MNIST_CNN_99.5%.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib.image as mpimg\n",
    "import seaborn as sns\n",
    "%matplotlib inline\n",
    "\n",
    "np.random.seed(2)\n",
    "\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import confusion_matrix\n",
    "import itertools\n",
    "\n",
    "# 转换为独热编码\n",
    "from keras.utils.np_utils import to_categorical\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPool2D, BatchNormalization\n",
    "from keras.optimizers import RMSprop\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "from keras.callbacks import ReduceLROnPlateau\n",
    "\n",
    "# 设置显示样式的参数\n",
    "# http://seaborn.pydata.org/generated/seaborn.set.html\n",
    "sns.set(style='white', context='notebook', palette='deep')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# import warnings\n",
    "# warnings.filterwarnings('ignore')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 加载数据\n",
    "train = pd.read_csv(\"./train.csv\")\n",
    "test = pd.read_csv(\"./test.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(42000,)\n",
      "(42000, 784)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "1    4684\n",
       "7    4401\n",
       "3    4351\n",
       "9    4188\n",
       "2    4177\n",
       "6    4137\n",
       "0    4132\n",
       "4    4072\n",
       "8    4063\n",
       "5    3795\n",
       "Name: label, dtype: int64"
      ]
     },
     "execution_count": 72,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYgAAAEFCAYAAAD5bXAgAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ngg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x24c951d3dd8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 'label'\n",
    "Y_train = train[\"label\"]\n",
    "print(Y_train.shape)\n",
    "\n",
    "# 删除 'label' 列\n",
    "X_train = train.drop(labels = [\"label\"],axis = 1) \n",
    "print(X_train.shape)\n",
    "\n",
    "# 释放一些空间\n",
    "del train\n",
    "\n",
    "# 使用条形图显示每个分类数据集合中的观测值\n",
    "# https://seaborn.pydata.org/generated/seaborn.countplot.html?highlight=countplot\n",
    "g = sns.countplot(Y_train)\n",
    "\n",
    "# 对训练集中的元素计数\n",
    "# https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.Series.value_counts.html\n",
    "Y_train.value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count       784\n",
       "unique        1\n",
       "top       False\n",
       "freq        784\n",
       "dtype: object"
      ]
     },
     "execution_count": 73,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 检查数据\n",
    "X_train.isnull().any().describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count       784\n",
       "unique        1\n",
       "top       False\n",
       "freq        784\n",
       "dtype: object"
      ]
     },
     "execution_count": 74,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test.isnull().any().describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 对数据进行归一化，到[0, 1]范围内，减小光照的影响，并可加速CNN收敛速度\n",
    "X_train = X_train / 255.0\n",
    "test = test / 255.0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Reshape三维图像(height = 28px, width = 28px , canal = 1)\n",
    "X_train = X_train.values.reshape(-1,28,28,1)\n",
    "test = test.values.reshape(-1,28,28,1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 将标签编码为一个独热向量 (例如: 2 -> [0,0,1,0,0,0,0,0,0,0])\n",
    "Y_train = to_categorical(Y_train, num_classes = 10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 设置随机种子\n",
    "random_seed = 2\n",
    "\n",
    "# 分割出训练集和验证集\n",
    "X_train, X_val, Y_train, Y_val = train_test_split(X_train, Y_train, test_size = 0.1, random_state=random_seed)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPkAAAD3CAYAAADfRfLgAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nGEfIAeMIOWAcIQeM+z9cEMcG4LbjQgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x24ca9d7f0f0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 一些例子\n",
