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