简述
用Tf做线性回归有点小题大做了么，其实关键点只是练习优化器的使用，
loss = tf.reduce_mean((y-y_data)**2)
optimizer = tf.train.GradientDescentOptimizer(0.01)
train = optimizer.minimize(loss)
tf的这三句话比自己造轮子至少省了一个月时间。
代码
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
DATA_NUMBER = 100
#1 构造数据集
x_data = np.linspace(-5,5,DATA_NUMBER)
y_data = 0.37*X + +2.7+1.6*np.sin(np.random.random((DATA_NUMBER)))*np.cos(np.random.random((DATA_NUMBER)))
#2 构造graph
w = tf.Variable(np.random.random(),dtype=np.float32)
b = tf.Variable(np.random.random(),dtype=np.float32)
y = w*x_data + b
#3 优化器
loss = tf.reduce_mean((y-y_data)**2)
optimizer = tf.train.GradientDescentOptimizer(0.01)
train = optimizer.minimize(loss)
#4 会话
with tf.Session() as sess:
init_var = tf.global_variables_initializer()
sess.run(init_var)
print(w.eval(),b.eval(),loss.eval())
for i in range(50000):
sess.run(train)
w_value = w.eval()
b_value = b.eval()
if i%5000 == 0:print(w_value,b_value,loss.eval())
#5 展示
plt.scatter(x_data,y_data,c='r')
plt.plot(x_data,w_value*x_data + b_value)
展示
线性回归
