{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from keras.datasets import mnist\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense\n",
    "from keras.utils import np_utils"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'2.0.5'"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import keras\n",
    "keras.__version__"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "np.random.seed(42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading data from https://s3.amazonaws.com/img-datasets/mnist.npz\n",
      "10903552/11490434 [===========================>..] - ETA: 0s \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b"
     ]
    }
   ],
   "source": [
    "(X_train, y_train), (X_test, y_test) = mnist.load_data()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Картинки 24х24, развернем каждую в вектор длины 784."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((60000, 28, 28), (10000, 28, 28))"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.shape, X_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "X_train = X_train.reshape(X_train.shape[0], 784).astype('float32')\n",
    "X_test = X_test.reshape(X_test.shape[0], 784).astype('float32')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Масштабируем данные."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "scaler = StandardScaler()\n",
    "X_train_scaled = scaler.fit_transform(X_train)\n",
    "X_test_scaled = scaler.transform(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "y_train = np_utils.to_categorical(y_train, 10)\n",
    "y_test = np_utils.to_categorical(y_test, 10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 0.,  0.,  0.,  0.,  0.,  1.,  0.,  0.,  0.,  0.],\n",
       "       [ 1.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.],\n",
       "       [ 0.,  0.,  0.,  0.,  1.,  0.,  0.,  0.,  0.,  0.]])"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_train[:3, :]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "model = Sequential()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "model.add(Dense(800, input_dim=784, kernel_initializer='normal', activation='relu'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "model.add(Dense(10, kernel_initializer='normal', activation='softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "model.compile(loss='categorical_crossentropy', optimizer='SGD', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_1 (Dense)              (None, 800)               628000    \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 10)                8010      \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 800)               8800      \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 10)                8010      \n",
      "=================================================================\n",
      "Total params: 652,820\n",
      "Trainable params: 652,820\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n"
     ]
    }
   ],
   "source": [
    "print(model.summary())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Визуализация модели (нужно выполнить pip install pydot-ng)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "image/svg+xml": [
       "<svg height=\"387pt\" viewBox=\"0.00 0.00 309.00 387.00\" width=\"309pt\" xmlns=\"http://www.w3.org/2000/svg\" xmlns:xlink=\"http://www.w3.org/1999/xlink\">\n",
       "<g class=\"graph\" id=\"graph0\" transform=\"scale(1 1) rotate(0) translate(4 383)\">\n",
       "<title>G</title>\n",
       "<polygon fill=\"white\" points=\"-4,4 -4,-383 305,-383 305,4 -4,4\" stroke=\"none\"/>\n",
       "<!-- 139891033145040 -->\n",
       "<g class=\"node\" id=\"node1\"><title>139891033145040</title>\n",
       "<polygon fill=\"none\" points=\"0,-332.5 0,-378.5 301,-378.5 301,-332.5 0,-332.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"81.5\" y=\"-351.8\">dense_1_input: InputLayer</text>\n",
       "<polyline fill=\"none\" points=\"163,-332.5 163,-378.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"190.5\" y=\"-363.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"163,-355.5 218,-355.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"190.5\" y=\"-340.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"218,-332.5 218,-378.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"259.5\" y=\"-363.3\">(None, 784)</text>\n",
