merge_to_max_k

merge_to_max_k(point_labels, embeddings, max_k=12)

Merge communities to ≤max_k using complete linkage on community centroids.

Returns point_labels unchanged if already ≤ max_k. Otherwise: compute community centroids → L2-normalize → complete linkage → fcluster at max_k → remap point labels → reassign points.

Parameters

Name	Type	Description	Default
point_labels		int32 array (N,) of cluster IDs.	required
embeddings		(N, D) embedding matrix (float32).	required
max_k		Maximum number of clusters to keep (default 12).	12

Returns

Name	Type	Description
		int32 array (N,) of merged cluster IDs (0-based contiguous).