How Universal Polynomial Bases Enhance Spectral Graph Neural Networks: Heterophily, Over-smoothing, and Over-squashing
UniFilter is a polynomial graph filter utilizing a novel universal polynomial basis called UniBasis. This repository contains the source codes for UniFilter. For further details, please refer to our paper in ICML 2024 (https://arxiv.org/abs/2405.12474). Should you encounter any issues, please reach out to Keke Huang, thanks!
- pytorch 1.7.0
- torch-geometric 1.6.1
- scipy 1.9.3
- seaborn 0.12.0
- scikit-learn 1.1.3
- ogb 1.3.1
- gdown
Please acquire all the data from ChebNet II and put the data in the subfolder './data'. The ogb datasets (ogbn-arxiv and ogbn-papers100M) and non-homophilous datasets (from LINKX ) can be downloaded automatically.
Please create a folder named 'pretrained' before running.
Please cite our paper if it is relevant to your work, thanks!