       "<polyline fill=\"none\" points=\"218,-355.5 301,-355.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"259.5\" y=\"-340.3\">(None, 784)</text>\n",
       "</g>\n",
       "<!-- 139891033144464 -->\n",
       "<g class=\"node\" id=\"node2\"><title>139891033144464</title>\n",
       "<polygon fill=\"none\" points=\"30.5,-249.5 30.5,-295.5 270.5,-295.5 270.5,-249.5 30.5,-249.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"81.5\" y=\"-268.8\">dense_1: Dense</text>\n",
       "<polyline fill=\"none\" points=\"132.5,-249.5 132.5,-295.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"160\" y=\"-280.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"132.5,-272.5 187.5,-272.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"160\" y=\"-257.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"187.5,-249.5 187.5,-295.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"229\" y=\"-280.3\">(None, 784)</text>\n",
       "<polyline fill=\"none\" points=\"187.5,-272.5 270.5,-272.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"229\" y=\"-257.3\">(None, 800)</text>\n",
       "</g>\n",
       "<!-- 139891033145040&#45;&gt;139891033144464 -->\n",
       "<g class=\"edge\" id=\"edge1\"><title>139891033145040-&gt;139891033144464</title>\n",
       "<path d=\"M150.5,-332.366C150.5,-324.152 150.5,-314.658 150.5,-305.725\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"154,-305.607 150.5,-295.607 147,-305.607 154,-305.607\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 139890994950288 -->\n",
       "<g class=\"node\" id=\"node3\"><title>139890994950288</title>\n",
       "<polygon fill=\"none\" points=\"30.5,-166.5 30.5,-212.5 270.5,-212.5 270.5,-166.5 30.5,-166.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"81.5\" y=\"-185.8\">dense_2: Dense</text>\n",
       "<polyline fill=\"none\" points=\"132.5,-166.5 132.5,-212.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"160\" y=\"-197.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"132.5,-189.5 187.5,-189.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"160\" y=\"-174.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"187.5,-166.5 187.5,-212.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"229\" y=\"-197.3\">(None, 800)</text>\n",
       "<polyline fill=\"none\" points=\"187.5,-189.5 270.5,-189.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"229\" y=\"-174.3\">(None, 10)</text>\n",
       "</g>\n",
       "<!-- 139891033144464&#45;&gt;139890994950288 -->\n",
       "<g class=\"edge\" id=\"edge2\"><title>139891033144464-&gt;139890994950288</title>\n",
       "<path d=\"M150.5,-249.366C150.5,-241.152 150.5,-231.658 150.5,-222.725\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"154,-222.607 150.5,-212.607 147,-222.607 154,-222.607\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 139890995173520 -->\n",
       "<g class=\"node\" id=\"node4\"><title>139890995173520</title>\n",
       "<polygon fill=\"none\" points=\"30.5,-83.5 30.5,-129.5 270.5,-129.5 270.5,-83.5 30.5,-83.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"81.5\" y=\"-102.8\">dense_3: Dense</text>\n",
       "<polyline fill=\"none\" points=\"132.5,-83.5 132.5,-129.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"160\" y=\"-114.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"132.5,-106.5 187.5,-106.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"160\" y=\"-91.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"187.5,-83.5 187.5,-129.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"229\" y=\"-114.3\">(None, 10)</text>\n",
       "<polyline fill=\"none\" points=\"187.5,-106.5 270.5,-106.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"229\" y=\"-91.3\">(None, 800)</text>\n",
       "</g>\n",
       "<!-- 139890994950288&#45;&gt;139890995173520 -->\n",
       "<g class=\"edge\" id=\"edge3\"><title>139890994950288-&gt;139890995173520</title>\n",
       "<path d=\"M150.5,-166.366C150.5,-158.152 150.5,-148.658 150.5,-139.725\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"154,-139.607 150.5,-129.607 147,-139.607 154,-139.607\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 139890995174288 -->\n",
