integration.md

🔨 Integrating Integrate Your own Query Methods, Datasets, Trainings & Models

Query Method

To add your own custom query method familiarize yourself with the class QuerySampler in /src/query/query.py.

For uncertainty based query methods operating purely on model outputs and ranking of specific scores s.a. Entropy, BALD, check out the examples in /src/query_uncertainty.

For diversity based query methods that require some form of intermediate representations s.a. Core-Set or BADGE, check out the examples in /src/query_diversity.py

Datasets

To add a new dataset please check out the class BaseDataModule in /src/data/base_datamodule.py

Trainings

To add a new training strategy check out the class AbstractClassifier in /src/models/abstract_classifier.py and its corresponding inheritors.

You also might have to add a new trainer class in a new file /src/trainer_{training}.py (see /src/trainer.py and /src/trainer_fix.py for details).

Final touch for running experiments

Finally you would need to add a src/run_training_{training}.py and src/main_{training}.py alongside additional launchers /launchers/exp_{dataset}_{training}.py.