GraphEm: geometric graph embedding and radial node ranking

GraphEm builds a low-dimensional graph layout whose radial coordinate can be used as an empirical node-ranking score. This site documents the portable JAX reference package, distributed as graphem-jax.

The H100-qualified production CUDA implementation is maintained separately in graphem-rapids. It uses the same GraphEm algorithmic family but has a different execution and evidence contract.

Important

Radial rank is a heuristic proxy. It is not guaranteed to reproduce every centrality measure and it is not a guaranteed influence-maximization method. Validate it against task-appropriate baselines on held-out graphs or seeds.

Documentation

User guide

• Installation
  • Reference JAX package
  • JAX accelerator installation
  • Verification
  • Troubleshooting
• Quick start
  • Create an embedding
  • Visualize a two-dimensional layout
  • Evaluate radial seed selection
  • Measure centrality correlation
  • Next steps
• Tutorials
  • Graph families
  • Radial ranking and ties
  • Centrality evaluation
  • Influence experiments
  • Real-world datasets
  • Replacing one benchmark cell
• API reference
  • Core embedder
  • Nearest-neighbor index
  • Graph generators
  • Radial selection and influence evaluation
  • Datasets
  • Visualization and statistics
  • Benchmark helpers
• Contributing to GraphEm
  • Set up a development environment
  • Run the core Python and Sphinx gates
  • Documentation rules
  • Benchmark evidence
  • Submitting changes

Project links

• Source repository

• CUDA implementation

• Issue tracker

• Manuscript

• PyPI package

Indices

• Index

• Module Index

• Search Page