+gulk+umZ\n2z87fQ+/ceb236O9v++a8z7GAP7wJXwO/+3p6/yXVzymv+70ff8xAH/x9DPszv7dZizf+rf+EOaE\noE/vUwBqp8/3H1zhWP7Hp3//V07/v45LTPmZ8bye8dTeq35NMN0k/6gZy7u1zsLsm5c9N7/v9Ll+\n5oz7f9vp/f/buT/Pq/oyzHmzeQB/f2ayDwH8cQCxS/xiPD3j/r97+prfOeeL8SfP+J0/ffo733vG\n/X/19MsePP3/l06f7xfnPNYB4BtnfDESAJ4ASFzwM/jNp8+/DSB6xeP5EzNjeQDgX7uk5743Ywlg\n7fR1h3hdc/Mj2mLyB65oHD/G9KT757TbropQmfG8wvHEtGDk3z8dPw+AFQA/iGnxyKVs/vdpLN/w\nGVznOmv2zcubm//26ev8L2fc/6+e3v9z5/08r6vK71swrVAoAfgOAP8CQAzA78A0wvH9Sqlvty6n\n8uUfnnH73wPwXQC+ZeYxFqahyHlg/va7lVL/ypz7MwCcmA7qVwF86+ntr5WbW5Y1UUr9IwAP59zX\nwFRbcW6chkwZKfptlmUdXuT53gbLsv4ggD+olPJj+vf/MICfV0r955Zl/cQlvczCj6VlWbtKqR8H\n8EcB/F9KqZ/GlJx+K6b6jH8B4JswneiXCqWUC8D/evp6P3LZzz8HZjyvcDwty/qZmZv2AfyUUuqr\nAH4JwA8rpf6UdTlVaQs/lvNwneus2TevfjwvG1dOqJRSTgD/O4AkgF9hWVb19K4+gD+hlMoB+L2Y\nph3+6CW8ZPmM20uYhtyjZ9w3D9Qe/PAbXs/CtJQWp89tveU9XDqUUr8a0/zvMaanglkx7JXBmgok\n/wWAf/dUq/FfKKV+4ZLew70YS8uy/phS6l9iOg9+I6aLzT8//fdvwHQDrlzW62n4QwC+GdNUny68\nVVfwWoAZz6sez7Pez1eVUv8vppGAX42pRuWiuBdjqeM611mzb77T67wvSH7n/S367a3zvsB1RKg+\nxNTP459pXwodfxfA78PUp+UykD3j9hymgzbvRGGd8Tt8bMSyrN47vPYhpl++N72HS4VS6jsxXSCP\nAXyPdREPjYvj5zEVIX4ZwGUsNvdmLC3L+msA/trs7UqpP3T6z6sY12/B9G/8+0q9xqEsAN+hlJoA\naFmWlbiE1zPjebXj+SZw7Q1e0vPdm7EEbmSdNfum/T1cBp6e/nxyxv2PT38+O+8LXEeVn/f0Z+qM\n+1n2Obcc/xz4jjNu/8rpz6++x3P90unPd7W8//9Of3559g6llANnv7dzQSn1azE9MY0w1S/dJJkC\nppoN4PIqX+7NWM7DaVXKlwD8smVZ//IKXuIXAPwkppU6+vWXMF3gSqf//wuX9HpmPK92PM96XRde\npVVeqz48J+7NWN7QOmv2TVzueFqW9QLALoAnSqn1OQ/5PkxJ4i9e5EWu9MJUHNnAVFT2O2fui2Hq\nXTQG8Ltm7qPw7Uff8XX0aoX/aOa+33J632dzXuNMwzBMqyxGmFYgPJ5zvxszxmra3/ObZ27/vTi7\nWiF5+lrJ9/hcfz2AHqapg2++6nHUxvK1qo7T+74d05PGEYAnZizfayzDc25LYrowjXEJ1TzvOc5v\nFKWb8bx944lp+uTJnNvdmFaGTQB83Yzl7V9nT1/X7JtXM5401/3LANScv/WXLzJuV57ysyzrSCn1\n+wD8FID/QSn1b2HKduOYVkukMPXSmG1X4sCULb5vtOPnAfxJpdT3YqrteQzgt2JaOvyD7/nen576\nafwkgE+UUj+PaTjQjWk1z3diOtE+1n7td2J68v9ppdRfxbTdwxcx9bb4OQDfM+elfg+AHwXwY3iH\nfPip59PPYDrp/gam4sTvn/P+f/yd/tB3hx/AP1dK/TKmTr37mLYp+QjTv88C8MOWZc2GTM1Yvhk/\nqpT6HkznQQXAMqZzIwrgP7Es6xfe/S+9FpjxfDNuYjyTAD5VSv1TTDenIqZRjK9gahBZwbTKaRZm\nLM/ADa6zZt+8urn5pzDVMv52AP+PUurvYHqA/O2Y2v6819/6Gq6RcX8HgL+CacXLCEAbUx3Bf4o5\n7UwwLa08BrD5nkz7RzE1AvsFTMVlh5gOyLfO+Z03Mm3tcZ/D9Iv7EtMvWA3AL2PqRfLdcx7/LZga\nox2eXn/r9D3NfT28v5/Gl/G638zsdXIFY+jCVMj8tzANnfYxPb09A/A/Afj2M37PjOWb39P3Ydpf\nqoRpSXQB07Tb3M/zqi9MF5gxgL9vxvNujCeAMKZ9yf7x6esNMV1jv4ppRVjKjOXdWGdn3oPZNy9x\nbp7+ng9TEvb09H2VT+fnhxcdL3X6ArcOSqkqgL9tWda8U9W8x38ZU6Hej1mWdRlVDwaXBDOWiwUz\nnosDM5aLBTOeN4vraj3zXlBKfR5T067L8jMyuCGYsVwsmPFcHJixXCyY8bx5XIux5/vCsqyvY+rd\nYnDHYcZysWDGc3FgxnKxYMbz5nErI1QXgIWzvTEM7hbMWC4WzHguDsxYLhbMeF4Sbq2GysDAwMDA\nwMDgruA6Ws8YxnbDsCzrUlqJmLG8HTDjuTgwY7lYMOO5ODjPWC5ays/AwMDAwMDA4NphCJWBgYGB\ngYGBwQVhCJWBgYGBgYGBwQVhCJWBgYGBgYGBwQVhCJWBgYGBgYGBwQVhCJWBgYGBgYGBwQVhCJWB\ngYGBgYGBwQVhCJWBgYGBgYGBwQVhCJWBgYGBgYGBwQVhCJWBgYGBgYGBwQVx5a1nbgoOhwNOpxMO\nh0MupRSUUhiPx5hMJhiPx/JY3jeZTGBZlvzkvw0MDAwMDAwMzsJCECql1Gv/9/v9cnm9Xrjdbrjd\nbjidTgwGAwwGA/T7fSilbPcdHR3JdXx8jOPjY5ycnAj5MjAwMDAwMDCYxZ0nVIwszd7m8/kQjUYR\njUYRCoWEXHk8HrRaLbkAyH0ulwv9fh/9fh+9Xg/D4VCiVoZQGRgYGBgYGJyFO02oSKZmCZXD4YDP\n50MkEkE6nUY8Hkc4HEY4HIbP50O5XIbH44FlWVBKIRwOIxQKwePxoN1u4/DwUNKAk8lEIlUGBgYG\nBgYGBvOgLMu62hdQ6lJfQCdRkUgE0WgUkUgEXq9Xbnc4HEgmk0gmk0ilUgiFQvD5fPD5fHC73Tg8\nPES73Ua73YbD4YDf70cgEIDL5cLh4SEODw/RarVsPzudjhCr4+NjjMdjua76M7woLMtSb3/U23HZ\nY2lwPpjxXBzcxrFUSsHlcokUwufz2eQTs7pUYjKZ2CQTurRiNBpd1tu71biN43nT4J6slLpTuuTz\njOWdi1AppURsnkwmsbq6itXVVUQiEdtEj8ViiMViiEajsgiQcKVSKdFFOZ1OuN1ueDweKKWEUB0e\nHqJer6NaraJWq6HRaEg6kAvE0dERRqORSQcaGBgsDJRS8Hq9CAQCCAQCiMfjckCNRqNwu91CuHRC\nNR6P0el05Go0GqhWq7AsC0dHRwBw6w+fBpcL7tfcsymfIbFaNNwpQkWmywq+RCKBBw8e4POf/zwy\nmYzc7nQ6EQqFEAqFEAwG4XA4bJElr9crl14JaFmWLSpVLpdRKpUQDocRDAbldp1t8zkNDAwMFgFK\nKXg8HoRCIUSjUSwtLWFtbQ2rq6vIZrOydvp8PhuhOj4+Rr1eR61WQ61Ww/7+PizLQq/XQ7vdXsgN\n1ODN4J7NwAX3SlbRLxpuPaHSSRLDzz6fD16vF2tra3jw4AEeP36MbDYrj3M6nRKm9vl8sCwLw+EQ\no9EIx8fHCAaDCAaDCIVCcLlefQTj8RjhcBjRaBTtdhvBYBCBQAB+vx+hUAj1eh31eh0+n09E7Scn\nJzg5OQFgTl8GBncJDofDltriv7kmsNL36OgI4/FY5voiznM9kuD1epFIJJDJZJDJZLC6uoq1tTWs\nr69jaWlJ1t95hKpWqyGRSCAej8PhcKDf78u6OZlM5DK4/WDql0EHfaz14MbsxfuB6Rzj/PJ4PDg5\nObFV0OvfCd3OiP++a9+XW0+o9NBzKBRCPB5HLBZDPB7H48eP8eDBA2QyGZnAvDiAuvjc6XRiPB6L\nlmpedaDH40EgEJD7XC4XQqEQksmknLzq9TqKxSLcbrcsuBz4RVxsDQwWES6XS2QB/EldpmVZaDQa\nqNfraDab6HQ66Ha76Ha7C1mg4na75QAaiUSwvr4uVzabFXIVj8dtG6S+hpJsut1uhMNhHB0dodFo\noFgsIhAI2DSoZp28/XC73QgEAggGg/D7/QBeaZhdLpfsr7y8Xi88Ho+NfOl7MSNUus5uNBpJsGM4\nHKLf72MwGNhuu0v6u1tPqDweD8LhMOLxOFKpFPL5PJaWlrC0tIR8Po98Po9MJoNQKGQTrOvRKsuy\nJMJlWZaw7nmEyu12A4B8CUKhEFKpFLrdruipqtUqPB4Pjo+P0Wq10G635TlM+s/A4G7A7XYjGo1i\neXnZtq4sLS1hMplgZ2cHOzs72N3dRbVaBQAMh8OFJVSBQEAqozc2NvDhhx/igw8+QDKZlCrpQCBg\nM02eFaVzvR6NRuj3+yiVSojFYggEAhgOh1I1bXD7wWBCIpFANBq1icv1QAdJFy99b50NbjBCdXR0\nhOFwKIeUbrdrk9u02210Oh2Mx2NDqC4TXq8XkUgEqVQKq6urePjwoVz0maLw/DLA0xcAhMNhqUgY\njUZIpVKoVqtIJpOYTCY4PDxEoVCQFIE5dRkY3B243W7EYjHk83k8fvwYm5ubePToETY3N3FycoJP\nPvkEwWBQHj8YDNBsNm/wHV8dXC4XgsEg4vE4crkcNjY28NFHH+GLX/wiIpGIHEKdTucbnyccDsu/\nO50OdnZ2EI1G4ff7hUxRf2pwu+F2uyU7k06nbRkgkm9eeqSX+yEJmB7FOjk5kYKufr+PZrMpV7Va\nRaVSkcgn9927hFtPqMLhMJaXl/HkyRPJ4edyOTn1UFg+G206D2afQ8/nsgy40+nIoppOp/G5z30O\n6XQajUZDLlYQmhTgu4MhZN3V3uVyweVyvXFsZ8PNemTyMr4Ts+j3++h2u+h0OmL+yrD0Xcr131Zw\n7PW0ksfjgcvlsrWDGg6HUpbPCrKzoJ+sfT6f6CLT6TQ2NzeFSOVyOUQiEemm0O/30W63UavV0G63\nJcKyiPD7/Ugmk1hfX8eDBw+Qz+cRi8Xg9Xrhcrnm2iS8DW63W0haKpUSyUW/37/Cv8TgfaETJXoy\nhsNhJJNJ2W/T6bTN89Hr9YqVRiAQkAhmOByWAi/ufZZl2cadES6u036/H7FYTNLtlPUUCgUopXB0\ndGRLMXL+8+Jr3Ya99s4Qqo8++giPHj16zXuKOdurAL8Ix8fHGI1G6PV6ODw8RKPRAABkMhkkk0l0\nOh1sb2/j5cuXGAwGIn43WoF3BxdfVlTqrYPeNL76SSkYDMp34qq+F7VaDcViEYVCAeVyWcLUFFga\nnB/UMHKRZuEIvw+691u73Rah85sIlV5l5HQ6EYlEkEgkkEwmkc/nhUxxbeGGcHR0hF6vh1arZSNU\ni5rS9/l8SKVSWF9fx+bmJpaXlxGLxWQenWcuMY1I24WTkxP0+304HI6F/RzvGnR5jNvtFhLFK5vN\nIpfLIZVKAYAQFz6eREdP/wEQTfHJyYmN/LhcLluVqN/vRzQaxcnJiWijE4mEbR52Oh0pDAuFQphM\nJraiMF3UftO49YQqEolgZWUFn/vc5/D48WNbJc5sZcFlYx6harfbaDQaCAQCyGQySKVSOD4+htfr\nxWAwQLlcli8do1QGbwfDy/qE4qVXYs4iGo0inU4jnU4jkUjIZuz3+9+anjgPdnd38ezZM4RCIVmI\nOOkNLgYSqmAwKCdVXuFwGCcnJ9Jbs1KpYDweo9frvfWz18u2w+EwstksVlZWsL6+Lim+R48e2SKb\nOqGqVqvodrsYDAYLO5/9fr8QqsePH0ulHqO+54n26mnEVCqFwWCAVqt1JZFjg/OBhMrtdsPr9SKV\nSuHBgwd48uQJVlZWkMvlkM1mReYyW3xlWRYcDofNisiyLKmIZXDh5OQEvV4PPp9PBOqBQMCWUWi3\n25JeDIVCOD4+RrvdRqVSQSwWk+/keDyGy+XCyckJut2u+FrdhlTyrSdUFDnS9VzHVU9MPZSoE6vB\nYCAbN9/TwcEBIpEIfD6flFlTL6Bf+nPe9ODfFjAEHI1GZQLrOfk3EaNYLCaEKh6PyynpbZGt88Lr\n9dpKed1uNyaTCYbDIXq9niwki7rxXjb08mtqmjKZDLLZLFKplCyirBqj/kIphU6nI2Lxs0ANB0/D\nqVQKy8vLePDgATY2NsRbKZFIyPOTSPFidIrVvIsIrrPpdBpLS0tiGUPxucHdh16hp0eXWN0ZCAQk\n5fvo0SOsrKzYDF1Zdce5wGs8Htt0Unq7tuFwiEajgWaziUajIQR7OBwiFotJO7hwOCz742QyQTwe\nRzabRbvdxtHRkRCqRCIhqfdutwuXyyWkzkSo3gEM5XGj0h3Prxqz6QJdQ6CTLH4RdM+Nk5MTGWz9\n9/m3GI3VqwlOkSMLD9bW1oRQRaPRN0aoZlN+FzlRvwv8fj+y2awYxDISNh6P0Ww20e12RVtl8Hbo\notVAIIB8Pi9FJ6lUytaDk7omaqfK5bIUkLzp+anRiEajWFlZwdraGh4+fIi1tTUkk0kEg0EopTAa\njaTKqFgsol6vo91uYzAYiIHvos5XRhloT0Mt40XmEaMSzWZT0qaj0WhhP8PbDD2bw9S2foVCIUQi\nEWxsbIiGLpFIIBgMSjSo3W6LgJxVeN1uF8PhUFJ4Xq/XZo0wGAxs3Uf0LEQ6nZYK22AwKBV9/X4f\n4/EY8XgcT548QTablXRfOBxGq9XCYDBArVaD0+kUjmAiVO8AEhcSqqsUHM9C3/DPIlQ09ZxMJjZC\ndXx8LJYNusBaJ2B62PQ+Qo9OBINBJJNJrKysYHNz0+YN9KYI1Vmi9KtCIBBANptFIBCQ6NnJyQkG\ng4EtXWQI1buBRpKBQEAsDJ48eYIvfOELSKVSYuTrcrleK7EOhUJvJVROp1PGKpvNYnl5Gevr63j4\n8CGWl5clmqmUwnA4lA4JxWJRSABTfYtMqJxOp5BaGh5fdJ09Pj6WSq56vY5Op7PQwv7bCq6zHNNw\nOCy+YslkUtLqiUQCuVxOhOicXy6XS9oKVSoVHBwcoFKpiGFru922aV7H47EU6rCQixezTYlEAsvL\ny5hMJggGg1haWrIRqpOTE8TjcWQyGTH0JmkrlUqo1WrY2dmRw+xtiaLeGULFECLw6gsy+7i3PY+O\nsxYK/XH6xNcfz7AkCdU8V1heTGdx0+dGO6uvWtSF+k3Qyarf70cikcDKyopYYrwLoZqHi6ZU5303\nSIKZUuSCRKFtu90WjxWjp3p3MDLCk+vy8jIeP36Mb/7mb5bKMEZ6dZ+aUqn0XoSKdgDLy8vSYSGX\ny8njLMuSCFWpVLJFqO4DOSah8vv9NquIWbxpTs0eEEejEbrdrkSoer2eiVDdAKiTYqovEokgm81i\nbW0NS0tLyGQySKfTSKVS0gM3Ho9LZS2JDufd9vY29vb2UCgUUCwWRVPMSxeiDwYD9Ho9uSKRiMh3\nDg8PhUxRwM6o1mQysRE8vUWc3+/Hy5cvRcd6nVmrt+HWEyp6vxwcHEgVGEOWOvQu1rP29YwY0JRv\ntoJstiKBKYXRaCRXp9PBwcEBDg4OUCqVJM87mUzgdrttTZZZbRYOh+F2u23iWp5+i8WibMJ8r/cN\netdx+pIcHh6i2WxKKsjv978x5acLJWfHj8+th4Jnw8K6Iz4XHabyeCrie50FjWD9fr9UJ7Ic2ODd\nEAwGRSD+4MEDfPDBB9LehIcmphAqlQr29vawt7eHzz77DAcHB+j1eq89p16BFIvFkMvlRGy9uroq\nYmse1ngwajQaKBQK2Nrawvb2NiqVyr0p8efc6XQ6aLVarx0KZ8H1Vne7ptN1r9dDv9/Hs2fP8I1v\nfEM+x9FotLCte24znE6n7J3hcFjm2sOHD5HNZkVaQd9FagdZmNHv99HpdFAoFHBwcIBCoYBqtYpG\no4HDw0NJ0VEzNc8NnfsbD6CtVguhUEishprNJo6Pj+HxeJBIJABM9bHco/mck8kEnU5H7FL0ff42\nfK9uPaGiqK1QKCAYDEoIcB6h0lODepsDpggODw/R6/VE3JZIJCRlw4oE3WiMX6bBYIBut2sbfI/H\ng36/j8lkIhsxTwA6mw6Hw1hZWcHy8jJWVlbw9OlTqQjkyZfv/b6BhIoLM20pms2mjai8aaJwIs8b\nP6Zj3wZGSRhFZHSMGy8wP2rFiKQhVOdHIBDA8vIyvvCFL+ALX/gCstksstnsa4ed4XCISqWC58+f\n45NPPsHOzg4KhcJcQqWTYkamSKhyuRwSiYQQKt25mW1Stra28PLlS9Tr9bnPv4hg2rrdbqPVakm0\nYd5hRp+3TJPy99hNolarYW9vT4ipXrBhcL2g43kqlUIqlRIN4ePHj5FOp239cfViDArJ9Yu30YeP\nRJlkSo9qcT/m+mxZlqSBJ5MJ/H6/PGer1ZJ5m0wmhQRyHeD6zuCGTtBvC5kC7gChGgwGaDQaODg4\nkGhFJBJ57XGzqUHdcLHZbKJcLotvUD6fF9G42+2WBbXX60n0qFgs4vDwcK6J42AwQDgcxmAwwHg8\nFt0OqybYnNnn8yGTyeDJkydy+f1+DAYDFAoFNBqNe0umgFenXAqC9QgVScrbTBvZS5HVWXoE8G2/\nS1AUz8ommthRT3JWKNlEqC4ORqi+8IUv4Etf+pJNE8dFkpoMEqqvfvWrKJfLUgAwC84/ts0goXry\n5Ami0agYApNQUevRbDYlQvXy5UuZ7/cB4/FY9C6tVktadLGHmw49E0BCValUUCqVsLu7K1etVpPD\nTb/ftxExg+sD7SuoUWWEanNzE6lUyiah4Xhtb29jf3/f5rnHyL9epMHokF7JDthlF/qY069vOBzC\n4/