| # Copyright (c) Microsoft Corporation. All rights reserved. | |
| # Licensed under the MIT License. | |
| """This module contains algorithms implementing the reductions approach to disparity mitigation. | |
| In this approach, disparity constraints are cast as Lagrange multipliers, which cause the | |
| reweighting and relabelling of the input data. This *reduces* the problem back to standard machine | |
| learning training. | |
| """ | |
| from ._reduction import Reduction # noqa: F401 | |
| from ._exponentiated_gradient import ExponentiatedGradient # noqa: F401 | |
| from ._exponentiated_gradient import ExponentiatedGradientResult # noqa: F401 | |
| from ._grid_search import GridSearch, GridSearchResult # noqa: F401 | |
| from ._moments import AbsoluteLoss, Moment, ConditionalSelectionRate # noqa: F401 | |
| from ._moments import DemographicParity, EqualizedOdds, ErrorRate # noqa: F401 | |
| from ._moments import GroupLossMoment, SquareLoss, ZeroOneLoss # noqa: F401 | |
| from ._moments import ClassificationMoment, LossMoment # noqa: F401 | |
| _exponentiated_gradient = [ | |
| "ExponentiatedGradient", | |
| "ExponentiatedGradientResult" | |
| ] | |
| _grid_search = [ | |
| "GridSearch", | |
| "GridSearchResult" | |
| ] | |
| _moments = [ | |
| "AbsoluteLoss", | |
| "Moment", | |
| "ClassificationMoment", | |
| "ConditionalSelectionRate", | |
| "DemographicParity", | |
| "EqualizedOdds", | |
| "ErrorRate", | |
| "GroupLossMoment", | |
| "LossMoment", | |
| "SquareLoss", | |
| "ZeroOneLoss" | |
| ] | |
| __all__ = ["Reduction"] + _exponentiated_gradient + _grid_search + _moments | |