liang1232015的博客

backward函数，如果没有参数，则默认是标量比如：x = Variable(torch.ones(2, 2), requires_grad=True)y = x + 2z = y * y * 3out = z.mean()out.backward()则，此时可以或得到x,y,z的outgrade值。而，backward实际参数，则是和调用其的tensor保持相同的维度，例如x = torch....

一个简单的例子import torchfrom torch.autograd import Variableimport torch.nn as nnimport torch.nn.functional as Fclass Net(nn.Module): def __init__(self): super(Net, self).__init__() ...