holisticai.bias.mitigation.DisparateImpactRemover

class holisticai.bias.mitigation.DisparateImpactRemover(*args, **kargs)

Disparate impact remover [1] edits feature values to increase group fairness while preserving rank-ordering within groups.

Parameters

repair_levelfloat, optional

The amount of repair to be applied. It should be between 0.0 and 1.0. Default is 1.

Examples

>>> from holisticai.bias.mitigation import DisparateImpactRemover
>>> mitigator = DisparateImpactRemover()
>>> train_data_transformed = mitigator.fit_transform(train_data, group_a, group_b)
>>> test_data_transformed = mitigator.transform(test_data, group_a, group_b)

References

[1]

Feldman, Michael, et al. “Certifying and removing disparate impact.” proceedings of the 21th ACM SIGKDD international conference on knowledge discovery and data mining. 2015.

fit()

Fit the model

Returns

Self

fit_transform(X: ndarray, group_a: ndarray, group_b: ndarray)

Fit the model and transform data

Parameters

Xarray-like

Input data

group_aarray-like

mask vector

group_barray-like

mask vector

Returns

array-like

Repaired input data matrix

repair_data(X, group_a, group_b)

Repair data to a fair representation

Parameters

Xarray-like

Input data

group_aarray-like

mask vector

group_barray-like

mask vector

Returns

array-like

Repaired input data matrix

transform(X: ndarray, group_a: ndarray, group_b: ndarray)

Transform data to a fair representation

Parameters

Xarray-like

Input data

group_aarray-like

mask vector

group_barray-like

mask vector

Returns

Self