Representing Long-Range Context for Graph Neural Networks with Global Attention

@inproceedings{Wu2021GraphTrans,
  title={Representing Long-Range Context for Graph Neural Networks with Global Attention},
  author={Wu, Zhanghao and Jain, Paras and Wright, Matthew and Mirhoseini, Azalia and Gonzalez, Joseph E and Stoica, Ion},
  booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
  year={2021}
}

Overview

We release the PyTorch code for the GraphTrans [paper]

Installation

To setup the Python environment, please install conda first. All the required environments are in requirement.yml.

conda env create -f requirement.yml

How to Run

To run the experiments, please refer to the commands below (taking OGBG-Code2 as an example):

# GraphTrans (GCN-Virtual)
python main.py --configs configs/code2/gnn-transformer/JK=cat/pooling=cls+norm_input.yml --runs 5
# GraphTrans (GCN)
python main.py --configs configs/code2/gnn-transformer/no-virtual/pooling=cls+norm_input.yml --runs 5
# Or to use slurm
sbatch ./slurm-run.sh ”configs/code2/gnn-transformer/JK=cat/pooling=cls+norm_input.yml --runs 5”

The config path for each dataset/model can be found in the result table below.

Results