# -*- coding: utf-8 -*-
"""
Created on Sat May 26 14:31:09 2018
@author: 被遗弃的庸才
"""
import tensorflow as tf
import numpy as np
from tensorflow.examples.tutorials.mnist import input_data
#28*28=784的像素点,y是10个位置上面的数据
mnist=input_data.read_data_sets('MNIST_data',one_hot=True)#如果没有数据源会帮我们下载,如果有会用本地的数据源
def compute_accuracy(v_xs,v_ys):
#global prediction
y_pre=sess.run(prediction,feed_dict={xs:v_xs,keep_prob:1.0})
correct_prediction=tf.equal(tf.argmax(y_pre,1),tf.arg_max(v_ys,1))
accuracy=tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
result=sess.run(accuracy,feed_dict={xs:v_xs,ys:v_ys})
return result
def add_layer(inputs,in_size,out_size,activiation_function=None):
Weight=tf.Variable(tf.random_normal([in_size, out_size]))*0.1 #如果relu参数初始化不理想,前向运算结果为负值,则进过梯度计算时候全部变零,在反向运算的时候没有响应,这就是“死区”现象。
#如果使用sigmoid就不会出现这样的问题。。。。。。。。。。。当然前提是网络还不是很深的情况,如果很深可能会出现vanishing grading
biases=tf.Variable(tf.zeros([1]))
Wx_plus_b=tf.matmul(inputs,Weight)+biases
Wx_plus_b=tf.nn.dropout(Wx_plus_b,keep_prob)
if activiation_function is None:
outputs=Wx_plus_b
else:
outputs=activiation_function(Wx_plus_b)
return outputs
#定义输入的x,y变量
keep_prob=tf.placeholder(tf.float32)
xs=tf.placeholder(tf.float32,[None,784])
ys=tf.placeholder(tf.float32,[None,10])
#构建神经网络的结构
l1=add_layer(xs,784,200,activiation_function=tf.nn.relu)
l2=add_layer(l1,200,200,activiation_function=tf.nn.relu)
l3=add_layer(l2,200,200,activiation_function=tf.nn.relu)
prediction=add_layer(l3,200,10,activiation_function=tf.nn.softmax)
#loss function 交叉熵
loss=tf.reduce_mean(-tf.reduce_sum(ys*tf.log(prediction),reduction_indices=[1]))
#训练
train=tf.train.AdamOptimizer(0.008).minimize(loss)#手动生成学习率
with tf.Session() as sess:
sess.run(tf.initialize_all_variables())#初始化所有的参数
for i in range(501):
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(compute_accuracy(mnist.test.images,mnist.test.labels))
"""
Created on Sat May 26 14:31:09 2018
@author: 被遗弃的庸才
"""
import tensorflow as tf
import numpy as np
from tensorflow.examples.tutorials.mnist import input_data
#28*28=784的像素点,y是10个位置上面的数据
mnist=input_data.read_data_sets('MNIST_data',one_hot=True)#如果没有数据源会帮我们下载,如果有会用本地的数据源
def compute_accuracy(v_xs,v_ys):
#global prediction
y_pre=sess.run(prediction,feed_dict={xs:v_xs,keep_prob:1.0})
correct_prediction=tf.equal(tf.argmax(y_pre,1),tf.arg_max(v_ys,1))
accuracy=tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
result=sess.run(accuracy,feed_dict={xs:v_xs,ys:v_ys})
return result
def add_layer(inputs,in_size,out_size,activiation_function=None):
Weight=tf.Variable(tf.random_normal([in_size, out_size]))*0.1 #如果relu参数初始化不理想,前向运算结果为负值,则进过梯度计算时候全部变零,在反向运算的时候没有响应,这就是“死区”现象。
#如果使用sigmoid就不会出现这样的问题。。。。。。。。。。。当然前提是网络还不是很深的情况,如果很深可能会出现vanishing grading
biases=tf.Variable(tf.zeros([1]))
Wx_plus_b=tf.matmul(inputs,Weight)+biases
Wx_plus_b=tf.nn.dropout(Wx_plus_b,keep_prob)
if activiation_function is None:
outputs=Wx_plus_b
else:
outputs=activiation_function(Wx_plus_b)
return outputs
#定义输入的x,y变量
keep_prob=tf.placeholder(tf.float32)
xs=tf.placeholder(tf.float32,[None,784])
ys=tf.placeholder(tf.float32,[None,10])
#构建神经网络的结构
l1=add_layer(xs,784,200,activiation_function=tf.nn.relu)
l2=add_layer(l1,200,200,activiation_function=tf.nn.relu)
l3=add_layer(l2,200,200,activiation_function=tf.nn.relu)
prediction=add_layer(l3,200,10,activiation_function=tf.nn.softmax)
#loss function 交叉熵
loss=tf.reduce_mean(-tf.reduce_sum(ys*tf.log(prediction),reduction_indices=[1]))
#训练
train=tf.train.AdamOptimizer(0.008).minimize(loss)#手动生成学习率
with tf.Session() as sess:
sess.run(tf.initialize_all_variables())#初始化所有的参数
for i in range(501):
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(compute_accuracy(mnist.test.images,mnist.test.labels))