tensorflow/practice/convnets/ConvnetsCIFAR10.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "\n",
    "import os\n",
    "\n",
    "import tensorflow.keras as keras\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from tensorflow.keras.layers import Dense, Conv2D, MaxPooling2D, Flatten, Input, Activation, add, Add, Dropout, BatchNormalization\n",
    "from tensorflow.keras.callbacks import EarlyStopping\n",
    "from tensorflow.keras.models import Sequential, Model\n",
    "from tensorflow.keras.datasets import cifar10\n",
    "from tensorflow.keras.datasets import fashion_mnist\n",
    "from tensorflow.keras.utils import to_categorical\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "from IPython.display import SVG\n",
    "from tensorflow.python.keras.utils.vis_utils import model_to_dot\n",
    "\n",
    "random_state = 42"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(50000, 32, 32, 3)\n",
      "(50000, 1)\n"
     ]
    }
   ],
   "source": [
    "(x_train, y_train), (x_test, y_test) = cifar10.load_data()\n",
    "\n",
    "print(x_train.shape)\n",
    "print(y_train.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train = x_train.astype('float32') / 255\n",
    "y_train = np.eye(10)[y_train.astype('int32').flatten()]\n",
    "x_test = x_test.astype('float32') / 255\n",
    "y_test = np.eye(10)[y_test.astype('int32').flatten()]\n",
    "\n",
    "x_train, x_valid, y_train, y_valid = train_test_split(x_train, y_train, test_size=10000)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(40000, 32, 32, 3)\n",
      "(10000, 32, 32, 3)\n"
     ]
    }
   ],
   "source": [
    "print(x_train.shape)\n",
    "print(x_valid.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x1080 with 9 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(9, 15))\n",
    "fig.subplots_adjust(left=0, right=1, bottom=0, top=0.5, hspace=0.05, wspace=0.05)\n",
    "\n",
    "for i in range(9):\n",
    "    ax = fig.add_subplot(1, 9, i+1, xticks=[], yticks=[])\n",
    "    ax.imshow(x_train[i])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From C:\\Users\\suizh\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\ops\\resource_variable_ops.py:435: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Colocations handled automatically by placer.\n"
     ]
    }
   ],
   "source": [
    "model = Sequential()\n",
    "\n",
    "# 32x32x3\n",
    "model.add(Conv2D(6, kernel_size=(5, 5), activation='relu', \n",
    "                 kernel_initializer='he_normal', input_shape=(32, 32, 3)))\n",
    "# 28x28x6\n",
    "model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "# 14x14x6\n",
    "model.add(Conv2D(16, kernel_size=(5, 5), activation='relu', \n",
    "                 kernel_initializer='he_normal'))\n",
    "# 10x10x16\n",
    "model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "# 5x5x16\n",
    "model.add(Flatten())\n",
    "# 400\n",
    "model.add(Dense(120, activation='relu', kernel_initializer='he_normal'))\n",
    "# 120\n",
    "model.add(Dense(84, activation='relu', kernel_initializer='he_normal'))\n",
    "# 84\n",
    "model.add(Dense(10, activation='softmax'))\n",
    "# 10\n",
    "\n",
    "model.compile(loss='categorical_crossentropy',\n",
    "              optimizer='adam',\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/svg+xml": [
       "<svg height=\"709pt\" viewBox=\"0.00 0.00 392.00 709.00\" width=\"392pt\" 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(4 705)\">\n",
       "<title>G</title>\n",
       "<polygon fill=\"white\" points=\"-4,4 -4,-705 388,-705 388,4 -4,4\" stroke=\"none\"/>\n",
       "<!-- 2289219578792 -->\n",
       "<g class=\"node\" id=\"node1\"><title>2289219578792</title>\n",
