Find flow-based communities in complex networks

infomap.run() is now the canonical, fully typed entry point for community detection — common options live on the signature, and results come back as rich Result objects with summary(), to_series(), and one-call to_networkx() and to_igraph() conversions, plus a summary card that renders inline in notebooks. Also new: bundled example networks in infomap.datasets, named state nodes across the Python and R APIs, the infomap-network v1.0 JSON input format, and exposed flow-iteration and tolerance controls. Options are now validated with guided error messages, and engine logging routes through Python's standard logging module (Python docs, changelog).

Stop trials on a codelength plateau with --converge, compact --multilayer-relax-to-self coupling, multilayer cluster-data in the Python/R API, cooperative interrupt hooks for long-running runs, and pretty console output. Million-scale runs now use roughly half the peak memory and finish about 25% faster than v2.12 with no loss in solution quality, and peak memory is now largely independent of hierarchy depth — so deep hierarchical networks are far cheaper to cluster (changelog).

Automatic thread counts for clusters and schedulers, reproducible run metadata, JSON run reports, distributed trial sharding, concurrent trials, a Python GraphRAG adapter, and lower peak memory for multi-trial runs (HPC notebook, changelog).

Drop-in NetworkX find_communities(), igraph and scipy sparse input, AnnData/Scanpy integration, and revamped Python documentation with tutorial notebooks (changelog).

Per-level module counts in JSON output, idiomatic R package with SWIG bindings, library-safe no-output mode (changelog).