手写数字预测的CNN模型搭建_与训练和评估.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "d4c08f48-fe23-4ddb-ac46-d97f05397514",
    "_uuid": "f2156d1dd26a1243e18512002e10872c5bd7271e"
   },
   "source": [
    "* **1. Introduction**\n",
    "* **2. Data preparation**\n",
    "    * 2.1 Load data\n",
    "    * 2.2 Check for null and missing values\n",
    "    * 2.3 Normalization\n",
    "    * 2.4 Reshape\n",
    "    * 2.5 Label encoding\n",
    "    * 2.6 Split training and valdiation set\n",
    "* **3. CNN**\n",
    "\n",
    "* **4. Evaluate the model**\n",
    "\n",
    "* **5. Prediction and submition**\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "eb88b372-a6e5-40c8-a1c6-c03799165490",
    "_uuid": "e9aff3cf1bb8daa73bec67b970d12195677679f3"
   },
   "source": [
    "# 1. Introduction\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "_cell_guid": "f67b9393-8ea1-4e23-b856-2ce149cfe421",
    "_execution_state": "idle",
    "_uuid": "72334cb006d02a4bcfc2a2fe622524eba824c6f8"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "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",
    "\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import confusion_matrix\n",
    "import itertools\n",
    "\n",
    "from keras.utils.np_utils import to_categorical # convert to one-hot-encoding\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPool2D\n",
    "from keras.optimizers import RMSprop\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "from keras.callbacks import ReduceLROnPlateau\n",
    "import keras\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPool2D\n",
    "from keras.layers.normalization import BatchNormalization\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "from keras.callbacks import ReduceLROnPlateau\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "sns.set(style='white', context='notebook', palette='deep')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "6d2fb3e6-ab71-4974-b5a2-4af1ebdb99f4",
    "_execution_state": "idle",
    "_uuid": "86061d98eccaa02efe0dab0fa3884e71fcf4c310"
   },
   "source": [
    "# 2. Data preparation\n",
    "## 2.1 Load data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "_cell_guid": "5e51d00e-62fd-4141-bf73-50ac4f2da7d0",
    "_execution_state": "idle",
    "_uuid": "84bbd5ab8d7895bd430d5ecfe2f7ddf77baa7b74"
   },
   "outputs": [],
   "source": [
    "# >>>>>填写<<<< 利用pandas的load_csv函数，读取我们的train 和 test数据集合 变量已经给出 >>>>>填写<<<< ######\n",
    "train = pd.read_csv(\"D:/AI培训/作业/实验7/subset_train.csv\")\n",
    "test = pd.read_csv(\"D:/AI培训/作业/实验7/Small_test.csv\")\n",
    "#####train validation test(完全独立的，与训练过程无关的)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "_cell_guid": "86570a36-5c20-460a-9dfd-2070548532a7",
    "_execution_state": "idle",
    "_uuid": "1213b979d5ed3e0d13824d17d694c79d2ece92fa"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "3    227\n",
       "7    220\n",
       "2    220\n",
       "1    218\n",
       "6    214\n",
       "9    209\n",
       "4    209\n",
       "5    201\n",
       "0    199\n",
       "8    183\n",
       "Name: label, dtype: int64"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# >>>>>填写<<<< 利用pandas的header选择，将label列传递给Y_train >>>>>填写<<<< \n",
    "Y_train = train[\"label\"]\n",
    "Y_test = test['label']\n",
    "# 因为train.csv中，第一列label在上述代码已经传递给Y_label，这里对于x_train 我们不需要训练集的第一列 #####\n",
    "X_train = train.drop(labels = [\"label\"],axis = 1) \n",
    "X_test = test.drop(labels = [\"label\"],axis = 1)\n",
    "\n",
    "\n",
    "g = sns.countplot(Y_train)\n",
    "\n",
    "Y_train.value_counts()\n",
    "Y_train\n",
    "Y_test.value_counts()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "5aea4062-1790-4987-b739-c4bebd79030f",
    "_uuid": "b7b1b1d36243c885e57374c8b60c5a7e10abe922"
   },
   "source": [
