Small-text

Changed

- An [errata](https://small-text.readthedocs.io/en/v1.3.3/errata.html) section was added to the documentation.

Fixed

- Fixed a deviation from the paper, where `DeltaFScore` also took into account the agreement in predictions of the negative label. ([51](https://github.com/webis-de/small-text/pull/51))
- Fixed a bug in `KappaAverage` that affected the stopping behavior. ([52](https://github.com/webis-de/small-text/pull/52))

---

Fixed

- Fixed a bug in `TransformerBasedClassification`, where `validations_per_epoch>=2` left the model in eval mode. ([40](https://github.com/webis-de/small-text/issues/40))

---

Fixed

- Fixed a bug where parameter groups were omitted when using `TransformerBasedClassification`'s layer-specific fine-tuning functionality. ([36](https://github.com/webis-de/small-text/issues/36), [#38](https://github.com/webis-de/small-text/pull/38))
- Fixed a bug where class weighting resulted in `nan` values. ([39](https://github.com/webis-de/small-text/issues/39))

Contributors

[JP-SystemsX](https://github.com/JP-SystemsX)

---

Added

- Added dropout sampling to `SetFitClassification <https://github.com/webis-de/small-text/blob/v1.3.0/small_text/integrations/transformers/classifiers/setfit.py>`__.

Fixed

- Fixed broken link in README.md.
- Fixed typo in README.md. ([26](https://github.com/webis-de/small-text/pull/26))

Changed

- The `ClassificationChange <https://github.com/webis-de/small-text/blob/v1.3.0/small_text/stopping_criteria/change.py>`__ stopping criterion now supports multi-label classification.
- Documentation:
- Updated the active learning setup figure.
- The documentation of integrations has been reorganized.

Contributors

[rmitsch](https://github.com/rmitsch)

---

Added

- Added new classifier: [SetFitClassification](https://github.com/webis-de/small-text/blob/v1.2.0/small_text/integrations/transformers/classifiers/setfit.py) which wraps [huggingface/setfit](https://github.com/huggingface/setfit).
- Active Learner:
- [PoolBasedActiveLearner](https://github.com/webis-de/small-text/blob/v1.2.0/small_text/active_learner.py) now handles keyword arguments passed to the classifier's `fit()` during the `update()` step.
- Query Strategies:
- New strategy: [BALD](https://github.com/webis-de/small-text/blob/v1.2.0/small_text/query_strategies/bayesian.py).
- [SubsamplingQueryStrategy](https://github.com/webis-de/small-text/blob/v1.2.0/small_text/query_strategies/strategies.py) now uses the remaining unlabeled pool when more samples are requested than are available.
- Notebook Examples:
- Revised both existing notebook examples.
- Added a notebook example for active learning with SetFit classifiers.
- Added a notebook example for cold start initialization with SetFit classifiers.
- Documentation:
- A [showcase](https://small-text.readthedocs.io/en/v1.2.0/showcase.html) section has been added to the documentation.

Fixed

- Distances in [lightweight_coreset](https://github.com/webis-de/small-text/blob/v1.2.0/small_text/query_strategies/coresets.py) were not correctly projected onto the [0, 1] interval (but ranking was unaffected).

Changed

- [Coreset implementations](https://github.com/webis-de/small-text/blob/v1.2.0/small_text/query_strategies/coresets.py) now use the distance-based (as opposed to the similarity-based) formulation.

---

Fixed

- Model selection raised an error in cases where no model was available for selection ([21](https://github.com/webis-de/small-text/issues/21)).

---