HumanCompatible · Bias Detection

A toolbox for measuring bias in data & models

Maximum Subgroup Discrepancy (MSD) - bias metric with linear sample complexity
…with a MILP formulation that also tells you which subgroup is most affected.

Quick install & 60-second demo

python -m pip install git+https://github.com/humancompatible/detect.git

from humancompatible.detect import detect_bias_csv

msd_val, rule = detect_bias_csv(
    csv_path="census.csv",                  # any CSV file
    target_col="income_50k",                # binary target
    protected_list=["race", "age"],         # columns to audit
    method="MSD",                           # chosen method
)

print(f"MSD = {msd_val:.3f}", "Rule ->", rule) 

The function returns

msd_val – the maximum gap (in percentage‐points) between any subgroup and its complement

rule – the raw subgroup encoding as a list of (feature_index, Bin) pairs.
To get a human‐readable description, do the following:

pretty = " AND ".join(str(cond) for _, cond in rule)
print("Subgroup:", pretty)
# -> "Subgroup: Race = Blue AND Age = 0-18"

Contents

Detect API

Tutorial -> Your first audit in 5 minutes
Examples -> Start with a simple example notebook, or go directly to a realistic example using Folktables

MSD as a distance?

Bias detection can be understood as measuring some distance between two distributions (positive X negative samples, some training dataset X general population data…).

However, most distances have exponential sample complexity, whereas MSD requires a linear number of samples (w.r.t. the dimension) to achieve the same error.

Classical metric	Needs full d‐dim joint?	Sample cost	Drawbacks
Wasserstein, TV, MMD, …	yes	Ω(2^d)	exponential samples, no subgroup info
MSD (ours)	only protected attrs	O(d)	✓ returns exact subgroup & gap

MSD maximises the absolute difference in probability over all protected‐attribute combinations (subgroups), yet is solvable in practice through an exact Mixed‐Integer optimization that scans the doubly‐exponential space effectively.

Citation

If you use MSD, please cite:

@inproceedings{MSD,
  author = {Jiří Němeček and Mark Kozdoba and Illia Kryvoviaz and Tomáš Pevný and Jakub Mareček},
  title = {Bias Detection via Maximum Subgroup Discrepancy},
  year = {2025},
  booktitle = {Proceedings of the 31st ACM SIGKDD International Conference on Knowledge Discovery \& Data Mining},
  series = {KDD '25}
}

Looking for the installation matrix, solver details or developer setup?
Head to the README -> Installation section.