tensorflow-正则化+指数衰减+滑动平均

Posted 2021-01-27

#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
Created on Tue Sep 19 09:42:22 2017

@author: myhaspl
"""
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data

INPUT_NODE=784
OUTPUT_NODE=10

LAYER1_NODE=500
BATCH_SIZE=100

LEARNING_RATE_BASE=0.8
LEARNING_RATE_DECAY=0.99

REGULARIZATION_RATE=0.0001
TRANING_STEPS=30000
MOVING_AVERAGE_DECAY=0.99

def inference(input_tensor,avg_class,weights1, biases1,weights2,biases2):
    if avg_class==None:#非滑动平均
        layer1=tf.nn.relu(tf.matmul(input_tensor,weights1)+biases1)
        return tf.matmul(layer1,weights2)+biases2
    else:#滑动平均
        layer1=tf.nn.relu(tf.matmul(input_tensor,avg_class.average(weights1))+avg_class.average(biases1))
        return tf.matmul(layer1,avg_class.average(weights2))+avg_class.average(biases2)

def train(mnist):
    #样本数据与样本标签
    x_=tf.placeholder(tf.float32,[None,INPUT_NODE],name=‘x_-input‘)
    y_=tf.placeholder(tf.float32,[None,OUTPUT_NODE],name=‘y_-input‘)
    #参数初始值
    weights1=tf.Variable(tf.truncated_normal([INPUT_NODE,LAYER1_NODE],stddev=0.1))
    biases1=tf.Variable(tf.constant(0.1,shape=[LAYER1_NODE]))
    weights2=tf.Variable(tf.truncated_normal([LAYER1_NODE,OUTPUT_NODE],stddev=0.1))
    biases2=tf.Variable(tf.constant(0.1,shape=[OUTPUT_NODE]))
    global_step=tf.Variable(0,trainable=False)

    #非滑动平均    
    y_nohd=inference(x_,None,weights1,biases1,weights2,biases2)

    #滑动平均
    variable_averages=tf.train.ExponentialMovingAverage(MOVING_AVERAGE_DECAY,global_step)
    #滑动平均更新变量的操作
    variable_averages_op=variable_averages.apply(tf.trainable_variables())
    y_hd=inference(x_,variable_averages,weights1,biases1,weights2,biases2)

    #交叉嫡损失函数,使用softmax归一化
    cross_entropy=tf.nn.sparse_softmax_cross_entropy_with_logits(logits=y_nohd,labels=tf.arg_max(y_,1))
    cross_entropy_mean=tf.reduce_mean(cross_entropy)
    #加入L2正则化损失
    regularizer=tf.contrib.layers.l2_regularizer(REGULARIZATION_RATE)
    regularization=regularizer(weights1)+regularizer(weights2)
    loss=cross_entropy_mean+regularization

    #设置指数衰减的学习率
    learning_rate=tf.train.exponential_decay(
            LEARNING_RATE_BASE,
            global_step,
            mnist.train.num_examples/BATCH_SIZE,
            LEARNING_RATE_DECAY)

    train_step=tf.train.GradientDescentOptimizer(learning_rate).minimize(loss,global_step=global_step)
    #训练与更新参数的滑动平均值
    #将2大步操作打包在train_op中，第1大步操作是使用正则化和指数衰减更新参数值
    #第2大步操作是使用滑动平均再次更新参数值。
    #每次训练都完成这2大步操作。
    train_op=tf.group(train_step,variable_averages_op)
    #检验滑动平均平均模型的神经网络前向传播结果是否正确
    correct_predection=tf.equal(tf.argmax(y_hd,1),tf.argmax(y_,1))
    accuracy=tf.reduce_mean(tf.cast(correct_predection,tf.float32))

    #开始训练过程
    with tf.Session() as sess:
        tf.initialize_all_variables().run()
        #训练样本集
        validate_feed={x_:mnist.validation.images,
                       y_:mnist.validation.labels
                      }
        #测试集
        test_feed={x_:mnist.test.images,
                   y_:mnist.test.labels
                   }
        for i  in range(TRANING_STEPS):
            if i%1000==0:
                #每1000轮计算当前训练的结果
                validate_acc=sess.run(accuracy,feed_dict=validate_feed)
                print("%d次后=>正确率%g"%(i,validate_acc))
            #每一轮使用的样本，然后开始训练
            xs,ys=mnist.train.next_batch(BATCH_SIZE)
            sess.run(train_op,feed_dict={x_:xs,y_:ys})

        #TRANING_STEPS次训练结束，对测试数据进行检测，检验神经网络准确度
        test_acc=sess.run(accuracy,feed_dict=test_feed)
        print("正确率:%g"%test_acc)

def main(argv=None):
    mnist=input_data.read_data_sets("/tmp/data",one_hot=True)
    train(mnist)

if __name__==‘__main__‘:
    tf.app.run()

使用了非线性激活函数relu，防止梯度消失。