meanap.pipeline.modularity¶

Consensus-clustering modularity, port of mod_consensus_cluster_iterate.m.

Runs Louvain community detection (louvain.py) many times, builds a consensus co-classification matrix (Lancichinetti & Fortunato 2012), and repeats on the thresholded consensus matrix until it becomes block-diagonal (stable partition).

Not bit-reproducible against MATLAB. Louvain’s local-moving phase visits nodes in a random order each pass (randperm in MATLAB, a different RNG stream than Python’s), so the specific community labels and even the partition itself can differ between MATLAB and Python runs — same situation as Step 3’s probabilistic thresholding. What’s expected to match is quality: modularity Q should land in a similar range, and consensus clustering should still converge to a stable, block-diagonal partition.

Functions

`consensus_coclassify`(partitions)	Co-classification matrix: fraction of partitions where i, j share a module.
`consensuscheck`(d)	True if `d` is binary (only 0s and 1s) and symmetric (block-diagonal).
`mod_consensus_cluster_iterate`(adj_m[, ...])	Returns (Ci, Q, num_repeats): consensus community affiliation + modularity.

meanap.pipeline.modularity.consensus_coclassify(partitions)[source]¶

Co-classification matrix: fraction of partitions where i, j share a module.

Parameters:: partitions (list[ndarray])
Return type:: ndarray

meanap.pipeline.modularity.consensuscheck(d)[source]¶

True if d is binary (only 0s and 1s) and symmetric (block-diagonal).

Parameters:: d (ndarray)
Return type:: bool

meanap.pipeline.modularity.mod_consensus_cluster_iterate(adj_m, threshold=0.4, rep_num=50, rng=None, max_outer_iterations=50)[source]¶

Returns (Ci, Q, num_repeats): consensus community affiliation + modularity.

max_outer_iterations is a safety cap not present in MATLAB (which loops unconditionally until block-diagonal) — consensus clustering on real data converges in a handful of iterations; this just prevents a pathological input from hanging forever.

Parameters:

adj_m (ndarray)
threshold (float)
rep_num (int)
rng (Generator | None)
max_outer_iterations (int)

Return type:

tuple[ndarray, float, int]