Skip to contents

Measure modularity of a file-backed fuzzy graph and disk-backed labels

Source: R/massive_graph_modularity.R
massive_graph_modularity.Rd

Scans each undirected edge of a symmetric CSR fuzzy graph once, using bounded buffers. Labels are one-based little-endian uint32 values and are memory-mapped; their pages may contribute to process RSS. This is a CPU-only quality diagnostic, not a full-graph clustering algorithm.

Usage

massive_graph_modularity(
  graph,
  membership,
  n_communities = NULL,
  resolution = 1,
  chunk_rows = 8192L,
  memory_limit = "1GB",
  backend = "cpu"
)

Arguments

graph

A result of massive_umap_fuzzy_graph().

membership

Path to a uint32 label file, or a fastEmbedR_massive_clusters descriptor.

n_communities

Number of labels for a file path; inferred from a cluster descriptor.

resolution

Positive modularity resolution.

chunk_rows

Maximum CSR rows scanned per chunk.

memory_limit

RAM budget for native scan buffers. Memory-mapped label pages are tracked separately and are not a strict RSS cap.

backend

Only "cpu" is supported; other requests fail explicitly.

Value

Modularity, internal and total edge weights, edge-pair count, backend, and a buffer/mapping resource estimate.