pytorch深度学习实践4——反向传播

利用pytorch实现反向传播，简单代码

# -*- coding: utf-8 -*-
"""
Spyder Editor

This is a temporary script file.
"""

import torch

x_data = [1.0, 2.0, 3.0]
y_data = [2.0, 4.0, 6.0]

w = torch.Tensor([1.0])
w.requires_grad = True

def forward(x):
    return x * w

def loss(x,y):
    y_pred = forward(x)
    return (y_pred - y) ** 2

print("predict (before training)", 4, forward(4).item())


for epoch in range(100):
    for x, y in zip(x_data, y_data):
        l = loss(x,y)
        l.backward()
        print('\tgrad:', x, y, w.grad.item())
        w.data = w.data - 0.01*w.grad.data
        w.grad.data.zero_()
        
    print("progress:", epoch, l.item())
print("predict ( after training)", 4, forward(4).item()) 

课后习题