holisticai.bias.mitigation.FairletClusteringPreprocessing#

class holisticai.bias.mitigation.FairletClusteringPreprocessing(*args, **kargs)[source]#

Fairlet decomposition [1] is a pre-processing approach that computes fair micro-clusters where fairness is guaranteed. They then use the fairlet centers as a newly transformed dataset from the original. This transformed fairlet-based dataset is then provided to vanilla clustering algorithms, and hence, we obtain approximately fair clustering outputs as a result of the fairlets themselves being fair.

Parameters

decompositionstr, optional: Fairlet decomposition strategy, available: Vanilla, Scalable, MCF. Default is Vanilla.
pint, optional: fairlet decomposition parameter for Vanilla and Scalable strategy. Default is 1.
qint, optional: fairlet decomposition parameter for Vanilla and Scalable strategy. Default is 3.
seedint, optional: Random seed. Default is None.

Examples

>>> from holisticai.bias.mitigation import FairletClusteringPreprocessing
>>> mitigator = FairletClusteringPreprocessing()
>>> train_data_transformed = mitigator.fit_transform(train_data, group_a, group_b)

References

fit_transform(X: ndarray, group_a: ndarray, group_b: ndarray, sample_weight: ndarray | None = None)[source]#

Fits the model by learning a fair cluster.

Parameters

Xmatrix-like: input matrix
group_aarray-like: binary mask vector
group_barray-like: binary mask vector
sample_weightarray-like, optional: Samples weights vector. Default is None.

Returns

matrix: Transformed matrix

transform(X)[source]#

Transforms the model by learning a fair cluster.

Parameters

Xmatrix-like: input matrix

Returns

matrix: Transformed matrix