meanap.pipeline.louvain

Louvain modularity optimization, port of community_louvain.m (BCT).

Single-run community detection. Inherently stochastic (random node processing order each pass) — used as the building block for modularity.py’s consensus-clustering wrapper, which stabilizes the randomness across many runs. Not bit-reproducible against MATLAB (different RNG stream) — see modularity.py’s module docstring.

Functions

community_louvain(w[, gamma, rng])

Returns (M, Q): community affiliation vector (1-indexed, like MATLAB) and modularity.
meanap.pipeline.louvain.community_louvain(w, gamma=1.0, rng=None)[source]

Returns (M, Q): community affiliation vector (1-indexed, like MATLAB) and modularity.

Direct port of community_louvain.m’s default (‘modularity’) path with no initial partition — matches how mod_consensus_cluster_iterate.m always calls it (community_louvain(adjM), no extra args). Renumbering of module labels happens exactly once per hierarchical level, after the local-moving phase fully converges — matching MATLAB’s structure, not after every node sweep.

Parameters:

  • w (ndarray)

  • gamma (float)

  • rng (Generator | None)

Return type:

tuple[ndarray, float]