RuntimeError: weight tensor should be defined either for all 4 classes or no classes but got weight tensor of shape: [3]

[HonLZL]

When I set train_batch_size=1, I got this error. It happen in

 n_loss = compute_crossentropy_loss(n_scores.to(device), tg.ndata['label'].to(device)) 
==>  
def compute_crossentropy_loss(scores: torch.Tensor, labels: torch.Tensor):
    w = class_weight.compute_class_weight(class_weight='balanced', classes=np.unique(labels.cpu().numpy()),
                                          y=labels.cpu().numpy())
    return torch.nn.CrossEntropyLoss(weight=torch.tensor(w, dtype=torch.float32).to('cuda:0'))(scores, labels)

RuntimeError: weight tensor should be defined either for all 4 classes or no classes but got weight tensor of shape: [3]

I think it may be because there are only three labels in one image that caused this error. Could you tell me the reason and how to fix it? Thank you so much.

[TranQuocDat0405]

I have the same problem as you, have you fixed this error yet?

[HonLZL]

I have the same problem as you, have you fixed this error yet?

Sorry, I couldn't find a solution. And you?

[HonLZL]

I have the same problem as you, have you fixed this error yet?

Hi， I fixed this bug by changing some code.

First: in utils compute_crossentropy_loss, add

    new_labels = labels.cpu().numpy().tolist()

    if len(set(new_labels)) == 3:
        for i in range(4):
            if i not in new_labels:
                w = w.tolist()
                w.insert(i, len(new_labels))
                break

    if len(set(new_labels)) == 1:
        for i in range(2):
            if i not in new_labels:
                w = w.tolist()
                w.insert(i, len(new_labels))
                break

Second: in utils compute_auc_mc function, changing a line

# from
labels = F.one_hot(labels).cpu().numpy()

# to 
labels = F.one_hot(labels, num_classes=2).cpu().numpy()

I hope I can help you

[TranQuocDat0405]

I have the same problem as you, have you fixed this error yet?

Hi， I fixed this bug by changing some code.

First: in utils compute_crossentropy_loss, add

    new_labels = labels.cpu().numpy().tolist()

    if len(set(new_labels)) == 3:
        for i in range(4):
            if i not in new_labels:
                w = w.tolist()
                w.insert(i, len(new_labels))
                break

    if len(set(new_labels)) == 1:
        for i in range(2):
            if i not in new_labels:
                w = w.tolist()
                w.insert(i, len(new_labels))
                break

Second: in utils compute_auc_mc function, changing a line

# from
labels = F.one_hot(labels).cpu().numpy()

# to 
labels = F.one_hot(labels, num_classes=2).cpu().numpy()

I hope I can help you

Thank you so much for your help! I really appreciate it.