       "<polygon fill=\"none\" points=\"51.5,-581.5 51.5,-627.5 332.5,-627.5 332.5,-581.5 51.5,-581.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"108\" y=\"-600.8\">conv2d: Conv2D</text>\n",
       "<polyline fill=\"none\" points=\"164.5,-581.5 164.5,-627.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"192.5\" y=\"-612.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"164.5,-604.5 220.5,-604.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"192.5\" y=\"-589.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"220.5,-581.5 220.5,-627.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"276.5\" y=\"-612.3\">(None, 32, 32, 3)</text>\n",
       "<polyline fill=\"none\" points=\"220.5,-604.5 332.5,-604.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"276.5\" y=\"-589.3\">(None, 28, 28, 6)</text>\n",
       "</g>\n",
       "<!-- 2289219579744 -->\n",
       "<g class=\"node\" id=\"node2\"><title>2289219579744</title>\n",
       "<polygon fill=\"none\" points=\"11,-498.5 11,-544.5 373,-544.5 373,-498.5 11,-498.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"108\" y=\"-517.8\">max_pooling2d: MaxPooling2D</text>\n",
       "<polyline fill=\"none\" points=\"205,-498.5 205,-544.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"233\" y=\"-529.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"205,-521.5 261,-521.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"233\" y=\"-506.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"261,-498.5 261,-544.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"317\" y=\"-529.3\">(None, 28, 28, 6)</text>\n",
       "<polyline fill=\"none\" points=\"261,-521.5 373,-521.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"317\" y=\"-506.3\">(None, 14, 14, 6)</text>\n",
       "</g>\n",
       "<!-- 2289219578792&#45;&gt;2289219579744 -->\n",
       "<g class=\"edge\" id=\"edge2\"><title>2289219578792-&gt;2289219579744</title>\n",
       "<path d=\"M192,-581.366C192,-573.152 192,-563.658 192,-554.725\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"195.5,-554.607 192,-544.607 188.5,-554.607 195.5,-554.607\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 2289219658248 -->\n",
       "<g class=\"node\" id=\"node3\"><title>2289219658248</title>\n",
       "<polygon fill=\"none\" points=\"41,-415.5 41,-461.5 343,-461.5 343,-415.5 41,-415.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"104.5\" y=\"-434.8\">conv2d_1: Conv2D</text>\n",
       "<polyline fill=\"none\" points=\"168,-415.5 168,-461.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"196\" y=\"-446.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"168,-438.5 224,-438.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"196\" y=\"-423.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"224,-415.5 224,-461.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"283.5\" y=\"-446.3\">(None, 14, 14, 6)</text>\n",
       "<polyline fill=\"none\" points=\"224,-438.5 343,-438.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"283.5\" y=\"-423.3\">(None, 10, 10, 16)</text>\n",
       "</g>\n",
       "<!-- 2289219579744&#45;&gt;2289219658248 -->\n",
       "<g class=\"edge\" id=\"edge3\"><title>2289219579744-&gt;2289219658248</title>\n",
       "<path d=\"M192,-498.366C192,-490.152 192,-480.658 192,-471.725\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"195.5,-471.607 192,-461.607 188.5,-471.607 195.5,-471.607\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 2289219660040 -->\n",
       "<g class=\"node\" id=\"node4\"><title>2289219660040</title>\n",
       "<polygon fill=\"none\" points=\"0,-332.5 0,-378.5 384,-378.5 384,-332.5 0,-332.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"104.5\" y=\"-351.8\">max_pooling2d_1: MaxPooling2D</text>\n",
