TensorFlow_Random_linear_modeul

tensorflow_Linear_regression_demo

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt

#產生一個含有隨機的線性模型
x_data = np.random.rand(100) #均勻隨機產生100個點
noise = 0.1*np.random.randn(100) #隨機項
y_data = x_data * 1 + 3 + noise  # y_data = 1 * x_data + 3  + noise

#宣告 tensorflow 中的變數
# y = m*x + b
m = tf.Variable(0.0)
b = tf.Variable(0.0)
y = m*x_data + b

#代價函數 : loss = mean((y-y_data)^2)
#其中 tf.reduce_mean 計算 tensor中每一個 dimension 的平均值
# tf.square 計算 tensor 中每一個元的平方
loss = tf.reduce_mean(tf.square(y_data - y))

#Gradient desent method  (learning rate = 0.1)
gd = tf.train.GradientDescentOptimizer(0.3)

#最小化 代價函數 (operator)
train = gd.minimize(loss)

#初始化變數 operator
init = tf.global_variables_initializer()

with tf.Session() as sess:
    sess.run(init)
    for step in range(50):
        sess.run(train)
        print("iter=", step, ", m=",sess.run(m), ", b=", sess.run(b))
    #將 train 後的結果存下來
    m = sess.run(m);
    b = sess.run(b);
    #繪製結果圖
    plt.figure()
    plt.scatter(x_data, y_data)
    plt.plot([0, 1], [b, m*1+b], '-r', lw=3)
    plt.show()