nimnet

A comprehensive network science library for Nim, inspired by Python’s NetworkX.

Get started API Reference View on GitHub

---

Why nimnet?

Performance — Nim compiles to C with zero-overhead abstractions. nimnet leverages cached edge counts (O(1)), func/{.inline.} annotations, and pre-sized allocations for serious speed.

Ergonomics — Nim-idiomatic API with operator overloading (in, [], +, -), iterators (for n in g), and batch operations (addEdgesFrom).

Extensibility — Generic node types (Graph[int], Graph[string], or any hashable type), typed edge tuples, and a modular architecture.

Quick example

import nimnet

var g = newGraph[int]()
g.addEdgesFrom([(1, 2), (2, 3), (3, 1)])

assert 1 in g
assert g.len == 3
echo g  # Graph(nodes=3, edges=3)

for node in g:
  echo node, " has degree ", g.degree(node)

Features at a glance

Category	Modules
Core	Graph[N], DiGraph[N], MultiGraph[N], MultiDiGraph[N], generic node types, cached metrics
Algorithms (48)	BFS/DFS traversal, Dijkstra, Bellman-Ford, A*, Floyd-Warshall, Johnson’s, PageRank, eigenvector/Katz/HITS/betweenness centrality, Tarjan SCC, Kruskal/Prim MST, Edmonds-Karp max flow, min-cost flow, topological sort, Louvain/greedy modularity communities, VF2 isomorphism, planarity testing, TSP heuristics, k-core decomposition, Bron-Kerbosch cliques, graph coloring, bipartite matching, Euler/Hamiltonian paths, link prediction, connectivity, tree decomposition, bridges/articulation points, ego graphs, distance measures, simple paths, efficiency, rich-club coefficient, Wiener index, cycle basis, matching, graph products, small-world metrics, WL graph hashing, SimRank similarity, spectral analysis, triads, Voronoi partitions, LCA, layout (spring, circular, shell), minors, cuts, parallel algorithms
Generators (10)	Classic (complete, cycle, path, star, wheel, grid, barbell, lollipop, ladder), random (Erdős-Rényi, Barabási-Albert, Watts-Strogatz, regular, SBM), small/famous (Petersen, karate club), trees, line graph, lattice (grid2d, triangular, hypercube), geometric (random geometric, Waxman), community (caveman, planted partition), degree sequence (configuration model, Havel-Hakimi), directed (GN, GNR, GNC, random k-out)
I/O (10)	Edge list, adjacency list, JSON node-link, DOT/Graphviz, GML, GraphML, GEXF, Graph6/Sparse6, Pajek, SVG export
Operators	Union, complement, compose, intersection, difference, disjoint union, node relabeling, directed ↔ undirected conversion
Other	Builder DSL, graph views (lazy subgraph/filter), CompactGraph (CSR), StaticGraph (compile-time), built-in datasets (dolphins, Florentine families, les misérables), adjacency matrix/edge list conversion, parallel algorithms (malebolgia threading)

Performance

NimNet is compiled Nim with zero C/Fortran dependencies — yet it outperforms NetworkX (Python + scipy/numpy C backend) on most benchmarks.

Large graph (10,000 nodes, 50,000 edges) — Erdős-Rényi random graph:

Benchmark	NimNet	NetworkX	Result
Graph creation	0.014s	0.051s	NimNet 3.6× faster
BFS traversal	0.008s	0.015s	NimNet 1.9× faster
DFS traversal	0.006s	0.010s	NimNet 1.7× faster
Dijkstra	0.040s	0.012s	NetworkX 3.3× faster
PageRank	0.079s	0.035s	NetworkX 2.3× faster*
Connected components	0.006s	0.005s	~1× (tied)
MST (Kruskal)	0.077s	0.088s	NimNet 1.1× faster

*NetworkX PageRank uses scipy sparse matrix operations (C/Fortran BLAS); NimNet is pure Nim.

Medium graph (1,000 nodes, 5,000 edges):

Benchmark	NimNet	NetworkX	Result
Graph creation	0.001s	0.005s	NimNet 5× faster
BFS traversal	0.001s	0.002s	NimNet 2× faster
DFS traversal	0.001s	0.001s	~1× (tied)
Dijkstra	0.002s	0.001s	NetworkX 2× faster
PageRank	0.007s	0.004s	NetworkX 1.8× faster*
Connected components	0.001s	0.000s	—
MST (Kruskal)	0.005s	0.005s	~1× (tied)
Clustering	0.013s	0.015s	NimNet 1.2× faster
Triangles	0.004s	0.004s	~1× (tied)

Compiled with nim c -d:release -d:danger --opt:speed. See benchmarks/ for reproduction scripts.