    "## We have similar counts for the 10 digits."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "5b7d4b66-a140-4fcc-a889-bcef007c880a",
    "_uuid": "5d77934302869925c19128c77e247b3c8ca84d71"
   },
   "source": [
    "## 2.2 Check for null and missing values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "_cell_guid": "ececaa00-2ae3-4d13-b631-438df085b030",
    "_execution_state": "idle",
    "_uuid": "cdf27c27e2a5b15e6d7bfc70de7a18c08f3feb7a"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count       784\n",
       "unique        1\n",
       "top       False\n",
       "freq        784\n",
       "dtype: object"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 检查训练数据是否有空值\n",
    "X_train.isnull().any().describe() "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "_cell_guid": "92572e80-8543-4e72-8767-5c9be8381b04",
    "_execution_state": "idle",
    "_uuid": "a0089bb7ec9aec76373db475399aea24699ae989"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count       784\n",
       "unique        1\n",
       "top       False\n",
       "freq        784\n",
       "dtype: object"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# >>>>填写<<<< 检查训练数据是否有空值 >>>>填写<<<< ###\n",
    "X_test.isnull().any().describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "09d04cae-4245-4659-85dd-ef48531da295",
    "_uuid": "c0bee59691c2df0b275c78e38e7f9907d02ac038"
   },
   "source": [
    "I check for corrupted images (missing values inside).\n",
    "\n",
    "There is no missing values in the train and test dataset. So we can safely go ahead."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "6812040d-80ad-43d2-a571-275f4f20067b",
    "_uuid": "2954681f25f0dcbe986e6914396cdbce61db591f"
   },
   "source": [
    "## 2.3 Normalization"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "159d5854-437a-4d0f-bc1e-fc3f7e43d178",
    "_uuid": "0ecf4b52510ab7957d0d4eb646c0aa1ba5986273"
   },
   "source": [
    "We perform a grayscale normalization to reduce the effect of illumination's differences. \n",
    "\n",
    "Moreover the CNN converg faster on [0..1] data than on [0..255].\n",
    "标准化，将灰度值 0-255 映射到0 - 1区间"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "_cell_guid": "cdc4340b-6e24-4e12-be99-ac806098ff17",
    "_execution_state": "idle",
    "_uuid": "b5d4f8fcf2a967e2c7d57daedf95aa8c5ab7f8cb"
   },
   "outputs": [],
   "source": [
    "# Normalize the data\n",
    "X_train = X_train / 255.0\n",
    "###### >>>填写<<< 标准化测试集合 #######\n",
    "X_test = X_test / 255.0"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "7413df94-bcb9-4f75-b174-c127d4445766",
    "_uuid": "a66741bf1ac597094f3a3166877008feef27c519"
   },
   "source": [
    "## 2.3 Reshape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "_cell_guid": "34b6a5f7-8fd2-4387-8ef4-c9dc19584fed",
    "_execution_state": "idle",
    "_uuid": "f0a6ad80dab8e0f2c2e46165ccd9cd82dd162bc3"
   },
   "outputs": [],
   "source": [
    "# >>>>填写<<<<< 利用 reshape 函数， 将X_train变换成 (height = 28px, width = 28px , canal = 1)>>>>填写<<<<< ######\n",
    "X_train = X_train.values.reshape(-1,28,28,1)\n",
    "X_test = X_test.values.reshape(-1,28,28,1) \n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "8decd1ce-7b7e-431d-8458-eaca18e0e1f7",
    "_uuid": "f4fb5553e188d9956f5d8b3a5d275ab00ea667ce"
   },
   "source": [
    "Train and test images (28px x 28px) has been stock into pandas.Dataframe as 1D vectors of 784 values. We reshape all data to 28x28x1 3D matrices. \n",
    "\n",
    "Keras requires an extra dimension in the end which correspond to channels. MNIST images are gray scaled so it use only one channel. For RGB images, there is 3 channels, we would have reshaped 784px vectors to 28x28x3 3D matrices. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "bdb422e2-bdec-444f-97a5-283a1e54bf2c",
    "_uuid": "39b7a31e843bac6b705461bcce89da216b91799e"
