Meeting 5 minutes

Minutes – Differential Fairness Toolkit Project¶

Meeting #5

Date: 27th December 2025 Time: 10:00 AM Location: Online

The purpose of the meeting was to:

Review and compare each member’s prototype pipeline

Agree on a shared design for the fairness toolkit

Decide which components should be merged into the common development branch

Each group member presented a minimal working pipeline covering:

Dataset loading

Preprocessing steps

Fairness metric computation

Model training and evaluation

Clarity and usability of outputs

All pipelines successfully demonstrated.

Common patterns were identified across pipelines, particularly in data loading, preprocessing, and evaluation alignment.

Some duplication of logic was noted, especially around group handling and evaluation data structures.

Agreement that a standardised pipeline structure would improve clarity and reuse.

Following discussion, the group agreed on the following design principles:

A clean, modular package structure separating data loading, preprocessing, grouping, metrics, and visualisation.

A minimal, well-documented v0.1 feature set focused on correctness and clarity rather than breadth.

Inclusion of example notebooks and documentation as components of the toolkit.

The group agreed that:

Only reviewed, working components would be merged into the shared main branch.

Focus would be on core functionality rather than optimisation.

The following responsibilities were agreed:

Becky & Kayla: Continue development of fairness metric functions, including group-level and intersectional metrics.
Nick. Develop project documentation. Finalise and maintain the overall package structure. Ensure examples and tutorials clearly demonstrate intended usage.
Raiet. Develop visualisation components to support interpretation of fairness metrics.
All: use docstrings to explain the scope of each function

Additionally:

The group agreed to publish the package initially to TestPyPI to validate installation and imports.
Pending successful testing, the package will later be published to PyPI.

Final meeting scheduled for: