Samuel Hoffman

Dataset Improvements

12

Major improvements: - Changed how `prot_attr` arguments are handled. Now, when processing a dataset and running metrics, an explicit array (or list of arrays) containing protected attribute values per sample...

Tutorial on dataset loading and running metrics

Port pre-processing algorithms to sklearn-compatible API

10

- [ ] DisparateImpactRemover - [x] LearnedFairRepresentation - [ ] OptimizedPreprocessing - [X] Reweighing

* Clean up bias scan/score * Additional bias scan notebook reproducing COMPAS analysis from original paper * Add differential fairness metrics (`smoothed_edf`, `df_bias_amplification`) #150 * Add metrics from [AWS](https://pages.awscloud.com/rs/112-TZM-766/images/Amazon.AI.Fairness.and.Explainability.Whitepaper.pdf) (`class_imbalance`,...

Port micellaneous items to sklearn-compatible API

8

- [x] MEPS dataset - [x] differential fairness metrics - [ ] rich subgroup fairness

- [X] AdversarialDebiasing - [ ] ARTClassifier - [ ] MetaFairClassifier - [ ] PrejudiceRemover - [ ] GerryFairClassifier

- [X] CalibratedEqualizedOdds - [ ] EqualizedOdds - [x] RejectOptionClassifier

Fixed memory leak issue

2

See #74. Removed `'previous'` field from metadata.

Unable to replicate MetaFairClassifier notebook

1

The predicted labels and scores in MetaFairClassifier need to be transposed to be `(n, 1)` instead of `(1, n)`: https://github.com/IBM/AIF360/blob/master/aif360/algorithms/inprocessing/meta_fair_classifier.py#L87-L88 I tried to fix this and the associated notebook but...

Wraps [`inFairness.fairalgo`s](https://ibm.github.io/inFairness/reference/algorithms.html) with skorch to provide sklearn-compatible interface.