local_curvature_profile

Requirements

To configure and activate the conda environment for this repository, run

conda env create -f environment.yml
conda activate borf 
pip install -r requirements.txt

Experiments

1. For graph classification

To run experiments for the TUDataset benchmark, run the file run_graph_classification.py. The following command will run the benchmark with the LCP based on the ORC:

python run_graph_classification.py --encoding LCP

To use a different model or add more layers, add the --layer_type and --num_layers options

python run_graph_classification.py --encoding LCP --layer_type GIN \
	--num_layers 8

2. For node classification

To run node classification, simply change the script name to run_node_classification.py. For example:

python run_node_classification.py --encoding LCP

Other encoding methods

To compare the LCP against other encoding methods, simply run

# runs graph classification with Laplacian Eigenvector Positional Encodings
python run_graph_classification.py --encoding LAPE

Citation and reference

For technical details and full experiment results, please check our paper.

@inproceedings{fesser2023effective,
  title={Effective Structural Encodings via Local Curvature Profiles},
  author={Fesser, Lukas and Weber, Melanie},
  booktitle={The Twelfth International Conference on Learning Representations},
  year={2023}
}