Tensorflow 使用CNN 进行mnist 训练代码示例

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

mnist = input_data.read_data_sets('MNIST/',one_hot = True)   #这里向文件夹读取资源

def calculate_accuracy(v_xs,v_ys):    #这里用 test 资源来进行准确率的训练

    global prediction                          #指全局的预测模型
    y_prediction = sess.run(prediction,feed_dict={xs:v_xs,keep_prob : 1})       #进行测试文件预测   
    num_correct = tf.equal(tf.argmax(y_prediction,1),tf.argmax(v_ys,1))         #这里预测结果和 原来结果对比
    accuary = tf.reduce_mean(tf.cast(num_correct,tf.float32))                          #计算准确的概率
    result = sess.run(accuary,feed_dict= {xs:v_xs , ys:v_ys ,keep_prob : 1})       
    return result

def convn_layer(batch_x,batch_y,in_size,out_size,in_resource, name):             #这里卷积和池化层一起

    #这里初始化卷积层变量，卷积采样为batch_x*batch_y，输入层数为 in_size，输出层数out_size

    W= tf.Variable( tf.truncated_normal([batch_x,batch_y,in_size,out_size], stddev = 0.1,dtype = tf.float32))   

    b=tf.Variable(tf.constant(0.1,shape = [out_size]))
    conv = tf.nn.relu(tf.nn.conv2d(in_resource,W ,strides=[1,1,1,1],padding = 'SAME')+b)     #这里卷积
    pool = tf.nn.max_pool(conv,ksize=[1,2,2,1],strides=[1,2,2,1],padding = 'SAME')               #这里池化，层数*2

    return pool

def func_layer(in_size,out_size,in_resource,name,if_drop,m_function = None):       #这里为添加神经网络函数

    W= tf.Variable( tf.truncated_normal([in_size,out_size], stddev = 0.1,dtype =tf.float32))
    b= tf.Variable( tf.constant(0.1,shape = [out_size]))

    if m_function == None:
        func = tf.matmul(in_resource,W)+b
    else:
        func = m_function(tf.matmul(in_resource,W)+b)
    if if_drop:
        output = tf.nn.dropout(func,keep_prob)
    else:
        output = func
    return output

xs = tf.placeholder(tf.float32,[None,784])/255          #mnist 的图片为28*28，所以为784，/255使得数值减小，利于快速下降
ys = tf.placeholder(tf.float32,[None,10])

keep_prob = tf.placeholder(tf.float32,name = 'dropout_value')

image_in = tf.reshape(xs, [-1,28,28,1])

conv_layer_1 = convn_layer(5,5,1,32,image_in,name = 'conv_layer_1')        #添加卷积，池化层 
conv_layer_2 = convn_layer(5,5,32,64,conv_layer_1,name = 'conv_layer_2')     #添加卷积，池化层 

conv_layer_2_flat  = tf.reshape(conv_layer_2,[-1,7*7*64])

#添加神经网络层

layer1_out  = func_layer(7*7*64,1024,conv_layer_2_flat,name = 'function_layer1',if_drop=1,m_function =tf.nn.relu) 

prediction = func_layer(1024,10,layer1_out,name = 'function_layer_out',if_drop=0,m_function =tf.nn.softmax )

cross_entropy = tf.reduce_mean(-tf.reduce_sum(ys*tf.log(prediction),reduction_indices = [1]))

train = tf.train.AdamOptimizer(0.001).minimize(cross_entropy)       #采用Adam的训练方式

with tf.Session() as sess:

    init = tf.global_variables_initializer()
    sess.run(init)

    for i in range(1000):
        batch_x,batch_y = mnist.train.next_batch(100)
        sess.run(train,feed_dict = {xs:batch_x, ys:batch_y, keep_prob: 0.5})

        if i % 50==0:
            print(calculate_accuracy(mnist.test.images[:1000], mnist.test.labels[:1000]))    #这里输入测试样本进行准确率运算