   },
   "source": [
    "## 2.5 Label encoding"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "_cell_guid": "4b7f3e78-44dc-4561-b1f0-9429ee024cf4",
    "_execution_state": "idle",
    "_uuid": "cabefd1478d5c1bdfe57fd6a34395340916a854c"
   },
   "outputs": [],
   "source": [
    "# 利用0 1编码 将0-9数字标签编码成10维向量 (ex : 9 -> [0,0,0,0,0,0,0,0,0,1])\n",
    "##\n",
    "Y_train = to_categorical(Y_train, num_classes = 10)\n",
    "Y_test = to_categorical(Y_test, num_classes = 10)\n",
    "## one-hot encoding"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "ae068bd8-b12e-4768-8a7e-0fc865dd7562",
    "_uuid": "dcfb688587dfc6feafd27442a3505e35dc01b82d"
   },
   "source": [
    "Labels are 10 digits numbers from 0 to 9. We need to encode these lables to one hot vectors (ex : 2 -> [0,0,1,0,0,0,0,0,0,0])."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "32152fc3-a570-4d64-8a7d-6c689a4acd33",
    "_uuid": "d8abbbf31483b94e1b29d07c4c8253d1311648a7"
   },
   "source": [
    "## 2.6 Split training and valdiation set "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "_cell_guid": "3a698301-9759-4279-ae48-fd980f89ea53",
    "_execution_state": "idle",
    "_uuid": "6e51c925c6e0f1b936679c9649fef345c853555f"
   },
   "outputs": [],
   "source": [
    "# Set the random seed\n",
    "random_seed = 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "_cell_guid": "dcd25ebb-d845-4d32-9867-082e352b1396",
    "_execution_state": "idle",
    "_uuid": "b779ac76d8317647db92d5a88b4098d212d72884"
   },
   "outputs": [],
   "source": [
    "# 将训练集合按照9:1 分成训练集合 和验证集合 validation 10折交叉验证 10-fold validation  ####\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": "markdown",
   "metadata": {
    "_cell_guid": "adbeacf0-0dc0-4675-b2df-9c9663750f32",
    "_uuid": "60eed15ec5bc0d354385301789ecb8538fc02267"
   },
   "source": [
    "We can get a better sense for one of these examples by visualising the image and looking at the label."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "_cell_guid": "5f76131b-4ba0-45f1-a98c-bd4e7d561793",
    "_execution_state": "idle",
    "_uuid": "e0dae8943d3d35f075dba3d7ba31bde1d4bf2ff4"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Some examples #x-train里面第一个sample的 0:最大 0:最大 0 [:,:,0] \n",
    "g = plt.imshow(X_train[0][:,:,0],cmap='gray') #plt为什么把灰度可以生"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "d5265777-aeb3-449d-b171-d88cad74c0a4",
    "_uuid": "5fa18b37a9acd9e098bac1d12264b0dd4310fdd3"
   },
   "source": [
    "# 3. CNN\n",
    "## 3.1 Define the model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "504fa00e-148c-4364-9b68-218b3aaedfdb",
    "_uuid": "7697570491420f957f6e4d3569d51410b5277250"
   },
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "#####  填写 batch_size epoch 请根据traindata 总量填写合适的值 ####\n",
    "##### 我们的分配数量num_classes,提示 我们的任务是手写体0-9的识别 ####### \n",
    "\n",
    "batch_size = 20\n",
    "num_classes = 10\n",
    "epochs = 20\n",
    "\n",
    "input_shape = (28,28,1)\n",
    "\n",
    "#构建CNN 模型 这里我们利用Sequential 序列累加 ######\n",
    "model = Sequential()\n",
    "## 第一个 卷积层 32个kernel kernel大小3*3 输出的激活函数relu kernel利用 He-正态分布 生成  ####\n",
    "model.add(Conv2D(32, kernel_size=(3, 3),activation='relu',kernel_initializer='he_normal',input_shape=input_shape))\n",
    "\n",
    "###  请自行构建第二个卷积层，此时kernel的初始尝试用全零初始/全1初始/正态初始\n",
    "model.add(Conv2D(32, kernel_size=(3, 3),activation='relu',kernel_initializer='he_normal'))\n",
    "### 构建一个最大池化层 \n",
    "model.add(MaxPool2D((2, 2),strides=2))\n",
    "\n",
    "model.add(Dropout(0.20))\n",
    "###  在下述卷积层内 构建一个padding, 在之后构建一个kernel size = 2 *2 的池化层 \n",
    "model.add(Conv2D(64, (3, 3), activation='relu',padding='same',kernel_initializer='he_normal'))\n",
    "model.add(MaxPool2D(pool_size=(2, 2)))\n",
    "\n",
    "model.add(Dropout(0.25))\n",
