Compliance Tools for statistical analysis and debiasing for LLM powered applicant tracking systems.
Project description
fairground
"There is nothing that surpasses the joy of creation, if only because through it one wins hours of self-forgetfulness, when one lives in a world of sound." – Clara Schuman
Compliance Tools for statistical analysis and debiasing for LLM powered applicant tracking systems.
Project cheatsheet
- pre-commit:
pre-commit run --all-files
- pytest:
pytest
orpytest -s
- coverage:
coverage run -m pytest
orcoverage html
- poetry sync:
poetry install --sync
- updating requirements: see docs/updating_requirements.md
Initial project setup
- See docs/getting_started.md or docs/quickstart.md for how to get up & running.
- Check docs/project_specific_setup.md for project specific setup.
- See docs/using_poetry.md for how to update Python requirements using Poetry.
- See docs/detect_secrets.md for more on creating a
.secrets.baseline
file using detect-secrets.
fairground Statistical Analysis Utils
fairground provides a comprehensive suite of visuaisation and statistical analysis tools for assessing fairness and bias in selection processes, particularly useful for applicant tracking systems and hiring decisions.
Core Features
Selection Rate Analysis
The selection rate analysis tools help visualise and analyse how different groups are selected within your process:
plot_selection_rate()
: Creates a horizontal bar plot showing selection rates across selected protected characteristics groupsplot_job_selection_rate()
: Generates faceted plots to compare selection rates across different job roles
Parity Analysis
The parity analysis functions help identify and visualize disparities between groups:
parity_difference()
: Calculates the arithmetic difference in selection rates between groupsplot_parity_difference()
: Creates a multi-panel visualisation showing parity differences using each group as a baselineplot_parity_difference_scatter()
: Generates a scatter plot heatmap showing parity differences between all group combinations
Disparate Impact Analysis
Tools for analysing disparate impact and selection rate deviations:
disparate_impact_ratio()
: Calculates the ratio of selection rates between groups (also known as the adverse impact ratio)plot_selection_rate_deviation_with_disparate_impact_ratio_value()
: Creates a comprehensive visualisation showing selection rate deviations from equal treatment and disparate impact ratio thresholds
Key Features
- Flexible Group Analysis: Support for both single and multiple demographic group analysis
- Regulatory Compliance: Built-in support for standard fairness metrics like the 4/5ths rule (disparate impact ratio)
- Visual Insights: Rich visualizations with detailed annotations including: -- Group sizes (n values) -- Statistical significance indicators -- Disparate impact ratio values -- Equal treatment thresholds
Example Usage
import pandas as pd
from fairground.utils import plot_selection_rate, plot_parity_difference
# Load your data
df = pd.read_csv("hiring_data.csv")
# Plot selection rates by gender
plot_selection_rate(df, "Gender", "Selected")
# Analyze parity differences across multiple demographics
plot_parity_difference(df, ["Gender", "Race", "Age"], "Selected")
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
File details
Details for the file fairground-0.0.0a11.tar.gz
.
File metadata
- Download URL: fairground-0.0.0a11.tar.gz
- Upload date:
- Size: 536.6 kB
- Tags: Source
- Uploaded using Trusted Publishing? Yes
- Uploaded via: twine/5.1.1 CPython/3.12.7
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 40888cb21fc896d629149c458e2c4e0c16a2ec24d63b3285d9a67d4c05ac5792 |
|
MD5 | 5519fb4c85b90c03ab46b7b3a7557b83 |
|
BLAKE2b-256 | d3b966f9e44f19cd162a62a68ab49e4fdee99e79e6a85c7800d7a8732b61b483 |
Provenance
The following attestation bundles were made for fairground-0.0.0a11.tar.gz
:
Publisher:
main.yaml
on CoefficientSystems/fairground
-
Statement type:
https://in-toto.io/Statement/v1
- Predicate type:
https://docs.pypi.org/attestations/publish/v1
- Subject name:
fairground-0.0.0a11.tar.gz
- Subject digest:
40888cb21fc896d629149c458e2c4e0c16a2ec24d63b3285d9a67d4c05ac5792
- Sigstore transparency entry: 146820071
- Sigstore integration time:
- Predicate type:
File details
Details for the file fairground-0.0.0a11-py3-none-any.whl
.
File metadata
- Download URL: fairground-0.0.0a11-py3-none-any.whl
- Upload date:
- Size: 542.1 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? Yes
- Uploaded via: twine/5.1.1 CPython/3.12.7
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 7b544cb5ee2f1e8ce8661b3935634fc4990d0e2d891e5f4a22c678465d499e61 |
|
MD5 | 06084850916b5a9adcda814b055b0ffc |
|
BLAKE2b-256 | ff5fccf41e34ad5a0f2cdff346e4cb9ea2f266809bcbe05798878ef06e7877f7 |
Provenance
The following attestation bundles were made for fairground-0.0.0a11-py3-none-any.whl
:
Publisher:
main.yaml
on CoefficientSystems/fairground
-
Statement type:
https://in-toto.io/Statement/v1
- Predicate type:
https://docs.pypi.org/attestations/publish/v1
- Subject name:
fairground-0.0.0a11-py3-none-any.whl
- Subject digest:
7b544cb5ee2f1e8ce8661b3935634fc4990d0e2d891e5f4a22c678465d499e61
- Sigstore transparency entry: 146820072
- Sigstore integration time:
- Predicate type: