vanilla_squeezenet.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import keras\n",
    "from keras import optimizers\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "from keras.callbacks import EarlyStopping, ReduceLROnPlateau\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "\n",
    "from squeezenet import SqueezeNet, preprocess_input"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "data_dir = '/home/ubuntu/data/'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 25600 images belonging to 256 classes.\n",
      "Found 5120 images belonging to 256 classes.\n"
     ]
    }
   ],
   "source": [
    "data_generator = ImageDataGenerator(\n",
    "    data_format='channels_last',\n",
    "    preprocessing_function=preprocess_input\n",
    ")\n",
    "\n",
    "train_generator = data_generator.flow_from_directory(\n",
    "    data_dir + 'train', \n",
    "    target_size=(299, 299),\n",
    "    batch_size=64\n",
    ")\n",
    "\n",
    "val_generator = data_generator.flow_from_directory(\n",
    "    data_dir + 'val', shuffle=False,\n",
    "    target_size=(299, 299),\n",
    "    batch_size=64\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "853824"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = SqueezeNet(weight_decay=1e-4, image_size=299)\n",
    "model.count_params()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model.compile(\n",
    "    optimizer=optimizers.SGD(lr=1e-2, momentum=0.9, nesterov=True), \n",
    "    loss='categorical_crossentropy', metrics=['accuracy', 'top_k_categorical_accuracy']\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/30\n",
      "400/400 [==============================] - 36s - loss: 2.3837 - acc: 0.4822 - top_k_categorical_accuracy: 0.7059 - val_loss: 1.8188 - val_acc: 0.5773 - val_top_k_categorical_accuracy: 0.8105\n",
      "Epoch 2/30\n",
      "400/400 [==============================] - 35s - loss: 1.1869 - acc: 0.7113 - top_k_categorical_accuracy: 0.8940 - val_loss: 1.7067 - val_acc: 0.6201 - val_top_k_categorical_accuracy: 0.8400\n",
      "Epoch 3/30\n",
      "400/400 [==============================] - 35s - loss: 0.8814 - acc: 0.7823 - top_k_categorical_accuracy: 0.9320 - val_loss: 1.6830 - val_acc: 0.6330 - val_top_k_categorical_accuracy: 0.8463\n",
      "Epoch 4/30\n",
      "400/400 [==============================] - 35s - loss: 0.7074 - acc: 0.8240 - top_k_categorical_accuracy: 0.9532 - val_loss: 1.6935 - val_acc: 0.6385 - val_top_k_categorical_accuracy: 0.8453\n",
      "Epoch 5/30\n",
      "400/400 [==============================] - 35s - loss: 0.5857 - acc: 0.8561 - top_k_categorical_accuracy: 0.9655 - val_loss: 1.7206 - val_acc: 0.6309 - val_top_k_categorical_accuracy: 0.8533\n",
      "Epoch 6/30\n",
      "400/400 [==============================] - 35s - loss: 0.5124 - acc: 0.8749 - top_k_categorical_accuracy: 0.9740 - val_loss: 1.7514 - val_acc: 0.6336 - val_top_k_categorical_accuracy: 0.8459\n",
      "Epoch 7/30\n",
      "400/400 [==============================] - 35s - loss: 0.3407 - acc: 0.9335 - top_k_categorical_accuracy: 0.9868 - val_loss: 1.5808 - val_acc: 0.6660 - val_top_k_categorical_accuracy: 0.8619\n",
      "Epoch 8/30\n",
      "400/400 [==============================] - 36s - loss: 0.2986 - acc: 0.9481 - top_k_categorical_accuracy: 0.9888 - val_loss: 1.5681 - val_acc: 0.6648 - val_top_k_categorical_accuracy: 0.8637\n",
      "Epoch 9/30\n",
      "400/400 [==============================] - 36s - loss: 0.2862 - acc: 0.9540 - top_k_categorical_accuracy: 0.9891 - val_loss: 1.5635 - val_acc: 0.6674 - val_top_k_categorical_accuracy: 0.8670\n",
      "Epoch 10/30\n",
      "400/400 [==============================] - 36s - loss: 0.2789 - acc: 0.9559 - top_k_categorical_accuracy: 0.9896 - val_loss: 1.5620 - val_acc: 0.6689 - val_top_k_categorical_accuracy: 0.8646\n",
      "Epoch 11/30\n",
      "400/400 [==============================] - 36s - loss: 0.2660 - acc: 0.9611 - top_k_categorical_accuracy: 0.9907 - val_loss: 1.5574 - val_acc: 0.6709 - val_top_k_categorical_accuracy: 0.8654\n",
      "Epoch 12/30\n",
      "400/400 [==============================] - 36s - loss: 0.2645 - acc: 0.9608 - top_k_categorical_accuracy: 0.9913 - val_loss: 1.5712 - val_acc: 0.6693 - val_top_k_categorical_accuracy: 0.8629\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7f0b0cfaa390>"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit_generator(\n",
    "    train_generator, \n",
    "    steps_per_epoch=400, epochs=30, verbose=1,\n",
    "    callbacks=[\n",
    "        EarlyStopping(monitor='val_acc', patience=4, min_delta=0.01),\n",
    "        ReduceLROnPlateau(monitor='val_acc', factor=0.1, patience=2, epsilon=0.007)\n",
    "    ],\n",