    "g = plt.imshow(X_train[0][:,:,0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_5\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_18 (Conv2D)           (None, 28, 28, 32)        832       \n",
      "_________________________________________________________________\n",
      "batch_normalization_20 (Batc (None, 28, 28, 32)        128       \n",
      "_________________________________________________________________\n",
      "conv2d_19 (Conv2D)           (None, 28, 28, 32)        25632     \n",
      "_________________________________________________________________\n",
      "batch_normalization_21 (Batc (None, 28, 28, 32)        128       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_3 (MaxPooling2 (None, 14, 14, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_20 (Conv2D)           (None, 14, 14, 64)        18496     \n",
      "_________________________________________________________________\n",
      "batch_normalization_22 (Batc (None, 14, 14, 64)        256       \n",
      "_________________________________________________________________\n",
      "conv2d_21 (Conv2D)           (None, 14, 14, 64)        36928     \n",
      "_________________________________________________________________\n",
      "batch_normalization_23 (Batc (None, 14, 14, 64)        256       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_4 (MaxPooling2 (None, 7, 7, 64)          0         \n",
      "_________________________________________________________________\n",
      "flatten_4 (Flatten)          (None, 3136)              0         \n",
      "_________________________________________________________________\n",
      "dense_7 (Dense)              (None, 256)               803072    \n",
      "_________________________________________________________________\n",
      "dropout_10 (Dropout)         (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_8 (Dense)              (None, 10)                2570      \n",
      "=================================================================\n",
      "Total params: 888,298\n",
      "Trainable params: 887,914\n",
      "Non-trainable params: 384\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# 设置CNN模型\n",
    "model = Sequential()\n",
    "\n",
    "model.add(Conv2D(filters = 32, kernel_size = (5,5),padding = 'Same', \n",
    "                 activation ='relu', input_shape = (28,28,1)))\n",
    "model.add(BatchNormalization())\n",
    "model.add(Conv2D(filters = 32, kernel_size = (5,5),padding = 'Same', \n",
    "                 activation ='relu'))\n",
    "model.add(BatchNormalization())\n",
    "model.add(MaxPool2D(pool_size=(2,2)))\n",
    "# model.add(Dropout(0.25))\n",
    "\n",
    "\n",
    "model.add(Conv2D(filters = 64, kernel_size = (3,3),padding = 'Same', \n",
    "                 activation ='relu'))\n",
    "model.add(BatchNormalization())\n",
    "model.add(Conv2D(filters = 64, kernel_size = (3,3),padding = 'Same', \n",
    "                 activation ='relu'))\n",
    "model.add(BatchNormalization())\n",
    "model.add(MaxPool2D(pool_size=(2,2), strides=(2,2)))\n",
    "# model.add(Dropout(0.25))\n",
    "\n",
    "\n",
    "model.add(Flatten())\n",
    "model.add(Dense(256, activation = \"relu\"))\n",
    "model.add(Dropout(0.5))\n",
    "model.add(Dense(10, activation = \"softmax\"))\n",
    "\n",
    "# 输出模型各层的参数状况\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 用adam优化器和交叉熵损失进行编译\n",
    "model.compile(optimizer=\"adam\", loss=\"categorical_crossentropy\", metrics=[\"accuracy\"])\n",
