FairDo Documentation#
Fairness-Agnostic Data Optimization (FairDo) is a Python package for mitigating bias in datasets. It provides robust fairness-agnostic methods to pre-process data. Machine learning models trained on these datasets do not come with compromises in performance but significantly discriminate less.
The pipeline to mitigate bias in datasets consists of three main steps:
fairdo.metrics: Select a fairness metric to evaluate the dataset.fairdo.optimize: Select optimization method.fairdo.preprocessing: Choose pre-processing method with selected metric and optimizer and apply it to the dataset. All pre-processors come with.fit(),.transform(),.fit_transform()interfaces.
Contents:
- fairdo.metrics package
- Submodules
- fairdo.metrics.dependence module
- fairdo.metrics.group module
average_odds_difference()average_odds_error()disparate_impact_ratio()disparate_impact_ratio_deviation()disparate_impact_ratio_objective()equal_opportunity_abs_diff()equal_opportunity_difference()mean_difference()predictive_equality_abs_diff()predictive_equality_difference()statistical_parity_abs_diff()statistical_parity_abs_diff_intersectionality()statistical_parity_abs_diff_max()statistical_parity_abs_diff_mean()statistical_parity_abs_diff_multi()statistical_parity_abs_diff_sum()statistical_parity_difference()
- fairdo.metrics.individual module
- fairdo.metrics.penalty module
- fairdo.optimize package
- fairdo.preprocessing package
- fairdo.utils package