       "<polyline fill=\"none\" points=\"209,-332.5 209,-378.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"237\" y=\"-363.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"209,-355.5 265,-355.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"237\" y=\"-340.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"265,-332.5 265,-378.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"324.5\" y=\"-363.3\">(None, 10, 10, 16)</text>\n",
       "<polyline fill=\"none\" points=\"265,-355.5 384,-355.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"324.5\" y=\"-340.3\">(None, 5, 5, 16)</text>\n",
       "</g>\n",
       "<!-- 2289219658248&#45;&gt;2289219660040 -->\n",
       "<g class=\"edge\" id=\"edge4\"><title>2289219658248-&gt;2289219660040</title>\n",
       "<path d=\"M192,-415.366C192,-407.152 192,-397.658 192,-388.725\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"195.5,-388.607 192,-378.607 188.5,-388.607 195.5,-388.607\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 2289219658696 -->\n",
       "<g class=\"node\" id=\"node5\"><title>2289219658696</title>\n",
       "<polygon fill=\"none\" points=\"64,-249.5 64,-295.5 320,-295.5 320,-249.5 64,-249.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"111.5\" y=\"-268.8\">flatten: Flatten</text>\n",
       "<polyline fill=\"none\" points=\"159,-249.5 159,-295.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"187\" y=\"-280.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"159,-272.5 215,-272.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"187\" y=\"-257.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"215,-249.5 215,-295.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"267.5\" y=\"-280.3\">(None, 5, 5, 16)</text>\n",
       "<polyline fill=\"none\" points=\"215,-272.5 320,-272.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"267.5\" y=\"-257.3\">(None, 400)</text>\n",
       "</g>\n",
       "<!-- 2289219660040&#45;&gt;2289219658696 -->\n",
       "<g class=\"edge\" id=\"edge5\"><title>2289219660040-&gt;2289219658696</title>\n",
       "<path d=\"M192,-332.366C192,-324.152 192,-314.658 192,-305.725\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"195.5,-305.607 192,-295.607 188.5,-305.607 195.5,-305.607\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 2289219962192 -->\n",
       "<g class=\"node\" id=\"node6\"><title>2289219962192</title>\n",
       "<polygon fill=\"none\" points=\"77.5,-166.5 77.5,-212.5 306.5,-212.5 306.5,-166.5 77.5,-166.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"122.5\" y=\"-185.8\">dense: Dense</text>\n",
       "<polyline fill=\"none\" points=\"167.5,-166.5 167.5,-212.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"195.5\" y=\"-197.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"167.5,-189.5 223.5,-189.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"195.5\" y=\"-174.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"223.5,-166.5 223.5,-212.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"265\" y=\"-197.3\">(None, 400)</text>\n",
       "<polyline fill=\"none\" points=\"223.5,-189.5 306.5,-189.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"265\" y=\"-174.3\">(None, 120)</text>\n",
       "</g>\n",
       "<!-- 2289219658696&#45;&gt;2289219962192 -->\n",
       "<g class=\"edge\" id=\"edge6\"><title>2289219658696-&gt;2289219962192</title>\n",
       "<path d=\"M192,-249.366C192,-241.152 192,-231.658 192,-222.725\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"195.5,-222.607 192,-212.607 188.5,-222.607 195.5,-222.607\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 2289220198016 -->\n",
       "<g class=\"node\" id=\"node7\"><title>2289220198016</title>\n",
