Code for paper "Strengthening structural baselines for graph classification using Local Topological Profile" J. Adamczyk, W. Czech, accepted at ICCS 2023 (International Conference on Computational Science). Preprint is available at arXiv.
In this paper, we present Local Topological Profile (LTP), method optimizing and extending popular Local Degree Profile (LDP) baseline for topological graph classification.
To reproduce results from the paper, refer to main.py
and function perform_experiment()
in perform_experiment.py
. Change values as needed. Datasets are downloaded automatically upon the first use.
Dependencies:
- Numpy
- NetworKit
- NetworkX
- Pandas
- PyTorch
- PyTorch Geometric
- Scikit-learn
- tqdm
- zstd
Optionally, to speed up estimators on Intel architectures: Intel(R) Extension for Scikit-learn