Introduction

graphlet is a petgraph-native library for the structural / network-science mining subfield: graphlet and orbit statistics, subgraph census, and network-motif detection. It is the home for the pieces genuinely absent from the Rust graph ecosystem — the naive rest of the domain (traversal, shortest-path, flow, MST, centrality, isomorphism, planarity) is already served by petgraph, rustworkx-core, and graphalgs, which graphlet builds on rather than rebuilding.

It depends only on petgraph and rand, owning every small, well-understood algorithm rather than relabelling a transitive trust chain through a facade.

The shape

• Census substrate: the organizing center — enumerate connected k-subsets → canonically label → fold — with lazy instance enumeration and streaming counting as two readouts of one pass.
• Orbits (GDV/GDD): per-node attribution to all 73 automorphism orbits at k ≤ 5 (undirected).
• Named motifs: a catalog (seeded with the diamond) that threads induced vs. non-induced semantics honestly.
• Template matching: a thin parallel arm over petgraph's VF2, induced-native.

Scope today

Undirected graphlets at k ≤ 5 are implemented and verified. Null-model generators, graph kernels, motif significance, neighborhood statistics, scalable k=5, and directed k ≥ 4 are documented as future work (see the crate's rim module).

Genericity

Every entry point is generic over Graph / StableGraph, directed / undirected, and arbitrary node/edge weights through one trait-bound set. Directed graphs are analyzed on their underlying undirected structure.