Overhaul of Distance Based Scoring Functions

[Yejashi]

Motivation

We discovered that the current implementation of Bhattacharyya and Hellinger scoring functions does not fully capture the intended behavior. These distance metrics only return a magnitude, providing no indication of the direction or preference in the comparison. As a result, it becomes unclear what the values represent in terms of scoring.

To address this, we've introduced functionality to assign a sign to the distance metrics. This allows the scoring functions to not only return the magnitude of the difference between two samples but also indicate a preference or direction through the signage. This enhancement gives users a clearer interpretation of the results.

Key Updates

• Split Bhattacharyya and Hellinger Metrics: The Bhattacharyya and Hellinger distance metrics have been separated into individual functions for better modularity and maintainability.
• Scoring Function Built on Top of Metrics: A new scoring functionality has been implemented to leverage these metrics, providing both the magnitude and the signage for improved decision-making.