tensorflow中训练mnist数据集－－－训练

训练

import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt
from tensorflow.examples.tutorials.mnist import input_data
print("packs loaded")

mnist = input_data.read_data_sets('./data/',one_hot=True)
#如果该位置下没有数据，第一次会自动下载

input_num = 784
hidden1 = 256
hidden2 = 128
output_num = 10
x = tf.placeholder(tf.float32, [None, input_num], name = 'input-x')
y = tf.placeholder(tf.float32, [None, output_num], name = 'input-y')

W = {
    'w1':tf.Variable(tf.random_normal([input_num, hidden1],stddev=0.1)),
    'w2':tf.Variable(tf.random_normal([hidden1, hidden2], stddev=0.1)),
    'out':tf.Variable(tf.random_normal([hidden2, output_num]))
}
b = {
    'b1':tf.Variable(tf.random_normal([hidden1])),
    'b2':tf.Variable(tf.random_normal([hidden2])),
    'b_out':tf.Variable(tf.random_normal([output_num]))
}
print "network init finish"
#获得x,y的名字，在加载模型测试时要用到
print(x.name)
print(y.name)

”“”
以下是变量名称，在notebook中多次运行会产生不一样的变量名，可能是input-x:1,input-y:1.要明确知道你的模型中保存的变量名称。
network init finish
input-x:0
input-y:0

“”“

def net(weight,b):
    layer1 = tf.nn.sigmoid(tf.add(tf.matmul(x,weight['w1']), b['b1']))
    layer2 = tf.nn.sigmoid(tf.add(tf.matmul(layer1,weight['w2']),b['b2']))
    return tf.add(tf.matmul(layer2,weight['out']) , b['b_out'])

#经过前向传播得到的预测值，是一个10维tf.float32向量
pred = net(W,b)

#计算交叉熵损失函数，y是标签，pred是预测值
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=pred,labels=y))
#梯度下降
optm = tf.train.GradientDescentOptimizer(0.05).minimize(cost)
"""
tf.argmax(pred,1),获得pred中最大值的下标，１表示在哪个维度上计算
tf.argmax(y,1)同理，
tf.equal(),对应坐标对比，如果相同，为True,返回的是数组
"""
corr = tf.equal(tf.argmax(pred,1), tf.argmax(y,1))
#计算准确率，tf.cast()更改类型，bool-->tf.float32,tf.reduce_mean()计算均值
accr = tf.reduce_mean(tf.cast(corr,tf.float32))
#初始化变量
init = tf.global_variables_initializer()
print(pred.name)
print(accr.name)

“”“
Add_2:0
Mean_1:0
”“”

training_epochs = 50
batch_size = 128
display_step = 5

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

#保存模型对象初始化
saver = tf.train.Saver()

for epcho in range(training_epochs):
    avg_cost = 0
    num_batch = int(mnist.train.num_examples/batch_size)
    
    for i in range(num_batch):
        
        batch_xs,batch_ys = mnist.train.next_batch(batch_size)
        sess.run(optm,feed_dict = {x:batch_xs,y:batch_ys})
        feeds = {x:batch_xs, y:batch_ys}
        avg_cost= sess.run(cost,feed_dict = feeds)/num_batch
    if epcho % 5 == 0:
        feeds_train = {x:batch_xs,y:batch_ys}
        feeds_test = {x:mnist.test.images,y:mnist.test.labels}
        train_acc = sess.run(accr, feed_dict = feeds_train)
        test_acc = sess.run(accr, feed_dict = feeds_test)
        
        print("Epoch %03d/%03d cost: %0.9f train_acc: %0.3f test_acc: %0.3f"
              %(epcho,training_epochs,avg_cost,train_acc,test_acc))
#保存模型  
saver.save(sess, "./model/model.ckpt")
sess.close()
print "DONE"