使用tensorflow实现线性回归

# -*- coding: UTF-8 -*-

'''使用tensorflow实现线性回归
reference：https://www.cnblogs.com/selenaf/p/9102398.html
思路：在数据上选择一条直线y=Wx+b，在这条直线附近随机生成一些数据点，让TensorFlow建立回归模型，去学习什么样的W和b能更好去拟合这些数据点。
'''

'''
1）随机生成1000个数据点，围绕在y=0.1x+0.3 周围，设置W=0.1，b=0.3，届时看构建的模型是否能学习到w和b的值。
'''
import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt
num_points=1000
vectors_set=[]
for i in range(num_points):
    x1=np.random.normal(0.0,0.55)   #横坐标，进行随机高斯处理化，以0为均值，以0.55为标准差
    y1=x1*0.1+0.3+np.random.normal(0.0,0.03)   #纵坐标，数据点在y1=x1*0.1+0.3上小范围浮动
    vectors_set.append([x1,y1])
    x_data=[v[0] for v in vectors_set]
    y_data=[v[1] for v in vectors_set]
plt.scatter(x_data,y_data,c='r')
plt.show()

'''
2）构造线性回归模型，学习上面的数据是符合一个怎么样的W和b
'''
W = tf.Variable(tf.random_uniform([1], -1.0, 1.0), name='W')  # 生成1维的W矩阵，取值是[-1,1]之间的随机数
b = tf.Variable(tf.zeros([1]), name='b') # 生成1维的b矩阵，初始值是0
y = W * x_data + b     # 经过计算得出预估值y
loss = tf.reduce_mean(tf.square(y - y_data), name='loss') # 以预估值y和实际值y_data之间的均方误差作为损失
optimizer = tf.train.GradientDescentOptimizer(0.5) # 采用梯度下降法来优化参数  学习率为0.5
train = optimizer.minimize(loss, name='train')  # 训练的过程就是最小化这个误差值
sess = tf.Session()
init = tf.global_variables_initializer()
sess.run(init)
print ("W =", sess.run(W), "b =", sess.run(b), "loss =", sess.run(loss))  # 初始化的W和b是多少
for step in range(20):   # 执行20次训练
  sess.run(train)
  print ("W =", sess.run(W), "b =", sess.run(b), "loss =", sess.run(loss)) # 输出训练好的W和b

'''
打印每一次结果，如下，随着迭代进行，训练的W、b越来越接近0.1、0.3，说明构建的回归模型确实学习到了之间建立的数据的规则。
loss一开始很大，后来慢慢变小，说明模型表达效果随着迭代越来越好。

W = [-0.68837476] b = [0.] loss = 0.29411852
W = [-0.43151537] b = [0.3049943] loss = 0.09167927
W = [-0.26009744] b = [0.30306205] loss = 0.042414192
W = [-0.14415416] b = [0.30177253] loss = 0.019876134
W = [-0.06573282] b = [0.30090034] loss = 0.0095652975
W = [-0.01269045] b = [0.30031043] loss = 0.004848239
W = [0.02318618] b = [0.2999114] loss = 0.0026902542
W = [0.04745229] b = [0.29964152] loss = 0.0017030073
W = [0.06386533] b = [0.29945898] loss = 0.0012513561
W = [0.07496673] b = [0.2993355] loss = 0.0010447322
W = [0.08247545] b = [0.299252] loss = 0.00095020473
W = [0.08755418] b = [0.29919553] loss = 0.0009069599
W = [0.09098931] b = [0.29915732] loss = 0.000887176
W = [0.09331276] b = [0.29913148] loss = 0.0008781252
W = [0.09488428] b = [0.299114] loss = 0.00087398454
W = [0.09594722] b = [0.2991022] loss = 0.0008720903
W = [0.09666617] b = [0.29909417] loss = 0.0008712237
W = [0.09715245] b = [0.29908878] loss = 0.0008708272
W = [0.09748136] b = [0.2990851] loss = 0.00087064586
W = [0.09770382] b = [0.29908264] loss = 0.0008705628
W = [0.09785429] b = [0.29908097] loss = 0.00087052496
'''