Global Interpretable Graph-level Anomaly Detection Via Prototype

This is the code for GLADPro - KDD-2025. Thanks to the PyG library.

The paper is now available at ACM. If you use our code or results, please kindly cite our paper.

@inproceedings{yang2025global,
  title={Global Interpretable Graph-level Anomaly Detection via Prototype},
  author={Yang, Zhenyu and Zhang, Ge and Wu, Jia and Yang, Jian and Xue, Shan and Beheshti, Amin and Peng, Hao and Sheng, Quan Z},
  booktitle={Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2},
  pages={3586--3597},
  year={2025}
}

System requirement

Programming language

Python 3.10.0

Python Packages

PyTorch 1.12.0
CUDA 11.6.1
torch-geometric 2.2.0
numpy   1.22.3
scikit-learn 1.1.1
scipy   1.8.0
networkx  2.8.4

Run the demo code

For Mutagen datasets, run with defualt setting

python3 main.py

For BA-TYPE dataset, run with

python3 main.py --dataset BA-TYPE --n_prot 3 --regular 500 --hidden_dim 128 --out_dim 64

For MUTAG dataset, run with

python3 main.py --dataset MUTAG --epochs 500 --lr 1e-4

For PROTEIN dataset, run with

python3 main.py --dataset PROTEINS --hidden_dim 128 --out_dim 64 --epochs 500 

For DD dataset, run with

python3 main.py --dataset DD --regular 100 --epochs 500 

For IMDB-BINARY dataset, run with

python3 main.py --dataset IMDB-BINARY --regular 100 --epochs 100