       "<polygon fill=\"none\" points=\"70.5,-83.5 70.5,-129.5 313.5,-129.5 313.5,-83.5 70.5,-83.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"122.5\" y=\"-102.8\">dense_1: Dense</text>\n",
       "<polyline fill=\"none\" points=\"174.5,-83.5 174.5,-129.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"202.5\" y=\"-114.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"174.5,-106.5 230.5,-106.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"202.5\" y=\"-91.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"230.5,-83.5 230.5,-129.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"272\" y=\"-114.3\">(None, 120)</text>\n",
       "<polyline fill=\"none\" points=\"230.5,-106.5 313.5,-106.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"272\" y=\"-91.3\">(None, 84)</text>\n",
       "</g>\n",
       "<!-- 2289219962192&#45;&gt;2289220198016 -->\n",
       "<g class=\"edge\" id=\"edge7\"><title>2289219962192-&gt;2289220198016</title>\n",
       "<path d=\"M192,-166.366C192,-158.152 192,-148.658 192,-139.725\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"195.5,-139.607 192,-129.607 188.5,-139.607 195.5,-139.607\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 2289220198352 -->\n",
       "<g class=\"node\" id=\"node8\"><title>2289220198352</title>\n",
       "<polygon fill=\"none\" points=\"73.5,-0.5 73.5,-46.5 310.5,-46.5 310.5,-0.5 73.5,-0.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"125.5\" y=\"-19.8\">dense_2: Dense</text>\n",
       "<polyline fill=\"none\" points=\"177.5,-0.5 177.5,-46.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"205.5\" y=\"-31.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"177.5,-23.5 233.5,-23.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"205.5\" y=\"-8.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"233.5,-0.5 233.5,-46.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"272\" y=\"-31.3\">(None, 84)</text>\n",
       "<polyline fill=\"none\" points=\"233.5,-23.5 310.5,-23.5 \" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"272\" y=\"-8.3\">(None, 10)</text>\n",
       "</g>\n",
       "<!-- 2289220198016&#45;&gt;2289220198352 -->\n",
       "<g class=\"edge\" id=\"edge8\"><title>2289220198016-&gt;2289220198352</title>\n",
       "<path d=\"M192,-83.3664C192,-75.1516 192,-65.6579 192,-56.7252\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"195.5,-56.6068 192,-46.6068 188.5,-56.6069 195.5,-56.6068\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 2289219657968 -->\n",
       "<g class=\"node\" id=\"node9\"><title>2289219657968</title>\n",
       "<polygon fill=\"none\" points=\"140,-664.5 140,-700.5 244,-700.5 244,-664.5 140,-664.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times New Roman,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"192\" y=\"-678.8\">2289219657968</text>\n",
       "</g>\n",
       "<!-- 2289219657968&#45;&gt;2289219578792 -->\n",
       "<g class=\"edge\" id=\"edge1\"><title>2289219657968-&gt;2289219578792</title>\n",
       "<path d=\"M192,-664.254C192,-656.363 192,-646.749 192,-637.602\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"195.5,-637.591 192,-627.591 188.5,-637.591 195.5,-637.591\" stroke=\"black\"/>\n",
       "</g>\n",
       "</g>\n",
       "</svg>"
      ],
      "text/plain": [
       "<IPython.core.display.SVG object>"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "SVG(model_to_dot(model, show_shapes=True).create(prog='dot', format='svg'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 40000 samples, validate on 10000 samples\n",
      "WARNING:tensorflow:From C:\\Users\\suizh\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\ops\\math_ops.py:3066: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.cast instead.\n",
