ChebNetII: A Python repository from ivam-he

Convolutional Neural Networks on Graphs with Chebyshev Approximation, Revisited

This repository contains a PyTorch implementation of our NeruIPS2022 paper "Convolutional Neural Networks on Graphs with Chebyshev Approximation, Revisited."

Environment Settings

pytorch 1.11.0
numpy 1.23.4
torch-geometric 1.7.2
tqdm 4.64.1
scipy 1.9.3
seaborn 0.12.0
scikit-learn 1.1.3
ogb 1.3.5

Datasets

We provide the small datasets in the folder '/main/data'. The ogb datasets (ogbn-arxiv and ogbn-papers100M) and non-homophilous datasets (from LINKX ) can be downloaded automatically.

Code Structure

The folder "main" is the code for the main results from Tables 1-3 and 6-7 in the paper; The folder "non-homophilous" is the code for the results from Table 8; The folder "ogb" is the code for the results from Table 9.

Citation

@inproceedings{he2022chebnetii,
  title={Convolutional Neural Networks on Graphs with Chebyshev Approximation, Revisited},
  author={He, Mingguo and Wei, Zhewei and Wen, Ji-Rong},
  booktitle={NeurIPS},
  year={2022}
}

Contact