GgXq8LqeLcDAQCopukPRLTjtyP+/2+RKhuUx/dW0+omLstl8sIh8OIxWJzFzmdUOlNFimKrFar\noovgY3q9nvgI6b2nSqUSyuWyDBzTf3wdvfSU5aYs8+x2u+Kx4ff7kclksL6+jtXVVSwvLyOdTiMc\nDtsc3m9D7vcmMatf8Xg8EhVst9tv1MmQUDFlyPErlUrv3DPM6XRKBUkoFBKtFFO6utP2rPuv7qBP\njch9Jcjngd58PJfL2RZFLtpM2Xc6HYkiNRoNOfnOe06/349IJCJiW5Z/BwIBiSIfHx/LczYaDezv\n76NUKqE6v4/eAAAgAElEQVRareLw8PBepeKp/avVaiiVSnJQ1JvictOd1Zkyyqen/PRNj3OIzzP7\nHAZXC6a+8/k8Hj16JNop9mhkRInRo2q1iu3tbezs7MhaWq1WX4s2vQt0/ZaeyeF7YhSU3xHunT6f\nT56DFgmVSgWVSgU7OzuoVCrodDoimbkt36dbT6j0ic4vxbz+Pnr4nlEknmApZm00GqhWqxgOh6jV\natjd3ZWqLP4eRa/MIeuLKk/OXq9X/HG4aOfzeQCQKBojVNFoVJo3G5wNlsEqpdDtdlEoFN7JNkFP\n85KUcQzfNb3AqCcvr9crXimsLAUgVZu6fqTdbqNcLmN7exuFQgHNZvPO9Z+6TZglrbpQlRFnLupn\nnUq5KNMcNBwOw+/3SwsbvRqT1UI7Ozt48eKFnMT1k+99AA+uxWJRNlma7QKQzXBWAKxvmGwmH4/H\n0e/35XbgdUmG8eK7Png8HsTjcayuruKjjz5CNptFPB6H2+1+bd9sNpsolUrY2dnB3t6eZGl4uHnf\nSJDD4bC1e2IWgPKdtbU1pFIpmZ9ut1uIux55ajQa2N7exosXL7C1tYWdnR00m02joXpfMGdaq9UQ\njUbFy2QWs18Mveca00gkVPV6XU5cepqQkSqKLGcXCy4w4XAYiURCqtDi8TiUUtIPi9EriqrpkXTf\nI1FvAglVv99HpVKx+Zq86XObLUTg2L1Pbz2eljiewWAQ2WwWg8FASJn+Hvh6dPItl8vY2dmRNLEh\nVO8P/fPVDf70lBzH9ejoSNo6zVtEXS6XWCUkEglEIhFJ81FkTUJVr9fx8uVLfO1rX8P+/r4IqPVN\n/z6Ah5FisSjpFqZMWQE425VCr9Kl/oWEip6BAIRE0ZBVJ6r35fO9SbAwanV1FR9//LGQGt2Ekxma\nZrOJYrGInZ0dHBwcyHrK+fY+Bwx+P7gHcq+kfjmTyYh56LwDj07C6/U6tre38bWvfQ1bW1uo1+ui\nk71N8/TWE6qjoyN0u10opYRQsZKAE3Z2AdaJEdN+vLrdrs0ckMI2PX3AxZ3MmmFJnnop7uPpNxqN\nwuv1SjqSC4+urWKKQe89ZPAKHC/2SbwOcJzdbrcsGuwTp+fo9Waf/I7p1Srlchm7u7uo1+sS7TR4\nP8xbFGdF4yRSDPPPg96+ZpZQMdXH1xqNRqjX69jZ2cEnn3wi4liKZu8Tjo6OhFDp3l2pVEpEwT6f\nb26nB0ogGKHinJmNNDBzwOebjQIab76rATsQLC8vY3Nz0xYFPjo6skWBDw8PUa/XUSwWUS6Xbc+j\njzd/vunfnIsMQKTTaWQyGemGwXZhbBlGQsX3x8rT0WiEarUq83R7e/u16sHbgltPqPihOhwOcWqt\n1Wool8u26gSW2TMSNCsWpvfGxsaGrWpBJ1ij0Ugq9JxOpywo6XRa7Pd1x+V0Oi2mgCzVBiBsn+kJ\nvbR/a2sL5XIZvV7PFkY1uHroE14vCw8Gg1heXpaLpo+RSEQ2YS4Q/X4fjUYD9XodhUIBL1++RLVa\nvRdu2lcBnpC52erzjxs1o5U83ASDQSG7PKHq2gwaAq6srMgJOBAIQCll09zplUtMbbxPZHORwIMr\nNzSaELfbbUQiEZvVjN63jVpUGhmTvKbTaemZyBQ8q8J6vZ5EHPXiIV76Afe2bZh3FXokfzbirxMg\ndhXI5XJQStlIL//NXnrce/UKaT3KRNNeymN46dkAWjlQKqObgtZqNTEP/fTTT3FwcIBOpyMk8Dau\ns7eeUDGNM5lMcHh4KGm7crmMaDSKSCQiA8j0kE6mjo6OJLxIkTP9pFhp1263JSqhb7SsDlpbWxN7\n/FAoJKffeDwu4jnqbBwOh6Qcm80mOp2OzafFEKqbBU/Oeof0eDyO9fV1bG5uYnNzE/l8XvxZ9Cbc\nANDr9cTc7uXLl68RKhN9fD+wlJqHEF1HQXdnps7ZPzMYDNrSupPJxLbAs0DkLELFTV0v2WY7i0Xu\n1/cmUKDP9fbk5EScscPhsHzubFzNi5G+o6MjsYnhOPCzZpaAbtlch3l1Oh35STH7cDi81xXQlw09\nwq53/NCht2nKZrMSSSS51a2IfD6fpPFYzMOLqXX20KQ2lVV8nMd6QIQ9Uhkta7VaYrtBjSMJ1W3e\nN289oWKK7OjoSLRQjFBxIWUPN5IpCtT4k8I1st/nz59Lmo4GjtQ96dqd9fV1fPjhh/joo4+wvr5u\ns7+ncSA1GWTkNEbr9Xrix6IvHHt7e1Ly/b4VEwYXg97qhtFMFg1wrD//+c8jkUgI2WLUkQsQKwm/\n8Y1v4NNPP0WhUECtVhNCBZh0xftAj1CxtJ7zeDZCxYU4EAig1+vJAmxZllTcsvmqHqGiZgR41Q6l\n1WqJDoOk6j6n4xnxo+t/p9NBuVxGJBKRQyT7vdErLJvN2ipw6cLN1iJ6nzUWcNAPsNFoyPpYr9fl\np15N+65VugZvhx6hAjCXVM1GqNxut0QNdRnNaDRCMBiUpvSUvzASpZMlCtH5HWLwQ+8tyItZIlqY\n7O3t4dNPP8Unn3wi9gqMUPFvum249YSKhIg5eFYhbG9vYzwei8knCRVPtWeBCypFduFwGKlUSkiQ\nHr5cXV3FxsYGVlZW5As2zz1YNzIbj8fiy7K7uyuO6Dx90ZuFFWS3pTphkcHvhR6eTiaT0j6GDXlX\nVlakEScnPjdtXRxZLpclylmv19Htdg0xPifoE7e/v4/nz5+LxQHnMUkVWxNRBxKNRiXacXR0JNHq\nSCSCBw8eYHl5GalUSqo2OWdZNVyv11GtVmXeMzJ1X8eQmhXg1SGW3m46kQ2FQhJharVaIlZniy1K\nKIbDoc1LCHhVLMADKLtJcMyZgtXXS1qR6MUIvEz06t3AQAK/9wwMcM+kXIU64OXlZYzHY6TTaZvb\nuZ6iJaGinpgpvFgsJgEHPqeeqgfsHlV61eisNQ3lGCz6YqRaN8QGYPtOcB3WgyrXeUi69YSKYJn6\n4eEhCoWCfBkCgQASiQR8Pp9Eid7UKJFh6VwuB5fLhXQ6LSmA0WgkUSeXy4VkMimpH94277n53vhl\nI5l6/vw59vf3Xwt7s1LxNhmSLSJ0bQAnJ3P2+Xwea2tr4hG2tLQkDt1M3+rOvhy/crmMSqUirReY\n0jVjeD4MBgMUi0V89tlnmEwm2NjYkJQCo0o06sxkMtjc3ITD4RCvKFZV6hqN1dVVrK6uyuJOcgxM\nCVy73RZPm1arJeklM4ZT6MUAuvkpi3qYLi0UCkJ6ueHpkYpQKCQbqc/ns81HVtYy4qU3vuZFPyu9\n+wFJNNOTBm/HyckJut0uarUa9vf3hfxQ0kD7BKUU0uk0HA4HQqGQVLuyEl439mQXgng8bjPkDAQC\nkqmZB36fZrWP1ExSPxkIBKSHXyKRsH0HBoOBLYWpN19moEJ3aed1HfP7zhAqYMpEW60WCoWC9ORL\nJBLI5/PSN+9tXaeVUgiHw3C5XIhGo7bKIZb68mIKgSWm85oyA5DTHCd6uVzG3t4enj9/jt3dXVvI\nVBfFkkyZhfzqoFtf8LSTSqWwsbEh7vUbGxuiD2EPOd0agV5mNJhle416vS6n5/uYJroMDAYDlEol\nWJYl+giSJ5rf0mAyk8lItS9TAI1GA8PhEKlUSpquMuIRj8elYIXzlqkn6jAPDw+FUBlMwawA1ysW\n6/Cq1+viaK+njrimUi9D8bFuR8LUj66l0btV0KCXh9x2u439/X2JCFerVQCQiInB20FCVa1Wsb+/\nL956wWDQFijg/hkOh5HNZm0pcDrp83DJ6j0WfekSGL2X32xloK6T0jtlnEWoEokENjY2RJxer9el\nCpfvq1qtyprMlCD3W1ZoX1cG4c4Rqna7jclkgl6vh1QqhXw+j8PDQ6lE4RfjLDgcDtk8LwvUeHHj\nrVarKBQKIqYjWzYnqusFQ9k0eqRYOZfLYWNjA5ubm6KPA+whZE56VvVR78E2DPV63eambTbk82E4\nHKJarUoRRzAYlNQe9VNerxcul0sOQqy+1QkV9RzpdFqIMVN9XLipEaLVRblclgiVgR3ncYlnZIPE\nKZFISEqIRFfvoUoRczQaBfBq09Vbhx0eHiIej4v+hulDHoBnW4+YA+rrGI/HQqh2d3cBQAgRyRTX\nShZZzeqrJpOJzYKIaXj6u+mgDkrXwek9V/WCEr4ePSE5nqzW