单个神经元的网络分析

import tensorflow as tf

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

mnist = input_data.read_data_sets("dataset/",one_hot=True)

   #print(mnist.train.num_examples,mnist.validation.num_examples,mnist.test.num_examples)

#print(mnist.train.images.shape,mnist.train.labels.shape)

x=tf.placeholder(tf.float32,[None,784],name="X")   #输入

y=tf.placeholder(tf.float32,[None,10],name="Y")    #输出

W=tf.Variable(tf.random_normal([784,10],name="W"))   #构建权重

b=tf.Variable(tf.zeros([10]),name="b")                 #构建偏置

forward=tf.matmul(x,W)+b                  #构建前向乘

pred=tf.nn.softmax(forward)               #softmax分类

train_epochs=50                        #训练轮数

batch_size=100                             #单次训练样本数

total_batch=int(mnist.train.num_examples/batch_size)          #一轮训练有多少次

display_step=1                 #显示粒度

learning_rate=0.01            #学习率

loss_function=tf.reduce_mean(-tf.reduce_sum(y*tf.log(pred),reduction_indices=1))   #定义损伤函数

optimizer=tf.train.GradientDescentOptimizer(learning_rate).minimize(loss_function)  #选择梯度下降优化器

 

#检查预测类别tf.argmax(pred,1)与实际类别tf.argmax(y,1)的匹配情况

correct_prediction=tf.equal(tf.argmax(pred,1),tf.argmax(y,1))

#准确率，将布尔值装换为浮点数，并计算平均值

accuracy=tf.reduce_mean(tf.cast(correct_prediction,tf.float32))

sess=tf.Session()#声明会话

init=tf.global_variables_initializer()#变量初始化

sess.run(init)

#开始训练

for epoch in range (train_epochs):

    for batch in range(total_batch):

        xs,ys=mnist.train.next_batch(batch_size) #读取批次数据

        sess.run(optimizer,feed_dict={x: xs,y: ys}) #执行批次训练

        #total_batch个批次训练完成后，使用验证数据计算误差与准确率；验证集没有分批

    loss,acc=sess.run([loss_function,accuracy],feed_dict={x:mnist.validation.images,y:mnist.validation.labels})

       #打印训练过程中的详细信息

    if (epoch+1) % display_step == 0:

        print("Train Epoch:", '%02d' % (epoch+1),"Loss=","{:.9f}".format(loss),\

               "Accuracy=","{:.4f}".format(acc))

print("Train Finished!")