    "    validation_data=val_generator, validation_steps=80, workers=4\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Loss/epoch plots"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYUAAAEKCAYAAAD9xUlFAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nUBARkTb/HyCUaifa1LDQAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f0b0cfc4978>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(model.history.history['loss'], label='train');\n",
    "plt.plot(model.history.history['val_loss'], label='val');\n",
    "plt.legend();\n",
    "plt.xlabel('epoch');\n",
    "plt.ylabel('logloss');"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYUAAAEKCAYAAAD9xUlFAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzt3Xd8XNWd9/HPT73ZlizJVbblinEBF2HTbRzIUhIgoZgQ\nWhJwNtSwyT4L2YQQlmzYfbLZJE8ooUPooSQmcSABy6aDZWzARXIvclOxZatYdc7zxx3LY1nGY6PR\n1cx836/XvGbunTszvzHofOeee+855pxDREQEIMHvAkREpOdQKIiISDuFgoiItFMoiIhIO4WCiIi0\nUyiIiEg7hYKIiLRTKIiISDuFgoiItEvyu4AjlZeX5woLC/0uQ0QkqixevLjKOZd/uO2iLhQKCwsp\nKSnxuwwRkahiZhvD2U7dRyIi0k6hICIi7RQKIiLSLuqOKXSmpaWF8vJyGhsb/S4lotLS0igoKCA5\nOdnvUkQkRsVEKJSXl9OrVy8KCwsxM7/LiQjnHNXV1ZSXlzN8+HC/yxGRGBUT3UeNjY3k5ubGbCAA\nmBm5ubkxvzckIv6KiVAAYjoQ9omH7ygi/oqJ7iMRkUhpaQtQ09DCroZmdtY3U9PQzM56b7mppc3b\nyAzz7jAseB9cDv6Y2/fc/scHbu+tD32fkHXB5ekjchnTv1dEv69CoQvU1NTwzDPPcP311x/R6849\n91yeeeYZsrOzI1SZiIT6vAZ+V30zO4P3u0K2qW1sPeT7mUF3TnN/94UTFArRoKamhvvuu++gUGht\nbSUp6dD/xPPmzYt0aSJRq7UtQHNbgKYW7765NUBTaxtNrfsee/ftj9vaaGhu8xr9o2jgM1ISyclI\nISczmZyMFIblZnjLGSn0zUwmOyOFvpkpB2yTlpzY/nrnHM6B2/cYgsvugOAIXddxWzps3/H5rNTI\nN9kKhS5w2223sXbtWiZNmkRycjJpaWnk5ORQWlrKqlWruPDCC9m8eTONjY3ccsstzJkzB9g/ZEdd\nXR3nnHMOp556Ku+99x6DBw/mz3/+M+np6T5/M5HDc85R19RKVV0z1XVNVNU1U1XXRHVdMzvrm9jb\n0nZwA94aoKktQFNLW0iDH7pNG4Ev8Au8YwM/tG9Ge4MeTgN/NPZ18wSXvtB7+SnmQuFnry5nxdY9\nXfqe4wb15qdfHX/I5++55x6WLVvG0qVLWbBgAeeddx7Lli1rP3X00UcfpW/fvuzdu5cTTjiBiy66\niNzc3APeY/Xq1Tz77LM89NBDXHrppbz00ktcccUVXfo9RMLV2hZgZ0MzVbXNVNd7DXxVsMH3Gv4m\nquubqaptoqq+mebWQKfv0yc9mYyURFKSEkhNSiAlKYGUxARSkxLJTkkmpVdq+3Op+55LTiQlMeHA\n13R4LrXDc6lJ3mekJSd0SQMfz2IuFHqCadOmHXAtwW9/+1teeeUVADZv3szq1asPCoXhw4czadIk\nAKZOncqGDRu6rV6JH1V1TWysbqA6pFGvrm+msq7JWxds/Hc1tHT6+uREIy8rldysFPKyUhndrxd5\nwce5WSnkZqW2L/fNTCE5MWZOcIwbMRcKn/eLvrtkZma2P16wYAFvvPEG77//PhkZGcycObPTaw1S\nU1PbHycmJrJ3795uqVViV3NrgBXb9rBk0y6WbKphyeZdbN558P9XvdKSyAs25iPzs5g+oi+5makh\njb33ODcrld5pSTo1OsbFXCj4oVevXtTW1nb63O7du8nJySEjI4PS0lI++OCDbq5O4oFzjq27G/cH\nwKZdLNu6p71bZ0DvNCYPzebKE4cFf93v+2WfQmqSulpkP4VCF8jNzeWUU05hwoQJpKen079///bn\nzj77bB544AGOPfZYjjnmGE488UQfK5VY0dDcyqflu1m6uaY9CCpqmwBITUrguII+XHNyIZOHZDNp\naDYD++ikBQmPue48ybYLFBUVuY6T7KxcuZJjjz3Wp4q6Vzx9V/EEAo711fXtewBLNtVQtqOWtuDp\nOYW5GUwemsPkodlMHpLD2IG91JcvBzGzxc65osNtpz0FkR5md0MLS8v3B8DSzTXs3usd+M1KTWLS\nkGyunzmSyUOzmTQkh76ZKT5XLLFEoSDiI+ccK7fVsmTz/mMBayvrAe9q2TH9enHOhAHeXsDQHEbm\nZ5GYoAO9EjkKBRGfNLa08cM/fsJfPt0GQN/MFKYMzebrUwqYPCSbiQV96JWmuTOkeykURHywq76Z\nOX8oYdGGXXz/zNF8fXIBQ/qm63RP8Z1CQaSbbapu4JrHPqK8Zi+/u3wyXzlukN8libRTKIh0o6Wb\na/jO44toc46nr53OCYV9/S5J5AA6b80HWVlZfpcgPvj78u1c9uD7ZKQm8tL3TlYgSI+kPQWRbvD4\nu+v52V9WcFxBNo9cXUReVurhXyTiA4VCF7jtttsYMmQIN9xwAwB33nknSUlJFBcXs2vXLlpaWrj7\n7ru54IILfK5Uulsg4PjPeSt5+J31nDWuP7+9bDLpKRpWQnqu2AuFv90G2z/r2vccMBHOueeQT8+e\nPZvvf//77aHwwgsv8Prrr3PzzTfTu3dvqqqqOPHEEzn//PN1dkkcaWxp49bnl/K3Zdu55uRCfvKV\ncbrGQHq82AsFH0yePJmKigq2bt1KZWUlOTk5DBgwgFtvvZW33nqLhIQEtmzZwo4dOxgwYIDf5Uo3\n2FnfzLVPLGLJ5hp+fN6xfOfU4fpBIFEh9kLhc37RR9Ill1zCiy++yPbt25k9ezZPP/00lZWVLF68\nmOTkZAoLCzsdMltiz4aqeq557CO27m7k3suncO7EgX6XJBK22AsFn8yePZvrrruOqqoqFi5cyAsv\nvEC/fv1ITk6muLiYjRs3+l2idIPFG3dx3ZMlOOd49rrpTB2mM4wkukT0lFQzO9vMysxsjZnd1snz\nw8zsTTP71MwWmFlBJOuJpPHjx1NbW8vgwYMZOHAg3/zmNykpKWHixIk8+eSTjB071u8SJcJeW7aN\nyx/6gF5pSbx8/SkKBIlKEdtTMLNE4F7gLKAcWGRmc51zK0I2+yXwpHPuCTObBfwCuDJSNUXaZ5/t\nP8Cdl5fH+++/3+l2dXV13VWSdJNH3lnP3X9dwaQh2Tx8VRG5OuVUolQk9xSmAWucc+ucc83Ac0DH\nczLHAfODj4s7eV6kR2sLOH726nL+4y8r+PK4/jx73YkKBIlqkQyFwcDmkOXy4LpQnwBfDz7+GtDL\nzHI7bIOZzTGzEjMrqaysjEixIkdqb3Mb1z+9mMfe3cC3TxnOfd+cSlqyrkGQ6Ob3MBc/BGaY2RJg\nBrAFaOu4kXPuQedckXOuKD8/v9M3irYZ5I5GPHzHaFFd18Q3HvqAv6/YwR1fGccdX9U1CBIbInn2\n0RZgSMhyQXBdO+fcVoJ7CmaWBVzknKs50g9KS0ujurqa3NzcmD0X3DlHdXU1aWlpfpcS99ZV1vGt\nxxexfXcj939zKmdP0LUnEjsiGQqLgNFmNhwvDC4DLg/dwMzygJ3OuQBwO/Do0XxQQUEB5eXlxHrX\nUlpaGgUFUXuCVkwo2bCT654swcx4ds6JTBma43dJIl0qYqHgnGs1sxuB14FE4FHn3HIzuwsocc7N\nBWYCvzAzB7wF3HA0n5WcnMzw4cO7qHKRzv31023c+sJSBmen8/i3TmBYbqbfJYl0OYu2fuqioiJX\nUlLidxkSR5xzPPz2en4+byVTh+Xw0FVF9M1M8bsskSNiZoudc0WH205XNIt8jraA465Xl/PE+xs5\nd+IAfnXpJJ1hJDFNoSByCA3Nrdz87FLeWLmDOaeP4Lazx5KgM4wkxikURDpRWdvEtU8s4rMtu/nZ\n+eO5+uRCv0sS6RYKBZEO1lbWcc1jH1FZ28TvryzirHH9/S5JpNsoFESA7bsbeWtVJQtXVbKgrIL0\nlESem3MSk4Zk+12aSLdSKEhcampto2TDrvYgKN1eC0D/3qmcd9xAbpo1miF9M3yuUqT7KRQkbmys\nrmfhqkoWllXy3tpq9ra0kZxonFDYl9vPGcuMY/I5pn+vmL0qXiQcCgWJWQ3NrXywrpqFZd7ewIbq\nBgCG9s3gkqICZozJ58QRuWSm6s9AZB/9NUjMcM6xakdde5fQR+t30twWID05kZNG5vKtU4