tensorflow mnist一层网络的逻辑回归的手写数字识别

下面是根据一个学习视频，自己写的一个一层网络的tensorflow来训练手写数字识别的逻辑回归的训练模型，一些理解见代码的注释，直接上代码，方便自己以后回顾。

from tensorflow.examples.tutorials.mnist import input_data
import tensorflow as tf
"""
*****************定义操作******************
"""
# 定义了像素总值，和lable的种类数
x = tf.placeholder("float", [None, 784])
y = tf.placeholder("float", [None, 10])

w = tf.Variable(tf.zeros([784, 10]))
b = tf.Variable(tf.zeros([10]))

# sofmax 将而分类回归为10类的模型
actv = tf.nn.softmax(tf.matmul(x, w) + b)

# reduction_indices 0 是按列压 1时 是横向压缩 https://www.cnblogs.com/likethanlove/p/6547405.html
# y =[0, 1, 0, 0, 0, 0, 0, 0, 0, 0]  y*tf.log(actv) -> one values
# softmax 求与label的差值为-ylog(actv)
cost = tf.reduce_mean(-tf.reduce_sum(y*tf.log(actv), reduction_indices=1))

# 梯度下降，让cost最小
learn_rate = 0.01
optm = tf.train.GradientDescentOptimizer(learn_rate)
train = optm.minimize(cost)

# tf.arg_max(actv, 1) 取actv（计算的y）最大值 与 label中的值是否相同，相同则转float为1，不同为0，
# 相加求平均则为准确率
pre = tf.equal(tf.arg_max(actv, 1), tf.arg_max(y, 1))
accr = tf.reduce_mean(tf.cast(pre, "float"))

init = tf.global_variables_initializer()
# 读取data值
mnist = input_data.read_data_sets("data/", one_hot=True)

"""
****************
"""
# 迭代次数，即训练所有图片的次数
training_epochs = 50
# 每次处理的图片个数
batch_size = 100
# 每迭代5次 显示训练效果
display_step = 5

with tf.Session() as sess:
    # 变量初始化
    sess.run(init)


    for epoch in range(training_epochs):
        # 迭代一次的batch数 ， 并读取一次处理的batch的数据batch_xs, batch_ys
        num_batch = int(mnist.train.num_examples / batch_size)
        batch_xs, batch_ys = mnist.train.next_batch(batch_size)

        sum_cost = 0
        for i in range(num_batch):
            feed_train = {x: batch_xs, y: batch_ys}
            sess.run(train, feed_dict=feed_train)
            sum_cost += sess.run(cost, feed_dict=feed_train)
        # 求一个batch训练后的平均损失率
        avg_cost = sum_cost / num_batch

        # 每隔迭代5次 dispaly，显示训练效果
        if epoch % display_step == 0:  
            # 测试训练数据正确率时，是将当前batch的训练数据来计算正确率         
            train_acc = sess.run(accr, feed_dict=feed_train)
            # 测试正确率时，是将整个测试数据跑完，算正确率
            feed_test = {x: mnist.test.images, y: mnist.test.labels}
            test_acc = sess.run(accr, feed_dict=feed_test)
            print("Epoch: %03d/%03d cost: %.9f train_acc: %.3f test_acc: %.3f,"
                  % (epoch, training_epochs, avg_cost, train_acc, test_acc))
       print(w.eval())
       print(b.eval())