TF——深度学习逻辑回归源码（加各种用到算法的解析）

解析在注释里：
 1 import tensorflow as tf
 2 from tensorflow.examples.tutorials.mnist import input_data
 3 import os
 4 os.environ["CUDA_VISIBLE_DEVICES"]="0"
 5 mnist = input_data.read_data_sets("./data/MNIST_data/", one_hot=True)
 6 learning_rate=0.01
 7 training_epochs=25
 8 batch_size=100  #一次训练所选取的样本数。
 9 display_step=1
10 
11 if __name__ =='__main__':
12     x=tf.placeholder(tf.float32,[None,784])
13     y=tf.placeholder(tf.float32,[None,10])
14 
15     print("PPPPPP")
16     batch_xs, batch_ys = mnist.train.next_batch(batch_size)
17     print(batch_xs.shape)
18     print("LLLLLLLLLLLL")
19     print(batch_ys.shape)
20 
21     '''Variable 初始化'''
22     W=tf.Variable(tf.zeros([784,10]))
23     b=tf.Variable(tf.zeros([10]))
24 
25     '''逻辑回归模型'''
26     pred=tf.nn.softmax(tf.matmul(x,W)+b)
27 
28     '''构造代价函数'''
29     '''
30     tf.reduce_mean 函数用于计算张量tensor沿着指定的数轴（tensor的某一维度）上的的平均值，主要用作降维或者计算tensor（图像）的平均值。
31     reduce_mean(input_tensor,
32                     输入的待降维的tensor
33                     axis=None,
34                     指定的轴，如果不指定，则计算所有元素的均值
35                     keep_dims=False,
36                     是否降维度
37                     name=None,
38                     操作的名称
39                     reduction_indices=None
40                     用来指定轴
41                     )
42     '''
43     cost = tf.reduce_mean(-tf.reduce_sum(y * tf.log(pred), reduction_indices=1))
44 
45     '''梯度下降最小值'''
46     optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost)
47 
48     init = tf.global_variables_initializer()
49     with tf.Session() as sess:
50         sess.run(init)
51 
52         '''训练模型'''
53         for epoch in range(training_epochs):
54             avg_cost = 0
55         total_batch = int(mnist.train.num_examples / batch_size)
56         for i in range(total_batch):
57             batch_xs, batch_ys = mnist.train.next_batch(batch_size)
58             _, c = sess.run([optimizer, cost], feed_dict={x: batch_xs, y: batch_ys})
59             # Conpute average loss
60             avg_cost += c / total_batch
61         if (epoch + 1) % display_step == 0:
62             print("第:", '%04d步' % (epoch + 1), "标准差:", "{:.09f}".format(avg_cost))
63 
64 
65         '''测试模型'''
66         '''
67         argmax(f(x))是使得 f(x)取得最大值所对应的变量点x(或x的集合)
68         test = np.array([
69             [1, 2, 3],
70              [2, 3, 4], 
71              [5, 4, 3], 
72              [8, 7, 2]])
73             np.argmax(test, 0)　　　＃输出：array([3, 3, 1]
74             np.argmax(test, 1)　　　＃输出：array([2, 2, 0, 0]
75 
76             axis = 0:
77             　　axis=0时比较每一列的元素，将每一列最大元素所在的索引记录下来，最后输出每一列最大元素所在的索引数组。
78             axis = 1:
79     　　axis=1的时候，将每一行最大元素所在的索引记录下来，最后返回每一行最大元素所在的索引数组。
80         '''
81         correct_prediction = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))
82         accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
83         print("准确的:", accuracy.eval({x: mnist.test.images[:3000], y: mnist.test.labels[:3000]}))