       "<g class=\"node\" id=\"node5\"><title>139890995174288</title>\n",
       "<polygon fill=\"none\" points=\"30.5,-0.5 30.5,-46.5 270.5,-46.5 270.5,-0.5 30.5,-0.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"81.5\" y=\"-19.8\">dense_4: Dense</text>\n",
       "<polyline fill=\"none\" points=\"132.5,-0.5 132.5,-46.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"160\" y=\"-31.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"132.5,-23.5 187.5,-23.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"160\" y=\"-8.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"187.5,-0.5 187.5,-46.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"229\" y=\"-31.3\">(None, 800)</text>\n",
       "<polyline fill=\"none\" points=\"187.5,-23.5 270.5,-23.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"229\" y=\"-8.3\">(None, 10)</text>\n",
       "</g>\n",
       "<!-- 139890995173520&#45;&gt;139890995174288 -->\n",
       "<g class=\"edge\" id=\"edge4\"><title>139890995173520-&gt;139890995174288</title>\n",
       "<path d=\"M150.5,-83.3664C150.5,-75.1516 150.5,-65.6579 150.5,-56.7252\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"154,-56.6068 150.5,-46.6068 147,-56.6069 154,-56.6068\" stroke=\"black\"/>\n",
       "</g>\n",
       "</g>\n",
       "</svg>"
      ],
      "text/plain": [
       "<IPython.core.display.SVG object>"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from IPython.display import SVG\n",
    "from keras.utils.vis_utils import model_to_dot\n",
    "\n",
    "SVG(model_to_dot(model, show_shapes=True).create(prog='dot', format='svg'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## Callbacks\n",
    "### Ранняя остановка\n",
    "Нужно также сказать несколько слов о такой важной особенности Keras, как колбеки. Через них реализовано много полезной функциональности. Например, если вы тренируете сеть в течение очень долгого времени, вам нужно понять, когда пора остановиться, если ошибка на вашем датасете перестала уменьшаться. По-английски описываемая функциональность называется \"early stopping\" (\"ранняя остановка\")."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "from keras.callbacks import EarlyStopping  \n",
    "early_stopping = EarlyStopping(monitor='loss')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "### Tensorboard\n",
    "Еще в качестве колбека можно использовать сохранение логов в формате, удобном для Tensorboard — это специальная утилита для обработки и визуализации информации из логов Tensorflow.\n",
    "После того, как обучение закончится (или даже в процессе!), вы можете запустить Tensorboard, указав абсолютный путь к директории с логами:\n",
    "tensorboard --logdir=/path/to/logs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "from keras.callbacks import TensorBoard  \n",
    "tensorboard = TensorBoard(log_dir='../logs/', write_graph=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 48000 samples, validate on 12000 samples\n",
      "Epoch 1/100\n",
      "4s - loss: 0.5569 - acc: 0.8703 - val_loss: 0.6286 - val_acc: 0.8392\n",
      "Epoch 2/100\n",
      "4s - loss: 0.5331 - acc: 0.8825 - val_loss: 0.6521 - val_acc: 0.8056\n",
      "Epoch 3/100\n",
      "4s - loss: 0.5124 - acc: 0.8879 - val_loss: 0.5859 - val_acc: 0.8623\n",
      "Epoch 4/100\n",
      "4s - loss: 0.4892 - acc: 0.8970 - val_loss: 0.5624 - val_acc: 0.8705\n",
      "Epoch 5/100\n",
      "4s - loss: 0.4695 - acc: 0.9018 - val_loss: 0.5454 - val_acc: 0.8783\n",
      "Epoch 6/100\n",
      "4s - loss: 0.4479 - acc: 0.9068 - val_loss: 0.5229 - val_acc: 0.8814\n",
      "Epoch 7/100\n",
      "4s - loss: 0.4141 - acc: 0.9184 - val_loss: 0.5296 - val_acc: 0.8748\n",
      "Epoch 8/100\n",
      "4s - loss: 0.3972 - acc: 0.9211 - val_loss: 0.5017 - val_acc: 0.8826\n",
      "Epoch 9/100\n",
      "4s - loss: 0.3762 - acc: 0.9268 - val_loss: 0.4880 - val_acc: 0.8828\n",
      "Epoch 10/100\n",
      "4s - loss: 0.3564 - acc: 0.9335 - val_loss: 0.4799 - val_acc: 0.8811\n",
      "Epoch 11/100\n",
      "4s - loss: 0.3405 - acc: 0.9365 - val_loss: 0.4492 - val_acc: 0.8964\n",
      "Epoch 12/100\n",
