I'm creating a simple 2 class sentiment classifier using bert, but i'm getting an error related to output and label size. I cannot figure out what I'm doing wrong. Below are the required code snippets.
My custom dataset class:
class AmazonReviewsDataset(torch.utils.data.Dataset):
def __init__(self, df):
self.df = df
self.maxlen = 256
self.tokenizer = DistilBertTokenizer.from_pretrained("distilbert-base-uncased")
def __len__(self):
return len(self.df)
def __getitem__(self, index):
review = self.df['reviews'].iloc[index].split()
review = ' '.join(review)
sentiment = int(self.df['sentiment'].iloc[index])
encodings = self.tokenizer.encode_plus(
review,
add_special_tokens=True,
max_length=self.maxlen,
padding='max_length',
truncation=True,
return_attention_mask=True,
return_tensors='pt'
)
return {
'input_ids': encodings.input_ids.flatten(),
'attention_mask': encodings.attention_mask.flatten(),
'labels': torch.tensor(sentiment, dtype=torch.long)
}
output of dataloader:
for batch in train_loader:
print(batch['input_ids'].shape)
print(batch['attention_mask'].shape)
print(batch['labels'])
print(batch['labels'].shape)
break
torch.Size([32, 256])
torch.Size([32, 256])
tensor([2, 2, 2, 2, 1, 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 1, 1, 2, 2, 1, 2, 1, 2,
2, 2, 2, 2, 2, 1, 1, 2])
torch.Size([32])
My nn:
criterion = nn.CrossEntropyLoss().to(device)
class SentimentClassifier(nn.Module):
def __init__(self):
super(SentimentClassifier, self).__init__()
self.distilbert = DistilBertModel.from_pretrained("distilbert-base-uncased")
self.drop0 = nn.Dropout(0.25)
self.linear1 = nn.Linear(3072, 512)
self.relu1 = nn.ReLU()
self.drop1 = nn.Dropout(0.25)
self.linear2 = nn.Linear(512, 2)
self.relu2 = nn.ReLU()
def forward(self, input_ids, attention_mask):
outputs = self.distilbert(input_ids, attention_mask)
last_hidden_state = outputs[0]
pooled_output = torch.cat(tuple([last_hidden_state[:, i] for i in [-4, -3, -2, -1]]), dim=-1)
x = self.drop0(pooled_output)
x = self.relu1(self.linear1(x))
x = self.drop1(x)
x = self.relu2(self.linear2(x))
return x
Train loop:
for batch in loop:
optimizer.zero_grad()
input_ids = batch['input_ids'].to(device)
attention_mask = batch['attention_mask'].to(device)
labels = batch['labels'].to(device)
output = model(input_ids, attention_mask)
print(output.size(), labels.size())
loss = criterion(output, labels) # ERROR
loss.backward()
optimizer.step()
Error:
torch.Size([32, 2]) torch.Size([32])
---------------------------------------------------------------------------
IndexError Traceback (most recent call last)
<ipython-input-19-6268781f396e> in <module>()
12 print(output.size(), labels.size())
13 # output_class = torch.argmax(results, dim=1)
---> 14 loss = criterion(output, labels)
15 train_loss += loss
16 loss.backward()
2 frames
/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
1049 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
1050 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1051 return forward_call(*input, **kwargs)
1052 # Do not call functions when jit is used
1053 full_backward_hooks, non_full_backward_hooks = [], []
/usr/local/lib/python3.7/dist-packages/torch/nn/modules/loss.py in forward(self, input, target)
1119 def forward(self, input: Tensor, target: Tensor) -> Tensor:
1120 return F.cross_entropy(input, target, weight=self.weight,
-> 1121 ignore_index=self.ignore_index, reduction=self.reduction)
1122
1123
/usr/local/lib/python3.7/dist-packages/torch/nn/functional.py in cross_entropy(input, target, weight, size_average, ignore_index, reduce, reduction)
2822 if size_average is not None or reduce is not None:
2823 reduction = _Reduction.legacy_get_string(size_average, reduce)
-> 2824 return torch._C._nn.cross_entropy_loss(input, target, weight, _Reduction.get_enum(reduction), ignore_index)
2825
2826
IndexError: Target 2 is out of bounds.
I read a tutorial which says that, do not use softmax when applying a nn.CrossEntropyLoss, as I have 2 classes. What is wrong can anyone guide me! Thank you!
You have two classes, which means the maximum target label is 1 not 2 because the classes are indexed from 0 (see official documentation). You essentially have to subtract 1 to your labels tensor, such that class n°1 is assigned the value 0, and class n°2 value 1.
In turn the labels of the batch you printed would look like:
tensor([1, 1, 1, 1, 0, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 0, 1,
1, 1, 1, 1, 1, 0, 0, 1])