CODEWITHSUNDEEP
pytorchtensor
Hal
Hal

Reputation: 372

Getting error when trying to torch stack a list of tensors

I am trying to get cuda to work but I need to change my training input into a tensor. When I tried to do that, I am getting an error when I tried to stack a list of tensor into one tensor.

Code

for epoch in:
        alst = []
        for x, y in loader:
            x = torch.stack(x)
            #x = torch.Tensor(x)
            #x = torch.stack(x).to(device,dtype=float)

Shape of x: List of tensors

[tensor([[[0.325],
     [ 0.1257],
     [ 0.1149],
     ...,
     [-1.572],
     [-1.265],
     [-3.574]],
]), tensor([1,2,3,4,5]), tensor(6,5,4,3,2])]

Error I got

     22             alst = []
     23             for x, y in loader:
---> 24                 x_list = torch.stack(x)
     25 #                 x = torch.Tensor(x)
     26 #                 x = torch.stack(x).to(device,dtype=float)

RuntimeError: Expected object of scalar type Float but got scalar type Long for sequence element 1 in sequence argument at position #1 'tensors'

Not sure what I am doing wrong. I tried x = torch.stack(x).to(device,dtype=float) as well but it still didn't work.

Upvotes: 0

Views: 1016

Answers (1)

Szymon Maszke
Szymon Maszke

Reputation: 24884

First tensor in your output is of float type with values to input your network with, second looks like labels (of type long).

Furthermore the first one is a tensor while the second and third element are vectors (with 6 and 9 elements respectively).

You cannot stack tensors of different shape hence this won't work no matter the types.

Unpack your x via 

matrix, vector1, vector2 = x

To remove type warnings cast vector1 and vector2 to float via 

vector1 = vector1.float()

Check their shapes via .shape attribute and act accordingly. Probably though you already have batches of data as you are using train_loader. See: DataLoader documentation for more information and check whether you are using one.

Upvotes: 1

Related Questions