nx-rustworkx

A NetworkX 3.x backend that runs selected graph algorithms through rustworkx. Keep import networkx as nx, keep your original node IDs, and let NetworkX dispatch the work.

• 111 NetworkX algorithms are implemented.
• Measured cutoffs avoid dispatch when conversion would cost more than it saves.
• Unsupported calls stay on NetworkX when the input is an nx.Graph.
• A rustworkx-backed graph object avoids repeated conversion for supported workloads.

Best for CPU-heavy whole-graph algorithms on graphs from a few hundred nodes up; deliberately not for tiny graphs or one-off linear-time calls, which stay on NetworkX. Method and full numbers: benchmarks.

Quick start

pip install nx-rustworkx

Set rustworkx as a preferred backend:

NETWORKX_BACKEND_PRIORITY=rustworkx python your_script.py

Or configure it after importing NetworkX:

import networkx as nx

G = nx.erdos_renyi_graph(2_000, 0.01, seed=1)
nx.config.backend_priority = ["rustworkx"]

scores = nx.betweenness_centrality(G)

NetworkX sees a normal nx.Graph. For large enough supported calls, nx-rustworkx converts it, runs the rustworkx kernel, and remaps the answer. Small or unsupported calls remain on NetworkX.

Choose a mode

Automatic dispatch

Best default. NetworkX asks the backend whether each call is worth running.

Explicit backend

Use backend="rustworkx" on one call to try the kernel directly.

Native backend graph

Build a RustworkxGraph once when repeated conversion is the bottleneck.

Continue with usage and configuration, or see the full algorithm list and caveats.