Segmentation

Segmentation primitives used to divide data into advantaged and disadvantaged groups for bias analysis. Import from fair_perf_ml.bias.

Enums

fair_perf_ml.bias.segmentation.BiasSegmentationType

Bases: str, Enum

Two segmentation types are supported. Label implies segmentation based on a equality with the segmentation value. Threshold implies segmentation based on the ordering with respect to the segmentation value.

fair_perf_ml.bias.segmentation.BiasSegmentationThresholdType

Bases: str, Enum

When the segmentation type is BiasSegmentationType.Threshold. This enum defines the ordering logic to derive segmentation group.

Segmentation classes

fair_perf_ml.bias.segmentation.LabeledBiasSegmentation

Bases: BiasSegmentationProtocol

Implements the BiasSegmentationProtocol for assigning segmentation values for equality types.

fair_perf_ml.bias.segmentation.ThresholdBiasSegmentation

Bases: BiasSegmentationProtocol

Implements the BiasSegmentationProtocol for assigning segmentation values for ordered types.

Factory

fair_perf_ml.bias.segmentation.bias_segmentation_criteria_factory

bias_segmentation_criteria_factory(value: Any, segmentation_type: BiasSegmentationType | str, segmentation_threshold_type: BiasSegmentationThresholdType | str | None = None) -> BiasSegmentationProtocol

Factory method to define the segmentation class, based on arguments provided. There are two sets of valid states: segmentation_type = BiasSegmentationType.Label segmentation_threshold_type = None

OR

segmentation_type = BiasSegmentationType.Threshold
segmentation_threshold_type = BiasSegmentationThresholdType

Data payload

fair_perf_ml.bias.segmentation.BiasDataPayload

Convience type that groups a dataset and the segmentation criteria as a single unit.

factory `classmethod`

factory(data_payload: Sequence[T], value: T, segmentation_type: BiasSegmentationType | str, segmentation_threshold_type: BiasSegmentationThresholdType | str | None = None) -> Self

Factory method to derive the segmentation criteria.

Exceptions

fair_perf_ml.bias.segmentation.BiasPayloadTypeException

Bases: Exception

fair_perf_ml.bias.segmentation.InvalidBiasSegmentationConfig