#!/usr/bin/env python

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data

# In[2]:

mnist = input_data.read_data_sets(‘MNIST_data‘, one_hot=True)

# 每個批次的大小
batch_size = 100
# 計算一共有多少個批次
n_batch = mnist.train.num_examples // batch_size

# 初始化權值
def weight_variable(shape):
    initial = tf.truncated_normal(shape, stddev=0.1)  # 生成一個截斷的正態分布
    return tf.Variable(initial)


# 初始化偏置
def bias_variable(shape):
    initial = tf.constant(0.1, shape=shape)
    return tf.Variable(initial)


# 卷積層
def conv2d(x, W):
    # x input tensor of shape `[batch, in_height, in_width, in_channels]`
    # W filter / kernel tensor of shape [filter_height, filter_width, in_channels, out_channels]
    # `strides[0] = strides[3] = 1`. strides[1]代表x方向的步長，strides[2]代表y方向的步長
    # padding: A `string` from: `"SAME", "VALID"`
    return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding=‘SAME‘)


# 池化層
def max_pool_2x2(x):
    # ksize [1,x,y,1]
    return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding=‘SAME‘)
# 定義兩個placeholder
x = tf.placeholder(tf.float32, [None, 784])
y = tf.placeholder(tf.float32, [None, 10])
# 改變x的格式轉為4D的向量[batch, in_height, in_width, in_channels]`
x_image = tf.reshape(x, [-1, 28, 28, 1])
# 初始化第一個卷積層的權值和偏置
W_conv1 = weight_variable([5, 5, 1, 16])  # 5*5的采樣窗口，32個卷積核從1個平面抽取特征
b_conv1 = bias_variable([16])  # 每一個卷積核一個偏置值
# 把x_image和權值向量進行卷積，再加上偏置值，然後應用於relu激活函數
conv2d_1 = conv2d(x_image, W_conv1) + b_conv1
h_conv1 = tf.nn.relu(conv2d_1)
h_pool1 = max_pool_2x2(h_conv1)  # 進行max-pooling
    # 初始化第二個卷積層的權值和偏置
W_conv2 = weight_variable([5, 5, 16, 32])  # 5*5的采樣窗口，64個卷積核從32個平面抽取特征
b_conv2 = bias_variable([32])  # 每一個卷積核一個偏置值
    # 把h_pool1和權值向量進行卷積，再加上偏置值，然後應用於relu激活函數
conv2d_2 = conv2d(h_pool1, W_conv2) + b_conv2
h_conv2 = tf.nn.relu(conv2d_2)
h_pool2 = max_pool_2x2(h_conv2)  # 進行max-pooling
# 28*28的圖片第一次卷積後還是28*28，第一次池化後變為14*14
# 第二次卷積後為14*14，第二次池化後變為了7*7
# 進過上面操作後得到64張7*7的平面
    # 初始化第一個全連接層的權值
W_fc1 = weight_variable([7 * 7 * 32, 512])  # 上一場有7*7*64個神經元，全連接層有1024個神經元
b_fc1 = bias_variable([512])  # 1024個節點
    # 把池化層2的輸出扁平化為1維
h_pool2_flat = tf.reshape(h_pool2, [-1, 7 * 7 * 32])
    # 求第一個全連接層的輸出
wx_plus_b1 = tf.matmul(h_pool2_flat, W_fc1) + b_fc1
h_fc1 = tf.nn.relu(wx_plus_b1)
    # keep_prob用來表示神經元的輸出概率
keep_prob = tf.placeholder(tf.float32)
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)
    # 初始化第二個全連接層
W_fc2 = weight_variable([512, 10])
b_fc2 = bias_variable([10])
wx_plus_b2 = tf.matmul(h_fc1_drop, W_fc2) + b_fc2
        # 計算輸出
prediction = tf.nn.softmax(wx_plus_b2)
# 交叉熵代價函數
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y, logits=prediction))
# 使用AdamOptimizer進行優化
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
        # 結果存放在一個布爾列表中
correct_prediction = tf.equal(tf.argmax(prediction, 1), tf.argmax(y, 1))  # argmax返回一維張量中最大的值所在的位置
        # 求準確率
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    for epoch in range(11):
        for batch in range(n_batch):
            batch_xs, batch_ys = mnist.train.next_batch(batch_size)
            sess.run(train_step, feed_dict={x: batch_xs, y: batch_ys, keep_prob: 0.7})
        acc = sess.run(accuracy, feed_dict={x: mnist.test.images, y: mnist.test.labels, keep_prob: 1.0})
        print("Iter " + str(epoch) + ", Testing Accuracy= " + str(acc))

Iter 0, Testing Accuracy= 0.9366
Iter 1, Testing Accuracy= 0.9602
Iter 2, Testing Accuracy= 0.9677
Iter 3, Testing Accuracy= 0.9735
Iter 4, Testing Accuracy= 0.9769
Iter 5, Testing Accuracy= 0.9803
Iter 6, Testing Accuracy= 0.9783
Iter 7, Testing Accuracy= 0.9842
Iter 8, Testing Accuracy= 0.9839
Iter 9, Testing Accuracy= 0.9853
Iter 10, Testing Accuracy= 0.9848

卷積+池化+卷積+池化+全連接2