README.md

Artifact of Clenshaw Graph Neural Networks

[Under Construction]

This repository includes the implementation for ClenshawGCN(KDD'23). By equipping GCN convolution layers with a simple residual connection, our model simulates any polynomial based on the second kind of Chebyshev polynomials!

A previous version can be viewed here. A new version is coming soon!

Install

conda env create -f environment.yml  # create a conda environment for ClenshawGNN
conda activate checkclenshaw         # enter conda environment

Reproducing Results for Geom-GCN and Citation datasets

Run the scripts altogether:

bash scripts/run_all.sh

or run one command in the script at a time.

Reproducing Results for LINKX datasets

ClenshawGCN achieves state-of-the-art results on genius and twitch-gamer. The results are as below; they will be released with the KDD'23 paper.

bash scripts/run_all_linkx.sh