      "Epoch 1/100\n",
      "40000/40000 [==============================] - 4s 107us/sample - loss: 1.7317 - acc: 0.3754 - val_loss: 1.5099 - val_acc: 0.4518\n",
      "Epoch 2/100\n",
      "40000/40000 [==============================] - 3s 71us/sample - loss: 1.4319 - acc: 0.4851 - val_loss: 1.3665 - val_acc: 0.5113\n",
      "Epoch 3/100\n",
      "40000/40000 [==============================] - 3s 71us/sample - loss: 1.3301 - acc: 0.5213 - val_loss: 1.3355 - val_acc: 0.5262\n",
      "Epoch 4/100\n",
      "40000/40000 [==============================] - 3s 73us/sample - loss: 1.2606 - acc: 0.5503 - val_loss: 1.2764 - val_acc: 0.5478\n",
      "Epoch 5/100\n",
      "40000/40000 [==============================] - 3s 72us/sample - loss: 1.2025 - acc: 0.5721 - val_loss: 1.2616 - val_acc: 0.5488\n",
      "Epoch 6/100\n",
      "40000/40000 [==============================] - 3s 70us/sample - loss: 1.1507 - acc: 0.5920 - val_loss: 1.2126 - val_acc: 0.5701\n",
      "Epoch 7/100\n",
      "40000/40000 [==============================] - 3s 71us/sample - loss: 1.1111 - acc: 0.6073 - val_loss: 1.2064 - val_acc: 0.5761\n",
      "Epoch 8/100\n",
      "40000/40000 [==============================] - 3s 72us/sample - loss: 1.0702 - acc: 0.6217 - val_loss: 1.1992 - val_acc: 0.5765\n",
      "Epoch 9/100\n",
      "40000/40000 [==============================] - 3s 70us/sample - loss: 1.0303 - acc: 0.6360 - val_loss: 1.2022 - val_acc: 0.5758\n",
      "Epoch 00009: early stopping\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x2151092a048>"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "early_stopping = EarlyStopping(patience=1, verbose=1)\n",
    "model.fit(x_train, y_train,\n",
    "          batch_size=128, epochs=100, verbose=1,\n",
    "          validation_data=(x_valid, y_valid),\n",
    "          callbacks=[early_stopping])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test loss: 1.1864906604766845\n",
      "Test accuracy: 0.5804\n"
     ]
    }
   ],
   "source": [
    "score = model.evaluate(x_test, y_test, verbose=0)\n",
    "print('Test loss:', score[0])\n",
    "print('Test accuracy:', score[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "datagen = ImageDataGenerator(\n",
    "    width_shift_range=0.2,\n",
    "    height_shift_range=0.2,\n",
    "    horizontal_flip=True,\n",
    "    samplewise_center=False,\n",
    "    samplewise_std_normalization=False,\n",
    "    zca_whitening=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/30\n",
      "10000/10000 [==============================] - 0s 35us/sample - loss: 1.4806 - acc: 0.5042\n",
      "400/400 [==============================] - 15s 37ms/step - loss: 1.4519 - acc: 0.4817 - val_loss: 1.4806 - val_acc: 0.5042\n",
      "Epoch 2/30\n",
      "10000/10000 [==============================] - 0s 35us/sample - loss: 1.2984 - acc: 0.5482\n",
      "400/400 [==============================] - 15s 37ms/step - loss: 1.3985 - acc: 0.5006 - val_loss: 1.2984 - val_acc: 0.5482\n",
      "Epoch 3/30\n",
      "10000/10000 [==============================] - 0s 35us/sample - loss: 1.2492 - acc: 0.5650\n",
      "400/400 [==============================] - 15s 38ms/step - loss: 1.3709 - acc: 0.5106 - val_loss: 1.2492 - val_acc: 0.5650\n",
      "Epoch 4/30\n",
      "10000/10000 [==============================] - 0s 41us/sample - loss: 1.1685 - acc: 0.58570s - loss: 1.1739 - acc:\n",
      "400/400 [==============================] - 17s 42ms/step - loss: 1.3494 - acc: 0.5164 - val_loss: 1.1685 - val_acc: 0.5857\n",
      "Epoch 5/30\n",
      "10000/10000 [==============================] - 0s 36us/sample - loss: 1.1648 - acc: 0.5920\n",