    "model.add(Flatten())\n",
    "###  构建一个全联接 其中包含128个神经元 并使用relu激活函数\n",
    "\n",
    "\n",
    "model.add(Dense(128, activation='relu'))\n",
    "model.add(BatchNormalization())\n",
    "\n",
    "model.add(Dropout(0.25))\n",
    "\n",
    "###  构建一个全联接，该全联接需要用特定的激活函数和适当的神经元个数 来实现我们的分类目标  提示：我们有多少个标签？什么激活适合最后的输出？\n",
    "model.add(Dense(num_classes, activation='softmax'))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "_cell_guid": "a4c55409-6a65-400a-b5e8-a1dc535429c0",
    "_execution_state": "idle",
    "_uuid": "420c704367b397b8255fefe9d882b35ac8929b95"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_1 (Conv2D)            (None, 26, 26, 32)        320       \n",
      "_________________________________________________________________\n",
      "conv2d_2 (Conv2D)            (None, 24, 24, 32)        9248      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 12, 12, 32)        0         \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 12, 12, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_3 (Conv2D)            (None, 12, 12, 64)        18496     \n",
      "_________________________________________________________________\n",
      "max_pooling2d_2 (MaxPooling2 (None, 6, 6, 64)          0         \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 6, 6, 64)          0         \n",
      "_________________________________________________________________\n",
      "flatten_1 (Flatten)          (None, 2304)              0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 128)               295040    \n",
      "_________________________________________________________________\n",
      "batch_normalization_1 (Batch (None, 128)               512       \n",
      "_________________________________________________________________\n",
      "dropout_3 (Dropout)          (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 10)                1290      \n",
      "=================================================================\n",
      "Total params: 324,906\n",
      "Trainable params: 324,650\n",
      "Non-trainable params: 256\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "### 运行model.summary（）回答下列问题 第二天课上一起讨论 ####\n",
    "### 能否画出这个模型的概括图?  >>>\n",
    "\n",
    "### 这个模型有几个卷积层？ 3\n",
    "### 这个模型最大的参数量是哪一层？full - connection\n",
    "### 第一层卷积层为什么有320个实际变量需要调节       32 * 9 + 32 * 1 (W,bias)  y=wx+b\n",
    "\n",
    "\n",
    "#最后一层 max——pooling完 有64个6*6 feature maps 64*6*6 = 2304\n",
    "\n",
    "### para = 2304*128 + 128 #128个w,b\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "#优化器  尝试使用不同的优化器 至少以下三种\n",
    "## 中文参考 https://keras.io/zh/optimizers/\n",
    "## \n",
    "## SGD(lr=0.01, momentum=0.0, decay=0.0, nesterov=False)\n",
    "## RMSprop(lr=0.001, rho=0.9, epsilon=None, decay=0.0)\n",
    "## Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=None, decay=0.0, amsgrad=False)\n",
    "\n",
    "sgd=keras.optimizers.SGD(lr=0.01, momentum=0.0, decay=0.0, nesterov=False)\n",
    "rmsprop=keras.optimizers.RMSprop(lr=0.001, rho=0.9, epsilon=None, decay=0.0)\n",
    "adam=keras.optimizers.Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=None, decay=0.0, amsgrad=False)\n",
    "### 将模型compile 编译\n",
    "### 调节loss 参数，即loss function\n",
    "### mean_squared_error\n",
    "### categorical_crossentropy/为什么不用binary_crossentropy\n",
    "### mean_absolute_error \n",
    "model.compile(optimizer ='sgd'  , loss ='mean_squared_error'   , metrics=[\"accuracy\"])\n",
    "\n",
    "### training 过程中的 自动调节函数\n",
    "### Reduce LR On Plateau = 减少学习率，当某一个参数达到一个平台期 自动的 把上面优化器中的 lr 减小\n",
    "\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": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 3780 samples, validate on 420 samples\n",
      "Epoch 1/20\n",
