Mitigating Oversmoothing Through Reverse Process of GNNs for Heterophilic Graphs

This repository contains a PyTorch implementation of "Mitigating Oversmoothing Through Reverse Process of GNNs for Heterophilic Graphs" (ICML, 2024).

Python environment setup with Conda

conda create --name rep python=3.8
conda activate rep
conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.6 -c pytorch -c nvidia
conda install pyg -c pyg
pip install seaborn
pip install wandb
conda install pytorch-sparse -c pyg
pip install torchdiffeq
pip install networkx

Running the code

You can run the code using script files in script/ directory.
For example, you can run GRAND-rep for the squirrel dataset with ./script/grand_rep.sh.
Also, you can run GCN-rep for the squirrel dataset with ./script/gcn_rep.sh.

Plot the graph

If you want to plot the graph smoothing level, use the --plot-gsl argument.
If you want to visualize a result for the minesweeper dataset, use the --plot-ms argument.
You can directly run ./script/plot.sh.