How trusted-ai/aif360 works
AIF360 is a Python toolkit for detecting and mitigating bias across the machine-learning lifecycle. Users supply datasets, protected attributes, and typically model predictions; they can transform data or train/post-process models, then evaluate fairness and inspect or detect problematic subgroups. The graph reflects the documented capability areas and sampled interfaces; unsampled algorithm wiring is intentionally left general rather than presented as a specific call chain.
Main entry points: structured_dataset.py Dataset interfaces · classification_metric.py Dataset and model metrics · metrics.py Sklearn fairness metrics · mdss_detector.py Bias scan
Components 17
Data and interfaces
- Dataset interfaces
aif360/datasets/structured_dataset.py - Dataset loaders
aif360/datasets - Scikit-learn API
aif360/sklearn
Bias mitigation
- Preprocessing mitigation
aif360/algorithms/preprocessing - In-processing mitigation
aif360/algorithms/inprocessing - Intersectional mitigation
aif360/algorithms/intersectional_fairness.py - Post-processing mitigation
aif360/algorithms/postprocessing - Sklearn mitigation
aif360/sklearn
Fairness assessment
- Dataset and model metrics
aif360/metrics/classification_metric.py - Sklearn fairness metrics
aif360/sklearn/metrics/metrics.py - Other fairness metrics
aif360/metrics
Detection and explanation
- Metric explainers
aif360/explainers - Bias scan
aif360/detectors/mdss_detector.py - Recourse discovery
aif360/sklearn/detectors/facts
Other components
- Data scientist
- Model predictions
- Fairness findings
Connections 24
- Data scientist → Dataset loaders: select data
- Data scientist → Dataset interfaces: prepare data
- Dataset loaders → Dataset interfaces: build datasets
- Data scientist → Scikit-learn API: use API
- Dataset interfaces → Preprocessing mitigation: transform data
- Dataset interfaces → In-processing mitigation: train fairly
- Dataset interfaces → Intersectional mitigation: mitigate bias
- Dataset interfaces → Post-processing mitigation: adjust predictions
- Dataset interfaces → Sklearn mitigation: transform data
- Model predictions → Post-processing mitigation: provide scores
- Dataset interfaces → Dataset and model metrics: compare labels
- Model predictions → Dataset and model metrics: evaluate predictions
- Dataset interfaces → Sklearn fairness metrics: measure fairness
- Model predictions → Sklearn fairness metrics: measure outcomes
- Dataset interfaces → Other fairness metrics: measure fairness
- Dataset and model metrics → Metric explainers: explain metrics
- Dataset and model metrics → Fairness findings: report metrics
- Sklearn fairness metrics → Fairness findings: report metrics
- Dataset interfaces → Bias scan: scan outcomes
- Model predictions → Bias scan: provide expectations
- Dataset interfaces → Recourse discovery: find recourse
- Model predictions → Recourse discovery: score actions
- Bias scan → Fairness findings: report subgroups
- Recourse discovery → Fairness findings: report recourse