      "3780/3780 [==============================] - 6s 2ms/step - loss: 0.1048 - accuracy: 0.1368 - val_loss: 0.0865 - val_accuracy: 0.2643\n",
      "Epoch 2/20\n",
      "  90/3780 [..............................] - ETA: 6s - loss: 0.0954 - accuracy: 0.2222"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "D:\\anaconda\\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": [
      "3780/3780 [==============================] - 5s 1ms/step - loss: 0.0948 - accuracy: 0.2259 - val_loss: 0.0761 - val_accuracy: 0.3952\n",
      "Epoch 3/20\n",
      "3780/3780 [==============================] - 5s 1ms/step - loss: 0.0857 - accuracy: 0.3111 - val_loss: 0.0665 - val_accuracy: 0.4929\n",
      "Epoch 4/20\n",
      "3780/3780 [==============================] - 5s 1ms/step - loss: 0.0792 - accuracy: 0.3653 - val_loss: 0.0603 - val_accuracy: 0.5405\n",
      "Epoch 5/20\n",
      "3780/3780 [==============================] - 5s 1ms/step - loss: 0.0728 - accuracy: 0.4265 - val_loss: 0.0541 - val_accuracy: 0.5976\n",
      "Epoch 6/20\n",
      "3780/3780 [==============================] - 5s 1ms/step - loss: 0.0675 - accuracy: 0.4693 - val_loss: 0.0500 - val_accuracy: 0.6286\n",
      "Epoch 7/20\n",
      "3780/3780 [==============================] - 5s 1ms/step - loss: 0.0634 - accuracy: 0.5156 - val_loss: 0.0468 - val_accuracy: 0.6524\n",
      "Epoch 8/20\n",
      "3780/3780 [==============================] - 5s 1ms/step - loss: 0.0593 - accuracy: 0.5481 - val_loss: 0.0435 - val_accuracy: 0.6833\n",
      "Epoch 9/20\n",
      "3780/3780 [==============================] - 5s 1ms/step - loss: 0.0559 - accuracy: 0.5802 - val_loss: 0.0407 - val_accuracy: 0.6929\n",
      "Epoch 10/20\n",
      "3780/3780 [==============================] - 5s 1ms/step - loss: 0.0534 - accuracy: 0.5987 - val_loss: 0.0382 - val_accuracy: 0.7310\n",
      "Epoch 11/20\n",
      "3780/3780 [==============================] - 6s 1ms/step - loss: 0.0506 - accuracy: 0.6257 - val_loss: 0.0366 - val_accuracy: 0.7381\n",
      "Epoch 12/20\n",
      "3780/3780 [==============================] - 5s 1ms/step - loss: 0.0483 - accuracy: 0.6442 - val_loss: 0.0342 - val_accuracy: 0.7571\n",
      "Epoch 13/20\n",
      "3780/3780 [==============================] - 5s 1ms/step - loss: 0.0470 - accuracy: 0.6471 - val_loss: 0.0326 - val_accuracy: 0.7690\n",
      "Epoch 14/20\n",
      "3780/3780 [==============================] - 5s 1ms/step - loss: 0.0446 - accuracy: 0.6743 - val_loss: 0.0313 - val_accuracy: 0.7810\n",
      "Epoch 15/20\n",
      "3780/3780 [==============================] - 5s 1ms/step - loss: 0.0416 - accuracy: 0.6976 - val_loss: 0.0298 - val_accuracy: 0.7976\n",
      "Epoch 16/20\n",
      "3780/3780 [==============================] - 5s 1ms/step - loss: 0.0413 - accuracy: 0.7032 - val_loss: 0.0285 - val_accuracy: 0.8048\n",
      "Epoch 17/20\n",
      "3780/3780 [==============================] - 6s 1ms/step - loss: 0.0393 - accuracy: 0.7175 - val_loss: 0.0276 - val_accuracy: 0.8119\n",
      "Epoch 18/20\n",
      "3780/3780 [==============================] - 6s 2ms/step - loss: 0.0383 - accuracy: 0.7241 - val_loss: 0.0265 - val_accuracy: 0.8238\n",
      "Epoch 19/20\n",
      "3780/3780 [==============================] - 5s 1ms/step - loss: 0.0371 - accuracy: 0.7339 - val_loss: 0.0256 - val_accuracy: 0.8238\n",
      "Epoch 20/20\n",
      "3780/3780 [==============================] - 6s 1ms/step - loss: 0.0361 - accuracy: 0.7421 - val_loss: 0.0248 - val_accuracy: 0.8381\n"
     ]
    }
   ],
   "source": [
    "# 训练模型 注意填写缺失参数 记得用变量形式填写\n",
    "history = model.fit(X_train,Y_train, batch_size=30,\n",
    "                              epochs =20 , validation_data = (X_val,Y_val),callbacks=[learning_rate_reduction])\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 生成学习曲线 和损失函数 随着epoch的变化曲线\n",
    "# 模型的学习效果怎么样？ 能找到适合的epoch吗？\n",
    "# 简单的评价标准应该用什么？\n",
    "# 尝试改变模型参数 生成不同的学习曲线 比较\n",
    "# 提示 从epoch>优化器>损失函数>学习率>dropout有无 依次调试 \n",
    "fig, ax = plt.subplots(2,1)\n",
    "\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": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 生成10标签混淆矩阵\n",
    "\n",
    "def plot_confusion_matrix(cm, classes,\n",
    "                          normalize=False,\n",
    "                          title='Confusion matrix',\n",
    "                          cmap=plt.cm.Blues):\n",
    "    \"\"\"\n",
    "    This function prints and plots the confusion matrix.\n",
    "    Normalization can be applied by setting `normalize=True`.\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",
    "# Predict the values from the validation dataset\n",
    "Y_pred = model.predict(X_val)\n",
    "# Convert predictions classes to one hot vectors \n",
    "Y_pred_classes = np.argmax(Y_pred,axis = 1) \n",
    "# Convert validation observations to one hot vectors\n",
    "Y_true = np.argmax(Y_val,axis = 1) \n",
    "# compute the confusion matrix\n",
    "confusion_mtx = confusion_matrix(Y_true, Y_pred_classes) \n",
    "# plot the confusion matrix\n",
    "plot_confusion_matrix(confusion_mtx, classes = range(10)) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 9 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "### 打印出认错的数字\n",
    "\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",
    "    \"\"\" This function shows 6 images with their predicted and real labels\"\"\"\n",
    "    n = 0\n",
    "    nrows = 3\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",
    "# Probabilities of the wrong predicted numbers\n",
    "Y_pred_errors_prob = np.max(Y_pred_errors,axis = 1)\n",
    "\n",
    "# Predicted probabilities of the true values in the error set\n",
    "true_prob_errors = np.diagonal(np.take(Y_pred_errors, Y_true_errors, axis=1))\n",
    "\n",
    "# Difference between the probability of the predicted label and the true label\n",
    "delta_pred_true_errors = Y_pred_errors_prob - true_prob_errors\n",
    "\n",
    "# Sorted list of the delta prob errors\n",
    "sorted_dela_errors = np.argsort(delta_pred_true_errors)\n",
    "\n",
    "# Top 9 errors \n",
    "most_important_errors = sorted_dela_errors[-9:]\n",
    "\n",
    "# Show the top 9 errors\n",
    "display_errors(most_important_errors, X_val_errors, Y_pred_classes_errors, Y_true_errors)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "ename": "SyntaxError",
     "evalue": "invalid syntax (<ipython-input-19-46583a9aa1bf>, line 8)",
     "output_type": "error",
     "traceback": [
      "\u001b[1;36m  File \u001b[1;32m\"<ipython-input-19-46583a9aa1bf>\"\u001b[1;36m, line \u001b[1;32m8\u001b[0m\n\u001b[1;33m    fpr[i], tpr[i], _ =\u001b[0m\n\u001b[1;37m                        ^\u001b[0m\n\u001b[1;31mSyntaxError\u001b[0m\u001b[1;31m:\u001b[0m invalid syntax\n"
     ]
    }
   ],
   "source": [
    "#optional 计算每个标签的tpr fpr\n",
    "from sklearn.metrics import roc_curve, auc\n",
    "fpr = dict()\n",
    "tpr = dict()\n",
    "roc_auc = dict()\n",
    "y_score = model.predict(X_test)\n",
    "for i in range(num_classes):\n",
    "     fpr[i], tpr[i], _ = \n",
    "    # AUC Area Under the Curve\n",
    "     roc_auc[i] = \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#画出ROC\n",
    "for i in range(num_classes):\n",
    "    plt.plot()\n",
    "\n",
    "plt.plot([0, 1], [0, 1], 'k--', lw=2)\n",
    "plt.xlim([0.0, 1.0])\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('Some extension of Receiver operating characteristic to multi-class')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.utils import plot_model\n",
    "# 打印出模型图片 尝试自己改动模型 \n",
    "plot_model(model, to_file='model.png', show_shapes=True)"
   ]
  }
 ],
 "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
}