Code repository for our paper "The Expressive Power of Path based Graph Neural Networks" (ICML 2024). This repository contains only parts of all experiments, the code for EXP and SR can be found here.
Clone this repository and open the directory
Add this directory to the python path. Let $PATH be the path to where this repository is stored (i.e. the result of running pwd).
export PYTHONPATH=$PYTHONPATH:$PATH
Create a conda environment (this assume miniconda is installed)
conda create --name pathGNN
Activate environment
conda activate pathGNN
Install dependencies
conda install pytorch==1.12.0 torchvision==0.13.0 torchaudio==0.12.0 -c pytorch
conda install -c pyg pyg=2.2.0
python -m pip install -r requirements.txt
Training and different experiments are tracked via wandb. If you want to make use of the tracking you need a wandb account. The first time you train a model, you will be prompted to enter you wandb API key. If you want to disable tracking you can do this in the config Configs/config.yaml.
You can recreate the experiments with the following commands. The results will be stored in the Results folder.
- PAIN on CSL:
bash Scripts/run_csl_exp.sh
- PAIN on ZINC (this will also give you our runtime evaluation):
python Exp/run_experiment.py -dataset ZINC -grid Configs/Benchmark/ZINC_pathGNN.yaml --repeats 10
- PAIN on MOLHIV:
python Exp/run_experiment.py -dataset ogbg-molhiv -grid Configs/Benchmark/molhiv_pathGNN_V3.yaml --repeats 10
-
PAIN on EXP & SR: Please refer to this repository.
-
Runtime benchmark of GIN, DS and DSS on ZINC: Please refer to this repository.
If you use our code please cite us as
@inproceedings{pathGNNs2024,
title={The Expressive Power of Path based Graph Neural Networks},
author={Drucks,Tamara and Graziani, Caterina and Jogl, Fabian and Bianchini, Monica and Scarselli, Franco and Gärtner, Thomas },
booktitle={ICML},
year={2024},
url={https://openreview.net/forum?id=io1XSRtcO8}
}