Skip to content

The model does not train anymore when changing batch_size in test_loader in shapes.py #3

Description

@sadaszewski

Hello, thank you for your nice work. There is a problem with one of the examples. The model does not train anymore when changing batch_size in test_loader in shapes.py, i.e. it trains much worse when decrasing the batch_size to 10 or 5 (emphasis: in test_loader not train_loader!) and with test_loader at batch_size=1 the best performance is achieved after only a few epochs and it is at the level of 0.3. Is this to be expected? Is the role of the test dataset somehow different than normally and is supposed to leak information during training? Thanks in advance.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions