search

ContinualAI / avalanche

Avalanche: an End-to-End Library for Continual Learning based on PyTorch.
http://avalanche.continualai.org
MIT License
1.76k stars 289 forks source link

Define early stopping's patience in terms of number of iterations instead of epochs #674

Open AntonioCarta opened 3 years ago

AntonioCarta commented 3 years ago

See #670 for a possible usecase.

Add an option to the early stopping plugin to use the number of iterations instead of epochs.

AntonioCarta commented 3 years ago

I'm trying to close this issue.

I think right now it doesn't make sense to stop training before the end of an epoch because validation metrics are only updated at the end of an epoch. So even if we check the validation metrics at each iteration, they will be updated only at the end of an epoch.