PolyGCL: GRAPH CONTRASTIVE LEARNING via Learnable Spectral Polynomial Filters

This repository contains a PyTorch implementation of ICLR 2024 paper "PolyGCL: GRAPH CONTRASTIVE LEARNING via Learnable Spectral Polynomial Filters".

Environment Settings

• pytorch 1.11.0
• numpy 1.20.3
• torch-geometric 1.7.2
• dgl-cu113 0.8.2
• scipy 1.7.1
• seaborn 0.11.2
• scikit-learn 0.24.2

Datasets

We provide the small datasets in the folder 'data'. You can access the heterophilic datasets and the large heterophilic graph arXiv-year via heterophilous-graphs and LINKX respectively.

Reproduce the results

On real-world datasets

You can run the following commands directly.

sh exp_PolyGCL.sh

Heterophilic datasets:

cd HeterophilousGraph
sh exp_PolyGCL.sh

Large heterophilic graph arXiv-year:

cd non-homophilous
sh exp_PolyGCL.sh

On synthetic datasets

Generate the cSBM data firstly.

cd cSBM
sh create_cSBM.sh

Then run the following command directly.

sh run_cSBM.sh

Acknowledgements

This project includes code or ideas inspired by the following repositories:

• ChebNetII
• MVGRL
• DGI

Citation

@inproceedings{
    chen2024polygcl,
    title={Poly{GCL}: {GRAPH} {CONTRASTIVE} {LEARNING} via Learnable Spectral Polynomial Filters},
    author={Jingyu Chen and Runlin Lei and Zhewei Wei},
    booktitle={The Twelfth International Conference on Learning Representations},
    year={2024},
    url={https://openreview.net/forum?id=y21ZO6M86t}
}

Contact

If you have any questions, please feel free to contact me with [email protected].