YzY0w+\nhXm6ElnkUBQKEtV2723h3TVV7XsD2/d482CP6Z/F1ScPY8aYfhQV5uiCM5EwKRQkqgQCjmVbd7eH\nwJLNNbQFHL3SkjhtdB4zxuRz+ph8BvZJ97tUkaikUJAer6ahmeKyChaWVfLW6ip21jcDcFxBH66f\nOZIZY/KZNCSbpES/pwcRiX4KBemxtu3ey0NvrefZjzaxt6WN3MwUZozJZ8aYfE4dnUeepr0U6XIK\nBelxNlTV8/u31vLi4nICDi6cNJirTx7GhEF9NPaQSIQpFKTHKNtey30L1vDqJ1tJSkzgshOGMuf0\nEbqITKQbKRTEd0s313Bv8Rr+sWIHmSmJXHfaCL5z2nD69dLUoyLdTaEgvnDO8f66au4rXss7a6ro\nk57M988czTUnF5KdoQlsRPyiUJBu5ZxjfmkFvytew5JNNeT3SuVH547l8unDyNKVxSK+01+hdIu2\ngGPeZ9u4t3gNpdtrKchJ5z8unMAlUwt0YZlID6JQkIhqbg3wpyVbuH/hWtZX1TMyP5P/ueR4zp80\niGRdVyDS4ygUJCL2Nrfx/KJNPPjWOrbubmT8oN7c/80p/NP4ATqtVKQHUyhIl9rT2MJTH2zkkbfX\nU13fzAmFOfzn1ycyY0y+hqQWiQIKBekSO+ubeezd9Tz+3gZqG1uZMSafG84YxbThff0uTUSOgEJB\nvpDtuxt56O11PPPhJhpb2zh7/ACunzmKiQV9/C5NRI6CQkGOysbqeh5YuI6XFpfT5hwXTBrE92aM\nZHT/Xn6XJiJfgEJBjsiexhZ++XoZT32wkaTEBC49oYDvnj5SQ1GIxAiFgoTFOcdfP9vGXa+uoLKu\niStPHMaNZ4yiX28NRSESSxQKclibqhv4yZ+XsXBVJRMG9+bhq4s4riDb77JEJAIUCnJIza0BHnp7\nHb99czVJCcYdXxnHVScN02Q2IjFMoSCdWrRhJ//+ymes2lHHP43vz53nj9cUlyJxQKEgB6hpaOae\nv5Xy3KLNDM5O5+GrijhzXH+/yxKRbqJQEMA7kPzKki38/K8rqdnbwpzTR3DLl0aTqZFLReJKRP/i\nzexs4DdAIvCwc+6eDs8PBZ4AsoPb3OacmxfJmuRg6yrr+PGflvHe2momDcnmD1+byLhBvf0uS0R8\nELFQMLNE4F7gLKAcWGRmc51zK0I2+zHwgnPufjMbB8wDCiNVkxyoqbWN+xes5b7itaQmJ3D3hRO4\nfNpQDVgnEsciuacwDVjjnFsHYGbPARcAoaHggH0/SfsAWyNYj4R4b20VP35lGeuq6vnq8YP4yVeO\n1fSXIhLRUBgMbA5ZLgemd9jmTuDvZnYTkAmcGcF6BKiua+Ln81by8sdbGNo3gye+PY0ZY/L9LktE\negi/jyJ+A3jcOfc/ZnYS8Aczm+CcC4RuZGZzgDkAQ4cO9aHM6BcIOP64eDO/+Fsp9U2t3HjGKG6c\nNUqznonIASIZCluAISHLBcF1ob4DnA3gnHvfzNKAPKAidCPn3IPAgwBFRUUuUgXHqlU7avn3Vz5j\n0YZdTCvsy8+/NkED14lIpyIZCouA0WY2HC8MLgMu77DNJuBLwONmdiyQBlRGsKa40tjSxv+bv5rf\nL1xHVloS/33RcVw8tUAHkkXkkCIWCs65VjO7EXgd73TTR51zy83sLqDEOTcX+AHwkJndinfQ+Rrn\nnPYEusDCVZX85E/L2LSzgYumFPCjc8eSm5Xqd1ki0sNF9JhC8JqDeR3W3RHyeAVwSiRriDcVexq5\n6y8r+Mun2xiRl8kz103n5JF5fpclIlHC7wPN0kXaAo5nPtzIf79WRlNbgFvPHMM/zxxBapIOJItI\n+BQKMaC6ronrnizh4001nDIql7svnMjwvEy/yxKRKKRQiHJ7Glu46tGPWFNRx//OPp4LJw3GTAeS\nReToKBSiWENzK99+bBGrdtTy4FVFnHFMP79LEpEop9lSolRTaxvf/cNiPt60i1/PnqxAEJEuoT2F\nKNTaFuDmZ5fw9uoq/vvi4zjvuIF+lyQiMUJ7ClEmEHD8n5c+5fXlO7jjK+O4tGjI4V8kIhKmsELB\nzF42s/PMTCHiI+ccd766nJc/3sK/nDWGb5863O+SRCTGhNvI34c3RMVqM7vHzI6JYE1yCL/8exlP\nvr+ROaeP4KZZo/wuR0RiUFih4Jx7wzn3TWAKsAF4w8zeM7NvmVlyJAsUz/0L1nJv8Vq+MW0ot58z\nVqedikhEhN0dZGa5wDXAtcASvGk2pwD/iEhl0u4P72/gv14r5fzjB3H3hRMUCCISMWGdfWRmrwDH\nAH8Avuqc2xZ86nkzK4lUcQKvLCnnJ39ezpnH9uN/Lj2eRI1wKiIRFO4pqb91zhV39oRzrqgL65EQ\nry/fzg//+Cknjcjld5dPITlRx/lFJLLCbWXGmVn2vgUzyzGz6yNUkwDvrK7ipmeWMHFwHx66ukgz\npIlItwg3FK5zztXsW3DO7QKui0xJsnjjTq57soQR+Zk8/q0TyErVNYYi0j3CDYVECzm6aWaJQEpk\nSopvy7fu5prHFjGgTxpPfmca2Rn6ZxaR7hPuT9DX8A4q/z64/N3gOulCayrquOqRj+iVmsRT106n\nX680v0sSkTgTbij8G14QfC+4/A/g4YhUFKc272zgykc+xAyeunY6g7PT/S5JROJQWKHgnAsA9wdv\n0sUq9jRyxSMfUt/UyvPfPYkR+Vl+lyQicSrc6xRGA78AxgHtfRrOuRERqitu7Kpv5spHPqKytomn\nrp3OsQN7+12SiMSxcA80P4a3l9AKnAE8CTwVqaLiRV1TK9c89hHrq+t5+KoipgzN8bskEYlz4YZC\nunPuTcCccxudc3cC50WurNjX2NLGdx5fxLKte7jv8imcPCrP75JERMI+0NwUHDZ7tZndCGwB1PF9\nlJpbA3zvqcV8tGEnv549iTPH9fe7JBERIPw9hVuADOBmYCpwBXB1pIqKZW0Bx60vLKW4rJKfXziR\nCyYN9rskEZF2h91TCF6oNts590OgDvhWxKuKUc45fvTyZ/z102386NyxXD59qN8liYgc4LB7Cs65\nNuDUbqglpjnn+I+/rOT5ks3cPGsUc04f6XdJIiIHCfeYwhIzmwv8Eajft9I593JEqopBv35jNY++\nu55rTi7k1rPG+F2OiEinwg2FNKAamBWyzgEKhTA8/PY6fvPmai6ZWsAdXxmnSXJEpMcK94pmHUc4\nSs99tIm7/7qScycO4J6LjiNBk+SISA8W7hXNj+HtGRzAOfftLq8ohrz6yVZuf+UzZozJ59ezJ2vW\nNBHp8cLtPvpLyOM04GvA1q4vJ3YUl1Zw6/NLOWFYXx64YiopSZo1TUR6vnC7j14KXTazZ4F3IlJR\nDHDO8dO5yxnVL4tHrikiPUWzpolIdDjan6+jgX6H28jMzjazMjNbY2a3dfL8/5rZ0uBtlZnVdPY+\n0WZtZR2bdjZwxYnD6JWW7Hc5IiJhC/eYQi0HHlPYjjfHwue9JhG4FzgLKAcWmdlc59yKfds4524N\n2f4mYHL4pfdc80srADhj7GFzU0SkRwm3+6jXUbz3NGCNc24dgJk9B1wArDjE9t8AfnoUn9PjzC+t\nYOyAXpooR0SiTljdR2b2NTPrE7KcbWYXHuZlg4HNIcvlwXWdvf8wYDgwP5x6erI9jS2UbNjFzGO0\nlyAi0SfcYwo/dc7t3rfgnKuha3/VXwa8GBxS4yBmNsfMSsyspLKysgs/tuu9vaqK1oBjlrqORCQK\nhRsKnW13uK6nLcCQkOWC4LrOXAY8e6g3cs496Jwrcs4V5efnH+Zj/TW/tII+6clMGZrtdykiIkcs\n3FAoMbNfmdnI4O1XwOLDvGYRMNrMhptZCl7DP7fjRmY2FsgB3j+SwnuiQMCxcFUFp4/JJylR1yWI\nSPQJt+W6CWgGngeeAxqBGz7vBc65VuBG4HVgJfCCc265md1lZueHbHoZ8Jxz7qArpqPNp1t2U1XX\nzKyxPXtvRkTkUMI9+6geOOg6gzBeNw+Y12HdHR2W7zzS9+2p5pdWYAYzxuh4gohEp3DPPvqHmWWH\nLOeY2euRKys6LSirYPKQbPpmpvhdiojIUQm3+ygveMYRAM65XYRxRXM8qaht5NPy3TrrSESiWrih\nEDCz9rkjzayQTkZNjWcLyrxTZXUVs4hEs3BHSf134B0zWwgYcBowJ2JVRaHi0gr6905l3MDefpci\nInLUwtpTcM69BhQBZXjXE/wA2BvBuqJKc2uAt1dXccYx/TSrmohEtXAHxLsWuAXvArSlwIl41xXM\n+rzXxYuSDTupa2pV15GIRL1wu49uAU4APnDOnRG84Ow/I1dWdJlfWkFKYgKnjsrzuxQRAXAO2lqg\nrQlag7fQxx2XP++51kZoa+78OYDE5OAtJXg7iscJyZ+zTci6tN6QHNmBNsMNhUbnXKOZYWapzrlS\nMzsmopVFkfllFUwf0ZfM1HD/OUViSFsrtO6Flr3Q0gAtjcH7vd761mavUW1rCd6HPA60dL7+kI/D\n2La9we6Cc2EsAZLSvAY5KQ2SgveJqd5jLPg9Wg9da1fVAnDe/8AJ13bNex1CuK1YefA6hT8B/zCz\nXcDGyJUVPTZW17Ousp4rpg/zuxSJNi2NsPQpWPV3r/FJSISEpOAvx6T9t47L7esSvV+YnS0nJEHi\nvu2TD1wOtHm/fvc