CODEWITHSUNDEEP
pythonpytorchtensor
infiNity9819
infiNity9819

Reputation: 598

torch.Tensor() new() received an invalid combination of arguments - got (list, dtype=torch.dtype)

I am facing a very weird error when I am trying to run the following simple line :

a = torch.Tensor([0,0,0],dtype = torch.int64)

TypeError: new() received an invalid combination of arguments - got (list, dtype=torch.dtype), but expected one of:
 * (*, torch.device device)
      didn't match because some of the keywords were incorrect: dtype
 * (torch.Storage storage)
 * (Tensor other)
 * (tuple of ints size, *, torch.device device)
 * (object data, *, torch.device device)

whereas if we look at official documentation

torch.tensor(data, dtype=None, device=None, requires_grad=False, pin_memory=False) → Tensor

Parameters

data (array_like) – Initial data for the tensor. Can be a list, tuple, NumPy ndarray, scalar, and other types.

dtype (torch.dtype, optional) – the desired data type of returned tensor. Default: if None, infers data type from data.

Why is the code snippet not working even though the official documentation supports it?

Upvotes: 10

Views: 25094

Answers (2)

Don Feto
Don Feto

Reputation: 1524

As Jan's answer said but here I will explain why Tensor different from tensor

torch.Tensor is an alias for the default tensor type (torch.FloatTensor). which is a CPU tensor, 32-bit floating-point so the type and device are predetermined.

#This will cause an Error, because Tensor has no dtype and device properties on it's constructor

example_tensor = torch.Tensor(
    [
     [[1, 2], [3, 4]], 
     [[5, 6], [7, 8]], 
     [[9, 0], [1, 2]]
    ]
, dtype=torch.float32, device=device)

To move a tensor to a new device, you can write new_tensor = example_tensor.to(device) where the device will be either CPU or Cuda. and to set new type you can use new_tensor.dtype=torch.float64 for example

     example_tensor = torch.Tensor(
    [
     [[1, 2], [3, 4]], 
     [[5, 6], [7, 8]], 
     [[9, 0], [1, 2]]
    ]
)
    new_tensor = example_tensor.to(device)
    new_tensor.dtype  #set or get dtype

this is from the documentation

A tensor (case senstive tensor not Tensor) of specific data type can be constructed by passing a torch.dtype and/or a torch.device to a constructor or tensor creation op:

cuda0 = torch.device('cuda:0')
torch.ones([2, 4], dtype=torch.float64, device=cuda0)

.

Upvotes: 4

Jan Pokorný
Jan Pokorný

Reputation: 1868

Capitalization matters -- in your top example, you have Tensor with uppercase T, but the documentation excerpt is talking about tensor with lowercase t.

Upvotes: 15

Related Questions