      "4s - loss: 0.3232 - acc: 0.9414 - val_loss: 0.4373 - val_acc: 0.8974\n",
      "Epoch 13/100\n",
      "4s - loss: 0.3100 - acc: 0.9435 - val_loss: 0.4296 - val_acc: 0.8984\n",
      "Epoch 14/100\n",
      "4s - loss: 0.3022 - acc: 0.9439 - val_loss: 0.4260 - val_acc: 0.8979\n",
      "Epoch 15/100\n",
      "4s - loss: 0.2862 - acc: 0.9485 - val_loss: 0.4138 - val_acc: 0.9018\n",
      "Epoch 16/100\n",
      "4s - loss: 0.2721 - acc: 0.9518 - val_loss: 0.4070 - val_acc: 0.9023\n",
      "Epoch 17/100\n",
      "4s - loss: 0.2571 - acc: 0.9555 - val_loss: 0.4056 - val_acc: 0.9024\n",
      "Epoch 18/100\n",
      "4s - loss: 0.2493 - acc: 0.9563 - val_loss: 0.4482 - val_acc: 0.8796\n",
      "Epoch 19/100\n",
      "4s - loss: 0.2418 - acc: 0.9570 - val_loss: 0.3796 - val_acc: 0.9093\n",
      "Epoch 20/100\n",
      "5s - loss: 0.2295 - acc: 0.9616 - val_loss: 0.3772 - val_acc: 0.9117\n",
      "Epoch 21/100\n",
      "4s - loss: 0.2222 - acc: 0.9615 - val_loss: 0.4037 - val_acc: 0.8967\n",
      "Epoch 22/100\n",
      "4s - loss: 0.2116 - acc: 0.9644 - val_loss: 0.3631 - val_acc: 0.9149\n",
      "Epoch 23/100\n",
      "4s - loss: 0.2025 - acc: 0.9651 - val_loss: 0.3645 - val_acc: 0.9127\n",
      "Epoch 24/100\n",
      "4s - loss: 0.1950 - acc: 0.9668 - val_loss: 0.3583 - val_acc: 0.9146\n",
      "Epoch 25/100\n",
      "4s - loss: 0.1936 - acc: 0.9666 - val_loss: 0.3562 - val_acc: 0.9158\n",
      "Epoch 26/100\n",
      "4s - loss: 0.1830 - acc: 0.9693 - val_loss: 0.3564 - val_acc: 0.9127\n",
      "Epoch 27/100\n",
      "4s - loss: 0.1747 - acc: 0.9707 - val_loss: 0.3471 - val_acc: 0.9167\n",
      "Epoch 28/100\n",
      "4s - loss: 0.1705 - acc: 0.9712 - val_loss: 0.3482 - val_acc: 0.9179\n",
      "Epoch 29/100\n",
      "4s - loss: 0.1626 - acc: 0.9732 - val_loss: 0.3422 - val_acc: 0.9197\n",
      "Epoch 30/100\n",
      "4s - loss: 0.1581 - acc: 0.9746 - val_loss: 0.3423 - val_acc: 0.9192\n",
      "Epoch 31/100\n",
      "4s - loss: 0.1523 - acc: 0.9753 - val_loss: 0.3464 - val_acc: 0.9158\n",
      "Epoch 32/100\n",
      "4s - loss: 0.1467 - acc: 0.9762 - val_loss: 0.3396 - val_acc: 0.9194\n",
      "Epoch 33/100\n",
      "4s - loss: 0.1430 - acc: 0.9763 - val_loss: 0.3377 - val_acc: 0.9190\n",
      "Epoch 34/100\n",
      "4s - loss: 0.1371 - acc: 0.9783 - val_loss: 0.3325 - val_acc: 0.9211\n",
      "Epoch 35/100\n",
      "4s - loss: 0.1348 - acc: 0.9787 - val_loss: 0.3368 - val_acc: 0.9192\n",
      "Epoch 36/100\n",
      "4s - loss: 0.1287 - acc: 0.9800 - val_loss: 0.3310 - val_acc: 0.9213\n",
      "Epoch 37/100\n",
      "5s - loss: 0.1279 - acc: 0.9793 - val_loss: 0.3345 - val_acc: 0.9198\n",
      "Epoch 38/100\n",
      "5s - loss: 0.1243 - acc: 0.9800 - val_loss: 0.3408 - val_acc: 0.9179\n",
      "Epoch 39/100\n",
      "4s - loss: 0.1193 - acc: 0.9814 - val_loss: 0.3307 - val_acc: 0.9202\n",
      "Epoch 40/100\n",
      "4s - loss: 0.1152 - acc: 0.9823 - val_loss: 0.3295 - val_acc: 0.9209\n",
      "Epoch 41/100\n",
      "5s - loss: 0.1143 - acc: 0.9823 - val_loss: 0.3320 - val_acc: 0.9186\n",
      "Epoch 42/100\n",
      "5s - loss: 0.1097 - acc: 0.9831 - val_loss: 0.3315 - val_acc: 0.9213\n",
      "Epoch 43/100\n",
      "5s - loss: 0.1085 - acc: 0.9833 - val_loss: 0.3339 - val_acc: 0.9207\n",
      "Epoch 44/100\n",
      "5s - loss: 0.1060 - acc: 0.9836 - val_loss: 0.3398 - val_acc: 0.9178\n",
      "Epoch 45/100\n",
      "4s - loss: 0.1031 - acc: 0.9846 - val_loss: 0.3299 - val_acc: 0.9214\n",
      "Epoch 46/100\n",
      "4s - loss: 0.1003 - acc: 0.9850 - val_loss: 0.3347 - val_acc: 0.9179\n",
      "Epoch 47/100\n",
      "5s - loss: 0.0979 - acc: 0.9856 - val_loss: 0.3342 - val_acc: 0.9204\n",
      "Epoch 48/100\n",
      "5s - loss: 0.0970 - acc: 0.9858 - val_loss: 0.3282 - val_acc: 0.9215\n",
      "Epoch 49/100\n",
      "5s - loss: 0.0946 - acc: 0.9860 - val_loss: 0.3316 - val_acc: 0.9200\n",
      "Epoch 50/100\n",
      "4s - loss: 0.0924 - acc: 0.9861 - val_loss: 0.3309 - val_acc: 0.9210\n",
      "Epoch 51/100\n",
      "4s - loss: 0.0898 - acc: 0.9875 - val_loss: 0.3294 - val_acc: 0.9222\n",
      "Epoch 52/100\n",
      "5s - loss: 0.0878 - acc: 0.9874 - val_loss: 0.3385 - val_acc: 0.9185\n",
      "Epoch 53/100\n",
