holisticai.explainability.metrics.spread_divergence

holisticai.explainability.metrics.spread_divergence(feature_importance: ndarray | list[float] | Series)

Calculates the spread divergence metric based on the inverse of the Jensen-Shannon distance (square root of the Jensen-Shannon divergence), for a given feature importance. A lower ratio concentrates the importances and facilitates interpretability.

Parameters

feature_importance: ArrayLike

The feature importance values for the features.

Returns

float: The spread divergence metric value.

Example

>>> from holisticai.explainability.metrics import spread_divergence
>>> feature_importance = np.array([0.10, 0.20, 0.30])
>>> score = spread_divergence(feature_importance)
0.8196393599933761