梯度反传

pytorch 梯度反传详细说明：

import torch
w = torch.tensor([1.], requires_grad=True)
x = torch.tensor([2.], requires_grad=True)
a = torch.add(w, x)
b = torch.add(w, 1)
y = torch.mul(a, b)

print(w)
print(x)
print(a)
print(b)
print(y)
  
y.backward(retain_graph=True)
print(w.grad)
y.backward()
print(w.grad)

结果是：

 

import torch
w = torch.tensor([1.], requires_grad=True)
x = torch.tensor([2.], requires_grad=True)

a = torch.add(w, x)
b = torch.add(w, 1)

y0 = torch.mul(a, b)
y1 = torch.add(a, b)

loss = torch.cat([y0, y1], dim=0)
grad_tensors = torch.tensor([1., 1.])
loss.backward(gradient=grad_tensors)  # gradient 传入torch.autograd.backward()中的grad_tensors
print(w.grad)

结果：

 

转载自：https://blog.csdn.net/weixin_42154841/article/details/108503148