TensorFlow 2.0 最基础的线性回归

import tensorflow as tf
import numpy as np

opt=tf.compat.v1.train.AdamOptimizer(1e-1)

input_xs=np.random.rand(1000)
input_ys=3*input_xs+0.217

# y=3*x+0.217
# 使用tensorflow,numpy 迭代计算线性函数

weight=tf.Variable(1.0,dtype=tf.float32,name="weight")
bias=tf.Variable(1.,dtype=tf.float32,name="bias")

def model(xs):
    logits=tf.multiply(xs,weight)+bias
    return logits

for xs,ys in zip(input_xs,input_ys):
    xs=np.reshape(xs,[1])
    ys=np.reshape(ys,[1])
    with tf.GradientTape() as tape:
        _loss=tf.reduce_mean(tf.pow(model(xs)-ys,2))/(2*1000)
    grads=tape.gradient(_loss,[weight,bias])
    opt.apply_gradients(zip(grads,[weight,bias])) 
    #attention the s after the word apply_gradient
    print('Training loss is :',_loss.numpy())
    
print(weight)
print(bias)