Improvements and refactoring of loss weighting#148

Merged

davidbenjamin merged 30 commits intomasterfrom db_domain_adaptation

Oct 9, 2024

+313-286

Commits on Oct 2, 2024

evaluation_metrics.record_loss handles mixed labeled and unlabeled batches, accounts for weight properly
davidbenjamin
committed
artifact model learning ready for mixed labeled / unlabeled batches
davidbenjamin
committed
did the same for BaseModel learning and also replaced the posterior probability stuff with regular weighting
davidbenjamin
committed
Removed references to batch.is_labeled in base batch embedding metrics
davidbenjamin
committed
a little adjustment in filter variants
davidbenjamin
committed
labeled-only data loader for pruning; only one use of is_labeled() left
davidbenjamin
committed
got the last is_labeled and maybe also fixed bug where pruning deleted unlabeled data
davidbenjamin
committed
got rid of last references to base batch is_labeled
davidbenjamin
committed
base batch sampler mixes unlabeled and labeled
davidbenjamin
committed
artifact batch sampler also mixes unlabeled and labeled
davidbenjamin
committed
whoops
davidbenjamin
committed
whoops
davidbenjamin
committed

Commits on Oct 4, 2024

Commits on Oct 8, 2024