TensorFlow——LinearRegression简单模型代码

代码函数详解

tf.random.truncated_normal()函数

tf.truncated_normal函数随机生成正态分布的数据，生成的数据是截断的正态分布，截断的标准是2倍的stddev。

zip()函数

zip() 函数用于将可迭代对象作为参数，将对象中对应的元素打包成一个个元组，然后返回由这些元组组成的对象。如果各个可迭代对象的元素个数不一致，则返回的对象长度与最短的可迭代对象相同。利用 * 号操作符，与zip相反，进行解压。

1import tensorflow as tf
2import numpy as np
3import matplotlib.pyplot as plt
4
5train_x = np.linspace(-5, 3, 50)
6train_y = train_x * 5 + 10 + np.random.random(50) * 10 - 5
7
8plt.plot(train_x, train_y, 'r.')
9plt.grid(True)
10plt.show()
11
12X = tf.placeholder(dtype=tf.float32)
13Y = tf.placeholder(dtype=tf.float32)
14
15w = tf.Variable(tf.random.truncated_normal([1]), name='Weight')
16b = tf.Variable(tf.random.truncated_normal([1]), name='bias')
17
18z = tf.multiply(X, w) + b
19
20cost = tf.reduce_mean(tf.square(Y - z))
21learning_rate = 0.01
22optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost)
23
24init = tf.global_variables_initializer()
25
26training_epochs = 20
27display_step = 2
28
29
30with tf.Session() as sess:
31    sess.run(init)
32    loss_list = []
33    for epoch in range(training_epochs):
34        for (x, y) in zip(train_x, train_y):
35            sess.run(optimizer,feed_dict={X:x, Y:y})
36
37        if epoch % display_step == 0:
38            loss = sess.run(cost, feed_dict={X:x, Y:y})
39            loss_list.append(loss)
40            print('Iter: ', epoch, ' Loss: ', loss)
41
42    w_, b_ = sess.run([w, b], feed_dict={X: x, Y: y})
43    print(" Finished ")
44    print("W: ", w_, " b: ", b_, " loss: ", loss)
45    plt.plot(train_x, train_x*w_ + b_, 'g-', train_x, train_y, 'r.')
46    plt.grid(True)
47    plt.show()