卷积+池化+卷积+池化+全连接2

#!/usr/bin/env python

import tensorflow as tffrom tensorflow.examples.tutorials.mnist import input_data

# In[2]:

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

# 每个批次的大小batch_size = 100# 计算一共有多少个批次n_batch = mnist.train.num_examples // batch_size

# 初始化权值def weight_variable(shape):    initial = tf.truncated_normal(shape, stddev=0.1)  # 生成一个截断的正态分布    return tf.Variable(initial)

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

# 卷积层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方向的步长，strides[2]代表y方向的步长    # padding: A `string` from: `"SAME", "VALID"`    return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding=‘SAME‘)

# 池化层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‘)# 定义两个placeholderx = tf.placeholder(tf.float32, [None, 784])y = tf.placeholder(tf.float32, [None, 10])# 改变x的格式转为4D的向量[batch, in_height, in_width, in_channels]`x_image = tf.reshape(x, [-1, 28, 28, 1])# 初始化第一个卷积层的权值和偏置W_conv1 = weight_variable([5, 5, 1, 16])  # 5*5的采样窗口，32个卷积核从1个平面抽取特征b_conv1 = bias_variable([16])  # 每一个卷积核一个偏置值# 把x_image和权值向量进行卷积，再加上偏置值，然后应用于relu激活函数conv2d_1 = conv2d(x_image, W_conv1) + b_conv1h_conv1 = tf.nn.relu(conv2d_1)h_pool1 = max_pool_2x2(h_conv1)  # 进行max-pooling    # 初始化第二个卷积层的权值和偏置W_conv2 = weight_variable([5, 5, 16, 32])  # 5*5的采样窗口，64个卷积核从32个平面抽取特征b_conv2 = bias_variable([32])  # 每一个卷积核一个偏置值    # 把h_pool1和权值向量进行卷积，再加上偏置值，然后应用于relu激活函数conv2d_2 = conv2d(h_pool1, W_conv2) + b_conv2h_conv2 = tf.nn.relu(conv2d_2)h_pool2 = max_pool_2x2(h_conv2)  # 进行max-pooling# 28*28的图片第一次卷积后还是28*28，第一次池化后变为14*14# 第二次卷积后为14*14，第二次池化后变为了7*7# 进过上面操作后得到64张7*7的平面    # 初始化第一个全连接层的权值W_fc1 = weight_variable([7 * 7 * 32, 512])  # 上一场有7*7*64个神经元，全连接层有1024个神经元b_fc1 = bias_variable([512])  # 1024个节点    # 把池化层2的输出扁平化为1维h_pool2_flat = tf.reshape(h_pool2, [-1, 7 * 7 * 32])    # 求第一个全连接层的输出wx_plus_b1 = tf.matmul(h_pool2_flat, W_fc1) + b_fc1h_fc1 = tf.nn.relu(wx_plus_b1)    # keep_prob用来表示神经元的输出概率keep_prob = tf.placeholder(tf.float32)h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)    # 初始化第二个全连接层W_fc2 = weight_variable([512, 10])b_fc2 = bias_variable([10])wx_plus_b2 = tf.matmul(h_fc1_drop, W_fc2) + b_fc2        # 计算输出prediction = tf.nn.softmax(wx_plus_b2)# 交叉熵代价函数cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y, logits=prediction))# 使用AdamOptimizer进行优化train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)        # 结果存放在一个布尔列表中correct_prediction = tf.equal(tf.argmax(prediction, 1), tf.argmax(y, 1))  # argmax返回一维张量中最大的值所在的位置        # 求准确率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

原文地址：https://www.cnblogs.com/rongye/p/10012915.html