Why does torch.gt function turn requires_grad into False?

Question

The requires_grad of tensor b and c are True. But the requires_grad of tensor d is False. I am curious why this change happens because all the requires_grad of inputs are True.

However, the requires_grad of tensor e is True. I can still do backward() on e. But is there an error in this way?

I am using Python3.7 and Pytorch1.1.

import torch import torch.nn as nn net = nn.Conv2d(1, 1, 3, padding=1) a = torch.randn(1, 1, 10, 10) b = net(a) c = net(b) d = torch.gt(b, c) e = b - c e[e > 0] = 1.0 e[e < 0] = 0.0

eugen · Accepted Answer · 2019-09-09 04:40:15Z

I assume this is because you cannot take a gradient of greater than operation. The return type is boolean:

>>> torch.gt(torch.tensor([[1, 2], [3, 4]]), torch.tensor([[1, 1], [4, 4]])) tensor([[False, True], [False, False]])

Whereas minus or other arithmetic operation returns another numeral.

Collectives™ on Stack Overflow

Why does torch.gt function turn requires_grad into False?

1 Answer 1

Comments

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Related