icx 0.0.3

This package provides the functionality for an interactive Streamlit dashboard designed to support stakeholders in exploring individual fairness notions within algorithmic decision-making systems.

Individual Consistency eXplorer (icx) Python Package

This package provides the functionality for an interactive Streamlit dashboard designed to support stakeholders in exploring individual fairness notions within algorithmic decision-making systems.

The dashboard allows users to:

• Explore and operate on a tabular dataset of individuals provided with their corresponding binary classifications;
• Define how similarity between individuals is measured, by configuring categorisation of attributes and how distances between attribute values are computed;
• Compute and visualise five individual fairness metrics that summarise the consistency of classifications across the dataset; and
• Inspect attributes of specific individuals and of those individuals most similar to them, to explore variations in attribute values and allow like-for-like comparisons of classifications.

This package implements the functionality described in a submission to ECAI Demo Track 2025, and further details and documentation will be provided upon publication.

📦 Installation

It is recommended to install icx in a virtual environment (e.g., conda).

pip install icx

Basic Usage

from icx import dashboard

Run with default settings

Uses first 1000 individuals from pre-loaded 'adult.csv' dataset from UCI Machine Learning repository.

dashboard.run()

🗂️ Use Your Own Dataset

You can provide your own dataset, as long as it meets the following requirements:

• The first row contains column headers (attribute names).
• Each attribute is either numeric (int, float) or categorical (string).
• Each instance must be assigned a binary class: either positive or negative. You can specify the value representing the positive class — all other values will be treated as negative.
• The class column must be named "y" or explicitly defined in the parameters.

Example

dashboard.run(
    filepath="path/to/your/data.csv",   # Path to your CSV file
    y="class_column_name",             # Name of the class column (default: "y")
    positive="positive_label"           # Label for the positive class (default: 1)
)