RuntimeError: expected dtype Float but got dtype Long for my loss function despite converting all tensors to float

vision
1

Creating a basic CNN model for binary image classification. Tried evaluating my model before training but receive the runtime error for dtypes mismatch as the BCE loss function uses it’s weights in float form and it seems as though my inputs are Tensors containing Long data. I don’t believe I ever converted my data into Long but changed all the relevant tensors to float type anyways in the validation step method definition:

    def validation_step(self, batch):
        images, targets = batch 
        targets.to(torch.float32)
        targets = targets.view(-1,1)
        images.to(torch.float32)
        out = self(images)
        out.to(torch.float32)
        loss = F.binary_cross_entropy(out, targets)  # Calculate loss
        score = binary_acc(out, targets)
        return {'val_loss': loss.detach(), 'val_score': score.detach() }

However still receiving the same error message. This is the full notebook https://jovian.ml/moaaz645/malaria-cnn-basic
And this is the error message:

RuntimeError Traceback (most recent call last)
in
1 model = to_device(MalariaCnnModel(), device)
----> 2 evaluate(model, val_dl)

/opt/conda/lib/python3.7/site-packages/torch/autograd/grad_mode.py in decorate_context(*args, **kwargs)
13 def decorate_context(*args, **kwargs):
14 with self:
—> 15 return func(*args, **kwargs)
16 return decorate_context
17

in evaluate(model, val_loader)
2 def evaluate(model, val_loader):
3 model.eval()
----> 4 outputs = [model.validation_step(batch) for batch in val_loader]
5 return model.validation_epoch_end(outputs)
6

in (.0)
2 def evaluate(model, val_loader):
3 model.eval()
----> 4 outputs = [model.validation_step(batch) for batch in val_loader]
5 return model.validation_epoch_end(outputs)
6

in validation_step(self, batch)
17 out = self(images)
18 out.to(torch.float32)
—> 19 loss = F.binary_cross_entropy(out, targets) # Calculate loss
20 score = binary_acc(out, targets)
21 return {‘val_loss’: loss.detach(), ‘val_score’: score.detach() }

/opt/conda/lib/python3.7/site-packages/torch/nn/functional.py in binary_cross_entropy(input, target, weight, size_average, reduce, reduction)
2377
2378 return torch._C._nn.binary_cross_entropy(
-> 2379 input, target, weight, reduction_enum)
2380
2381

RuntimeError: expected dtype Float but got dtype Long

2

The to() operation is not an inplace operation for tensors, so you would need to reassign the result:

targets = targets.to(torch.float32)
...
4 Likes
3

Thank you very much it works fine now. I apologize, my fundamentals are still weak in pytorch as I’ve recently started and it was a silly mistake.

2 Likes