      "400/400 [==============================] - 17s 44ms/step - loss: 1.3181 - acc: 0.5283 - val_loss: 1.1648 - val_acc: 0.5920\n",
      "Epoch 6/30\n",
      "10000/10000 [==============================] - 0s 42us/sample - loss: 1.2320 - acc: 0.5754\n",
      "400/400 [==============================] - 17s 42ms/step - loss: 1.3113 - acc: 0.5320 - val_loss: 1.2320 - val_acc: 0.5754\n",
      "Epoch 7/30\n",
      "10000/10000 [==============================] - 0s 37us/sample - loss: 1.1345 - acc: 0.5984 loss: 1.2973 - acc\n",
      "400/400 [==============================] - 22s 54ms/step - loss: 1.2971 - acc: 0.5334 - val_loss: 1.1345 - val_acc: 0.5984\n",
      "Epoch 8/30\n",
      "10000/10000 [==============================] - 0s 40us/sample - loss: 1.2043 - acc: 0.5771 loss: 1.28 - ETA: 0s - loss: 1.2821 - ac\n",
      "400/400 [==============================] - 18s 46ms/step - loss: 1.2812 - acc: 0.5444 - val_loss: 1.2043 - val_acc: 0.5771\n",
      "Epoch 9/30\n",
      "10000/10000 [==============================] - 0s 38us/sample - loss: 1.2100 - acc: 0.5770\n",
      "400/400 [==============================] - 19s 47ms/step - loss: 1.2707 - acc: 0.5466 - val_loss: 1.2100 - val_acc: 0.5770\n",
      "Epoch 10/30\n",
      "10000/10000 [==============================] - 0s 38us/sample - loss: 1.2420 - acc: 0.5705\n",
      "400/400 [==============================] - 18s 46ms/step - loss: 1.2544 - acc: 0.5525 - val_loss: 1.2420 - val_acc: 0.5705\n",
      "Epoch 11/30\n",
      "10000/10000 [==============================] - 0s 38us/sample - loss: 1.1138 - acc: 0.6078\n",
      "400/400 [==============================] - 18s 46ms/step - loss: 1.2429 - acc: 0.5588 - val_loss: 1.1138 - val_acc: 0.6078\n",
      "Epoch 12/30\n",
      "10000/10000 [==============================] - 0s 39us/sample - loss: 1.1634 - acc: 0.5899\n",
      "400/400 [==============================] - 19s 49ms/step - loss: 1.2326 - acc: 0.5559 - val_loss: 1.1634 - val_acc: 0.5899\n",
      "Epoch 13/30\n",
      "10000/10000 [==============================] - 0s 44us/sample - loss: 1.1524 - acc: 0.5943\n",
      "400/400 [==============================] - 19s 49ms/step - loss: 1.2255 - acc: 0.5625 - val_loss: 1.1524 - val_acc: 0.5943\n",
      "Epoch 14/30\n",
      "10000/10000 [==============================] - 1s 55us/sample - loss: 1.0895 - acc: 0.6141\n",
      "400/400 [==============================] - 26s 64ms/step - loss: 1.2218 - acc: 0.5642 - val_loss: 1.0895 - val_acc: 0.6141\n",
      "Epoch 15/30\n",
      "10000/10000 [==============================] - 0s 50us/sample - loss: 1.1889 - acc: 0.5826 loss: 1.2\n",
      "400/400 [==============================] - 27s 67ms/step - loss: 1.2103 - acc: 0.5681 - val_loss: 1.1889 - val_acc: 0.5826\n",
      "Epoch 16/30\n",
      "10000/10000 [==============================] - 1s 62us/sample - loss: 1.0771 - acc: 0.6256\n",
      "400/400 [==============================] - 28s 71ms/step - loss: 1.2021 - acc: 0.5692 - val_loss: 1.0771 - val_acc: 0.6256\n",
      "Epoch 17/30\n",
      "10000/10000 [==============================] - 0s 44us/sample - loss: 1.1430 - acc: 0.6006\n",
      "400/400 [==============================] - 27s 69ms/step - loss: 1.1889 - acc: 0.5776 - val_loss: 1.1430 - val_acc: 0.6006\n",
      "Epoch 18/30\n",
      "10000/10000 [==============================] - 1s 62us/sample - loss: 1.1211 - acc: 0.6103\n",
      "400/400 [==============================] - 24s 59ms/step - loss: 1.1873 - acc: 0.5750 - val_loss: 1.1211 - val_acc: 0.6103\n",
      "Epoch 19/30\n",
      "10000/10000 [==============================] - 0s 46us/sample - loss: 1.1069 - acc: 0.6164\n",
