Problem with data loaded on GPU

1

Hi everyone.
A newbie issue.

I’m using CrossEntropyLoss with data on gpu but I still getting this error:
RuntimeError: dimension out of range (expected to be in range of [-1, 0], but got 1).

I think that my primitive way to load data on gpu is making this problem because when I run neural network on data from cpu everything is working correctly.

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

xy = np.loadtxt('.csv',
                        delimiter=',', dtype=np.float32)

x_datai = torch.Tensor(xy[:, 0:-1]).to(device)
y_datai = torch.Tensor(xy[:, [-1]]).to(device)

train_loader =TensorDataset(x_datai,y_datai)


-------------------

for i, data in enumerate(train_loader):
   inputs,target = data
   
   target = target.squeeze(1)
   
   optimizer.zero_grad()
   output = model(inputs).to(device)
   
   loss = criterion(output, target.long())
   loss.backward()
   optimizer.step()

I tried to solve this on my own a long time but it’s too much for me.
Can someone explain me how the dataloader should look to not making this issue?

Thanks for help.

2

Could you check the shapes of inputs and target?
I guess both might be missing the batch dimension at dim0.
If that’s the case, you could try to apply .unsqueeze(0) on both of them.

3

Now I getting this:
return torch._C._nn.nll_loss(input, target, weight, size_average, ignore_index, reduce)
RuntimeError: multi-target not supported at c:\pytorch\aten\src\thcunn\generic/ClassNLLCriterion.cu:16

4

Sorry for the confusion. The target shouldn’t be unsqueezed for nn.CrossEntropyLoss or nn.NLLLoss.

5

It’s working perfectly!
Thank You!