aif360Use IBM AI Fairness 360 for tabular fairness datasets, metrics, bias mitigation algorithms, sklearn-compatible workflows, subgroup detectors, and explainers.
Independent GitHub snapshot · Automated classification · Original documentation governs
Use this skill when a task involves the AI Fairness 360 (aif360) Python
package: fairness datasets, group-fairness metrics, bias-mitigation algorithms,
the preferred aif360.sklearn API, MDSS/FACTS subgroup detectors, or metric
explainers.
Base package install:
pip install aif360
python -c "import aif360; print(aif360.__version__)"
For local package development, use an editable install in an isolated Python
environment. For optional algorithms, install the smallest named extra that
matches the requested workflow rather than aif360[all] by default; read
install-data-and-optional-deps.md.
Run the bundled base smoke when importability or optional-warning noise is unclear:
python scripts/check_aif360_env.py --json
The smoke uses only synthetic in-memory data and does not download benchmark datasets or require optional extras.
| Task signal | Read |
|---|---|
Legacy BinaryLabelDataset, StructuredDataset, standard dataset wrappers, BinaryLabelDatasetMetric, ClassificationMetric, raw-data availability, or protected group dictionaries | datasets-and-metrics |
Legacy aif360.algorithms preprocessing, inprocessing, postprocessing, deterministic reranking, optional mitigation extras, or before/after metric validation | mitigation-algorithms |
Preferred pandas/sklearn API, aif360.sklearn.datasets.fetch_*, protected attributes in pandas indexes, sklearn metrics/scorers, sklearn estimators, or PostProcessingMeta | sklearn-interface |
MDSS subgroup bias scan, FACTS recourse subgroup reports, MetricTextExplainer, or MetricJSONExplainer | detectors-and-explainers |
| Install/import failures, raw dataset files, optional extras, warnings, Python versions, R wrapper, or MLOps sample boundaries | Root references below |
BinaryLabelDataset and
legacy algorithms; choose aif360.sklearn for pandas/sklearn pipelines and
future-facing DataFrame workflows.favorable_label, pos_label,
privileged group, unprivileged group, protected attributes, and row alignment
explicit before interpreting a metric or detector result.aif360.sklearn.