---

name: sklearn-interface description: "Use AIF360's preferred sklearn-compatible pandas API for datasets, fairness metrics, scorers, estimators, and pipeline caveats." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

AIF360 sklearn Interface

Use this sub-skill when the task is best handled through the preferred aif360.sklearn API: pandas DataFrame/Series inputs, protected attributes stored in indexes, sklearn-compatible fairness metrics/scorers, estimators, meta-estimators, and pipeline caveats.

Fast routing

• Dataset loading and conversion: start with API reference and workflows.
• Group, individual, MDSS-score, and scorer functions: use API reference and workflows.
• Reweighing, postprocessors, and PostProcessingMeta: use workflows and optional estimators.
• Missing extras, dropped indexes, scorer surprises, fetch/cache failures, or postprocessing errors: use troubleshooting.
• Quick self-check of the base metrics path: run scripts/sklearn_metric_smoke.py from this sub-skill directory.

Boundaries and handoffs

• Route legacy BinaryLabelDataset, StructuredDataset, StandardDataset, and legacy metric-class tasks to the sibling datasets-and-metrics sub-skill.
• Route legacy aif360.algorithms preprocessing/inprocessing/postprocessing class workflows to the sibling mitigation-algorithms sub-skill.
• Route MDSS subgroup scans, FACTS counterfactual subgroup rules, and metric text/JSON explainers to the sibling detectors-and-explainers sub-skill. This sub-skill only covers aif360.sklearn.metrics.mdss_bias_score as a metric helper.
• Do not rely on source notebooks, tests, or external raw-data locations at runtime. Use the bundled references and scripts here, plus user-provided data or cache settings.

Safety and verification status

• Base CPU imports and in-memory sklearn metric workflows are supported by this sub-skill.
• Dataset fetchers can require network access or a warmed cache; do not use them in no-network smoke checks.
• Optional extras were not installed in the verified base environment. Treat extra-gated estimators and ot_distance as optional/unverified until the matching extra is installed and checked.