      "4s - loss: 0.0859 - acc: 0.9879 - val_loss: 0.3258 - val_acc: 0.9207\n",
      "Epoch 54/100\n",
      "4s - loss: 0.0853 - acc: 0.9878 - val_loss: 0.3314 - val_acc: 0.9211\n",
      "Epoch 55/100\n",
      "5s - loss: 0.0824 - acc: 0.9885 - val_loss: 0.3322 - val_acc: 0.9214\n",
      "Epoch 56/100\n",
      "5s - loss: 0.0826 - acc: 0.9883 - val_loss: 0.3485 - val_acc: 0.9147\n",
      "CPU times: user 23min 40s, sys: 10min 57s, total: 34min 38s\n",
      "Wall time: 4min 38s\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7f3ae8f3e8d0>"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "%%time\n",
    "model.fit(X_train_scaled, y_train, batch_size=200, epochs=100, \n",
    "          validation_split=0.2, callbacks=[early_stopping, tensorboard], verbose=2);"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "C помощью Tensorboard удобно отслеживать процесс обучения нейронной сети. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<img src='../img/tensorboard_mnist.png'>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Оцениваем качество обучения сети на тестовых данных"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Доля верных ответов на тестовых данных: 91.04%\n"
     ]
    }
   ],
   "source": [
    "scores = model.evaluate(X_test_scaled, y_test, verbose=0)\n",
    "print(\"Доля верных ответов на тестовых данных: %.2f%%\" % (scores[1]*100))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## Сохраняем сеть\n",
    "**Архитектуру – в JSON-файл**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "model_json = model.to_json()\n",
    "with open(\"mnist_model.json\", \"w\") as json_file:\n",
    "    json_file.write(model_json)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "import pprint\n",
    "import json"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{u'backend': u'tensorflow',\n",
      " u'class_name': u'Sequential',\n",
      " u'config': [{u'class_name': u'Dense',\n",
      "              u'config': {u'activation': u'relu',\n",
      "                          u'activity_regularizer': None,\n",
      "                          u'batch_input_shape': [None, 784],\n",
      "                          u'bias_constraint': None,\n",
      "                          u'bias_initializer': {u'class_name': u'Zeros',\n",
      "                                                u'config': {}},\n",
      "                          u'bias_regularizer': None,\n",
      "                          u'dtype': u'float32',\n",
      "                          u'kernel_constraint': None,\n",
      "                          u'kernel_initializer': {u'class_name': u'RandomNormal',\n",
      "                                                  u'config': {u'mean': 0.0,\n",
      "                                                              u'seed': None,\n",
      "                                                              u'stddev': 0.05}},\n",
      "                          u'kernel_regularizer': None,\n",
      "                          u'name': u'dense_1',\n",
      "                          u'trainable': True,\n",
      "                          u'units': 800,\n",
      "                          u'use_bias': True}},\n",
      "             {u'class_name': u'Dense',\n",
      "              u'config': {u'activation': u'softmax',\n",
      "                          u'activity_regularizer': None,\n",
      "                          u'bias_constraint': None,\n",
      "                          u'bias_initializer': {u'class_name': u'Zeros',\n",
      "                                                u'config': {}},\n",
      "                          u'bias_regularizer': None,\n",
      "                          u'kernel_constraint': None,\n",
      "                          u'kernel_initializer': {u'class_name': u'RandomNormal',\n",
      "                                                  u'config': {u'mean': 0.0,\n",
      "                                                              u'seed': None,\n",
      "                                                              u'stddev': 0.05}},\n",
      "                          u'kernel_regularizer': None,\n",
      "                          u'name': u'dense_2',\n",
