Fairness metrics for continuous risk scores
Project description
Fair-Scoring
Fairness metrics for continuous risk scores.
The implemented algorithms are described in the paper [1].
Installation
Install with pip
directly:
pip install fair-scoring
Usage
The following example shows how compute the equal opportunity bias of the compas dataset
import pandas as pd
from fairscoring.metrics import bias_eo
# Load compas data
dataURL = 'https://raw.githubusercontent.com/propublica/compas-analysis/master/compas-scores-two-years.csv'
df = pd.read_csv(dataURL)
# Relevant data
scores = df['decile_score']
target = df['two_year_recid']
attribute = df['race']
# Compute the bias
bias = bias_eo(scores, target, attribute, groups=['African-American', 'Caucasian'],favorable_target=0,prefer_high_scores=False)
Further examples
Further examples - especially the experiments conducted for the publication - can be found in the documentation.
Development
Setup
Clone the repository and install from this source via
pip install -e .[dev]
Tests
To execute the tests install the package in development mode (see above)
pytest
Following the pytest framework, tests for each package are located in a subpackages named test
Docs
To build the docs move to the ./docs
subfolder and call
make clean
make html
References
[1] Becker, A.K. and Dumitrasc, O. and Broelemann, K.; Standardized Interpretable Fairness Measures for Continuous Risk Scores; Proceedings of the 41th International Conference on Machine Learning, 2024;
Bibtex
@inproceedings{Zern2023Interventional,
author = {Ann{-}Kristin Becker and Oana Dumitrasc and Klaus Broelemann}
title = {Standardized Interpretable Fairness Measures for Continuous Risk Scores},
booktitle={Proceedings of the 41th International Conference on Machine Learning},
year = {2024}
}
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for fair_scoring-0.1.1-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 4e995efe1a01a5f877cc40f62efdd1e39ab5f65762d8db151bdb3e8ec03de6de |
|
MD5 | 0c30888ea1ecd85974148f927a1969ad |
|
BLAKE2b-256 | 27151fdae99c89c72f6c90c1e38d8dcdb3b6a09d1379d64b4264d0fdea107096 |