Torch learning notes (3) -- univariate linear regression model (self training)

import torch
import numpy as np
import matplotlib.pyplot as plt

# ( One yuan ) Linear regression 
#  Linear regression is a method of analyzing the relationship between one variable and other variables 
#  Solving steps ：1、 determine model（y = wx + b）;2、 choice Loss function （MSE, Cross entropy ）;3、 Set up LR、 solve grad Used to update parameters w,b
# w = w-LR*W.grad b = b-LR*b.grad

#  Case study 
#  Set training parameters 
lr = 0.1  #  Learning rate 
x = torch.rand((20, 1)) * 10  #  adopt torch Create arguments x,x by 20 Distributed in [0,10） The number tensor
print(x.shape)
y = 2 * x + (5 + torch.randn((20, 1)))  #  adopt x establish y, here ,w=2,b = 5,torch.randn((20, 1)) = noise

#  Set linear regression parameters 
w = torch.randn((1), requires_grad=True)  #  Initialize by normal distribution W, Turn on automatic derivation 
b = torch.zeros((1), requires_grad=True)  #  adopt 0 initialization b, Turn on automatic derivation 

#  iteration 
for iteration in range(200):
    # forward y_pred = w*x + b
    wx = torch.mul(w, x)
    y_pred = torch.add(wx, b)
    # Loss(MSE)
    loss = (0.5 * (y - y_pred) ** 2).mean()
    # back
    loss.backward()
    # up data
    b.data.sub_(lr * b.grad)
    w.data.sub_(lr * w.grad)
    #  visualization 
    if iteration % 20 == 0:
        plt.scatter(x.data.numpy(), y.data.numpy())
        plt.plot(x.data.numpy(), y_pred.data.numpy(), r'-', lw=5)
        # plt.text(2, 20, 'loss=%.4f' % loss.data.numpy(), fontdict={'size': 20, 'color': 'red'})
        plt.xlim(1.5, 10)
        plt.pause(0.5)
        #  Stop conditions 
        if loss.data.numpy()<1:
            break