LargeGT: Graph Transformers for Large Graphs

Source code for the paper Graph Transformers for Large Graphs

by Vijay Prakash Dwivedi, Yozen Liu, Anh Tuan Luu, Xavier Bresson, Neil Shah and Tong Zhao.

The paper proposes LargeGT which is a scalable Graph Transformer framework designed to efficiently handle large-scale graphs, featuring a combination of fast neighborhood sampling and local-global attention mechanisms.

1. Installation

To setup the Python environment with conda, follow these instructions.

2. Download data

Download preprocessed data by running this script as:

cd data  
bash download_data.sh

3. Run experiments

To run an experiment, run the command:

python main.py --dataset <dataset name> --sample_node_len <value of K>

For example:

python main.py --dataset ogbn-products --sample_node_len 100

To reproduce results, follow these steps.

4. Acknowledgement

This code repository leverages the open-source codebases released by GOAT and NAGphormer.

5. Reference

📃 Paper on arXiv

@article{dwivedi2023graph,
  title={Graph Transformers for Large Graphs},
  author={Dwivedi, Vijay Prakash and Liu, Yozen and Luu, Anh Tuan and Bresson, Xavier and Shah, Neil and Zhao, Tong},
  journal={arXiv preprint arXiv:2312.11109},
  year={2023}
}

6. Contact

Please contact vijaypra001@e.ntu.edu.sg for any questions.