13O0NefDxAc+F3A5o+Pfub/wDLV3wj2GQlHroX8ihj5PTIa3Pwdvu+7WdlLr/\nlpjaoTEPadwP+VzIaxO76IdeoO0Igq/ZC9rO1g87uWvq+RzhXtH8teDDO82sGOgDvBaxqqJIcXBC\nHV2fIGFrrofFj8O7v4W67ZA7GlIyvIYg0Oo1soHWDsttXuOwbzlSLBFSMr3GMTkdkjMgOc27T8+B\n3oMgKf3g55LTg68JPt53S0oPNsCH6UJJiPEpaxMSISE94l0/XeGIY9A5tzAShUSr+WWVjMjLpDAv\n0+9SpKdr3A0fPQQf3AcN1TD8dLjoISg8DY70rLVAWzA49gVFa/jLCYn7G/WODXmipo+Nd+oE/wIa\nmlv5YF01V56oriP5HA074YP74cPfQ9NuGP1lOO2HMHT60b9nQqJ3S0rtujpFUCh8Ie+uqaa5NaCu\nI+lc7Q54/3ew6BFoqYdjv+qFwaBJflcmckgKhS9gfmkFWalJnFDY1+9SpCfZXQ7v/gY+ftI7ODjh\nYjjtX6DfsX5XJnJYCoWj5JxjQVkFp47KIyVJE+oIsHMdvPO/sPRZwMHx34BTb4XckX5XJhI2hcJR\nWrmtlm27G/n+mZpQJ+5VlMI7v4LP/uidFjn1GjjlFsgectiXivQ0CoWjVFzmnYp6xjE6nhC3tn0C\nb/0SVr7qncFz0g1w0o3Qa4DflYkcNYXCUZpfWsGEwb3p1zvN71Kku23+yAuD1a9Dam84/Ycw/XuQ\nmet3ZSJfmELhKOyqb2bJpl3ceMYov0uR7uIcbHgH3vq/sH4hpPeFWT+GE66D9OzDv14kSigUjsJb\nqysJuDiaUMc57yrchiqor/YuvGrdC70GQp8CyOofu1ekOgdr3vDCYPOH3nf98t0w9VuQmuV3dSJd\nTqFwFOaXVpCbmcLxBVH6C9E5aKwJNvBVUF8Vcr+zw7rgNq2Nh36/hCQvIHoPhj6Dg/cFIcsFkJl3\n5Fft+ikQgLK/emGw7RPoMwTO/SVMvtIb2kEkRikUjlBbwLFwVSWzxvYjIaEHNXINO6FuR4cGvvrg\n5X23QGvn75Oc6fWNZ+RBZj/oNw4ycr1GPSNv/31SKtRu887J37MFdm/x7rcs9g68tjUf+L6Jqd64\nOX0KDg6MfUGS1qd7giMQ8PZ0mhuCg7s17B/krbkBarfCBw9A5UroOwLO/x0cNzs4KqZIbFMoHKEl\nm3ZR09Di31XMzsHuzd6v122fwvZPvfvarZ1vn9Znf2OeMxwKikIa932Nf+7+dUcyYNfA4zpfHwh4\nQXRAYJTvD471b3uB4toOfF1K1iH2Ngq8uloagg35Xu8K4X0jdnZc11wfMrpn/f7GPnQ458PJPxa+\n/jCM/1rXjZQpEgX0f/sRml9aQWKCcdrobrg+IdAG1Wu8Rn/b0v0B0FjjPW8JkDcGCk+FARO9xjT0\n13xGX38GOEtIgKx+3m3wlM63aWv19mz2bOkkPMph+zKorwjzA80b2bN9YLcMb9TR5AzIGuCtS8kM\nGdkzo/N1+16TkuX9uybookSJPwqFIzS/tIKiYTn0Se/ixra1CSpWBAPgEy8Adiz3ft2C1/3SfxyM\nuwAGHu/d+o3zGrJolJjkhVifwTBkWufbtDbBnq1eYLQ2el1bnTXmSanRdbxCpAdTKByBrTV7Kd1e\ny+3njP1ib9S4B3YsC+n++QQqS/f386f29n75T7na66IZeLz3yzXehjVOSoW+w72biHQLhcIRaL+K\n+UiOJ9RVwvZPDgyAnev2P5/Zz2v4R395fwBkF6rrQkR8oVA4AsWlFQzOTmd0v8Ocn96wE+b/B5S9\nduAB4OxhXsN//OX7A0BDIohID6JQCFNjSxvvrqnm4qkF2KH6r52DZS/Ba7d5wTD+Qhg0xQuAARO9\n6QxFRHowhUKYPly/k70tbYc+FbVmM/z1B954OIOmwJWveEEgIhJFFAphKi6tIC05gZNGdhj0LNDm\nzbv75l2Ag3/6BUz/buwO+yAiMU2hEAbnHPNLKzh5ZB5pySGN/Y7lMPdm2FICo86E834FOZqvWUSi\nl0IhDGsr69m0s4HrTh/hrWhp9MbEeffX3hXDX38YJl6sc+VFJOpF9LxHMzvbzMrMbI2Z3XaIbS41\nsxVmttzMnolkPUeruNQ7FXXW2H6w4V144BR4+5cw8RK4YREcd4kCQURiQsT2FMwsEbgXOAsoBxaZ\n2Vzn3IqQbUYDtwOnOOd2mVmPHIt6fmkFU/oZg9/6N/j4Ce/U0itfgZGz/C5NRKRLRbL7aBqwxjm3\nDsDMngMIg2NiAAALK0lEQVQuAFaEbHMdcK9zbheAcy7cwW66zZ69zfTd+Df+K+MPsGQXnHwTzLzd\nG2pBRCTGRDIUBgObQ5bLgekdthkDYGbvAonAnc651zq+kZnNAeYADB06NCLFdmrPVhqevZF7k9+k\nvtd4uPglGDS5+z5fRKSb+T2WQhIwGpgJfAN4yMwOmrnGOfegc67IOVeUn98do5MGYNHDcO90+m5/\nh19xBanfW6BAEJGYF8k9hS3AkJDlguC6UOXAh865FmC9ma3CC4lFEazr81WWeaeZbv4AN3wml2y6\nhCEjx5GUrAlWRCT2RXJPYREw2syGm1kKcBkwt8M2f8LbS8DM8vC6k9bhh9YmWHAPPHAqVJXBhffz\n2azH+aQ+x78JdUREulnE9hScc61mdiPwOt7xgkedc8vN7C6gxDk3N/jcl81sBdAG/KtzrjpSNR3S\npg9h7k1eGEy4GM6+B7Lymf/GKsxgxphu6LISEekBInrxmnNuHjCvw7o7Qh474F+Ct+7XuAfe/Bks\nesSb8vHyP8KYL7c/XVxawaQh2eRmpfpSnohId4vfK5pL53kD2NVug+n/DLN+DKn7h8SurG3ik/Ld\n/OCsMT4WKSLSveIvFGp3wN/+FVb8GfqNh9l/8Caz72DB0UyoIyIS5eInFJyDj5+Ef/zEG7to1k/g\nlFsOOcVlcVkF/XunMn5Q724uVETEP/ETCgt+AQv/C4adCl/9DeSNOuSmLW0B3l5VxXnHDTz0hDoi\nIjEofkJh6jXeweRJVxx2/uNFG3ZS29SqriMRiTvxEwq9B8GUq8LadEFZJcmJxqmj8iJclIhIz+L3\nMBc90vzSCqYPzyUzNX4yU0QEFAoH2byzgTUVdeo6EpG4pFDoYH7ohDoiInFGodDB/NIKhudlMjxP\n8yWISPxRKIRoaG7l/XXVnHGM9hJEJD4pFEK8t6aa5taAuo5EJG4pFELML6sgMyWRacP7+l2KiIgv\nFApBzjmKSys4dXQeKUn6ZxGR+KTWL6h0ey3bdjeq60hE4ppCIag4OCrqTB1kFpE4plAIKi6tYPyg\n3vTvneZ3KSIivlEoADUNzSzeuEtdRyIS9xQKwMJVlQScJtQREVEo4HUd9c1M4fiCbL9LERHxVdyH\nQlvAsXBVJTPH5JOYoAl1RCS+xX0oLN28i10NLeo6EhFBocD80goSE4zTx+T7XYqIiO8UCqWVTB2W\nQ5/0ZL9LERHxXVyHwvbdjazctkenooqIBMV1KOy7ilmhICLiietQmF9aweDsdEb3y/K7FBGRHiFu\nQ6GptY1311Rxxth8zHQqqogIxHEofLhuJw3Nbeo6EhEJEbehML+0gtSkBE4aked3KSIiPUZchoJz\njuKyCk4emUt6SqLf5YiI9BhxGQrrqurZWN2griMRkQ4iGgpmdraZlZnZGjO7rZPnrzGzSjNbGrxd\nG8l69iku9U5F1dAWIiIHSorUG5tZInAvcBZQDiwys7nOuRUdNn3eOXdjpOrozPzSCsb0z6IgJ6M7\nP1ZEpMeL5J7CNGCNc26dc64ZeA64IIKfF5baxhYWbdipvQQRkU5EMhQGA5tDlsuD6zq6yMw+NbMX\nzWxIBOsB4N01VbS0OWZpLmYRkYP4faD5VaDQOXcc8A/gic42MrM5ZlZiZiWVlZVf6APnl1bQOy2J\nqcNyvtD7iIjEokiGwhYg9Jd/QXBdO+dctXOuKbj4MDC1szdyzj3onCtyzhXl5x/9ENeBgKO4rJLT\nx+STlOh3HoqI9DyRbBkXAaPNbLiZpQCXAXNDNzCzgSGL5wMrI1gPy7fuobK2iTPUdSQi0qmInX3k\nnGs1sxuB14FE4FHn3HIzuwsocc7NBW42s/OBVmAncE2k6gGv68gMZh6jCXVERDoTsVAAcM7NA+Z1\nWHdHyOPbgdsjWUOo+WUVHF+QTW5Wand9pIhIVImbjvWquiY+La/RVcwiIp8jbkJhQVklzmlCHRGR\nzxM3odAnPZmzxvVn/KDefpciItJjRfSYQk9y1rj+nDWuv99liIj0aHGzpyAiIoenUBARkXYKBRER\naadQEBGRdgoFERFpp1AQEZF2CgUREWmnUBARkXbmnPO7hiNiZpXAxqN8eR5Q1YXl9DSx/P303aJX\nLH+/aPpuw5xzhx0iOupC4YswsxLnXJHfdURKLH8/fbfoFcvfLxa/m7qPRESknUJBRETaxVsoPOh3\nAREWy99P3y16xfL3i7nvFlfHFERE5PPF256CiIh8jrgJBTM728zKzGyNmd3mdz1dxcyGmFmxma0w\ns+VmdovfNXU1M0s0syVm9he/a+lqZpZtZi+aWamZrTSzk/yuqauY2a3B/yeXmdmzZpbmd01fhJk9\namYVZrYsZF1fM/uHma0O3uf4WWNXiItQMLNE4F