Tensorflow学习笔记（6）——Logistic实现

Logistic回归可以实现分类的功能，本文以mnist手写体数字识别为例，用Tensorflow实现logistic回归进行识别。 
代码注释已经非常详细了，这里不在多介绍。

#coding=utf-8
'''
Created on 2016年4月23日
Logistic回归Tensorflow代码实现，mnist手写体数据库识别
@author: hanchao
'''

import tensorflow as tf
import numpy as np
#input_data文件为我自己的读取mnist数据库代码文件
import input_data

readData = input_data.read_data_sets("/home/hanchao/mnist" , one_hot=True)
learning_rate = 0.01
#[None,784]则代表行数不确定，列数为784(784 = 28*28，mnist数据库图像为28*28）
x = tf.placeholder("float",[None,784])
y = tf.placeholder("float",[None,10])

W = tf.Variable(tf.zeros([784,10], "float", "weight"))
b = tf.Variable(tf.zeros([10],"float","bais"))
#activation为softmax结果矩阵，每行中最大的下标即为预测结果
activation = tf.nn.softmax(tf.matmul(x, W)+b)
#cost结果为互熵cross entropy
cost = tf.reduce_mean(-tf.reduce_sum(y*tf.log(activation),1))
#梯度下降法最小化互熵
opt = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost)

init = tf.initialize_all_variables()

with tf.Session() as sess:
    sess.run(init)
    #train过程
    for step in range(25):
        avg_cost = 0.
        #分为多个batch训练
        totalStep = (int)(readData.train.num_examples/100)
        for i in range(totalStep):
            images,labels = readData.train.next_batch(100)
            sess.run(opt,feed_dict={x:images,y:labels})
            avg_cost += sess.run(cost,feed_dict={x:images,y:labels})/totalStep
        print "Step: ","%04d"%step," cost: ",avg_cost
        
    #test过程
    #读取测试样本
    testImages,testLabels = readData.test.next_batch(200)
    #初始化准确率
    accuracy = 0.
    #获得activation中每行最大值的下标作为预测结果
    predictResult = tf.arg_max(activation, 1)
    #测试样本共200个sample，遍历对每个sample进行测试
    for i in range(200):
        #p为第i个样本的分类结果
        p = sess.run(predictResult[i],feed_dict={x:testImages})
        #label为真实的样本标签
        label = np.argmax(testLabels, 1)
        l = label[i]
        #分类正确
        if(p == l):
            accuracy += 1.0/200.0
        print "Predict Result: ",p," True Label: ",l
    print "accuracy: ",accuracy