    "\n",
    "# 用sgd优化器\n",
    "# model.compile(optimizer=\"sgd\", loss=\"categorical_crossentropy\", metrics=[\"accuracy\"])\n",
    "\n",
    "# 定义优化器\n",
    "# optimizer = RMSprop(lr=0.001, rho=0.9, epsilon=1e-08, decay=0.0)\n",
    "# 编译模型\n",
    "# model.compile(optimizer = optimizer , loss = \"categorical_crossentropy\", metrics=[\"accuracy\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 设置一个学习率衰减\n",
    "learning_rate_reduction = ReduceLROnPlateau(monitor='val_acc', \n",
    "                                            patience=3, \n",
    "                                            verbose=1, \n",
    "                                            factor=0.5, \n",
    "                                            min_lr=0.00001)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 训练轮数，暂时设置为1，可以自己尝试调整\n",
    "epochs = 30\n",
    "# epochs = 30 时可以99.5%\n",
    "# 批大小\n",
    "batch_size = 86"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 增加数据以防止过拟合\n",
    "datagen = ImageDataGenerator(\n",
    "        featurewise_center=False,  \t\t\t\t# 在数据集上将输入平均值设置为0\n",
    "        samplewise_center=False,  \t\t\t\t# 将每个样本的平均值设置为0\n",
    "        featurewise_std_normalization=False,    # 将输入除以数据集的std\n",
    "        samplewise_std_normalization=False,  \t# 将每个输入除以它的std\n",
    "        zca_whitening=False,  \t\t\t\t\t# 使用ZCA白化\n",
    "        rotation_range=10,  \t\t\t\t\t# 在范围内随机旋转图像（0到180度）\n",
    "        zoom_range = 0.1, \t\t\t\t\t\t# 随机缩放图像\n",
    "        width_shift_range=0.1,  \t\t\t\t# 水平随机移动图像（总宽度的一部分）\n",
    "        height_shift_range=0.1,  \t\t\t\t# 垂直随机移动图像（总高度的一部分）\n",
    "        horizontal_flip=False,  \t\t\t\t# 随机翻转图像\n",
    "        vertical_flip=False)  \t\t\t\t\t# 随机翻转图像\n",
    "\n",
    "\n",
    "datagen.fit(X_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/30\n",
      " - 378s - loss: 0.3523 - accuracy: 0.8947 - val_loss: 0.6066 - val_accuracy: 0.8069\n",
      "Epoch 2/30\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Administrat\\Anaconda3\\lib\\site-packages\\keras\\callbacks\\callbacks.py:1042: RuntimeWarning: Reduce LR on plateau conditioned on metric `val_acc` which is not available. Available metrics are: val_loss,val_accuracy,loss,accuracy,lr\n",
      "  (self.monitor, ','.join(list(logs.keys()))), RuntimeWarning\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " - 385s - loss: 0.1344 - accuracy: 0.9612 - val_loss: 0.0394 - val_accuracy: 0.9876\n",
      "Epoch 3/30\n",
      " - 402s - loss: 0.1081 - accuracy: 0.9682 - val_loss: 0.0518 - val_accuracy: 0.9860\n",
      "Epoch 4/30\n",
      " - 405s - loss: 0.0933 - accuracy: 0.9729 - val_loss: 0.0393 - val_accuracy: 0.9871\n",
      "Epoch 5/30\n",
      " - 372s - loss: 0.0811 - accuracy: 0.9769 - val_loss: 0.0344 - val_accuracy: 0.9876\n",
      "Epoch 6/30\n",
      " - 360s - loss: 0.0743 - accuracy: 0.9791 - val_loss: 0.0371 - val_accuracy: 0.9888\n",
      "Epoch 7/30\n",
      " - 359s - loss: 0.0677 - accuracy: 0.9818 - val_loss: 0.0271 - val_accuracy: 0.9919\n",
      "Epoch 8/30\n",
      " - 368s - loss: 0.0615 - accuracy: 0.9829 - val_loss: 0.0363 - val_accuracy: 0.9907\n",
      "Epoch 9/30\n",
      " - 374s - loss: 0.0613 - accuracy: 0.9837 - val_loss: 0.0285 - val_accuracy: 0.9912\n",
      "Epoch 10/30\n",
      " - 360s - loss: 0.0540 - accuracy: 0.9853 - val_loss: 0.0336 - val_accuracy: 0.9914\n",
      "Epoch 11/30\n",
      " - 360s - loss: 0.0507 - accuracy: 0.9854 - val_loss: 0.0407 - val_accuracy: 0.9888\n",
      "Epoch 12/30\n",
      " - 358s - loss: 0.0533 - accuracy: 0.9854 - val_loss: 0.0212 - val_accuracy: 0.9945\n",
      "Epoch 13/30\n",
      " - 354s - loss: 0.0490 - accuracy: 0.9867 - val_loss: 0.0274 - val_accuracy: 0.9919\n",
