louvain_cluster_leaves

louvain_cluster_leaves(
    idx,
    coords,
    embeddings,
    leaf_k=10,
    similarity_threshold=0.5,
    resolution=1.0,
)

Cluster tree leaves via centroid KNN + Louvain community detection.

Finds natural communities without requiring a target k. Builds a KNN graph over leaf centroids, filters weak edges by cosine similarity threshold, then runs Louvain to discover communities.

Parameters

Name	Type	Description	Default
idx		An open LazyIndex handle.	required
coords		(N, 2-or-3) UMAP coordinates for every item.	required
embeddings		(N, D) embedding matrix (float32).	required
leaf_k		Number of nearest neighbors per leaf centroid (default 10).	10
similarity_threshold		Minimum cosine similarity to keep an edge (default 0.5).	0.5
resolution		Louvain resolution parameter; higher = more communities (default 1.0).	1.0

Returns

Name	Type	Description
		Same tuple as agglomerate_tree_leaves:
		(point_labels, lsh_names, lsh_label_data, item_leaf_map, | | | | tree_structure).
		Returns (None, {}, [], None, tree) when the tree has fewer than
		two leaves.