Confusing about torch.Tensor and torch.tensor!

1

torch.Tensor(2) returns tensor([-4.9389e+23, 4.5758e-41]), while torch.tensor(2) returns tensor(2). I coud’t find the manual for torch.Tensor, and it’s very confusing of these two object constructor. What’s the reason to define ‘torch.Tensor’ when ‘torch.tensor’ is already defined?

2

torch.Tensor uses uninitialized memory, so you shouldn’t use it.
torch.tensor creates a new tensor using the passed values.

3

I found the usage in torch.nn.batchnorm

        self.weight = Parameter(torch.Tensor(num_features))

So torch.Tensor is not recommended?

4

Yes, don’t use it.
nn.BatchNorm calls reset_parameters() afterwards, which will initialize these tensors.

5

Ok, thanks for the reply!