[FIXED] RuntimeError: Found dtype Char but expected Float

Issue

I am using PyTorch in my program(Binary Classification).

The output from my model and actual labels are

model outputs are: (tensor([[0.4512],
        [0.2273],
        [0.4710],
        [0.2965]], grad_fn=<SigmoidBackward0>), torch.float32), 
actuall labels are (tensor([[0],
        [1],
        [0],
        [1]], dtype=torch.int8), torch.int8)

When I calculate the Binary Cross Entropy, it gives me the error

RuntimeError: Found dtype Char but expected Float

I have no idea how it is finding the Char dtype.

Even If calculate it manually, it gives me this error.

import torch
cri = torch.nn.BCELoss()
cri(torch.tensor([[0.4470],[0.5032],[0.3494],[0.5057]], dtype=torch.float), torch.tensor([[0],[1],[0],[0]], dtype=torch.int8))

My DataLoader is

# CREATING DATA LOADER

class MyDataset(torch.utils.data.Dataset):
    def __init__(self, dataframe, subset='train'):
        self.subset = subset
        
        self.dataframe = dataframe
        self.transforms = transforms.Compose([
                                        transforms.RandomResizedCrop(224),
                                        transforms.RandomHorizontalFlip(),
                                        transforms.Grayscale(),
                                        transforms.ToTensor(),
                                        transforms.Normalize((0.5), (0.5))
                                    ])
        
    
    def __len__(self):
        return len(self.dataframe)

    def __getitem__(self, index):
        row = self.dataframe.iloc[index]
        img = Image.open(os.path.join('/kaggle/input/mura-v11',row['path']))
        
        if self.subset=='train':
#             print(torch.tensor(0, dtype=torch.int8) if row['labels'] == 'negative' else torch.tensor(1, dtype=torch.int8))
            return (self.transforms(img), torch.tensor(0, dtype=torch.int8) if row['labels'] == 'negative' else torch.tensor(1, dtype=torch.int8))
        else:
            tensor_img = torchvision.transforms.functional.to_tensor(img)
            return (tensor_img, torch.tensor(0, dtype=torch.int8) if row['labels'] == 'negative' else torch.tensor(1, dtype=torch.int8))

my training loop is

def train_model(model, criterion, optimizer, scheduler, num_epochs=25):
    since = time.time()

    best_model_wts = copy.deepcopy(model.state_dict())
    best_acc = 0.0

    for epoch in range(num_epochs):
        print(f'Epoch {epoch}/{num_epochs - 1}')
        print('-' * 10)

        # Each epoch has a training and validation phase
        for phase in ['train', 'val']:
            if phase == 'train':
                model.train()  # Set model to training mode
            else:
                model.eval()   # Set model to evaluate mode

            running_loss = 0.0
            running_corrects = 0

            # Iterate over data.
            for inputs, labels in dataloaders[phase]:
                inputs = inputs.to(device)
                labels = labels.to(device)

                # zero the parameter gradients
                optimizer.zero_grad()

                # forward
                # track history if only in train
                with torch.set_grad_enabled(phase == 'train'):
                    outputs = model(inputs)
                    _, preds = torch.max(outputs, 1)
                    print(f"model outputs are: {outputs, outputs.dtype}, \nmodel labels are {labels.view(labels.shape[0],1), labels.dtype}")
                    loss = criterion(outputs, labels.view(labels.shape[0], 1))

                    # backward + optimize only if in training phase
                    if phase == 'train':
                        loss.backward()
                        optimizer.step()

                # statistics
                running_loss += loss.item() * inputs.size(0)
                running_corrects += torch.sum(preds == labels.data)
            if phase == 'train':
                scheduler.step()

            epoch_loss = running_loss / dataset_sizes[phase]
            epoch_acc = running_corrects.double() / dataset_sizes[phase]

            print(f'{phase} Loss: {epoch_loss:.4f} Acc: {epoch_acc:.4f}')

            # deep copy the model
            if phase == 'val' and epoch_acc > best_acc:
                best_acc = epoch_acc
                best_model_wts = copy.deepcopy(model.state_dict())

        print()

    time_elapsed = time.time() - since
    print(f'Training complete in {time_elapsed // 60:.0f}m {time_elapsed % 60:.0f}s')
    print(f'Best val Acc: {best_acc:4f}')

