GCL-SPAN

This is the code for the paper "Spectral Augmentation for Self-Supervised Learning on Graphs" accepted by ICLR 2023.

Requirement

Code is tested in Python 3.10.10. Some major requirements are listed below:

pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu117
pip install torch_geometric
pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-1.13.0+cu117.html
pip install dgl
pip install networkx
pip install numba

Run the code

Both node classification (full batch) and graph classification (minibatch) tasks are supported:

to launch node classification

python unsupervised_node.py

to launch graph classification

python unsupervised_graph.py

Cite

Please cite our paper if you find this repo useful for your research or development.

@inproceedings{lin2022spectral,
  title={Spectral Augmentation for Self-Supervised Learning on Graphs},
  author={Lin, Lu and Chen, Jinghui and Wang, Hongning},
  booktitle={International Conference on Learning Representations},
  year={2023}
}