compute_medoid_indices_from_features

fast_plscan.compute_medoid_indices_from_features(data, core_distances, probabilities, labels, *, metric='euclidean', metric_kws=None)

Compute cluster medoid indices from feature vectors.

Assumes inputs are already validated. Uses probability-weighted mutual reachability distances within each cluster.

Parameters:

• data (ndarray[tuple[int, int], dtype[single]]) – Feature matrix of shape (n_samples, n_features).

• core_distances (ndarray[tuple[int], dtype[single]]) – Core distances for all points, shape (n_samples,).

• probabilities (ndarray[tuple[int], dtype[single]]) – Cluster membership probabilities for all points, shape (n_samples,).

• labels (ndarray[tuple[int], dtype[int_]]) – Dense cluster labels with shape (n_samples,).

• metric (str, default: 'euclidean') – Pairwise distance metric.

• metric_kws (dict[str, Any] | None, default: None) – Optional metric keyword arguments.

Return type:

ndarray[tuple[int], dtype[int_]]

Returns:

medoid_indices – Integer array of shape (n_clusters,) with indices into data.