torch.Tensor() with requires_grad parameter

Question

I can't use torch.Tensor() with requires_grad parameter (torch version : 0.4.1)

without requires_grad :

x = torch.Tensor([[.5, .3, 2.1]]) print(x) > tensor([[0.5000, 0.3000, 2.1000]])

with requires_grad=True or requires_grad=False :

x = torch.Tensor([[.5, .3, 2.1]], requires_grad=False) print(x) Traceback (most recent call last): File "D:/_P/dev/ai/pytorch/notes/tensor01.py", line 4, in <module> x = torch.Tensor([[.5, .3, 2.1]], requires_grad=False) TypeError: new() received an invalid combination of arguments - got (list, requires_grad=bool), but expected one of: * (torch.device device) * (torch.Storage storage) * (Tensor other) * (tuple of ints size, torch.device device) didn't match because some of the keywords were incorrect: requires_grad * (object data, torch.device device) didn't match because some of the keywords were incorrect: requires_grad

Tyler Zeller · Accepted Answer · 2018-11-08 05:30:54Z

13

You are creating the tensor x by using the torch.Tensor class constructor which doesn't take the requires_grad flag. Instead, you want to use torch.tensor() (lowercase 't') method

x = torch.tensor([[.5, .3, 2.1]], requires_grad=False)

Edit: adding a link to docs: torch.Tensor

answered Nov 8, 2018 at 5:30

Tyler Zeller

3642 silver badges8 bronze badges

Sign up to request clarification or add additional context in comments.

3 Comments

Eduardo Reis Over a year ago

Wondering why there are torch.tensor and torch.Tensor.

afarley Over a year ago

@EduardoReis I agree, this is very confusing behavior.

András Aszódi Aug 7 at 14:42

@Eduardo Reis, @afarley The torch.Tensor class constructor always creates an uninitialised tensor with the default torch.float32 data type. The torch.tensor() method, which is a factory method, tries to figure out the data type intelligently from its argument and can be used to create properly initialised tensors.

Collectives™ on Stack Overflow

torch.Tensor() with requires_grad parameter

1 Answer 1

3 Comments

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

3 Comments

Related