Torch.clamp() does not work properly

1

torch.clamp() does not work properly
The environment I used was google colaboratory

I ran torch.clamp() according to the official documentation
https://pytorch.org/docs/master/generated/torch.clamp.html

working program

print(torch.__version__)
a = torch.randn(4)
print(a)
torch.clamp(a, min=-0.5, max=0.5)
print(a)

The results obtained

1.5.1+cu101
tensor([-0.8323, -0.0443,  0.4641,  0.0824])
tensor([-0.8323, -0.0443,  0.4641,  0.0824])

If it works like the document, the original result is tensor([-0.5, -0.0443, 0.4641, 0.0824])

What’s wrong with that?
Is this a bug?

I want to set constraints on weight parameters like keras.constraints
I can’t do it because torch.clamp() doesn’t work properly.

2

Change

torch.clamp(a, min=-0.5, max=0.5)

To

a = torch.clamp(a, min=-0.5, max=0.5)

Or

a = a.clamp(-0.5, 0.5)

Or

a.clamp_(-0.5, 0.5)
2 Likes