神经网络4：卷积神经网络学习 2

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

mnist = input_data.read_data_sets("MNIST_data", one_hot=True)
in_x = tf.placeholder(shape=[None, 784], dtype=tf.float32)
in_label = tf.placeholder(shape=[None, 10], dtype=tf.float32)
keep_prob = tf.placeholder(shape=None, dtype=tf.float32)
sess = tf.Session()


def createWeightVariable(shape):
    initial = tf.truncated_normal(shape=shape, mean=0.0, stddev=0.1, dtype=tf.float32)
    return tf.Variable(initial_value=initial, dtype=tf.float32)


def createBiasesVariable(shape):
    initial = tf.constant(value=1.0, shape=shape, dtype=tf.float32)
    return tf.Variable(initial_value=initial)


def conv2d(inputs, Weight):
    return tf.nn.conv2d(input=inputs, filter=Weight, strides=[1, 1, 1, 1], padding="SAME")


def max_pool_2x2(inputs):
    return tf.nn.max_pool(value=inputs, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding="SAME")


def add_layer(inputs, in_size, out_size, activation_func=None):
    Weight = createWeightVariable(shape=[in_size, out_size])
    biases = createBiasesVariable(shape=[out_size])
    outputs = tf.matmul(inputs, Weight) + biases
    if activation_func is None:
        return outputs
    else:
        return activation_func(outputs)



def compute_accuracy(test_x, test_label):
    global prediction
    pre_label = sess.run(prediction, feed_dict={in_x: test_x, keep_prob: 1.0})
    ok_label = tf.equal(tf.argmax(pre_label, axis=1), tf.argmax(test_label, axis=1))
    accuracy = tf.reduce_mean(tf.cast(ok_label, dtype=tf.float32))
    result = sess.run(accuracy, feed_dict={in_x: test_x, keep_prob: 1.0})
    return result


x_inputs = tf.reshape(tensor=in_x, shape=[-1, 28, 28, 1])

Weights1 = createWeightVariable(shape=[5, 5, 1, 32])
biases1 = createBiasesVariable(shape=[32])

h_conv1 = tf.nn.relu(conv2d(x_inputs, Weights1))
h_pool1 = max_pool_2x2(h_conv1)

Weights2 = createWeightVariable(shape=[5, 5, 32, 64])
biases2 = createBiasesVariable(shape=[64])
h_conv2 = tf.nn.relu(conv2d(h_pool1, Weights2))
h_pool2 = max_pool_2x2(h_conv2)
h_pool2_flat = tf.reshape(tensor=h_pool2, shape=[-1, 7 * 7 * 64])

f1_outs = add_layer(h_pool2_flat, 7 * 7 * 64, 1024, tf.nn.relu)
f1_drop_outs = tf.nn.dropout(f1_outs, keep_prob)

prediction = add_layer(f1_drop_outs, 1024, 10, tf.nn.softmax)

cross_entropy = tf.reduce_mean(-tf.reduce_sum(in_label * tf.log(prediction), reduction_indices=[1]))
train = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
sess.run(tf.global_variables_initializer())
for i in range(1000):
    batch_x, batch_y = mnist.train.next_batch(100)
    sess.run(train, feed_dict={in_x: batch_x, in_label: batch_y, keep_prob: 0.5})
    if i % 50 == 0:
        print(compute_accuracy(mnist.test.images, mnist.test.labels))