7gHGAc8A0zG+dvVV2mFfiBc24ccCJwQwx9t31u\nAVb6XUSE/AZ4zTk3FjieGPmeZjYYuBkocs5NABKBy/yt6gt7HDi7w7rbgDedc6OBN4PLUS0uQgGY\nBqxxzq1zzjUDzwEX+FxTl3DObXPOfRx8XIvXqAz2t6quY2YFwHnAw37X0tXMrA9wOvAIgHOu2TlX\n429VXSoJSDezJCAD2OpzPV+Ic+4tYGeH1RcATwQfPwFc2K1FRUC8hMJgYHPIcjkx1HDuY2aFwGTg\nQ38r6VK/Bv4PEPC7kAgYDlQCjwW7xx42s0y/i+oKzrktwC+BTcA2YLdz7u/+VhUR/Z1z24KPtwNR\nP+dvvIRCzDOzLOAl4PvOuT1+19MVzOwrQIVzbrHftURIEjAFuN85NxmoJwa6HwCCfesX4AXfICDT\nzK7wt6rIct6pnFF/Ome8hMIWYEjIckFwXUwws2S8QHjaOfey3/V0oVOA881sA16X3ywze8rfkrpU\nOVDunNu3Z/ciXkjEgjOB9c65SudcC/AycLLPNUXCDjMbCBC8r/C5ni8sXkJhETDazIabWQreAa+5\nPtfUJczM8PqkVzrnfuV3PV3JOXe7c67AOVeI999svnMuZn5tOue2A5vN7Jjgqi8BK3wsqSttAk40\ns4zg/6NfIkYOoncwF7g6+Phq4M8+1tIlkvwuoDs451rN7EbgdbyzIB51zi33uayucgpwJfCZmS0N\nrvuRc26ejzVJ+G4Cng7+WFkHfMvnerqEc+5DM3sR+BjvDLklRPnVv2b2LDATyDOzcuCnwD3AC2b2\nHbzRmy/1r8KuoSuaRUSkXbx0H4mISBgUCiIi0k6hICIi7RQKIiLSTqEgIiLtFAoi3cjMZsbiaK8S\nOxQKIiLSTqEg0gkzu8LMPjKzpWb2++CcDnVm9r/BOQLeNLP84LaTzOwDM/vUzF7ZN6a+mY0yszfM\n7BMz+9jMRgbfPitkDoWng1f8ivQICgWRDszsWGA2cIpzbhLQBnwTyARKnHPjgYV4V7QCPAn8m3Pu\nOOCzkPVPA/c6547HG/dn32iak4Hv483tMQLvqnSRHiEuhrkQOUJfAqYCi4I/4tPxBjoLAM8Ht3kK\neDk4J0K2c25hcP0TwB/NrBcw2Dn3CoBzrhEg+H4fOefKg8tLgULgnch/LZHDUyiIHMyAJ5xztx+w\n0uwnHbY72jFimkIet6G/Q+lB1H0kcrA3gYvNrB+0z8M7DO/v5eLgNpcD7zjndgO7zOy04PorgYXB\nWfDKzezC4HukmllGt34LkaOgXygiHTjnVpjZj4G/m1kC0ALcgDcJzrTgcxV4xx3AGzL5gWCjHzrS\n6ZXA783sruB7XNKNX0PkqGiUVJEwmVmdcy7L7zpEIkndRyIi0k57CiIi0k57CiIi0k6hICIi7RQK\nIiLSTqEgIiLtFAoiItJOoSAiIu3+PwqtAKBxkUVNAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f0b0cefb208>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(model.history.history['acc'], label='train');\n",
    "plt.plot(model.history.history['val_acc'], label='val');\n",
    "plt.legend();\n",
    "plt.xlabel('epoch');\n",
    "plt.ylabel('accuracy');"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYwAAAEKCAYAAAAB0GKPAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzt3Xd8XPWZ7/HPo27JTbbc5QYYN4xtEDaBQOgYCCUktAAX\n0thsSELYTXJJlhuyJLvh7u5NNskl1BBIaCEEErJLMKZfQrMMErjgggFLcpPLyLKtPs/94xxZY1m2\nRrZGo5n5vl+vec2po2co5zvn/M75/czdERER6U5WsgsQEZHUoMAQEZG4KDBERCQuCgwREYmLAkNE\nROKiwBARkbgoMEREJC4KDBERiYsCQ0RE4pKT7AJ6U0lJiU+aNCnZZYiIpJQlS5ZscfcR3W2X0MAw\ns/uATwOb3f2oLtYb8HPgXGA3cK27vx2uuwa4Odz0x+7+QHd/b9KkSZSXl/dW+SIiGcHMPo5nu0Rf\nkrofWHCA9ecAU8LXdcAdAGY2DLgFmA/MA24xs+KEVioiIgeU0MBw91eAbQfY5ELgtx54AxhqZmOA\ns4FF7r7N3bcDizhw8IiISIIlu9F7HFAVM18dLtvf8n2Y2XVmVm5m5bW1tQkrVEQk06V8o7e73w3c\nDVBWVrZPX+0tLS1UV1fT2NjY57X1tYKCAkpLS8nNzU12KSKShpIdGDXA+Jj50nBZDXBKp+UvHcwf\nqK6uZtCgQUyaNImgjT09uTtbt26lurqayZMnJ7scEUlDyb4k9RTwPyxwPFDn7huAhcBZZlYcNnaf\nFS7rscbGRoYPH57WYQFgZgwfPjwjzqREJDkSfVvtIwRnCiVmVk1w51MugLvfCTxNcEvtGoLbar8Q\nrttmZj8CFocfdau7H6jxvLs6DnbXlJIp31NEkiOhgeHuV3Sz3oHr97PuPuC+RNQlItLb3J3mtijN\nrcGrpc1pjUZpizqtUScavrfFvAfT0T3T+9u2Yz7aad+O6cNGFHHhnC7vDeo1yW7DyAiRSISHH36Y\nr33taz3a79xzz+Xhhx9m6NChCapMJLncwwOeO9EotIXz0T3LPGZZzHrvOFBG91rG3uvbPyNc1hQe\nzNsP7Hvmw2VNLW3Be+t+1re27ZmPXdcUfl4ynTVjlAIjHUQiEX71q1/tExitra3k5Oz/X8HTTz+d\n6NIkzXh4UNzd3MauplZ2NbcG703t810v393cxs6mVhpb2oi6E3U63sODbdQ7DvBRd9yDA3g0PNi7\nhwfocLuos/e2MdPRcFvf577G5MgyyMvJIj8nm7ycLPKys8jPyQqXBe8DcrMZMiCXvOy9l+ft2S47\nWBauz83OIifbyMkyssNXMJ1FdhZkZ2WRk2VkmZGT3bG+fX7PtmZkx35OzHyWdSzvi0vSCow+cNNN\nN/HBBx8wZ84ccnNzKSgooLi4mPfff59Vq1Zx0UUXUVVVRWNjIzfccAPXXXcd0NHVyc6dOznnnHP4\n5Cc/yWuvvca4ceP485//zIABA5L8zSSRNtc38urqLWzb1Rwc2Pcc5GMP/OEBv6mVneGBvzUa31E4\ny6AoP4eB+TkU5mUzMD+H/NxscrKyyMqCLLPwxZ4DUpaFy7M61sVu17GOjmWx23axX/vBNJimi2XB\nQTIrq9P69mXhfNaeZeyzrP2z9jrQZ2eRnxu852Qn+/6f1JBRgfHPf1nG8vU7evUzZ4wdzC3nzzzg\nNrfddhtLly6loqKCl156ifPOO4+lS5fuuf31vvvuY9iwYTQ0NHDcccfx2c9+luHDh+/1GatXr+aR\nRx7hnnvu4dJLL+WPf/wjV111Va9+F0m+qm27WbhsI88s3ciSddv3+gVekJtFUV4ORTEH+CEDchk3\ntIDCvI4Df1F+DkXt7+2vPctzKMoPpvNzsnSjhPRIRgVGfzFv3ry9npX4xS9+wZNPPglAVVUVq1ev\n3icwJk+ezJw5cwA49thj+eijj/qsXkmsNZvreWbpRp5ZtpGlNcEPmhljBnPjGUdy5oxRjCseQGFu\ntn4FS9JlVGB0dybQV4qKivZMv/TSSzz33HO8/vrrFBYWcsopp3T5LEV+fv6e6ezsbBoaGvqkVul9\n7s7Smh08s2wDzyzdyAe1uwA4ZsJQvn/uNM6eOZqJw4u6+RSRvpdRgZEsgwYNor6+vst1dXV1FBcX\nU1hYyPvvv88bb7zRx9VJX2iLOks+3s4zSzeycNlGaiINZGcZxx82jGtPmMRZM0czanBBsssUOSAF\nRh8YPnw4J554IkcddRQDBgxg1KhRe9YtWLCAO++8k+nTpzN16lSOP/74JFYqvam5Ncoba7fyzLKN\nPLtsE1t2NpGXncVJU0r41hlTOGP6KIqL8pJdpkjczPvLfW29oKyszDsPoLRixQqmT5+epIr6XqZ9\n3/6mobmNV1bXsnDpRp5bsYkdja0U5mVz6rSRLJg5mlOnjWRgvn6nSf9iZkvcvay77fRfrsghqm9s\n4YX3N/PM0o28tLKWhpY2hgzI5ayZo1kwczSfnFJCQW52sssUOWQKDJGDsHVnE8+t2MQzSzfytzVb\naW6LMmJQPp89dhwLZo5h/mHDyNVdTZJmFBgicdjd3MrSmh1UVG3nhfc389aH24g6lBYP4JoTJrLg\nqNHMHV9MVpaea5D0pcAQ6aQt6qzeXE/FugiV1REqqupYtametvAJ6iNGDuT6U4/g7JmjmTl2sB5+\nk4yhwJCM5u6sr2uksipCZVWEd6oiLK2pY3dzGwCDC3KYPX4oZ0w/nDnjh3J06VBGDMrv5lNF0pMC\nQzJKXUML71YH4VBRVUdldYTa+iYA8rKzmDF2MJeWjWf2+CHMGV/MpOGFOoMQCSkw+qGBAweyc+fO\nZJeR8ppbo6zYsCO4rLQuQkV1hLXhU9UAh40o4qQjSpgzYSizS4cybcwg8nN0N5PI/igwJC24Ox