ContinualAI / avalanche

Avalanche: an End-to-End Library for Continual Learning based on PyTorch.
http://avalanche.continualai.org
MIT License
1.78k stars 291 forks source link

Unnecessarily strict requirement for batch size in MASPlugin #1548

Open niniack opened 11 months ago

niniack commented 11 months ago

Describe the bug

In the MASPlugin, the following check seems to be unnecessarily strict:

Link to code

for _, batch in enumerate(dataloader):
    # Get batch
    if len(batch) == 2 or len(batch) == 3:
        x, _, t = batch[0], batch[1], batch[-1]
    else:
        raise ValueError("Batch size is not valid")

As a result, if I have a custom dataset which returns additional elements (two images and a label), thi

Expected behavior

I am not sure MAS needs to follow this check, it can simply grab the needed information (image, label, and token) and continue onwards.

ghost commented 10 months ago

Vvo^3 +3/2KQq'arcon279º = Tuv . Fûv grr = goo - wo / -1.214^22/4