5frNyupGoyH6ST6efX7Zi5DPz/fb\n6/XgcDhsYvarOgDfKULF1NpgMAAA8cug4zlPpvP6T10lSKjYi67b7drK6E1a7/qgpxX0akxGMvP5\nPJaWlrC6uopsNitpJT1MzH5mbBDKExBPeIVCQVIOZmwvBkYBB4MBXC4XCoUCnj59CqfTiUwmI7qo\nQCBgC+Hz9MlNORaL2RzRmfpn9Jgn1kqlgmKxiIODAxSLRanuM2N4OSAZAiBl9/Q3qlQqNmNktgNK\nJBJyyA0GgxJN9Hg8khK0LAt+vx/pdBr5fB6VSsVW8EMZhdFXvY7j42Nxwfd4PDIXBoOBtEJjpR7B\nyJEOGrxS+3RWRogRJ2Z1Tk5OpCCBr80IFYXr1EaRsLF5MsXtHP9kMolIJGIrONMzVbQ+YQRTryJl\nB5RWq4VOp3Mln/WdIlRcfAGIQLhYLCIajYpGgpGI6wSJHiNUJFR680azYF8PdHFjKBSS8vnl5WXk\n83nxmkomk4jFYvD7/bYUB/2JaKtRKBRQrVZtuilqORiZMmN7fnBO8zMsFApQSqHX6yGXy0nUKR6P\n20wfKVpnQQqr0fx+v3iH6U1Vqb0hoSIxZtNzg4tD11vxcEIypZfH09qCKVp6/bFIhNEGbqbUyjIq\nydQP+7NynnY6HQwGA4leGExBQkWpDD3fqInL5XKSuXlTI3jdMZ86OF1Erj+WkUTOZV56gQEAm/Zq\nMBjI4yzLkmgnjWKpx+Nr8kokEkLEhsPha63keDBm8IUWIVeBO0eoqHlSSgmh4iJKBntT74sRKk5s\nGpAZXB8YnaJGIJvN4uHDh9jc3MTy8rKQKxrJuVwumwh3MBhIo9xnz57h5cuXImCuVCoiejQp3MsB\nCRWjT4VCAd1uF6VSCblcDisrKxJN1Bus0qyXpfxcbPWiArY50Rsrz0aobqufzV0E59F4PBYTXF2T\nqh92YrGYECoaLh8dHUkKihotbrapVEo2Qm6SlUoFz549g9PpFB0c12LTreAVqD0msdJTYYz0hsNh\nAPMjUwQPKTpZmtfsmo/R9UvULZLwMDDCqCRF8CxIAF5ZefBw7Pf7EQqF4PF45LVmzUe5jjOKrVty\nRCIRHB8fX2k3jjtFqAB7ewJWLuiTjykAvZExF9irfE9k/gw50huHhmVm4b4a6N5SrOTjyWZ9fR2P\nHj3CgwcPsLa2hkwmIyFjmtbxVMNej81mE7u7u3j58iW2t7exv78vWp33abhs8O7QT5sM9zPKQYFr\nq9WyVQSxApfFI16v11aGTej9FnWrC1ZnGlwu3jViq48t9TSM8uu+RuFw2EbETk5OZIMFIETK6/VK\nFLlaraLZbNqE7vc5iszPFpiSFBISFkYxGEE9FKODejpvMpnYWrjNdgDRL30+z5Iavbk5AHnd2ZQf\ni8cikYi0d2P6n4cmEjxaQNBIlGnJ8Xhss4QYDAYS3WSKmNdlfT/uHKHSwWgCF1KGiD0ej1SX8IO/\nSuh9h1hqStfZUCgkIU5zarp8kEhReJ7L5aRX1OrqKtbW1rC2toalpSWJZFCgyFPa4eEh9vf3sbe3\nh/39fREsU3zOtJAhxFcLPWXEOU2SValUbD0yc7mcNDBnio8CVz4XDzqtVgsHBwd49uwZ9vb20Gg0\nzFy8YfAww3/rOh+W47MPI123eUgmgfb7/cjlcvD7/VhaWkKxWMTe3h729vbE/4+HoPscTSa5IIE6\nPDwUkfl4PJbPlj3/4vG4dCgg0RiPx2g2m6hUKtLpQ/cFY9RerwzUG8jzor6NkSd9bPUKeMCeDiSZ\n4hrO9+v1em3O/cxW8f1TewkAnU5HitjYDJwNli/roHznCZV+qtUJFT0saIV/ldDbZ7ClDAkVc776\nKcHg8qBXp4TDYSwtLb3WQiabzSKRSNhKvU9OTtDr9WSRePHiBZ4+fYqnT59Kj0e2wbhOH5P7DpY3\nc5Pt9/sShdYLDg4PDwFADi1erxehUMims6K2USdUbDdlolM3CxIqXXtaLBZlg9T7vunGoslkUnR1\n9D9aWloCAOzv70sHDKfTiXK5bGupc1+hm2PTeJpVtaPRyGZ4nM/n4XK5bKk1zsdWqyX+ivSDYjWm\nblUwm4Kjd9VwOHxNd8oDkm5VQ7KlR54YqaJNg96Kiofo8XiMWCwmVhw0AgWmJqGs8OUe7Xa75fth\nCBUgkR/mXFkuzRwvzcd0D6m3+VSdF/rpmf2LGE7l5kyPDn5p9NDo7GVwNvQGuiStXGxXVlbw6NEj\nfPzxx8hmsxKpDIVCEo7WyRQ9w7a2tvD06VN8/etflwl230+2NwG9pJnebvPgcDjEXmE4HNoqu7iQ\nc5yptdze3hYnbtPW5GbBKMU8YksrBfpX6X3ilpaWcHJyIr0b9cfRCofzXI9EMxqjSy/u0zrLv5n2\nBbrWlEUBzWYTkUhEUrAkYhwnFus8e/YMlUpF0nNsGcPKOv1zZnECD0jvCt0/ULfYoJyDa34oFJKi\nEhIwWqXwO8IsBgshDg8P0W63ZQ+4TE5wpwkVJwSFp41GA3t7e7KwMjdME1CGii8b9MZZXV2Fw+EQ\nYbrePkF3/+XPwWBgKwXX/32fJvv7gJOLZb56KDiVSkkLGbYdoficuhyeqIrFInZ2drC7uyt+YbVa\nTRrDmsrM2w22LqH7PUP9AKSqr9PpiG9YrVbD4eGhrc2Mwe0E5ROMLLG/oM/nQ7fbFYNdpvW57tJV\n+8GDB/B6vTaD0WazKSX0sz5G9w0kKoxExeNxpNNpLC0tSS8+p9MpLWtoP8AeusViUdbKs1oC6W1f\nzqsfJqEDpnOaAnRmgpj29fv9iEajSKVSGA6HCAQCtp6FeusjAJJN0t/rZeFOEyqC4rdGoyEmZtTU\nUIAcjUZFwHbZIPt1OByIRCK2ViXU6fCioytbZ7CElV9c4HymevcFrEjJZDKSymOrCxq9so8bm2az\nko9GgXRf3trakko+ihX1kLUhVLcXZxEqpZSUS9M3rFQqiRErU7hmft1eME3EDV1vB9ZoNFAulxEO\nh5FOp9HtduF0OhGPx+Wnz+dDIpEQXU0sFkOxWESpVEKxWJSDq97z7T6B9gfspUhClcvl5hIqVsjq\nhKper9ui+G/KvJzHSJO/x38z8sUWSIxceTweRKNR+S7MRqv1ikPdFNYQqjPAD4OEin26AoEAotEo\nksmk9Pm7KsNPEqpIJIJ8Pi8MmF4YOrGiaNLv96NarUo6kKkNln2aBX8+GKHKZDLY2NhANpu1NenU\nzQJpPKeUkipMnrQODg7w4sULfPrpp3jx4oVEr0wbmbuBWUKlp/pHo5FEMVhkQELFjfS+baJ3Cdyk\nWXQw2wOQRSjJZFKiUuvr63KwymazGI/HNkJFLetgMJC1lqnh+wYSEb/fbyNUS0tLEvFxOp2yJuqE\nirYjzWbTNo9mfxIXmWd6hIqielbw6ilLmr3SBoIViMArKQ5TgHrhy1UcrO48oSI4OciWGRpuNpvS\nuuBtOVz9xEIROYkRHVnp7MrHzDJhiug8Hg9OTk5EOMcIFJuI0luFlSh0Eq5Wq+Jfdd8Xfb1HGE+o\nbGy9srKC9fV1W5SKbtkUJbKShSFimgAWi0VsbW3h4OBA+j/pbSwMbh+UUlJ04vP5kMvlpMs9I5EA\nRFNZr9dRKBTEJqHb7Uoq1+D24yzSywMne7/RL47iZK4Ffr8fXq8XyWTSpm9lyxVaoTQaDdmA7wvR\npraY7uLpdNo2j2jYySDF3t4ednZ2cHBwIK2erluDqI+N2+0WPVU4HMbq6qqMPaPVPFzpFYcMbrTb\nbRGnX3b6f2EIFWDXVHETbbVaaLfbiMfj73Qa0QWtNJGjZT4jIPS80T1O9PJPvdkjxfAUyCmlxCqf\nZZv0P/J6vWKCxvTffZjg86CfShmJCAQCsgisrq5iY2MDqVRK/EpCoZBUeTocDnFpbrVaqNfr2N3d\nlatQKKBUKhkydUfANjPcMGn2ycMS3ZNHo5FEIemgXa/XTXuZBYGeChoOhyiVSvjss88wGAzEJmV1\ndRXpdBrj8VgKlainjcViyGQyopvkJkvx9H34jtAAm6a5mUxGqiOZSmPqvFarYXt7G5999hn29vbQ\narVuPKrn8/mQzWbFqPnhw4d4+PDha1IPwN5DkM22W62WVHJfpgcVsICEik6vejkutRNv+yIwxMiU\nHdMGpVJJBOQejweJREIiWMz1szEzN3PgVUkooyXsT5RIJHB0dIROpyOmZ7VaDScnJ1I+zDJSvq/7\nBr0qk7n+WCwmE2ltbQ0bGxuIx+MSpmZqlwsCS+ZLpZKk+La2trC1tYVGoyHCZZ627uPnfFdAQpXJ\nZLCysoK1tTXRfDD6zKvT6Qih4qm63++b8V0A6L5Io9FI1uaDgwM8evQIvV4PAMRNn1ckErE1R/d4\nPNKYm1Gvq2yae5vA7AgJKHtm8sDPazQaCaH69NNPUa/XxbfpJkFC9eTJE3z88ce2Hq30LWOBCvfz\n4XAohIpBjH6/f+kSj4UiVARPqgz9JxIJKZOk1b7eDmH2dzmxmD5oNpvodrtSScCqBt2sjCFIurPq\n0SldGMd0oWVZEmZl6JKDXa1WAbwSp+umafflFMXPzul0SmQql8tJVU8+n0cul0MkEpFSWU4ignn3\nfr8vDTJ5OqFTNnPreqoXMOTqtoBzlP5E2WwW6+vrWFlZQTqdlvGnXrHb7dr0HtVqVea9wWKAc/Po\n6Ej6ajqdTvEWYpQ6k8mIEShdtGmxQE+iarUKp9MpRUP3IfVHuwn6N/FQMrt+MqrDQgD2qL0u0qk7\n5POgzC4Jy8vLePToET766CPRzOrZI75/ZqrYFo4X9+3LrqhfWEI1GAzQbDbh8XgQDAbFAZal9NRi\nzLqoM6IEvDINZJ8jl8uFo6MjGRC9vxtL96PRqGzwTAHqERQ9BElyFwqF4HQ68eDBAzEordfrorsa\nDAaSHmRLhUUGJxBJUiwWw8rKCh4/foxHjx4hn88jk8mIVoq+VLOgXiKdTtsEqG632xahYi6d12wL\nBYObATUv+jguLy9jY2MD+XwesVhMWjv1ej2pnj04OEC1WhWdBPsvmrFcPDCrYFkWOp0ODg4O4HK5\n0O/3sba2hm63i+PjY1nruZ4sLy+j1+tBKYWDgwOpAGSl+CJXWlN+QhnFWe3ZmB1gtTyjgmf1+rtM\n6B0wPB6P7MPhcFjIVD6fRzKZlKIUis91CyJKPphtajabst5fhaZyIQkVACmdtiwLPp9PTL1IqizL\ngsfjsX2RyIhZHUbTSBIqp9MphKpSqYiHUbfbRSwWkyoT9iYiIaNlAwmVniokoWJ0y+/3S1d1VgC2\n223s7e0BgCwQi7456CJ/enx99NFH+PDDD6WvE0vl50UaAYjDrt793OVySYUlrRKazaZUYbIKBMDC\nLqh3BazkCwQCon0hoeLJ2uv1ShFKtVrF7u7ua4TqMltLGNwu6IVEtEPp9/tSiMCuFdx42RduZWVF\n+sXR/FnvdAG8MghdNOgVssFgULRT8x7HgEQ4HJbP57oIlR6MoE0OU/4PHjzA8vIyEomELTjCdjej\n0UiCKpTVzBIq/eB8WVhIQsUIFRs6ejweaVmQSCQAvIpezIJhRfpY8ctkWZYQKupytre38ezZMzSb\nTemczua7nLz0p2KfMUam2MeKrxEIBODz+ZBKpbC+vi4Vis1mE7VaDUopdLtdlEolABAR7iJCj1B5\nvV7E43EhVN/0Td9kS9e+aXK73W45vTCkzRY1dE8nIWNJLYsM7qtHzW2CvvDHYjEpj97Y2EAikRBn\nZOola7Ua9vb2UCgUxOeNJNlgMaFHkdvtNvr9PsrlMgqFgpApaoOof6U3YTgcRj6fh9/vF71QpVIB\nsNiHKZfL9RqhelOEihq0Xq8n6+VVQ6+GD4fDyGaz2NjYwMbGBlZXV7G0tCQRKmYo6IxPzVS325W0\nbqlUeo1QXQUWklABr8RolmWh1WqhUCjA7/fj5OQEq6urGAwGAIBwOCyhxdkcMsszuXjTXZZh40wm\ng8FgINVlbrdbxObUXw0GA9EBMe2gVwHyOdmyhsSNZII9rNhzLhKJ2JpRLvLEJ/SS5/cxZiVRIvmM\nRqO27uqRSATJZBKNRsOWX591udcFz2Zzvj6QENMjR7dJoO8UF3e6a7MDgbEeuXnoNjI0XZ13COJB\nRnfaPk/KXbfOofaH6z6fj++DTYFjsRiWlpawsbGBXq8Hl8slkWs20V7E9P+sv9e8gynXTzqS01Lh\nqt6H3tcvFAohl8thaWkJuVxOROeUe9DAFYDYOLDNkL6Wk0iRZNfrdQlmXAUWllBxcjG3XigUpIqO\naQA2R2R/oLMIlVJKfjIywuf2+/2ia+Klp496vZ6Iqpl2YDiThIqbAk/kOoGgaL1cLuPg4ACxWEzC\nmbrxmcHr4HgBkDw8U7mRSETcdbvdrviBtdttVKtVmYTValXG02zQ1wumbJnqowhdL++mboLFG/SN\nM2N18+D6SW84PZKgb+BM2bIlF33/3ncMGVWmkS8rpnmgYsRF97ULBAJIp9N48OCBpAC3t7fhcrls\na+x9/D5xf6KgnxKZy0z56YVb1EvRtHVjY0Ma3VNOE4/HbZ0R2JqI5tis4CMpZmFKpVIReQctia4C\nC0uoOBEmk4ko+pvNJkqlEsbjsZhE6lVEs3C73RIS1fsTMYpEbRZ9jiiK7ff7Iibv9XpIJBLS1JO/\nexahUkpJdCoYDIrj7/7+PpLJJGKxGLrdrmh9Fl2gfhHoBJiTln2fdOHi0dGR9PnqdDrY29vDy5cv\npfrP4XDg5OTkSk82Bq9Dj1DRY4YpXD3iMWt3chUtJQzeHzzEcN3Sm93qmzLXZofDYRu38xwW+bsk\nVL19FhoAABV6SURBVA6Hw9Y8NxQKicg6EokgGAwinU7LoTkej4uovVKpiJ7qPh5c9QgVCRUj/pcB\nPSrFYAL1UMlkEmtra/jc5z6HL37xi6JL1uc+AFs0kinbSqWCcrmMSqUilkS1Wk2qfa9yz1xoQsXJ\nRbdsls2z71sul5P0wTw9FdNxfD69X5He7ZzeFwxdA698Umb7GjHdp9sp6Db683LZNGJLp9OSZgSw\n0KXgun0Fm93W63XRkL0v9PQqtVSczADkdNzr9cRTjONFkzg2AzW9/q4HjCBEo1HRvtAFWSll+370\n+32JNHLhvI+b4G0CCXEqlUI6nRbNm9frfY1Q1et10Z1yvWbEX1939etNFVrsRUcrhEgkgkqlYmtD\nwyrhSCQCp9MpWlYaQrNKnActrrt3HczecD17k3TkbWnBd8Vs1wte+v8pQPf7/cjn8+I1+PDhQyHi\nTqdTej2ypVupVLKl9kimKpWKrTr+OtzdF5ZQzWJ2g6bpZ7vdRiQSeacqID2FxPQgN+TxeCwLiL45\ne71erKysyAmNXyyGoM8q+Z993VAohGw2K9YKACRfvGjQIw4OhwP1eh1bW1uSXj0PnE6nFAvwZEp9\nGsPMFK+mUinxtKEVBv1N6GJ/E+0X7hv0LgN6ig94pZmimJhpWjZuZYWXwc1Bj1AtLy9LNTNtYoiT\nkxMhw/SHI6EiWdYbztNX6E2pG8oqSOAmkwk6nQ5KpZJkBUKhECaTiURHGPVfXV3FaDSCx+PBzs4O\n9vf3sbe3tzCEijrfRqMhlbLzAgqsmOPnfl77Ed33kQarTN3p7aR0vV0qlcLy8rJUaXNPIEnWyVO1\nWpVrVg/Lw9V1eWfdK0IFvDJ71NvS0MLgbSAZIjFi3pcLPtMTnPwUpCeTSUSjUZvhJ5/vrJL/2del\nS/SDBw9EfFer1S74qdxe6L5R9XodL1++xPHxMfb398/1fB6PR1yS9f5/et6e6VZaarC5Nhfkfr8P\nh8OBdrsthM/g6qATqlm/MS70jFySUJVKJdFJGKuEm4Ueocrn8zLnEomETdzMQ65OohidmvXgo0aG\nPnJngRFpRjV0QjWZTBAKhcSfjqSL36/V1VVpeM8Kb91w+a6DBVONRkP6381Lg5FQsT0Po1nnIVQc\nh0gkIq3Dstmszc2ej9FF6bOEStdEf+Mb38Dz589tmqlutysHXv17dF3ZhHtDqIBXUSq6qJNQvevi\nq5MhLvQApGs3X4Ndupny83q9MmHflUTNvi4JFQBpmcAqh0WDPnnG4zHq9TqOjo5QrVYRCATO9Zw+\nnw8PHjwQPR2r/Xhi1tO7TOUuLS1haWkJlmWh3++j0WhIdeWinFZvM5gG1wkV5w01be1220aoisWi\nFISYlOzNgtY0jDboB5rZAiCCxox6gQ/TOTTfLBaLcpB9E/SDKwnVYDDAcDhEOp2WdCDnPjU8JFOr\nq6twuVxoNpt4+fLlpX8+NwW9gTgdxucdDi+DUOlyFpfLhWg0iuXlZXzwwQfY2NgQLZtOnNi9gsVi\nrMrkRQPXzz77DF/96lfRarXkogyG7/G614B7RagA2FyVPR4PLMvCcDhEs9nE7u4ufD6fLZfL623l\n+jpBYpSDOhxO2Nk89LuSKormI5EIJpMJkskkQqGQaHsWEXobGEbkSFbPA4aT6VFWq9VQLBaRTqfF\n9I8pQafTKePG3otLS0sicgUgmp2bmriLCqYGaDUSCoUk7cqmpywSGA6H6HQ6ttA+ibgZj5sHhd1b\nW1uwLEvSMZRZMOXOyuazSuh178BgMChtqNrt9lvfg34A1qMfa2trIsOYfTz1PEqpN/o03VXo/e2G\nw6Ho1GbBA2cymUQul5PMjsfjsQUgqDWdJU9MpcbjccRiMcRiMdFF0ZyX0otgMAjglZkq13/KZ6hj\na7fbKBQKePbsGfb29qQtHHvt3nQvxntHqGhlwGbEvV4P9Xode3t70hOIVzKZRDKZlHLbd8Ws/5Ee\nygTenUjpYJ7b5XIhHo+LKeWighsiT0n893kJFcdjNBqh2WyiUCjIRKfGY2VlBcvLy1KSy1Mro1X0\npBkMBpJ2WESPmpuE3nKIpn5sG0VCxROrTqj6/b7NM8jg5tHr9VAsFjEej3F4eCjdCRqNBjKZDNLp\nNNLptKyNemNzncBEo1GJWMTjcdGnvi1KPHt4JWnz+XzIZDJIJBJzCRXnPU1BGR1dFLBPrS5Kn0dE\nSKjS6TQ6nY70P/R6vZIi5Nqn66SYQvX5fIhEIlhdXcXKyoqk+Tj2bB3Fwy7fm67d4sXopB6lLBaL\nODw8lNTeTZMp4B4SKkao2Ny4Wq0KS45EIraNdTgcyonmfUBCxUl40SoJRqhY9ZRIJOb6Zi0aOFk5\n4Y+Ojs7t0svy6WazKc7pvJLJJD7++GNYliV+Y7p9RSwWw9HREZxOJ0ajERqNBg4ODmR8b8NEXgTo\nDvk+n89GqFidxQqxeRGq69RKGLwdTNexs0StVkOj0UCr1UK328V4PJa5yAMPyZUepaJ9DavRzmtq\nrEe/KISelzakaJ2RKkZFFwWzESrajMxCty3h2lmpVMQ+QT9I6oSYLdtCoRBSqRQePXqEjz/+GB99\n9JHMY1bX65FDvUE9MwDUSG5tbeH58+d48eIFSqUSer2eRK/eperzunDvCBUAceWlIZhetqkLysfj\nsXxB+CXiY99EjmZPWJcBXaCrLzr3ATy1XBQMWTMUzVNUs9kUQs30AgC5PxQK4eTkBE6nE/V6HdFo\nVMr3AZPuu0zojVDpXxSJRETnNplMhEg1Gg2bEJ3tpgxuB7jOcmPUy/V5SCI55nzUTVv1yD6tTK56\nzdNJF4C5vll3HRwDkhJGd4+Pj22RPEYFk8kkxuMxWq2WLfWmk5hZs1Tqn9LpNDY3N/Ho0SNsbm5K\nIICkVW8Vo3sDdjodmdvlchkvXrzA1tYWXrx4IdmlWcPV27AO30tCpX/wuqM6ADQaDSilRKiu63io\n5aBvyXWCpoX0ZDJVTO8PXZfFaAY9xWq1GgqFgnRVpws+yTRd1qPRqK3VEADZGG7DhL7LcDgcEjHM\nZDLI5/Ni5kndBl3QS6USdnd38eLFCzx//hyVSkU2bYPbA726mlILRkj6/T6azSb29/clokHjTb/f\nLz95sJn1rzI4H46Pj21eW+12W4iVToxIjri+HR0dwePxIB6PS4SRe6d+QNUNOsPhMFZWVpDJZGzt\nokik6Q3JKDMtMpgirlarqNVq4jXFjhWzXoC3Ze29l4QKsE903bSz2WxiNBqh1WqJdofi8lwuB6fT\n+ZqPynW8V+aUedqjCM/g3aE73eteV0opEan7/X6JSiaTSTH25E+Ko0mo6KRucHHQbDGZTGJlZUWa\nn7LfJqu+ut0uyuUy9vf35dRKXY0hVLcPnGvchNlmhp0r2FaEWrlYLIZoNCo/eZhZ5CKc6wTJLCvj\nSKgGg4GQVgrLA4GAmFiTTK2srMj+w/6mOgnWCRmLqSKRCHw+n83UmmbbpVIJhULBVq2n22M0Gg1b\nayK+5m3Urt5bQgW80ujoUSjmiilg1tuW6GZw5x3I856wOAlofjccDo0T9HtCn4D6Zzcej1Gr1SSt\nQDI1Go2kIIFFCbMRKuNHdXlwOBwIBAJIpVJYXV21tZuhYzXtK/QI1YsXL276rRu8BYxQ9Xo9AJCu\nB6zipECd3SAymYzNkNHj8SAQCFz74eU2btoXBaUPDCCwITxba7E1GittaVVDMqX3raUYPBwOzzVt\nfRPYW7dYLOLFixdi0lmpVERrx+jVXcG9JlTzoE8eltfTRZebMH2hWM1Ats1LD3kyunGR/L9lWeh2\nu6hUKjg4OMDe3h7q9brxQrokcCOPx+NYWlqSTXzRRf+3BXp1UCQSQTKZRD6fRzablbZOelUfy+9N\nlPbuggbL3MTZ0JipqGq1ioODA5FYsC3NrIhcL9GfvfSDsA4eovWWNvNa24zHY3z66ad4+fLlO1k0\n3BWwis7hcIgfI9NqsVgMlmXB7Xa/VpWuey/qUSxmeeimzkIipgTp/cVm00Sv1xMX+v39fSF3tNa4\ni/PbEKozoPsVcTHnl0EpJQ6z1FNRqNdutxEOhyV0rQvUz5smpNVDpVLB9vY2dnd3DaG6RDAqRUKV\nTqclKmJwtdB7WOpGkPl8HplMRgjVrBidVgl3bcE1mIK9TklkSKZYIEJvolk9lb6GMg2vl97r/+eG\nrzf0ZcSJYujZazb6cnBwgK2trTsVJXkb+HlzX9EJFbtEUDw+a2ZNosVeifphR28+zJ+DwcDmcq9n\nBobDoTQ0rlartvZC9Me6a9F/Q6jmgMSJCzaFc8CrDaDf7yOVSskXkP3EqtUqUqmUtDTgpnyRUDUN\nzsrlMl6+fCkRqjf1sjJ4d+gRqlwuZwjVNUKvXqXXmu6s7fV6z4xQGUJ1d0EbFL3ijBu13sTc5XLZ\nerzp6ygPQiRbs//Wf0/PDpCc6x0taCHASAov2nIsYoSKPoyMCJZKJfGemq2e0y0t2MqH0hfdTb3Z\nbIq4fDAYiBEnfaN0gsRqPspY9KgWn/+uaSINoXoDOMAs6WQzR0akKJzzeDxSjVCr1SQdMRwOkUwm\nxX3d5/PZQssAXjP9JHiKIMt/8eIFdnZ2sLe3h3K5LIZmBhcHPadonbCI7si3FfQYCoVCSCQS4lwf\ni8VsnmAsv6/X6ygUClLVN68HmcHtx6x+dRazXlQkWjoxIqGavQKBgMgt2KtTj1DpvQPfRqoYJVkk\n4q7rwtgrsVQqwefzicfX8fExYrGYrVOIngrl4+gHx4BCrVYTgTsJlW7IqX+O4/H4zHTgXYUhVG8B\nB5kCukKhgJOTE5TLZVurEj3lxxBmsVhEIpGQx0UiEdvEBSBaq1kn9uFwKII8tsXZ3d1FsVhEs9lE\nv9+/c+HQ2wz9dKw7NhtcLTweDxKJBLLZLPL5PJaXlxGPx+XgwgPIaDTC4eEhSqUStre3USgUpCLX\nYPGgd0rg/2c3XFZfM8rV7/cl5cfo1jwipkdo3pb2my3PXzQcHR2h2WzC5XJhOBxKwcfOzo7NzT4e\nj9sIJ+0N6GWlV+SRiOrNrXUzV0Ifh0WBIVTvAE7aw8NDnJycoNVqyQlIF6XzpBONRlEsFhGPx20t\nbJLJpNgedLtdABATw1k3djJ7snu9pFQ/ORlcDvTUk+6VYnC18Hq9SCQSWF1dxaNHj7CysoJEIiEu\nyrOEqlwuY3t7G9VqFYPBwBCqBYWebtJF5LNzkt5kbxKkz3apmBWlz/u37r69iJV+xNHRERqNhnSA\nKJVKYlmRz+fx8OFDPHz4EJPJRFJznU7HZm9AMTlvIxFlJEsnYvrnqNsVLcrnawjVO2I8HssXCrB3\nMgdeTXzLshAMBiUqRV1ONptFNpu1fQmBaSkqBew6Go0Gtra28PLlS+zu7tom+aJ8+W4D9LSCrtvg\nQjxrHKdrB/TbDc4HetssLy/j4cOHWFpakh5fTqfT1naI0d+9vT3RNBosLt5ljplD5cVwfHwslXUA\nJMXn8/mwvr6O0Wgka6PuDUUxeaVSQa1Ws2mh7rOdjyFU7wkK9PSfvF33OGIfQBIumpgxRMoIFV1h\n50WoqtWqhEkXicXfBlDs6vV6kUwmhdSGw2EEAgGJPFqWZTthVatVtFotqWg5qw+WwbtBjwyyCa1O\nZhlN0ImsmQcGBlcD2lkopURq4nA40Gw2JbNC8kRJCjVTrBy8zzCE6hyY96XRb2NOX+9d1el0UK1W\nbV2+AeDw8FDKfHWQgJFQ6S7fBheHLoZOpVJIJBLieRMIBKTsejKZYDAYyCmuUqmIHoDtf0zU8Pw4\ni1ARs+kZ8zkbGFwddEsFGlz3ej3s7e3ZRPy6cF+3OLjv89MQqgvgrC8PI1QU5bXb7dfKTSm2PMt8\njmTMbNhXA7ZViEajSCaTSCQStggVx4vGdK1WS5x8WRpM0mzG5vzQfahm9Wuz0SnzWRsYXC1mBfvd\nbhfFYlEak+uXrjsz83MKQ6iuAG8rCTa4Geil2IxM5fN5rK2tIZfLIRaLSXWZ7q/SbDZRKBTE0bde\nr6PX6wnRve+LyEXgdDrFfyoWiyEYDEqZO8Xo9AJiZav5vA0Mrgaz0hVjS/J+MITK4F5Ab5XgcrkQ\nj8eRz+exubkp1WXRaFSiiDylUQi9u7uLp0+fYnd3VyrMzMZ+cbhcLvGgymaztgjhaDSSkuxyuSwV\nruagYmBgcBthCJXBvQHbm3i9XsRiMSwvL2NzcxMffPABUqkUYrGYhLYZnWq32yiXy9jZ2cHTp0/F\nwsIQqssBtWyJRAK5XE6MGJ1Op83Mk2a25nM3MDC4rTCEymDhweiU1+tFMBhEMBhEOp2WUv1Hjx7Z\neoWxkICNQyuVCvb397G1tSWeLSYUfjmgSz27CdCbhtpDRqdoaGsiVAYGBrcVhlAZLCz0thU+nw+5\nXA5LS0vI5XJ49OgRNjY2kEwmpe8XHZVHo5GNSOnpJtokmE39ctDpdPDixQv80i/90mv90jgGlUoF\npVIJhUIBh4eHxqbCwMDgVsIQKoOFBcvxPR4PQqEQ8vk8Hj9+jMePH2N1dRX5fB7JZBKBQMDWboaE\nqlQqYW9vT6wS6FBvCNXlodPpYGtrC