      "Epoch 14/30\n",
      " - 354s - loss: 0.0489 - accuracy: 0.9864 - val_loss: 0.0224 - val_accuracy: 0.9936\n",
      "Epoch 15/30\n",
      " - 354s - loss: 0.0478 - accuracy: 0.9871 - val_loss: 0.0300 - val_accuracy: 0.9921\n",
      "Epoch 16/30\n",
      " - 354s - loss: 0.0390 - accuracy: 0.9894 - val_loss: 0.0338 - val_accuracy: 0.9921\n",
      "Epoch 17/30\n",
      " - 354s - loss: 0.0400 - accuracy: 0.9892 - val_loss: 0.0363 - val_accuracy: 0.9931\n",
      "Epoch 18/30\n",
      " - 354s - loss: 0.0380 - accuracy: 0.9898 - val_loss: 0.0254 - val_accuracy: 0.9940\n",
      "Epoch 19/30\n",
      " - 355s - loss: 0.0410 - accuracy: 0.9890 - val_loss: 0.0331 - val_accuracy: 0.9914\n",
      "Epoch 20/30\n",
      " - 354s - loss: 0.0382 - accuracy: 0.9901 - val_loss: 0.0297 - val_accuracy: 0.9924\n",
      "Epoch 21/30\n",
      " - 354s - loss: 0.0342 - accuracy: 0.9905 - val_loss: 0.0219 - val_accuracy: 0.9917\n",
      "Epoch 22/30\n",
      " - 354s - loss: 0.0347 - accuracy: 0.9909 - val_loss: 0.0266 - val_accuracy: 0.9917\n",
      "Epoch 23/30\n",
      " - 355s - loss: 0.0336 - accuracy: 0.9909 - val_loss: 0.0219 - val_accuracy: 0.9936\n",
      "Epoch 24/30\n",
      " - 355s - loss: 0.0303 - accuracy: 0.9917 - val_loss: 0.0172 - val_accuracy: 0.9943\n",
      "Epoch 25/30\n",
      " - 355s - loss: 0.0341 - accuracy: 0.9910 - val_loss: 0.0259 - val_accuracy: 0.9938\n",
      "Epoch 26/30\n",
      " - 357s - loss: 0.0296 - accuracy: 0.9915 - val_loss: 0.0236 - val_accuracy: 0.9943\n",
      "Epoch 27/30\n",
      " - 357s - loss: 0.0332 - accuracy: 0.9914 - val_loss: 0.0336 - val_accuracy: 0.9912\n",
      "Epoch 28/30\n",
      " - 358s - loss: 0.0285 - accuracy: 0.9922 - val_loss: 0.0366 - val_accuracy: 0.9931\n",
      "Epoch 29/30\n",
      " - 358s - loss: 0.0311 - accuracy: 0.9919 - val_loss: 0.0252 - val_accuracy: 0.9929\n",
      "Epoch 30/30\n",
      " - 358s - loss: 0.0273 - accuracy: 0.9923 - val_loss: 0.0175 - val_accuracy: 0.9955\n"
     ]
    }
   ],
   "source": [
    "# 拟合模型\n",
    "history = model.fit_generator(datagen.flow(X_train,Y_train, batch_size=batch_size),\n",
    "                              epochs = epochs, validation_data = (X_val,Y_val),\n",
    "                              verbose = 2, steps_per_epoch=X_train.shape[0] // batch_size,\n",
    "                              callbacks=[learning_rate_reduction])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXIAAAD3CAYAAAAALt/WAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nhM5JkAshhM5JkAshhM5JkAshhM79f1fnzQvhpH2JAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x24cd1566eb8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 绘制训练和验证的损失和精度曲线\n",
    "fig, ax = plt.subplots(2,1)\n",
    "ax[0].plot(history.history['loss'], color='b', label=\"Training loss\")\n",
    "ax[0].plot(history.history['val_loss'], color='r', label=\"validation loss\",axes =ax[0])\n",
    "legend = ax[0].legend(loc='best', shadow=True)\n",
    "\n",
    "ax[1].plot(history.history['accuracy'], color='b', label=\"Training accuracy\")\n",
    "ax[1].plot(history.history['val_accuracy'], color='r',label=\"Validation accuracy\")\n",
    "legend = ax[1].legend(loc='best', shadow=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUcAAAEiCAYAAABqcBCCAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nrkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x24cd2008e48>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 看混淆矩阵\n",
    "def plot_confusion_matrix(cm, classes,\n",
    "                          normalize=False,\n",
    "                          title='Confusion matrix',\n",
    "                          cmap=plt.cm.Blues):\n",
    "    \"\"\"\n",
    "    此函数打印并绘制混淆矩阵\n",