      "400/400 [==============================] - 23s 58ms/step - loss: 1.1793 - acc: 0.5813 - val_loss: 1.1069 - val_acc: 0.6164\n",
      "Epoch 20/30\n",
      "10000/10000 [==============================] - 1s 53us/sample - loss: 1.1100 - acc: 0.6156\n",
      "400/400 [==============================] - 28s 69ms/step - loss: 1.1773 - acc: 0.5804 - val_loss: 1.1100 - val_acc: 0.6156\n",
      "Epoch 21/30\n",
      "10000/10000 [==============================] - 0s 50us/sample - loss: 1.1448 - acc: 0.6025\n",
      "400/400 [==============================] - 25s 63ms/step - loss: 1.1778 - acc: 0.5794 - val_loss: 1.1448 - val_acc: 0.6025\n",
      "Epoch 22/30\n",
      "10000/10000 [==============================] - 1s 89us/sample - loss: 1.1354 - acc: 0.6019\n",
      "400/400 [==============================] - 28s 70ms/step - loss: 1.1605 - acc: 0.5852 - val_loss: 1.1354 - val_acc: 0.6019\n",
      "Epoch 23/30\n",
      "10000/10000 [==============================] - 0s 50us/sample - loss: 1.0567 - acc: 0.6293\n",
      "400/400 [==============================] - 26s 66ms/step - loss: 1.1601 - acc: 0.5880 - val_loss: 1.0567 - val_acc: 0.6293\n",
      "Epoch 24/30\n",
      "10000/10000 [==============================] - 0s 49us/sample - loss: 1.0726 - acc: 0.6276\n",
      "400/400 [==============================] - 26s 66ms/step - loss: 1.1638 - acc: 0.5875 - val_loss: 1.0726 - val_acc: 0.6276\n",
      "Epoch 25/30\n",
      "10000/10000 [==============================] - 1s 51us/sample - loss: 1.0157 - acc: 0.6412\n",
      "400/400 [==============================] - 26s 65ms/step - loss: 1.1568 - acc: 0.5900 - val_loss: 1.0157 - val_acc: 0.6412\n",
      "Epoch 26/30\n",
      "10000/10000 [==============================] - 0s 48us/sample - loss: 1.1432 - acc: 0.6066\n",
      "400/400 [==============================] - 26s 65ms/step - loss: 1.1585 - acc: 0.5885 - val_loss: 1.1432 - val_acc: 0.6066\n",
      "Epoch 27/30\n",
      "10000/10000 [==============================] - 1s 51us/sample - loss: 1.0160 - acc: 0.6461\n",
      "400/400 [==============================] - 27s 67ms/step - loss: 1.1472 - acc: 0.5943 - val_loss: 1.0160 - val_acc: 0.6461\n",
      "Epoch 28/30\n",
      "10000/10000 [==============================] - 1s 52us/sample - loss: 1.0666 - acc: 0.6337\n",
      "400/400 [==============================] - 28s 70ms/step - loss: 1.1427 - acc: 0.5921 - val_loss: 1.0666 - val_acc: 0.6337\n",
      "Epoch 29/30\n",
      "10000/10000 [==============================] - 1s 55us/sample - loss: 1.0732 - acc: 0.6247\n",
      "400/400 [==============================] - 27s 68ms/step - loss: 1.1343 - acc: 0.5961 - val_loss: 1.0732 - val_acc: 0.6247\n",
      "Epoch 30/30\n",
      "10000/10000 [==============================] - 1s 58us/sample - loss: 1.0155 - acc: 0.6433\n",
      "400/400 [==============================] - 27s 69ms/step - loss: 1.1255 - acc: 0.6029 - val_loss: 1.0155 - val_acc: 0.6433\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x21587b5d7f0>"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit_generator(datagen.flow(x_train, y_train, batch_size=100),\n",
    "                    steps_per_epoch=x_train.shape[0] // 100, epochs=30,\n",
    "                    validation_data=(x_valid, y_valid))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test loss with data augmentation: 1.0099852146148682\n",
      "Test accuracy with data augmentation: 0.6444\n"
     ]
    }
   ],
   "source": [
    "score = model.evaluate(x_test, y_test, verbose=0)\n",
    "print('Test loss with data augmentation:', score[0])\n",
    "print('Test accuracy with data augmentation:', score[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}