modstablib 0.1.1

A library for evaluating and improving model stability.

Tools for stable ML models.

Installation

Install via pip:

pip install .

Usage

The main function provided is stability_index, which evaluates the stability of a sequence of metric values (e.g., accuracy over time or epochs).

Example

from modstablib.metrics._performance import stability_index

# Example: accuracy values over epochs
metric_vector = [0.85, 0.86, 0.84, 0.83, 0.82]

score = stability_index(metric_vector)
print(f"Stability index: {score}")

Parameters

• metric_vector: Sequence (list, numpy array, etc.) of metric values to evaluate stability.
• falling_rate_weight: (optional, default=12) Weight for penalizing negative trend (slope).
• variability_weight: (optional, default=0.5) Weight for penalizing variability (standard deviation of residuals).

Returns

• A float: Higher values indicate greater stability (less decrease and less variability).

License

MIT

Issues

For questions or issues, please use the GitHub issue tracker.