{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting MNIST_data/train-images-idx3-ubyte.gz\n",
      "Extracting MNIST_data/train-labels-idx1-ubyte.gz\n",
      "Extracting MNIST_data/t10k-images-idx3-ubyte.gz\n",
      "Extracting MNIST_data/t10k-labels-idx1-ubyte.gz\n"
     ]
    }
   ],
   "source": [
    "# Download MNIST datasource\n",
    "# 6w个 28 * 28个像素的手写数字图片集\n",
    "# 用 [60000, 784]的张量表示 [图片索引, 图片像素点索引]\n",
    "from tensorflow.examples.tutorials.mnist import input_data\n",
    "\n",
    "# `one-hot vectors`：向量中只有一个数据为 1，其余维度只能为 0\n",
    "# 转化为 [60000, 10]的张量表示 [图片索引, 图片表示的数值]\n",
    "mnist = input_data.read_data_sets(\"MNIST_data/\", one_hot = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 28 * 28 = 784的占位符\n",
    "# None表示可能是任何数值\n",
    "x = tf.placeholder(tf.float32, [None, 784])\n",
    "y = tf.placeholder(tf.float32, [None, 10])\n",
    "z = tf.placeholder(tf.float32)                        # 用于 drop_out操作时的依据 (0.8: 80%的神经元在工作)\n",
    "lr = tf.Variable(0.001, dtype = tf.float32)           # 用于不断递减的学习率，使得梯度下降到最低点时，能更好地命中"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 权重值(截断的随机正太分布) 和 偏置量 (0.1)\n",
    "W1 = tf.Variable(tf.truncated_normal([784, 600], stddev = 0.1))\n",
    "b1 = tf.Variable(tf.zeros([600]) + 0.1)\n",
    "L1 = tf.nn.tanh(tf.matmul(x, W1) + b1)\n",
    "L1_drop = tf.nn.dropout(L1, z)\n",
    "\n",
    "# 隐藏层\n",
    "W2 = tf.Variable(tf.truncated_normal([600, 400], stddev = 0.1))\n",
    "b2 = tf.Variable(tf.zeros([400]) + 0.1)\n",
    "L2 = tf.nn.tanh(tf.matmul(L1_drop, W2) + b2)\n",
    "L2_drop = tf.nn.dropout(L2, z)\n",
    "\n",
    "W3 = tf.Variable(tf.truncated_normal([400, 10], stddev = 0.1))\n",
    "b3 = tf.Variable(tf.zeros([10]) + 0.1)\n",
    "\n",
    "# softmax回归模型\n",
    "prediction = tf.nn.softmax(tf.matmul(L2_drop, W3) + b3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 二次 Loss Func\n",
    "# loss = tf.reduce_mean(tf.square(y - prediction))\n",
    "# 交叉熵 Loss Func\n",
    "# loss = tf.reduce_mean(-tf.reduce_sum(y * tf.log(prediction), reduction_indices=[1]))\n",
    "loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels = y, logits = prediction))\n",
    "\n",
    "# 梯度下降\n",
    "# train_step = tf.train.GradientDescentOptimizer(0.2).minimize(loss)\n",
    "train_step = tf.train.AdamOptimizer(lr).minimize(loss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Batch:  0 Accuracy: [ 0.9573 , 0.962128 ]\n",
      "Batch:  10 Accuracy: [ 0.9803 , 0.994455 ]\n",
      "Batch:  20 Accuracy: [ 0.9795 , 0.997073 ]\n",
      "Batch:  30 Accuracy: [ 0.9816 , 0.997709 ]\n",
      "Batch:  40 Accuracy: [ 0.9828 , 0.997946 ]\n",
      "Batch:  50 Accuracy: [ 0.9823 , 0.998128 ]\n",
      "Batch:  60 Accuracy: [ 0.9828 , 0.998237 ]\n",
      "Batch:  70 Accuracy: [ 0.9828 , 0.998309 ]\n",
      "Batch:  80 Accuracy: [ 0.9831 , 0.998346 ]\n",
      "Batch:  90 Accuracy: [ 0.9828 , 0.9984 ]\n",
      "Batch:  100 Accuracy: [ 0.9831 , 0.9984 ]\n"
     ]
    }
   ],
   "source": [
    "# 评估模型\n",
