Tensorflow入门（谭秉峰）（六）MNIST数据集手写体识别(参数改进，使用relu激活GPU训练)

# Author: Zubin
# -*- coding: utf-8 -*
from MNIST_defiction import *
import  tensorflow as tf

#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

#mnist数据集常数
INPUT_NODE=784
#第一层神经元的个数
L1_NODE=500
#输出类别数
OUTPUT_NODE=10



'''
定义两个placeholder.28x28的像素=784（将28*28像素的灰度图片转换成一个向量）
定义一层隐藏层，神经元个数为10个
'''

x=tf.placeholder(tf.float32,[None,784])
y=tf.placeholder(tf.float32,[None,10])
#辍学率
keep_prob=tf.placeholder(tf.float32)


#创建一个简单的神经网络,权重初始化为0
#第一层神经元
W1=tf.Variable(tf.truncated_normal([INPUT_NODE,L1_NODE],stddev=0.1))
b1=tf.Variable(tf.constant(0.1,shape=[L1_NODE]))
L1_Output=tf.nn.relu(tf.matmul(x,W1)+b1)
L1_drop=tf.nn.dropout(L1_Output,keep_prob)

#第二层神经元（输出层）
W2=tf.Variable(tf.truncated_normal([L1_NODE,OUTPUT_NODE],stddev=0.1))
b2=tf.Variable(tf.constant(0.1,shape=[OUTPUT_NODE]))
prediction=tf.nn.relu(tf.matmul(L1_drop,W2)+b2)

#二次代价函数
#loss=tf.reduce_mean(tf.square(y-prediction))
#交叉熵代价函数
loss=tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y,logits=prediction))
#使用Adam自适应优化算法
train_step=tf.train.AdamOptimizer(0.01).minimize(loss)

#初试化变量
init=tf.global_variables_initializer()

#将结果存放在一个布尔整型列表中
correct_prediction=tf.equal(tf.argmax(y,1),tf.argmax(prediction,1))
#求准确率函数:将布尔型转为整型
accuracy=tf.reduce_mean(tf.cast(correct_prediction,tf.float32))

with tf.Session() as sess:
    sess.run(init)
    #整个数据集迭代51次
    for epoch in range(51):
        #每一批训练多少张图片
        for batch in range (n_batch):
            #batch_x为图片，batch_ys为标签
             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})

        #迭代一次完求准确率
        train_acc=sess.run(accuracy,feed_dict={x:mnist.train.images,y:mnist.train.labels,keep_prob:1.0})
        test_acc=sess.run(accuracy,feed_dict={x:mnist.test.images,y:mnist.test.labels,keep_prob:1.0})

        print("After "+str(epoch)+" training steps,"+"training accuracy is: "+str(train_acc)+", testing accuracy is: "+str(test_acc))

50次迭代，输出结果： 测试集上准确率达到 97%

After 0 training steps,training accuracy is: 0.91596365, testing accuracy is: 0.9196
After 1 training steps,training accuracy is: 0.9622909, testing accuracy is: 0.9568
After 2 training steps,training accuracy is: 0.9655455, testing accuracy is: 0.9573
After 3 training steps,training accuracy is: 0.9729091, testing accuracy is: 0.9627
After 4 training steps,training accuracy is: 0.97621816, testing accuracy is: 0.9639
After 5 training steps,training accuracy is: 0.9763273, testing accuracy is: 0.9699
After 6 training steps,training accuracy is: 0.9782182, testing accuracy is: 0.9686
After 7 training steps,training accuracy is: 0.9811636, testing accuracy is: 0.9687
After 8 training steps,training accuracy is: 0.9788, testing accuracy is: 0.9669
After 9 training steps,training accuracy is: 0.9824909, testing accuracy is: 0.9689
After 10 training steps,training accuracy is: 0.97801816, testing accuracy is: 0.9662
After 11 training steps,training accuracy is: 0.9807818, testing accuracy is: 0.9672
After 12 training steps,training accuracy is: 0.9828909, testing accuracy is: 0.9701
After 13 training steps,training accuracy is: 0.9822364, testing accuracy is: 0.9701
After 14 training steps,training accuracy is: 0.98343635, testing accuracy is: 0.9692
After 15 training steps,training accuracy is: 0.98398185, testing accuracy is: 0.9702
After 16 training steps,training accuracy is: 0.985, testing accuracy is: 0.9704
After 17 training steps,training accuracy is: 0.9848727, testing accuracy is: 0.9681
After 18 training steps,training accuracy is: 0.98498183, testing accuracy is: 0.9719
After 19 training steps,training accuracy is: 0.9841818, testing accuracy is: 0.9693
After 20 training steps,training accuracy is: 0.98623633, testing accuracy is: 0.9699
After 21 training steps,training accuracy is: 0.98407274, testing accuracy is: 0.9646
After 22 training steps,training accuracy is: 0.98623633, testing accuracy is: 0.969
After 23 training steps,training accuracy is: 0.9865636, testing accuracy is: 0.9711
After 24 training steps,training accuracy is: 0.9869091, testing accuracy is: 0.9724
After 25 training steps,training accuracy is: 0.9888545, testing accuracy is: 0.9728
After 26 training steps,training accuracy is: 0.98734546, testing accuracy is: 0.9725
After 27 training steps,training accuracy is: 0.9864182, testing accuracy is: 0.9706
After 28 training steps,training accuracy is: 0.98881817, testing accuracy is: 0.9714
After 29 training steps,training accuracy is: 0.9884909, testing accuracy is: 0.9721
After 30 training steps,training accuracy is: 0.98905456, testing accuracy is: 0.9714
After 31 training steps,training accuracy is: 0.98547274, testing accuracy is: 0.9676
After 32 training steps,training accuracy is: 0.98636365, testing accuracy is: 0.9686
After 33 training steps,training accuracy is: 0.98745453, testing accuracy is: 0.9686
After 34 training steps,training accuracy is: 0.9914, testing accuracy is: 0.973
After 35 training steps,training accuracy is: 0.98963636, testing accuracy is: 0.9734
After 36 training steps,training accuracy is: 0.9895273, testing accuracy is: 0.971
After 37 training steps,training accuracy is: 0.9884, testing accuracy is: 0.9713
After 38 training steps,training accuracy is: 0.98725456, testing accuracy is: 0.9689
After 39 training steps,training accuracy is: 0.9899091, testing accuracy is: 0.97
After 40 training steps,training accuracy is: 0.98907274, testing accuracy is: 0.9712
After 41 training steps,training accuracy is: 0.9887091, testing accuracy is: 0.9691
After 42 training steps,training accuracy is: 0.9907454, testing accuracy is: 0.9725
After 43 training steps,training accuracy is: 0.9888909, testing accuracy is: 0.9715
After 44 training steps,training accuracy is: 0.9895818, testing accuracy is: 0.9717
After 45 training steps,training accuracy is: 0.98785454, testing accuracy is: 0.9695
After 46 training steps,training accuracy is: 0.99038184, testing accuracy is: 0.9718
After 47 training steps,training accuracy is: 0.9900727, testing accuracy is: 0.9726
After 48 training steps,training accuracy is: 0.99003637, testing accuracy is: 0.9719
After 49 training steps,training accuracy is: 0.98956364, testing accuracy is: 0.9708
After 50 training steps,training accuracy is: 0.98947275, testing accuracy is: 0.9719

Process finished with exit code 0