ce-mondrian-conditional

Subgroup-aware uncertainty calibration partitions calibration data by group to produce per-bin uncertainty intervals, exposing subgroup-specific estimates needed for fairness-aware model deployments.

To configure Mondrian conditional calibration, choose a binning approach, apply it to calibrate your subgrouped data, and validate the per-bin uncertainty estimates to ensure consistent group partitioning.

Mondrian calibration supports three bin specification options: inline bins for manual partitioning, MondrianCategorizer for continuous features, and a callable mc function for custom grouping logic.

No, you must maintain consistent use of bins between the calibrate and explain phases to ensure the subgroup-specific uncertainty estimates are accurately validated across deployment scenarios.

You validate per-bin uncertainty by applying Mondrian-based calibration to your subgrouped data and evaluating the generated uncertainty intervals for each group using the provided evaluation workflow.

Yes, Mondrian calibration handles continuous features by using the MondrianCategorizer bin specification option, which partitions continuous data into bins for conditional calibration.