    "\t可以通过设置 “normalize=true” 应用归一化\n",
    "    \"\"\"\n",
    "    plt.imshow(cm, interpolation='nearest', cmap=cmap)\n",
    "    plt.title(title)\n",
    "    plt.colorbar()\n",
    "    tick_marks = np.arange(len(classes))\n",
    "    plt.xticks(tick_marks, classes, rotation=45)\n",
    "    plt.yticks(tick_marks, classes)\n",
    "\n",
    "    if normalize:\n",
    "        cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n",
    "\n",
    "    thresh = cm.max() / 2.\n",
    "    for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):\n",
    "        plt.text(j, i, cm[i, j],\n",
    "                 horizontalalignment=\"center\",\n",
    "                 color=\"white\" if cm[i, j] > thresh else \"black\")\n",
    "\n",
    "    plt.tight_layout()\n",
    "    plt.ylabel('True label')\n",
    "    plt.xlabel('Predicted label')\n",
    "\n",
    "# 从验证数据集中预测值\n",
    "Y_pred = model.predict(X_val)\n",
    "# 将预测类转换为一个独热向量\n",
    "Y_pred_classes = np.argmax(Y_pred,axis = 1) \n",
    "# 将验证观测转换为一个独热向量\n",
    "Y_true = np.argmax(Y_val,axis = 1) \n",
    "# 计算混淆矩阵\n",
    "confusion_mtx = confusion_matrix(Y_true, Y_pred_classes) \n",
    "# 绘制混淆矩阵\n",
    "plot_confusion_matrix(confusion_mtx, classes = range(10)) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAW8AAAERCAYAAAC0FCalAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nIYSIEA3ehBAiQv8PYWoalyCABvgAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x24ca9d027b8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 显示一些错误结果\n",
    "\n",
    "# 错误是预测标签和真实标签之间的区别\n",
    "errors = (Y_pred_classes - Y_true != 0)\n",
    "\n",
    "Y_pred_classes_errors = Y_pred_classes[errors]\n",
    "Y_pred_errors = Y_pred[errors]\n",
    "Y_true_errors = Y_true[errors]\n",
    "X_val_errors = X_val[errors]\n",
    "\n",
    "def display_errors(errors_index,img_errors,pred_errors, obs_errors):\n",
    "    \"\"\"\n",
    "    此函数显示6个图像及其预测和实际标签\n",
    "    \"\"\"\n",
    "    n = 0\n",
    "    nrows = 2\n",
    "    ncols = 3\n",
    "    fig, ax = plt.subplots(nrows,ncols,sharex=True,sharey=True)\n",
    "    for row in range(nrows):\n",
    "        for col in range(ncols):\n",
    "            error = errors_index[n]\n",
    "            ax[row,col].imshow((img_errors[error]).reshape((28,28)))\n",
    "            ax[row,col].set_title(\"Predicted label :{}\\nTrue label :{}\".format(pred_errors[error],obs_errors[error]))\n",
    "            n += 1\n",
    "\n",
    "# 错误预测数的概率\n",
    "Y_pred_errors_prob = np.max(Y_pred_errors,axis = 1)\n",
    "\n",
    "# 误差集中真值的预测概率\n",
    "true_prob_errors = np.diagonal(np.take(Y_pred_errors, Y_true_errors, axis=1))\n",
    "\n",
    "# 预测标签概率与真实标签概率之差\n",
    "delta_pred_true_errors = Y_pred_errors_prob - true_prob_errors\n",
    "\n",
    "# 对预测标签概率与真实标签概率之差的列表进行排序\n",
    "sorted_dela_errors = np.argsort(delta_pred_true_errors)\n",
    "\n",
    "# Top 6错误\n",
    "most_important_errors = sorted_dela_errors[-6:]\n",
    "\n",
    "# 展示Top 6错误\n",
    "display_errors(most_important_errors, X_val_errors, Y_pred_classes_errors, Y_true_errors)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 预测结果\n",
    "results = model.predict(test)\n",
    "\n",
    "# 选择最大概率的整数\n",
    "results = np.argmax(results,axis = 1)\n",
    "results = pd.Series(results,name=\"Label\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "submission = pd.concat([pd.Series(range(1,28001),name = \"ImageId\"),results],axis = 1)\n",
    "\n",
    "# 转换成CSV格式，不保留索引\n",
    "submission.to_csv(\"cnn_mnist_datagen.csv\",index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "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": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}