Skip to content

Latest commit

 

History

History
62 lines (43 loc) · 2.33 KB

readme.md

File metadata and controls

62 lines (43 loc) · 2.33 KB
Raw

CULiNGAM - CUDA Accelerated LiNGAM Analysis

Introduction

CULiNGAM is a high-performance library that accelerates Linear Non-Gaussian Acyclic Model (LiNGAM) analysis on GPUs. It leverages the computing power of NVIDIA GPUs to provide fast and efficient computations for LiNGAM applications, making it ideal for large-scale data analysis in fields such as bioinformatics, economics, and machine learning. CULiNGAM can be more more than 32x faster than CPU implementations: https://arxiv.org/abs/2403.03772.

Installation

To install CULiNGAM, you need a system with NVIDIA CUDA installed and a compatible GPU. This library is tested with CUDA 12.2 and designed for GPUs with the architecture sm_86, although it may work with other versions and architectures with appropriate adjustments.

Prerequisites

  • NVIDIA GPU with CUDA Compute Capability 6.0 or higher
  • CUDA Toolkit 12.2 or compatible version
  • Python 3.6 or newer
  • Numpy
  • A C++17 compatible compiler

Installing from PyPI

CULiNGAM is available on PyPI and can be easily installed with pip:

pip install culingam

Installing CULiNGAM manually

  1. Clone the repository to your local machine:
git clone https://github.com/Viktour19/culingam
  1. Ensure that CUDA_HOME environment variable is set to your CUDA Toolkit installation path. If not, you can set it as follows (example for default CUDA installation path):
export CUDA_HOME=/usr/local/cuda-12.2
  1. Optionally, set the GPU_ARCH environment variable to match your GPU architecture if it differs from the default sm_86:
export GPU_ARCH=sm_xx  # Replace xx with your GPU's compute capability
  1. Install the library using pip:
pip install .

Usage

After installation, you can use CULiNGAM in your Python projects to accelerate LiNGAM analysis. Here's a simple example to get started:

https://github.com/Viktour19/culingam/blob/main/examples/basic.py

Support

For bugs, issues, and feature requests, please submit a report to the repository's issue tracker. Contributions are also welcome.

Author

Victor Akinwande

License and Acknowledgment

This project is licensed under the MIT License - see the LICENSE file for details. A good amount of the code is adapted from sequential implementations of the LiNGAM algorithms present in CULiNGAM: https://github.com/cdt15/lingam.