habit.kernels.cluster_selection.SCORE_DIRECTIONS

SCORE_DIRECTIONS: Mapping[str, str] = {'aic': 'minimize', 'bic': 'minimize', 'bic_elbow': 'bic_gradient', 'calinski_harabasz': 'maximize', 'davies_bouldin': 'minimize', 'elbow': 'knee', 'gap': 'maximize', 'inertia': 'knee', 'kneedle': 'knee', 'silhouette': 'maximize'}

Validation score -> selection rule. The rule is a property of the score itself (a silhouette is always maximised), so it does not depend on which clustering algorithm produced the curve; algorithms only differ in which scores they support.

elbow deliberately shares kneedle’s rule – see the module warning.