    "# 判断 一维张量 y、prediction中最大值的位置是否相等\n",
    "correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(prediction,1))\n",
    "# 准确率\n",
    "# 将 布尔型列表 corrent_prediction转化为 float32类型\n",
    "# [true, false, false, ...]  => [1.0, 0., 0., ...]\n",
    "accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))\n",
    "\n",
    "with tf.device('/gpu:0'):\n",
    "    with tf.Session() as sess:\n",
    "        tf.global_variables_initializer().run()\n",
    "\n",
    "        batch_size = 100\n",
    "        batch = (int) (60000 / batch_size)\n",
    "        # batch = mnist.train.num_examples\n",
    "\n",
    "        for _ in range(101):\n",
    "            sess.run(tf.assign(lr, 0.001 * (0.95 ** _)))\n",
    "            for batch_step in range(batch):\n",
    "                batch_xs, batch_ys = mnist.train.next_batch(batch_size)\n",
    "                sess.run(train_step, feed_dict = {x: batch_xs, y: batch_ys, z: 0.9973})\n",
    "\n",
    "            if (_ % 10) == 0:\n",
    "                test_accuracy = sess.run(accuracy, feed_dict = {x: mnist.test.images, y: mnist.test.labels, z: 1.0})\n",
    "                train_accuracy = sess.run(accuracy, feed_dict = {x: mnist.train.images, y: mnist.train.labels, z: 1.0})\n",
    "                print(\"Batch: \", _, \"Accuracy: [\", test_accuracy, \",\", train_accuracy, \"]\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 二次 Loss Func\n",
    "Batch:  0 Accuracy:  0.8394\n",
    "Batch:  10 Accuracy:  0.9067\n",
    "Batch:  20 Accuracy:  0.9142\n",
    "Batch:  30 Accuracy:  0.9187\n",
    "Batch:  40 Accuracy:  0.9199\n",
    "Batch:  50 Accuracy:  0.9219\n",
    "\n",
    "# 交叉熵 Loss Func\n",
    "Batch:  0 Accuracy:  0.8262\n",
    "Batch:  10 Accuracy:  0.9183\n",
    "Batch:  20 Accuracy:  0.9224\n",
    "Batch:  30 Accuracy:  0.9232\n",
    "Batch:  40 Accuracy:  0.9273\n",
    "Batch:  50 Accuracy:  0.9274\n",
    "\n",
    "# 隐藏层 + DropOut\n",
    "Batch:  0 Accuracy: [ 0.9176 , 0.915527 ]\n",
    "Batch:  10 Accuracy: [ 0.9565 , 0.963182 ]\n",
    "Batch:  20 Accuracy: [ 0.9669 , 0.975236 ]\n",
    "Batch:  30 Accuracy: [ 0.9718 , 0.982 ]\n",
    "Batch:  40 Accuracy: [ 0.9737 , 0.984836 ]\n",
    "Batch:  50 Accuracy: [ 0.9768 , 0.987036 ]\n",
    "\n",
    "# AdamOptimizer\n",
    "Batch:  0 Accuracy: [ 0.9573 , 0.962128 ]\n",
    "Batch:  10 Accuracy: [ 0.9803 , 0.994455 ]\n",
    "Batch:  20 Accuracy: [ 0.9795 , 0.997073 ]\n",
    "Batch:  30 Accuracy: [ 0.9816 , 0.997709 ]\n",
    "Batch:  40 Accuracy: [ 0.9828 , 0.997946 ]\n",
    "Batch:  50 Accuracy: [ 0.9823 , 0.998128 ]\n",
    "Batch:  60 Accuracy: [ 0.9828 , 0.998237 ]\n",
    "Batch:  70 Accuracy: [ 0.9828 , 0.998309 ]\n",
    "Batch:  80 Accuracy: [ 0.9831 , 0.998346 ]\n",
    "Batch:  90 Accuracy: [ 0.9828 , 0.9984 ]\n",
    "Batch:  100 Accuracy: [ 0.9831 , 0.9984 ]"
   ]
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "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": "3.5.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}