LMSPS

This repository contains a PyTorch implementation of the preprint paper: Long-range Meta-path Search on Large-scale Heterogeneous Graphs.

Requirements

All experiments are conducted on Tesla V100 GPU (16GB).

Please install the requirements using the following command. (The python version is 3.9.18)

pip install -r requirements.txt
git clone https://github.com/Yangxc13/sparse_tools.git --depth=1
cd sparse_tools
python setup.py develop
cd ..

Data preparation

For HGB datasets:

sh download_hgb_datasets.sh

For experiments on the large dataset ogbn-mag, the dataset will be automatically downloaded from OGB challenge.

Run LMSPS

You can run LMSPS on HGB and ogbn-mag based on the command in hgb and ogbn-mag, respectively.

Cite

If you use LMSPS in a scientific publication, we would appreciate citations to the following paper:

@misc{li2024longrange,
      title={Long-range Meta-path Search on Large-scale Heterogeneous Graphs}, 
      author={Chao Li and Zijie Guo and Qiuting He and Hao Xu and Kun He},
      year={2024},
      eprint={2307.08430},
      archivePrefix={arXiv},
      primaryClass={cs.AI}
}

Acknowledgment

This repository benefit from SeHGNN.