9t\n3U1F1XYqq+qoqIqwfP0OmtuCMQpKBuYzZ/xQLp47jtnhpaUhA3KTXLVIalFg9IGbbrqJ8ePHc/31\nweCCP/zhD8nJyeHFF19k+/bttLS08OMf/5gLL7wwyZWmlg+37OLZZRt5dc0W3q2uo66hBYABudnM\nKh3CF06cxOzxQ5k9fihjhxTo0pLIIcqswPjrTbDxvd79zNGz4JzbDrjJZZddxre+9a09gfHYY4+x\ncOFCvvnNbzJ48GC2bNnC8ccfzwUXXKCD2gFEo05ldYRnl29i0fJNrNkcXLY7ctRAzp01mtmlQThM\nGTlQPbuKJEBmBUaSzJ07l82bN7N+/Xpqa2spLi5m9OjR3HjjjbzyyitkZWVRU1PDpk2bGD16dLLL\n7VeaWtt47YOtPLtsE8+v2MTm+iays4z5k4dx5fwJnDF9FOOHFSa7TJGMkFmB0c2ZQCJdcsklPP74\n42zcuJHLLruMhx56iNraWpYsWUJubi6TJk3qslvzTFS3u4UXV27m2eUbeXllLbua2yjMy+aUqSM4\nc8YoTp06kqGF6rRPpK9lVmAk0WWXXcZXvvIVtmzZwssvv8xjjz3GyJEjyc3N5cUXX+Tjjz9OdolJ\nVRNpYNGyjTy7fBNvfbiN1qhTMjCfC+aM46wZo/jE4cPVH5NIkikw+sjMmTOpr69n3LhxjBkzhiuv\nvJLzzz+fWbNmUVZWxrRp05JdYp9yd5Zv2MGisD1iWTh07uEjivjKyYdx5oxRzCkdqq42RPoRBUYf\neu+9jgb3kpISXn/99S63S9dnMFraoiz+cNueRuuaSANmcMyEYr53zjTOnDGKw0YMTHaZIrIfCgxJ\nqF1Nrby8qpZFyzfxwvubqWtoIS8ni5OOKOEbpx3B6dNHqasNkRSR8MAwswXAz4Fs4F53v63T+onA\nfcAIYBtwlbtXh+vagPaf5evc/YJE1yuHbnN9I8+v2Myzyzbytw+20twaZWhhLqdPH8lZM0Zx0pQR\nFGkQIZGUk9D/a80sG7gdOBOoBhab2VPuvjxms/8AfuvuD5jZacBPgKvDdQ3uPudQ63D3jHi+Idmj\nJ26oa+AXz6/msfJq2qLO+GEDuGr+RM6cMYrjJhXr2QiRFJfon3nzgDXuvhbAzB4FLgRiA2MG8A/h\n9IvAn3qzgIKCArZu3crw4cPTOjTcna1bt1JQUNDnf3v7rmbufPkD7n/tI6LuXDV/ApfPm8C00YPS\n+p+5SKZJdGCMA6pi5quB+Z22qQQuJrhs9RlgkJkNd/etQIGZlQOtwG3u3uMwKS0tpbq6mtra2oP6\nAqmkoKCA0tLSPvt7u5pa+c3fPuSul9eys7mVz8wdx41nHKkH6UTSVH+4kPxt4P+a2bXAK0AN0Bau\nm+juNWZ2GPCCmb3n7h/E7mxm1wHXAUyYMGGfD8/NzWXy5MkJLD/zNLdGeeStdfzyhTVs2dnEmTNG\n8e2zpjJ19KBklyYiCZTowKgBxsfMl4bL9nD39QRnGJjZQOCz7h4J19WE72vN7CVgLvBBp/3vBu4G\nKCsrS+5F/DTXFnWeqqzhp4tWUbWtgfmTh3HX1cdy7MTiZJcmIn0g0YGxGJhiZpMJguJy4POxG5hZ\nCbDN3aPA9wjumMLMioHd7t4UbnMi8G8Jrle64O48v2Iz/75wJSs31TNz7GAe+OIsTp5SojYKkQyS\n0MBw91Yz+zqwkOC22vvcfZmZ3QqUu/tTwCnAT8zMCS5JXR/uPh24y8yiQBZBG8byff6IJNSba7fy\nbwtXsuTj7UwuKeKXV8zlvFlj9AS2SAayZN+K2ZvKysq8vLw82WWkhWXr6/j3hSt5aWUtowbnc8Pp\nR3JJWSm5ujVWJO2Y2RJ3L+tuu/7Q6C39yEdbdvF/Fq3iL5XrGTIgl++dM41rTpikjv9ERIEhgU07\nGvnF86v5/eIqcrOzuP7Uw7nu5MM1jKmI7KHAyHB1u1u44+UPuP+1D2ltcz4/fwJfP+0IRg7q+wcA\nRaR/U2BkqIbmNn7z2ofc+dIH1De1ctGc4KG7CcP10J2IdE2BkWFa2qI8uriKXzy/mtr6Jk6fNpJv\nnz2V6WMGJ7s0EennFBgZIhp1/vLuen66aBUfb93NcZOKuePKYyibNCzZpYlIilBgZIDX1mzhR/+9\nghUbdjBt9CB+c+1xnDJ1hB66E5EeUWCkuSferubbf6iktLiQn18+h/OPHquH7kTkoCgw0tiDb3zM\nzX9aygmHD+ee/1GmQYtE5JDoCJKm7n7lA/716fc5bdpIfnXlMXrwTkQOmQIjzbg7//ncan7+/GrO\nmzWGn102h7wcdechIodOgZFG3J1/+e8V3Pvqh3zu2FL+92ePJlvtFSLSSxQYaSIadW7+81IefnMd\n13xiIrecP1ON2yLSqxQYaaC1Lcp3Hn+XJ9+p4e9POZzvnj1Vt8yKSK9TYKS4ptY2vvnIOyxctonv\nnD2V6089ItkliUiaUmCksIbmNv7uwSW8sqqWW86fwRdO1NjlIpI4CowUVd/YwpceKGfxR9v4t88e\nzaXHje9+JxGRQ6DASEGR3c1cc99bLFu/g59fPpcLZo9NdkkikgEUGCmmtr6Jq3/9Jmtrd3HHVcdy\n5oxRyS5JRDKEAiOFrI80cNW9b7KhrpH7rj2OT04pSXZJIpJBFBgp4qMtu7jy3jfZ0dDC7740T92S\ni0ifU2CkgNWb6rny3jdpaYvy8FeOZ1bpkGSXJCIZSIHRzy2tqePqX79JTnYWv/+7T3DkqEHJLklE\nMpQCox9b8vE2rr1vMYMH5PLQl+czqaQo2SWJSAZTYPRTf1uzhS8/UM7oIQU8+OX5jBs6INkliUiG\nU2D0Q88t38TXHn6bycOL+N2X5zFyUEGySxIRIeEDJZjZAjNbaWZrzOymLtZPNLPnzexdM3vJzEpj\n1l1jZqvD1zWJrrU/+Evler764BKmjR7Eo9cdr7AQkX4joYFhZtnA7cA5wAzgCjOb0Wmz/wB+6+5H\nA7cCPwn3HQbcAswH5gG3mFlxIutNtsfKq7jh0XeYO2EoD315PsVFeckuSURkj0SfYcwD1rj7Wndv\nBh4FLuy0zQzghXD6xZj1ZwOL3H2bu28HFgELElxv0tz/tw/57uPvcuIRJTzwxXkMKshNdkkiIntJ\ndGCMA6pi5qvDZbEqgYvD6c8Ag8xseJz7poXbX1zDD/+ynLNmjOLea8oozFPTkoj0P/1hsOdvA58y\ns3eATwE1QFu8O5vZdWZWbmbltbW1iaoxIdydf1/4Pv++cCUXzhnL7VceQ35OdrLLEhHpUlyBYWaz\nDvLza4DYfrdLw2V7uPt6d7/Y3ecC/xQui8Szb7jt3e5e5u5lI0aMOMgy+1406vzzX5Zz+4sfcMW8\n8fz00jnkZveH/BYR6Vq8R6hfmdlbZvY1M+tJvxSLgSlmNtnM8oDLgadiNzCzEjNrr+N7wH3h9ELg\nLDMrDhu7zwqXpby2qHPTE+9y/2sf8aVPTuZfPzOLbI2/LSL9XFyB4e4nAVcS/OJfYmYPm9mZcezX\nCnyd4EC/AnjM3ZeZ2a1mdkG42SnASjNbBYwC/iXcdxvwI4LQWQzcGi5Leb/524c8Vl7NN0+fws3n\nTdf42yKSEszd4984uE32IuAXwA7AgO+7+xOJKa9nysrKvLy8PNlldOtL9y/m4227ee4fPpXsUkRE\nMLMl7l7W3XbxtmEcbWY/IzhLOA04392nh9M/O6RKM4y7U1EVYc74ockuRUSkR+K9f/OXwL0EZxMN\n7Qvdfb2Z3ZyQytJU9fYGtu5qVmCISMqJNzDOAxrcvQ0gbKQucPfd7v67hFWXhiqqIgAKDBFJOfHe\nJfUcENtdamG4THqosipCfk4WU0drXAsRSS3xBkaBu+9snwmnCxNTUnqrrI5w1LgheuZCRFJOvEet\nXWZ2TPuMmR0LNBxge+lCS1uU92rqmF2qy1EiknribcP4FvAHM1tPcCvtaOCyhFWVplZtqqexJcrs\n8RqTW0RST1yB4e6LzWwaMDVctNLdWxJXVnpqb/CeOz6te2kXkTTVk25RpxJ0RV4AHGNmuPtvE1NW\neqqsijCsKI/xwzTcqoiknrgCw8xuIejCYwbwNMGASK8CCoweqKyqY3bpEHUFIiIpKd5G788BpwMb\n3f0LwGxAF+J7YGdTK6s21zNbz1+ISIqKNzAa3D0KtJrZYGAze3c9Lt14r7oOdxQYIpKy4m3DKDez\nocA9wBJgJ/B6wqpKQ5XV4RPeuqVWRFJUt4FhwQX3n4SDGt1pZs8Ag9393YRXl0Yq1kWYOLyQ4qK8\nZJciInJQur0k5UH/50/HzH+ksOi5ymr1UCsiqS3eNoy3zey4hFaSxjbtaGRDXaOe8BaRlBZvG8Z8\n4Eoz+xjYRfC0t7v70QmrLI20P7CnBm/JaK3NsHsL7KqFXVtg91Zwh4LBkD943/es7GRXLJ3EGxhn\nJ7SKNFdZFSEny5g5dnCySxHpPa1N4YG/PQS2htPh/O6te0837ejZ5+cN7AiQgiFdh8p+l/dC6LhD\ntBXaWiDaAm2t4XtX87HbhfNYR+0FQ4P33AGQws9hxRsY8Y/jKvuoqIowfcxgCnL1i0n6qbZWaNkF\nTfXhQX5Lx8G/y0DYAs31XX9WVg4UlkBRCRQOh3HHQNGIcNnw8H1EsB6Dpjpo3BEEyp73ur2nm3YE\nNWz7oGN5W3P33ys2dPIHhSEQHuzbmg984I+29uo/4uCfTS4MCMNjz6vT/F7ri/del5Pcm2biDYz/\nJggNI+gaZDKwEpiZoLrSRjTqvFtdx0Vzxya7FEl10Si0NkDzruDVshuad0Pzzq6nW8LtupzeHX5G\nON3WtP+/m5UbHvzDA/7QieEBf3hHMMQGQsHQvvkV3dIYEzJdhc6Ovdc37QTLguzcINSyc4Pvts98\nTszyzvM92A8Pg6/9FemYboiZjlQF6xoiQVAdSG7h/sNm7FyYe2VC/5HH2/ngrNj5sKvzryWkojTz\nQe1Odja1MkcdDkpP1NXAOw/CsieCyzntAdET2XnBASavqOM9ryg4sA8tilleGPwSzy2E/IF7nwEU\nDg8ORv3xMkpuQfAaODLZlfQOd2ht3DtM9gqayL5hs3MjbFkZTO/e0j8CozN3f9vM5vd2MemoY0hW\n9aQi3Yi2wZrnYMn9sOoZ8ChMPhkmntD1gT83PNDnFXaxvCj4pSupwyxo48gdAIPHJLuaLsXb+eA/\nxMxmAccA6xNSUZqprI4wKD+Hw0oGJrsU6a92rA/OJt7+LdRVBb/uT7wBjrkGhk1OdnUie8R7hhE7\nAHUrQZvGH3u/nPRTURXh6PFDyMrqh6f0kjzRNljzfMzZRBscdiqc9WOYem7SGzdFuhJvG8Y/J7qQ\ndNTY0sb7G+q57uTDkl2K9Bc7NoRnEw90nE2c8A049hoYpv9OpH+L95LUIuCSsD8pzKwYeNTdu30+\nw8wWAD8HsoF73f22TusnAA8AQ8NtbnL3p81sErCC4G4sgDfc/avx1NtfLFtfR2vU1SVIpou2wQcv\nwpLfwMq/BmcTkz8FZ/0Ipp6nswlJGfFekhrRHhYA7r7dzLq9NcHMsoHbgTOBamCxmT3l7stjNrsZ\neMzd7zCz9gGaJoXrPnD3OXHW2O9UVNUBKDB6U1sL1L4P6ytg/TuwoSK4P3/kDBg9C0YdFbwXT4as\neHu+SZD6jfDO72DJb6FuXXD30QlfD9omhh+e3NpEDkK8gdFmZhPcfR2AmU0kvof55gFr3H1tuN+j\nwIVAbGA40P4I9BDSqDG9sirC2CEFjBxckOxSUlNbaxgOYTCsr4BNS4NbDwHyBsGY2TCuDDavgNWL\ngl/vENw9NGpmR4CMnhWESl5hYmuORmHtC1AeezZxMpz5zzDtPMjJT+zfF0mgeAPjn4BXzexlgof3\nTgKui2O/cUBVzHw1Qb9UsX4IPGtm3wCKgDNi1k02s3eAHcDN7v7/4qy3X6ioiqj/qHi1h0N7MKx/\np+twOO7LwQNKY+YE1/xjzyJaGqF2BWx8DzYuDd7f+wOU/zpYb1kw7PCOAGl/DRx16M8Z1G/saJuI\nrAueX/jE9XDstTqbkLQRb6P3M+HDeseHi77l7lt6qYYrgPvd/f+Y2SeA35nZUcAGYIK7bzWzY4E/\nmdlMd9+rQxozu44wvCZMmNBLJR26bbuaWbdtN5+f339q6jfaWoOHjWIvK21cGjzFDMHZQXs4jJkD\nY+cEB/ruLjHlFgRhMnZuxzJ3iHzcESCblkJNefBAXLvCkjA8joLRRwdnJSVTun+OIRqFtTFtE9FW\nmHQSnH4LTD9fZxOSduJt9P4M8IK7/1c4P9TMLnL3P3Wzaw17D+VaGi6L9SVgAYC7v25mBUCJu28G\nmsLlS8zsA+BIoDx2Z3e/G7gboKysrN/0ebVnhL3+cIbR3ldOdl7f9wDa1gpbVu19WWnje/uGQ9kX\nw4N9nOEQLzMonhS8pn+6Y3lDBDYtC0PkveD9zbs7usjIzoeR08J2kVkdgVIwBOo3QcWDsOSBIIwG\nDIPj/x6OuRZKjuidukX6oXgvSd3i7k+2z7h7xMxuAboLjMXAFDObTBAUlwOf77TNOuB04H4zm07Q\nV1WtmY0Atrl7m5kdBkwB1sZZb9JVrIuQZTBrXB8+4d1YB1vWBAfo2Ne2tR0dqVl2EBzZecEv6Jz8\n4H3Psph12Xmd1sdOt++bt+96HDaHbQ+dw2H00WE4zAnOHoYfkZzG6QFDYdKJwatdWwtsWR2chWx8\nNzgrWflMcKmp3ZDxUL8h5mziBzqbkIwRb2B09X90t/u6e6uZfR1YSHDL7H3uvszMbgXK3f0p4B+B\ne8zsRoIG8Gvd3c3sZOBWM2sBosBX3X1bnPUmXWV1hCkjB1GUf1C9r+xfNAo7asIwWB0TDKuDfmXa\nZeUEv9RLjgwaWwuGBAfE1qagl862lvC9KWa60/qWur3XtzaH69r3b9p/j557wuELHW0OyQqHeGXn\nwqgZwevoS4Nl7kH7RHuIbFoGg8YEbRMlU5Jarkhfi/doVm5mPyW4RRbgemBJPDu6+9PEDPEaLvtB\nzPRy4MQu9vsjKfo0ubtTWRXhrBmjD/5DWhqDrpy7CobYTugKhgShcMQZwQGs5MjgVTyxb/oSikbD\nrqFjAsXbgoNqOgyAYxb06zN4DEw5M9nViCRVvIHxDeB/Ab8P5xcRhIZ0Yd223Wzf3RLfHVK7tuwd\nBu3T2z+m485lg6HjgyCYeOLewVA0Irk9iWZlQVZ+cElGV2VE0lq8d0ntAm5KcC1po6OH2v0ExnuP\nw1v3BMHQEHOVLacAhk+BscfA7Cs6gmHY4Yl/fkBEpBvx3iU1AvguwYBJe55Cc/fTElRXSquoilCQ\nm8WRozr1UNu8C57+bnCHzcgZMOPCjjOFkilBg2p/vsYvIhkt3ktSDxFcjvo08FXgGqA2UUWlusqq\nCLPGDSEnO+bgv2kZ/OELwVnFyd+BT90UjNAlIpIi4v05O9zdfw20uPvL7v5FQGcXXWhujbJ0/Y6O\ny1HuUH4f3HMaNGyHq5+E025WWIhIyon3qNU+0OwGMzuPoL+nYYkpKbWt3FhPc2s0aPBurIOnvgnL\n/wSHnwafuSt9hpMUkYwTb2D82MyGEDwz8UuCzgJvTFhVKawifML7uNwP4c6/h7pqOOOHcMINap8Q\nkZQW711S/xVO1gGndl5vZt9z95/0ZmGpqvLjbdxQ+Awj//Bw8CzCF5+B8fOSXZaIyCHrrQvplwAK\njF1buWT1t5kfLYdpn4YL/y8MKE52VSIivaK3rpFowOqPXiV6xwnMaang5cO/A5c9qLAQkbTSW4HR\nb3qJ7XPRNnjpNnjgfJqsgIubb8Xm/11yn74WEUmA3roklZlHxx0b4ImvwEf/D46+nN8N+hrLnq9m\ndmk/6NJcRKSX9dYZxh966XNSx+pFcOeJULMELroDLr6LxRtaOKykiCGFfdDpn4hIHzvgGYaZlcSO\nrGdmVxGM070UuMfdHcDd/zWhVfYnrc3wwq3w2i9h5Ey45DcwYiruTkVVhE8eUZLsCkVEEqK7M4xn\n2yfM7GbgaoJuzc8EfprAuvqn7R/BbxYEYVH2JfjK8zBiKgAbdzRSW9/UP0bYExFJgO7aMGLbJi4G\nTnL3XWb2MPB24srqh5b9KXhqG+CSB2DmRXutrlgXPLAXV5fmIiIpqLvAGGBmcwnORLLDbs5x9xYz\na0t4df3ZOMEzAAAPAUlEQVRBSwMs/H7QH9S4Mvjcr4PxoTupqI6Ql53F9DGD+r5GEZE+0F1gbKDj\n0tM2Mxvj7hvMbDiwn7E500jtKnj8C8HwnCd8Mxi/eT+j2FWsizB97GDyc9JglDkRkS4cMDDcfZ9u\nQELbgZN7v5x+wh0qHoanvw25A+DKxw84PGdb1Hmvpo5Lji3twyJFRPpW3M9hmNnFwCcJHtJ71d2f\nTFhVydRUD//9j/Du72HSSXDxPcF4zgewZvNOdje3MWeC2i9EJH3FO+Ler4AjgEfCRX9nZme4e3qN\n673hXfjDtbD9Qzjl+3DytyGr+0tMFVXbAfTAnoiktXjPME4Dprc/d2FmDwDLElZVX3MPxth+9p+g\ncDhc8xeY9Mm4d6+oqmNwQQ6ThhclsEgRkeSK90nvNcCEmPnx4bL0sO4N+Ot34LBT4at/61FYQDCG\n9+zxQ8nKysweUkQkM8R7hjEIWGFmb4XzxwHlZvYUgLtfkIji+szET8DVf4LJn+rxIEe7m1tZtame\nM6YfnqDiRET6h3gD4wcJraI/OHx/N4Qd2LL1O2iLutovRCTtxTvi3stmNorgzALgLXffnLiyUoee\n8BaRTBHX9RczuxR4i2BkvUuBN83sc3Huu8DMVprZGjO7qYv1E8zsRTN7x8zeNbNzY9Z9L9xvpZmd\nHd9X6lsV1RHGDR3AiEH5yS5FRCSh4r0k9U/Ace1nFWY2AngOePxAO5lZNnA7QWeF1cBiM3vK3ZfH\nbHYz8Ji732FmM4CngUnh9OXATGAs8JyZHenu/apLkop1ET1/ISIZId4W3qxOl6C2xrnvPGCNu691\n92bgUeDCTts4MDicHgKsD6cvBB519yZ3/5Dgrqx5cdbbJ2rrm6iJNDBH7RcikgHiPcP4q5ktpOPB\nvcsIzgS6Mw6oipmvBuZ32uaHwLNm9g2gCDgjZt83Ou07Ls56+8S71Wq/EJHMEe8ZhgN3AUeHr7t7\nsYYrgPvdvRQ4F/idmcV9b6uZXWdm5WZWXltb24tlda+iKkJ2lnHUuMHdbywikuLiPTCf6e5PuPs/\nhK8ngXPi2K+G4CG/dqXhslhfAh4DcPfXgQKgJM59cfe73b3M3ctGjBgR59fpHRVVEY4cNYjCvN4a\nGl1EpP86YGCY2d+b2XvA1PAOpvbXh8C7cXz+YmCKmU