Change Log

Change Log#

[0.2.1] - 2024-08-15#

Bugfixes#

  • Getting started in the docs still used the old bias_eo object. Now it shows the correct name bias_metric_eo.

[0.2.0] - 2024-08-13#

Added#

Plots#

  • Calibration plot fairscoring.plots.plot_groupwise_score_calibration to visualize calibration bias.

  • Cumulative distribution plot fairscoring.plots.plot_groupwise_cdfs and the difference fairscoring.plots.plot plot_cdf_diffs. to visualize independence, equal opportunity and predictive equality bias.

  • Custom colormaps in fairscoring.plots.colors.

Bias Result Objects#

  • New result type IntegralBiasResult now contain the cdfs of the group distributions.

Changed#

  • Submodules of fairscoring.metrics are now public.

  • The class TwoGroupMixin is now called TwoGroupMetric. This allows for a more consistent naming scheme.

  • The default bias metrics where renamed from bias_-prefix to a bias_metric_-prefix. This was done, because the old names like bias_eo are ideal names for variables holding the result of a bias computation.

  • Extended documentation now available at readthedocs.

Deleted#

  • The independence bias has no default instance fairscoring.metrics.bias_ind anymore. The reason is, that independence bias is no recommended bias metric.

[0.1.1] - 2024-06-19#

Added#

  • Parameter prefer_high_scores to allow for both types of scores. This allowed for shorter dataset handling of the COMPAS examples.

  • Preparation to host docs on readthedocs.

Fixed#

  • Show correct project-link on pypi.org

  • README.md shows the correct installation command

[0.1.0] - 2024-06-03#

Added#

  • Bias Metric Classes for

    • Calibration: fairscoring.metrics.CalibrationMetric

    • Wasserstein Distance: fairscoring.metrics.WassersteinMetric

    • ROC-base Metrics: fairscoring.metrics.roc.ROCBiasMetric

  • Default Bias Metrics for

    • Equal Opportunity: fairscoring.metrics.bias_eo

    • Predictive Equality: fairscoring.metrics.bias_pe

    • Calibration: fairscoring.metrics.bias_cal

    • Independence: fairscoring.metrics.bias_ind

    • ROC: fairscoring.metrics.roc.bias_roc

    • xROC: fairscoring.metrics.roc.bias_xroc

  • Experiments on

    • Adult Dataset

    • COMPAS Dataset

    • German Credit Risk Dataset

  • Tests for compatibility with results in the publication

    • COMPAS Dataset

Contents