    # load best model weights
    model.load_state_dict(best_model_wts)
    return model

And my Model is

class MuraModel(torch.nn.Module):
    def __init__(self):
        """
        In the constructor we instantiate four parameters and assign them as
        member parameters.
        """
        super().__init__()
        self.inp = torch.nn.Conv2d(1, 3, 3) # Change the num of channels to 3
        self.backbone = models.resnet18(pretrained=True)
        num_ftrs = self.backbone.fc.in_features
        self.backbone.fc = nn.Linear(num_ftrs, 1)
        self.act = nn.Sigmoid()

    def forward(self, x):
        """
        In the forward function we accept a Tensor of input data and we must return
        a Tensor of output data. We can use Modules defined in the constructor as
        well as arbitrary operators on Tensors.
        """
        three_channel = self.inp(x)
        back_out = self.backbone(three_channel)
        return self.act(back_out)



# inp = nn.Conv2d(1, 3, 3)
# model_ft = models.resnet18(pretrained=True)(inp)
# num_ftrs = model_ft.fc.in_features
# model_ft.fc = nn.Linear(num_ftrs, 2)

# model_ft = model_ft.to(device)




criterion = nn.BCELoss()
model = MuraModel()

optimizer_ft = optim.SGD(model.parameters(), lr=0.001, momentum=0.9)

# Decay LR by a factor of 0.1 every 7 epochs
exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=7, gamma=0.1)

model = train_model(model, criterion, optimizer_ft, exp_lr_scheduler,
                       num_epochs=25)

How to overcome it.

EDIT

Trace back on train_model function:

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
/tmp/ipykernel_17/2718774237.py in <module>
      1 model = train_model(model, criterion, optimizer_ft, exp_lr_scheduler,
----> 2                        num_epochs=25)

/tmp/ipykernel_17/2670448577.py in train_model(model, criterion, optimizer, scheduler, num_epochs)
     33                     _, preds = torch.max(outputs, 1)
     34                     print(f"model outputs are: {outputs, outputs.dtype}, \nmodel labels are {labels.view(labels.shape[0],1), labels.dtype}")
---> 35                     loss = criterion(outputs, labels.view(labels.shape[0], 1))
     36 
     37                     # backward + optimize only if in training phase

/opt/conda/lib/python3.7/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1108         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1109                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1110             return forward_call(*input, **kwargs)
   1111         # Do not call functions when jit is used
   1112         full_backward_hooks, non_full_backward_hooks = [], []

/opt/conda/lib/python3.7/site-packages/torch/nn/modules/loss.py in forward(self, input, target)
    610 
    611     def forward(self, input: Tensor, target: Tensor) -> Tensor:
--> 612         return F.binary_cross_entropy(input, target, weight=self.weight, reduction=self.reduction)
    613 
    614 

/opt/conda/lib/python3.7/site-packages/torch/nn/functional.py in binary_cross_entropy(input, target, weight, size_average, reduce, reduction)
   3063         weight = weight.expand(new_size)
   3064 
-> 3065     return torch._C._nn.binary_cross_entropy(input, target, weight, reduction_enum)
   3066 
   3067 
RuntimeError: Found dtype Char but expected Float

Trace back on calculating loss individually

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
/tmp/ipykernel_17/4156819471.py in <module>
      1 import torch
      2 cri = torch.nn.BCELoss()
----> 3 cri(torch.tensor([[0.4470],[0.5032],[0.3494],[0.5057]], dtype=torch.float), torch.tensor([[0],[1],[0],[0]], dtype=torch.int8))

/opt/conda/lib/python3.7/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1108         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1109                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1110             return forward_call(*input, **kwargs)
   1111         # Do not call functions when jit is used
   1112         full_backward_hooks, non_full_backward_hooks = [], []

/opt/conda/lib/python3.7/site-packages/torch/nn/modules/loss.py in forward(self, input, target)
    610 
    611     def forward(self, input: Tensor, target: Tensor) -> Tensor:
--> 612         return F.binary_cross_entropy(input, target, weight=self.weight, reduction=self.reduction)
    613 
    614 

/opt/conda/lib/python3.7/site-packages/torch/nn/functional.py in binary_cross_entropy(input, target, weight, size_average, reduce, reduction)
   3063         weight = weight.expand(new_size)
   3064 
-> 3065     return torch._C._nn.binary_cross_entropy(input, target, weight, reduction_enum)
   3066 
   3067 

RuntimeError: Found dtype Char but expected Float

Solution

BCELoss() expects float labels. Yours are int8 (aka char). Converting them to float in the last line of__getitem__()should fix the issue.

Answered By - dx2-66