Leiden

Parallel implementation Leidem Algorithm on GPUs

Introduction

The Leiden algorithm is a popular method for community detection in networks, improving upon the Louvain algorithm by guaranteeing well-connected communities. While most existing implementations focus on undirected graphs and run on CPUs, this project introduces a GPU-accelerated implementation of the Leiden algorithm for directed graphs.

This implementation leverages CUDA to parallelize the key phases of the Leiden algorithm—local movement, refinement, and partition aggregation—making it significantly faster and scalable for large directed networks.

Prerequisites

• C++ compiler: g++ (version >= 9 recommended)
• CUDA Toolkit: version >= 12.0
• CMake (optional, if using CMake for building)
• Make utility
• Linux environment (tested on Ubuntu 20.04)
• Optional libraries:
  • libm (math library, usually comes with gcc)

Quick Start

Build and Run Instructions

1. Clone your repository

git clone <url_to_your_repository>
cd <repository_name>
make

./leiden <graph.txt> gpu    # Run GPU version
make clean
make
./leiden <graph.txt> cpu    # Run CPU version


Datasets and samlple outputs are also available in the repository.
For a large datasets please click on:  https://figshare.com/s/3b51e463a56e2a374bdf