      "                          u'trainable': True,\n",
      "                          u'units': 10,\n",
      "                          u'use_bias': True}},\n",
      "             {u'class_name': u'Dense',\n",
      "              u'config': {u'activation': u'relu',\n",
      "                          u'activity_regularizer': None,\n",
      "                          u'batch_input_shape': [None, 784],\n",
      "                          u'bias_constraint': None,\n",
      "                          u'bias_initializer': {u'class_name': u'Zeros',\n",
      "                                                u'config': {}},\n",
      "                          u'bias_regularizer': None,\n",
      "                          u'dtype': u'float32',\n",
      "                          u'kernel_constraint': None,\n",
      "                          u'kernel_initializer': {u'class_name': u'RandomNormal',\n",
      "                                                  u'config': {u'mean': 0.0,\n",
      "                                                              u'seed': None,\n",
      "                                                              u'stddev': 0.05}},\n",
      "                          u'kernel_regularizer': None,\n",
      "                          u'name': u'dense_3',\n",
      "                          u'trainable': True,\n",
      "                          u'units': 800,\n",
      "                          u'use_bias': True}},\n",
      "             {u'class_name': u'Dense',\n",
      "              u'config': {u'activation': u'softmax',\n",
      "                          u'activity_regularizer': None,\n",
      "                          u'bias_constraint': None,\n",
      "                          u'bias_initializer': {u'class_name': u'Zeros',\n",
      "                                                u'config': {}},\n",
      "                          u'bias_regularizer': None,\n",
      "                          u'kernel_constraint': None,\n",
      "                          u'kernel_initializer': {u'class_name': u'RandomNormal',\n",
      "                                                  u'config': {u'mean': 0.0,\n",
      "                                                              u'seed': None,\n",
      "                                                              u'stddev': 0.05}},\n",
      "                          u'kernel_regularizer': None,\n",
      "                          u'name': u'dense_4',\n",
      "                          u'trainable': True,\n",
      "                          u'units': 10,\n",
      "                          u'use_bias': True}}],\n",
      " u'keras_version': u'2.0.5'}\n"
     ]
    }
   ],
   "source": [
    "with open(\"mnist_model.json\", \"r\") as json_file:\n",
    "    pprint.pprint(json.loads(json_file.read()))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "**Веса сохраняем в бинарный hd5-файл**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "model.save_weights(\"mnist_model.h5\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "**Теперь сеть можно восстановить и использовать**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "from keras.models import model_from_json\n",
    "\n",
    "with open(\"mnist_model.json\") as json_file:\n",
    "    loaded_model_json = json_file.read()\n",
    "\n",
    "# Создаем модель на основе загруженных данных\n",
    "loaded_model = model_from_json(loaded_model_json)\n",
    "# Загружаем веса в модель\n",
    "loaded_model.load_weights(\"mnist_model.h5\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Перед использованием модели ее обязательно нужно скомпилировать. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "model.compile(loss='categorical_crossentropy', optimizer='SGD', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Доля верных ответов на тестовых данных: 91.04%\n"
     ]
    }
   ],
   "source": [
    "scores = model.evaluate(X_test_scaled, y_test, verbose=0)\n",
    "print(\"Доля верных ответов на тестовых данных: %.2f%%\" % (scores[1]*100))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda root]",
   "language": "python",
   "name": "conda-root-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}