02szyCRuynOm2zDjg9/HvTCQKjNtzucjPL\nN7PJwBSCO7X6BXensiqiEfZEJGN099P4YeCvwE+A2Fti6919W3cf7u6tZvZ1YCGQDdzn7svM7Fag\n3N2fAv4RuMfMbiS49HVt2GfVMjN7DFhOMPbG9f3pDqkPt+xiR2Mrc8YPSXYpIiJ9orvxMOqAOoJ2\nhoPi7k/TqYHc3X8QM70cOHE/+/4L8C8H+7cTqVIN3iKSYXrWcZLsUbEuQmFeNlNGakhWEckMCoyD\nVFFdx6xxQ8hWD7UikiEUGAehqbWNFet3qMFbRDKKAuMgrNhQT3NbVIEhIhlFgXEQKqvU4C0imUeB\ncRAqqiKMHJTPmCEFyS5FRKTPKDAOQmU4JKuZGrxFJHMoMHqobncLa7fsUvuFiGQcBUYPtT+wp8AQ\nkUyjwOihyqoIZjCrVF2CiEhmUWD0UGV1hMNHDGRwQW6ySxER6VMKjB5wdyqqIszWCHsikoEUGD1Q\nE2lgy85m9VArIhlJgdEDFVXtDd7FSa5ERKTvKTB6oLIqQl5OFlNHq4daEck8CoweqKyqY+bYweTl\n6B+biGQeHfni1NoW5b2aOj1/ISIZS4ERp1WbdtLQ0qbAEJGMpcCIU0eDtwJDRDKTAiNOlVURhhbm\nMmFYYbJLERFJCgVGnCqrgwf21EOtiGQqBUYcdjW1smpTvS5HiUhGU2DE4b2aOqKu9gsRyWwKjDhU\naEhWEREFRjwqqyJMGFbIsKK8ZJciIpI0Cow4tA/JKiKSyRIeGGa2wMxWmtkaM7upi/U/M7OK8LXK\nzCIx69pi1j2V6Fq7snlHI+vrGtV+ISIZLyeRH25m2cDtwJlANbDYzJ5y9+Xt27j7jTHbfwOYG/MR\nDe4+J5E1dqfjgT11aS4imS3RZxjzgDXuvtbdm4FHgQsPsP0VwCMJrqlHKqsj5GQZM8cqMEQksyU6\nMMYBVTHz1eGyfZjZRGAy8ELM4gIzKzezN8zsov3sd124TXltbW1v1b1HRVWEaWMGUZCb3eufLSKS\nSvpTo/flwOPu3hazbKK7lwGfB/7TzA7vvJO73+3uZe5eNmLEiF4tKBp13q2q05CsIiIkPjBqgPEx\n86Xhsq5cTqfLUe5eE76vBV5i7/aNhFu7ZSf1Ta1q8BYRIfGBsRiYYmaTzSyPIBT2udvJzKYBxcDr\nMcuKzSw/nC4BTgSWd943kSqq6gA94S0iAgm+S8rdW83s68BCIBu4z92XmdmtQLm7t4fH5cCj7u4x\nu08H7jKzKEGw3RZ7d1VfqKyKMDA/h8NGDOzLPysi0i8lNDAA3P1p4OlOy37Qaf6HXez3GjArocV1\no6IqwtGlQ8jOUg+1IiL9qdG7X2lsaWPFhh16wltEJKTA2I9l63fQGnW1X4iIhBQY+1GpIVlFRPai\nwNiPyuoIowcXMGpwQbJLERHpFxQY+1FRFdHZhYhIDAVGF7bvaubjrbvV4C0iEkOB0YWK6vYR9tTh\noIhIOwVGFyqrIpjB0epDSkRkDwVGFyqrIkwZOZCB+Ql/rlFEJGUoMDpxdzV4i4h0QYHRSdW2Brbv\nblGDt4hIJwqMTvY0eKv9QkRkLwqMTirWRSjIzWLq6EHJLkVEpF9RYHRSWR3hqLFDyM3WPxoRkVg6\nKsZoaYuytKZODd4iIl1QYMRYubGeptaoGrxFRLqgwIhRoR5qRUT2S4ERo6IqwvCiPEqLByS7FBGR\nfkeBEaOyKsLs8UMx05CsIiKdKTBC9Y0trKndqecvRET2Q4EReq+6DneYM0GBISLSFQVGqOMJb3Vp\nLiLSFQVGqGJdhMklRQwtzEt2KSIi/ZICI1RZHdHZhYjIASgwgA11DWza0aQH9kREDiDhgWFmC8xs\npZmtMbObulj/MzOrCF+rzCwSs+4aM1sdvq5JVI2NLVHOnTWaeZOHJepPiIikPHP3xH24WTawCjgT\nqAYWA1e4+/L9bP8NYK67f9HMhgHlQBngwBLgWHffvr+/V1ZW5uXl5b38LURE0puZLXH3su62S/QZ\nxjxgjbuvdfdm4FHgwgNsfwXwSDh9NrDI3beFIbEIWJDQakVEZL8SHRjjgKqY+epw2T7MbCIwGXih\np/uKiEji9adG78uBx929rSc7mdl1ZlZuZuW1tbUJKk1ERBIdGDXA+Jj50nBZVy6n43JU3Pu6+93u\nXubuZSNGjDjEckVEZH8SHRiLgSlmNtnM8ghC4anOG5nZNKAYeD1m8ULgLDMrNrNi4KxwmYiIJEFO\nIj/c3VvN7OsEB/ps4D53X2ZmtwLl7t4eHpcDj3rMLVvuvs3MfkQQOgC3uvu2RNYrIiL7l9Dbavua\nbqsVEem5/nJbrYiIpIm0OsMws1rg40P4iBJgSy+V09/ou6WudP5++m79w0R37/auobQKjENlZuXx\nnJalIn231JXO30/fLbXokpSIiMRFgSEiInFRYOzt7mQXkED6bqkrnb+fvlsKURuGiIjERWcYIiIS\nFwUG3Q/ylKrMbLyZvWhmy81smZndkOyaEsHMss3sHTP7r2TX0pvMbKiZPW5m75vZCjP7RLJr6i1m\ndmP43+RSM3vEzAqSXdOhMLP7zGyzmS2NWTbMzBaFA8AtCrs4SmkZHxjhIE+3A+cAM4ArzGxGcqvq\nNa3AP7r7DOB44Po0+m6xbgBWJLuIBPg58Iy7TwNmkybf0czGAd8Eytz9KIJugy5PblWH7H72Ha/n\nJuB5d58CPB/Op7SMDwx6PshTynD3De7+djhdT3DASasxRcysFDgPuDfZtfQmMxsCnAz8GsDdm909\ncuC9UkoOMMDMcoBCYH2S6zkk7v4K0LmvuwuBB8LpB4CL+rSoBFBgZMhATWY2CZgLvJncSnrdfwLf\nBaLJLqSXTQZqgd+El9vuNbOiZBfVG9y9BvgPYB2wAahz92eTW1VCjHL3DeH0RmBUMovpDQqMDGBm\nA4E/At9y9x3Jrqe3mNmngc3uviTZtSRADnAMcIe7zwV2kQaXNADCa/kXEoTiWKDIzK5KblWJFfbE\nnfK3pCowejbIU8oxs1yCsHjI3Z9Idj297ETgAjP7iOBS4mlm9mByS+o11UC1u7efET5OECDp4Azg\nQ3evdfcW4AnghCTXlAibzGwMQPi+Ocn1HDIFRpyDPKUiMzOCa+Ar3P2nya6nt7n799y91N0nEfx7\ne8Hd0+KXqrtvBKrMbGq46HRgeRJL6k3rgOPNrDD8b/R00qRBv5OngGvC6WuAPyexll6R0AGUUsH+\nBnlKclm95UTgauA9M6sIl33f3Z9OYk0Sv28AD4U/ZNYCX0hyPb3C3d80s8eBtwnu5HuHFH8q2swe\nAU4BSsysGrgFuA14zMy+RNCL9qXJq7B36ElvERGJiy5JiYhIXBQYIiISFwWGiIjERYEhIiJxUWCI\niEhcFBgi/YSZnZJuPe5KelFgiIhIXBQYIj1kZleZ2VtmVmFmd4Xjcew0s5+FYzw8b2Yjwm3nmNkb\nZvaumT3ZPiaCmR1hZs+ZWaWZvW1mh4cfPzBmDIyHwiehRfoFBYZID5jZdOAy4ER3nwO0AVcCRUC5\nu88EXiZ40hfgt8D/dPejgfdilj8E3O7uswn6UWrv1XQu8C2CsVkOI3haX6RfyPiuQUR66HTgWGBx\n+ON/AEGnclHg9+E2DwJPhGNaDHX3l8PlDwB/MLNBwDh3fxLA3RsBws97y92rw/kKYBLwauK/lkj3\nFBgiPWPAA+7+vb0Wmv2vTtsdbJ87TTHTbej/UelHdElKpGeeBz5nZiNhz7jNEwn+X/pcuM3ngVfd\nvQ7YbmYnhcuvBl4ORz+sNrOLws/IN7PCPv0WIgdBv15EesDdl5vZzcCzZpYFtADXEwxwNC9ct5mg\nnQOCbq3vDAMhtsfZq4G7zOzW8DMu6cOvIXJQ1FutSC8ws53uPjDZdYgkki5JiYhIXHSGISIicdEZ\nhoiIxEWBISIicVFgiIhIXBQYIiISFwWGiIjERYEhIiJx+f9nK3l+y7lojQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f0b0ce56c88>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(model.history.history['top_k_categorical_accuracy'], label='train');\n",
    "plt.plot(model.history.history['val_top_k_categorical_accuracy'], label='val');\n",
    "plt.legend();\n",
    "plt.xlabel('epoch');\n",
    "plt.ylabel('top5_accuracy');"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[1.5555390566587448, 0.67167968749999996, 0.865234375]"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.evaluate_generator(val_generator, 80)"
   ]
  }
 ],
 "metadata": {
  "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": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}