0dHR9jZ2ZHb6f9FZ2a9Sa0hVAYGBrcRhlAZLCwcDgfcbrdU\nkS0tLeGDDz7At37rtyKTyUh/Rb/fb2tNQUJVLBaxu7tri1Ax1Wd0PJeDbreLra0tHBwcwO122+6b\nbQNCCxFDqAwMDG4jDKEyWFjoppAUODcaDRSLRQwGA2lKPWuqWigU8OLFC2xtbWF/f186qJ+cnBgi\ndcmg11q/37/pt2JgYGBwIRhCZbCwoDkqME0t7e/vAwCazSYCgQDcbre4c+vQtTu1Wg2tVgv9ft+k\n+QwMDAwMzoS66hO3Usoc6W8YlmWptz/q7bhrY0kTT6b+QqEQQqEQwuGwOHLPc6kfjUbo9/vo9/sY\nDofScmE0Gt2KCNV9Hc9FhBnLxYIZz8XBecbSEKp7ADPJFwtmPBcHZiwXC2Y8FwfnGUvH2x9iYGBg\nYGBgYGDwJlx5hMrAwMDAwMDAYNFhIlQGBgYGBgYGBheEIVQGBgYGBgYGBheEIVQGBgYGBgYGBheE\nIVQGBgYGBgYGBheEIVQGBgYGBgYGBheEIVQGBgYGBgYGBheEIVQGBgYGBgYGBheEIVQGBgYGBgYG\nBheEIVQGBgYGBgYGBheEIVQGBgYGBgYGBheEIVQGBgYGBgYGBheEIVQGBgYGBgYGBheEIVQGBgYG\nBgYGBheEIVQGBgYGBgYGBheEIVQGBgYGBgYGBheEIVQGBgYGBgYGBheEIVQGBgYGBgYGBheEIVQG\nBgYGBgYGBheEIVQGBgYGBgYGBheEIVQGBgYGBgYGBhfE/w/5q7G/cOIoFgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x109807d10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "count = 0\n",
    "plt.figure(figsize=(10,5))\n",
    "for i in [73, 80, 111, 119, 151]:\n",
    "    count += 1\n",
    "    plt.subplot(1, 5, count)\n",
    "    plt.title(' %i, pred: %i ' % (test_label[index_error[i]], max_prob[index_error[i]]) , fontsize = 20)\n",
    "    displaychar(X_test[index_error[i]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 147,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>73</th>\n",
       "      <th>80</th>\n",
       "      <th>111</th>\n",
       "      <th>119</th>\n",
       "      <th>151</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>-4223.154046</td>\n",
       "      <td>-4235.493561</td>\n",
       "      <td>-4199.443401</td>\n",
       "      <td>-4189.854853</td>\n",
       "      <td>-4260.851682</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>-4189.005693</td>\n",
       "      <td>-4273.461298</td>\n",
       "      <td>-4080.113158</td>\n",
       "      <td>-4329.780832</td>\n",
       "      <td>-4477.407430</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>-4174.963632</td>\n",
       "      <td>-4222.829566</td>\n",
       "      <td>-4136.885762</td>\n",
       "      <td>-4149.806770</td>\n",
       "      <td>-4225.156626</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>-4175.487143</td>\n",
       "      <td>-4189.215302</td>\n",
       "      <td>-4149.401338</td>\n",
       "      <td>-4187.803792</td>\n",
       "      <td>-4223.117042</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>-4169.282148</td>\n",
       "      <td>-4142.716513</td>\n",
       "      <td>-4126.520789</td>\n",
       "      <td>-4172.388706</td>\n",
       "      <td>-4236.338734</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>-4203.749964</td>\n",
       "      <td>-4181.570761</td>\n",
       "      <td>-4178.808948</td>\n",
       "      <td>-4215.198234</td>\n",
       "      <td>-4223.095791</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>-4276.080889</td>\n",
       "      <td>-4278.602407</td>\n",
       "      <td>-4206.167506</td>\n",
       "      <td>-4228.735221</td>\n",
       "      <td>-4353.941623</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>-4111.870000</td>\n",
       "      <td>-4109.935105</td>\n",
       "      <td>-4089.337807</td>\n",
       "      <td>-4177.236776</td>\n",
       "      <td>-4294.949760</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>-4138.661136</td>\n",
       "      <td>-4179.540186</td>\n",
       "      <td>-4119.612502</td>\n",
       "      <td>-4156.345474</td>\n",
       "      <td>-4198.322281</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>-4117.665855</td>\n",
       "      <td>-4108.843635</td>\n",
       "      <td>-4110.986420</td>\n",
       "      <td>-4149.552638</td>\n",
       "      <td>-4215.944870</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           73           80           111          119          151\n",
       "0 -4223.154046 -4235.493561 -4199.443401 -4189.854853 -4260.851682\n",
       "1 -4189.005693 -4273.461298 -4080.113158 -4329.780832 -4477.407430\n",
       "2 -4174.963632 -4222.829566 -4136.885762 -4149.806770 -4225.156626\n",
       "3 -4175.487143 -4189.215302 -4149.401338 -4187.803792 -4223.117042\n",
       "4 -4169.282148 -4142.716513 -4126.520789 -4172.388706 -4236.338734\n",
       "5 -4203.749964 -4181.570761 -4178.808948 -4215.198234 -4223.095791\n",
       "6 -4276.080889 -4278.602407 -4206.167506 -4228.735221 -4353.941623\n",
       "7 -4111.870000 -4109.935105 -4089.337807 -4177.236776 -4294.949760\n",
       "8 -4138.661136 -4179.540186 -4119.612502 -4156.345474 -4198.322281\n",
