Official repository for NeurIPS'23 paper: GraphPatcher: Mitigating Degree Bias for Graph Neural Networks via Test-time Augmentation

Hi all, this is the official repository for NeurIPS 2023 paper: GraphPatcher: Mitigating Degree Bias for Graph Neural Networks via Test-time Augmentation. Our paper can be found at [OpenReview link]. We sincerely apprecaite your interests in our projects!

Instruction

To reproduce our experiment, you first need to train a GNN model that GraphPatcher improves on. We provide sample bash scripts for Cora, Citeseer, and Pubmed in bash_script and you can simply run:

bash bash_script/cora_gcn.sh 

The resultant model checkpoint will be saved in the temp directory.

Then to conduct the test-time augmentation from GraphPatcher, we have prepared bash scripts in the bash_script directory. Training a GraphPatcher can be done by:

bash bash_script/cora.sh <GPU_ID>

Predictions given by GraphPatcher with different numbers of patched node are saved in the outputs directory.

Dependencies and Environments

The package we use include:

* DGL 0.9.0
* PyTorch 1.12.0

Cite

If you find this repository useful in your research, please cite our paper:

@article{ju2023graphpatcher,
  title={GraphPatcher: Mitigating Degree Bias for Graph Neural Networks via Test-time Augmentation},
  author={Ju, Mingxuan and Zhao, Tong and Yu, Wenhao and Shah, Neil and Ye, Yanfang},
  journal={Advances in neural information processing systems},
  year={2023}
}

Contact

Mingxuan Ju (mju2@nd.edu)