识别手写数字.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 查看keras版本"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "'2.0.8'"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import keras\n",
    "keras.__version__"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 加载keras中的mnist训练集"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.datasets import mnist\n",
    "(train_images, train_labels), (test_images, test_labels) = mnist.load_data()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 查看训练集"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(60000, 28, 28)\n",
      "60000\n",
      "[5 0 4 ... 5 6 8]\n"
     ]
    }
   ],
   "source": [
    "print(train_images.shape)\n",
    "print(len(train_labels))\n",
    "print(train_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(10000, 28, 28)\n",
      "10000\n",
      "[7 2 1 ... 4 5 6]\n"
     ]
    }
   ],
   "source": [
    "print(test_images.shape)\n",
    "print(len(test_labels))\n",
    "print(test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 组织网络架构"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "784\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_3 (Dense)              (None, 512)               401920    \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 10)                5130      \n",
      "=================================================================\n",
      "Total params: 407,050\n",
      "Trainable params: 407,050\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "from keras import models\n",
    "from keras import layers\n",
    "\n",
    "network = models.Sequential()\n",
    "network.add(layers.Dense(512, activation='relu', input_shape=(28 * 28,)))\n",
    "network.add(layers.Dense(10, activation='softmax'))\n",
    "network.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 编译\n",
    "选择编译的步骤的三个参数\n",
    "* 损失函数(loss function): 网络如何衡量训练数据上的性能，即网络如何朝着正确的方向前进。\n",
    "* 优化器(optimizer): 基于训练数据和损失函数来更新网络的机制\n",
    "* 在训练和测试过程中需要监控的指标(metric): 本例只关心京都，即正确分类的图像所占的比例|"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "network.compile(optimizer='rmsprop',\n",
    "                loss='categorical_crossentropy',\n",
    "                metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 数据预处理\n",
    "对训练集中的数据进行预处理，将其变换为网络要求的形状，并缩放至所有值都在\\[0, 1\\]区间。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(60000, 784)\n",
      "(10000, 784)\n"
     ]
    }
   ],
   "source": [
    "train_images = train_images.reshape((60000, 28 * 28))\n",
    "train_images = train_images.astype('float32') / 255\n",
    "print(train_images.shape)\n",
    "\n",
    "test_images = test_images.reshape((10000, 28 * 28))\n",
    "test_images = test_images.astype('float32') / 255\n",
    "print(test_images.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 准备标签"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.utils import to_categorical\n",
    "train_labels = to_categorical(train_labels)\n",
    "test_labels = to_categorical(test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 拟合模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/5\n",
      "60000/60000 [==============================] - 3s - loss: 0.2589 - acc: 0.9248     \n",
      "Epoch 2/5\n",
      "60000/60000 [==============================] - 3s - loss: 0.1057 - acc: 0.9682     \n",
      "Epoch 3/5\n",
      "60000/60000 [==============================] - 3s - loss: 0.0695 - acc: 0.9790     \n",
      "Epoch 4/5\n",
      "60000/60000 [==============================] - 3s - loss: 0.0498 - acc: 0.9851     \n",
      "Epoch 5/5\n",
      "60000/60000 [==============================] - 3s - loss: 0.0369 - acc: 0.9886     \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x130a8dac8>"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "network.fit(train_images, train_labels, epochs=5, batch_size=128)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 检查在测试集上的性能"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " 9024/10000 [==========================>...] - ETA: 0stest_acc 0.9806\n"
     ]
    }
   ],
   "source": [
    "test_loss, test_acc = network.evaluate(test_images, test_labels)\n",
    "print('test_acc', test_acc)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 代码"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/5\n",
      "60000/60000 [==============================] - 3s - loss: 0.2540 - acc: 0.9278     \n",
      "Epoch 2/5\n",
      "60000/60000 [==============================] - 3s - loss: 0.1047 - acc: 0.9697     \n",
      "Epoch 3/5\n",
      "60000/60000 [==============================] - 3s - loss: 0.0679 - acc: 0.9797     \n",
      "Epoch 4/5\n",
      "60000/60000 [==============================] - 3s - loss: 0.0493 - acc: 0.9851     \n",
      "Epoch 5/5\n",
      "60000/60000 [==============================] - 3s - loss: 0.0370 - acc: 0.9889     \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x130a01e48>"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from keras.datasets import mnist\n",
    "from keras import models\n",
    "from keras import layers\n",
    "from keras.utils import to_categorical\n",
    "\n",
    "# 拉取数据集\n",
    "(train_images, train_labels), (test_images, test_labels) = mnist.load_data()\n",
    "\n",
    "# 构建网络\n",
    "network = models.Sequential()\n",
    "network.add(layers.Dense(512, activation='relu', input_shape=(28 * 28,)))\n",
    "network.add(layers.Dense(10, activation='softmax'))\n",
    "\n",
    "# 编译\n",
    "network.compile(optimizer='rmsprop',\n",
    "                loss='categorical_crossentropy',\n",
    "                metrics=['accuracy'])\n",
    "\n",
    "# 数据预处理\n",
    "train_images = train_images.reshape((60000, 28 * 28))\n",
    "train_images = train_images.astype('float32') / 255\n",
    "test_images = test_images.reshape((10000, 28 * 28))\n",
    "test_images = test_images.astype('float32') / 255\n",
    "\n",
    "# 准备标签\n",
    "train_labels = to_categorical(train_labels)\n",
    "test_labels = to_categorical(test_labels)\n",
    "\n",
    "# 拟合模型\n",
    "network.fit(train_images, train_labels, epochs=5, batch_size=128)\n",
    "\n",
    "# 检查性能\n",
    "test_loss, test_acc = network.evaluate(test_images, test_labels)\n",
    "print('test_acc', test_acc)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 总结\n",
    "* 神经网络的核心组件是层（layer）。本例中的Dense为全连接层，最后一个Dense是一个10路的softmax层，它将返回一个由10个概率值（总和为1）组成的数组，每个概率值表示10个数字类别中某一个的概率\n",
    "* 编译时需要的参数：损失函数（loss function）、优化器（optimizer）、监控指标（metrics）\n",
    "* 数据要制作成网络要求的形式，本例中数据集要进行缩放，数据值归一到\\[0, 1\\]区间。\n",
    "* 如果检查结果的性能低于训练结果的性能，为过拟合"
   ]
  }
 ],
 "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": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}