       "9 -4117.665855 -4108.843635 -4110.986420 -4149.552638 -4215.944870"
      ]
     },
     "execution_count": 147,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pandas as pd \n",
    "df = pd.DataFrame(prob)\n",
    "\n",
    "#[73,80,111,119,151]\n",
    "df = df[index_error[0:5]]\n",
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 144,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "predictions:  7 9 1 9 8\n"
     ]
    }
   ],
   "source": [
    "print \"predictions: \", max_prob[index_error[0]], max_prob[index_error[1]], max_prob[index_error[2]], max_prob[index_error[3]], max_prob[index_error[4]]\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h2 align=\"center\"> Actual (correspond to columns). Notice prediction is close to actual </h2>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 146,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>9</th>\n",
       "      <th>7</th>\n",
       "      <th>7</th>\n",
       "      <th>2</th>\n",
       "      <th>9</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>-4223.154046</td>\n",
       "      <td>-4235.493561</td>\n",
       "      <td>-4199.443401</td>\n",
       "      <td>-4189.854853</td>\n",
       "      <td>-4260.851682</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>-4189.005693</td>\n",
       "      <td>-4273.461298</td>\n",
       "      <td>-4080.113158</td>\n",
       "      <td>-4329.780832</td>\n",
       "      <td>-4477.407430</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>-4174.963632</td>\n",
       "      <td>-4222.829566</td>\n",
       "      <td>-4136.885762</td>\n",
       "      <td>-4149.806770</td>\n",
       "      <td>-4225.156626</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>-4175.487143</td>\n",
       "      <td>-4189.215302</td>\n",
       "      <td>-4149.401338</td>\n",
       "      <td>-4187.803792</td>\n",
       "      <td>-4223.117042</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>-4169.282148</td>\n",
       "      <td>-4142.716513</td>\n",
       "      <td>-4126.520789</td>\n",
       "      <td>-4172.388706</td>\n",
       "      <td>-4236.338734</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>-4203.749964</td>\n",
       "      <td>-4181.570761</td>\n",
       "      <td>-4178.808948</td>\n",
       "      <td>-4215.198234</td>\n",
       "      <td>-4223.095791</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>-4276.080889</td>\n",
       "      <td>-4278.602407</td>\n",
       "      <td>-4206.167506</td>\n",
       "      <td>-4228.735221</td>\n",
       "      <td>-4353.941623</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>-4111.870000</td>\n",
       "      <td>-4109.935105</td>\n",
       "      <td>-4089.337807</td>\n",
       "      <td>-4177.236776</td>\n",
       "      <td>-4294.949760</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>-4138.661136</td>\n",
       "      <td>-4179.540186</td>\n",
       "      <td>-4119.612502</td>\n",
       "      <td>-4156.345474</td>\n",
       "      <td>-4198.322281</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>-4117.665855</td>\n",
       "      <td>-4108.843635</td>\n",
       "      <td>-4110.986420</td>\n",
       "      <td>-4149.552638</td>\n",
       "      <td>-4215.944870</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "             9            7            7            2            9\n",
       "0 -4223.154046 -4235.493561 -4199.443401 -4189.854853 -4260.851682\n",
       "1 -4189.005693 -4273.461298 -4080.113158 -4329.780832 -4477.407430\n",
       "2 -4174.963632 -4222.829566 -4136.885762 -4149.806770 -4225.156626\n",
       "3 -4175.487143 -4189.215302 -4149.401338 -4187.803792 -4223.117042\n",
       "4 -4169.282148 -4142.716513 -4126.520789 -4172.388706 -4236.338734\n",
       "5 -4203.749964 -4181.570761 -4178.808948 -4215.198234 -4223.095791\n",
       "6 -4276.080889 -4278.602407 -4206.167506 -4228.735221 -4353.941623\n",
       "7 -4111.870000 -4109.935105 -4089.337807 -4177.236776 -4294.949760\n",
       "8 -4138.661136 -4179.540186 -4119.612502 -4156.345474 -4198.322281\n",
       "9 -4117.665855 -4108.843635 -4110.986420 -4149.552638 -4215.944870"
      ]
     },
     "execution_count": 146,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 135,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[73, 80, 111, 119, 151]"
      ]
     },
     "execution_count": 135,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "index_error[0:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}