holisticai.bias.mitigation.LPDebiaserMulticlass#

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

Linear Programming Debiaser [1] is a postprocessing algorithms designed to debias pretrained classifiers. The algorithm use constraints such as Equalized Odds and Equalized Opportunity. This technique extends LPDebiaserBinary for multiclass classification.

Parameters

constraintstr: Strategy used to evalute the cost function The available contraints are: “EqualizedOdds”, “EqualizedOpportunity”
lossstr: The loss function to optimize: “macro”, “micro”. The default is “macro”

Examples

>>> from holisticai.bias.mitigation import LPDebiaserMulticlass
>>> mitigator = LPDebiaserMulticlass(**params)
>>> mitigator.fit(y, y_pred, group_a, group_b)
>>> test_data_transformed = mitigator.transform(y_pred, group_a, group_b)

References

fit(y: ndarray, y_pred: ndarray, group_a: ndarray, group_b: ndarray)[source]#

Compute parameters for Linear Programming Debiaser. For binary classification y_pred or y_proba can be used. For Multiclass classification only y_pred must be used. .. rubric:: Parameters

yarray-like: Target vector
y_predarray-like: Predicted label vector (num_examples,).
group_aarray-like: Group membership vector (binary)
group_barray-like: Group membership vector (binary)

Returns

Self

fit_transform(y: ndarray, y_pred: ndarray, group_a: ndarray, group_b: ndarray)[source]#

Fit and transform

Parameters

yarray-like: Target vector
y_predarray-like: Predicted vector (nb_examples,)
group_aarray-like: Group membership vector (binary)
group_barray-like: Group membership vector (binary)

Returns

dict: A dictionnary with new predictions

transform(y_pred: ndarray, group_a: ndarray, group_b: ndarray)[source]#

Apply transform function to predictions and likelihoods

Parameters

y_predarray-like: Predicted vector (nb_examples,)
group_aarray-like: Group membership vector (binary)
group_barray-like: Group membership vector (binary)

Returns

dict: A dictionnary with new predictions