Tensorflow 简单实现MLP

MLP简介

代码及详细注释

#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
Created on Thu Apr  6 16:37:14 2017
@author: mml
"""
# tensorflow的MNIST数据加载模块
from tensorflow.examples.tutorials.mnist import input_data
# 载入tensorflow
import tensorflow as tf
mnist = input_data.read_data_sets("MNIST_data/",one_hot = True)
# 将session注册为默认session
sess = tf.InteractiveSession()

# 输入节点数
in_units = 784
# 隐含层节点数
h1_units = 300
# variable存储模型参数
# Variable长期存在并且每轮更新
# 隐含层初始化为截断的正态分布，标准差为0.1
w1 = tf.Variable(tf.truncated_normal([in_units,h1_units],stddev = 0.1))
# 隐含层偏置用0初始化
b1 = tf.Variable(tf.zeros([h1_units]))
# 输出层用0初始化，且维数为10
w2 = tf.Variable(tf.zeros([h1_units,10]))
b2 = tf.Variable(tf.zeros([10]))
# placeholder定义输入数据的地方
# 数据类型；[个数，数据维数]，none表示不限
x = tf.placeholder(tf.float32,[None,in_units])
# dropout保留节点比率
# dropout是防止过拟合的trick
keep_prob = tf.placeholder(tf.float32)

# 隐含层结构
# relu激活函数是梯度弥散的trick
hidden1 = tf.nn.relu(tf.matmul(x,w1)+b1)
# 对隐含层输出进行dropout
hidden1_drop = tf.nn.dropout(hidden1,keep_prob)
# 输出层结构
y = tf.nn.softmax(tf.matmul(hidden1_drop,w2)+b2)

# y_存储真实label
y_ = tf.placeholder(tf.float32,[None,10])
# 定义损失函数
# reduce_sum求和，reduce_mean对每一个batch数据求均值
cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y),reduction_indices = [1]))
# 只要定义好优化目标（损失函数）和优化算法，tensorflow就会自动求导进行反向传播
train_step = tf.train.AdagradOptimizer(0.3).minimize(cross_entropy)

# 全局参数初始化器
tf.global_variables_initializer().run()
# 前面只是定义了优化目标和优化算法，batch里必须feed数据才能进行训练
# 定义迭代次数和batch数据
for i in range(3000):
    batch_xs,batch_ys = mnist.train.next_batch(100)
    train_step.run({x:batch_xs,y_:batch_ys,keep_prob:0.75})
    
#测试
correct_prediction = tf.equal(tf.argmax(y,1),tf.argmax(y_,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
print accuracy.eval({x:mnist.test.images,y_:mnist.test.labels,keep_prob:1.0})

结果