Class balancing/resampling

[nshaud]

Current approach to class balancing is to use inverse median frequency loss reweighting.

Other options could be:

Resampling

Resample the dataset (e.g. upsample minority classes or downsamples majority classes)

• Static resampling before constructing the train Dataset
• Dynamic resampling in the DataLoader

Loss balancing

• IMF, this is what we do.
• IF, ignore the median cost.
• others?