---

Towards Graph-level anomaly detection via deep evolutionary mapping

Open-sourced implementation for GmapAD - KDD 2023.

GmapAD is a graph-level anomaly detection framework with specially designed explainable graph mapping that maps graphs into a latent space where anomalies can be effectively detected. GmapAD's framework is shown as above.

Python Dependencies

Our proposed GmapAD framework is implemented in Python 3.7 and major libraries include:

• Pytorch = 1.11.0+cu102
• [PyG] (https://pytorch-geometric.readthedocs.io/en/latest/) torch-geometric=2.1.0

More dependencies are provided in requirements.txt.

To Run

python src/main.py

Datasets

All datasets used in this paper are from previous works and the brain network datasets can be downloaded from BrainNetDatasets and graph classification datasets can be downloaded from GraphClsDatasets.

Baselines

As provided in the Appendix of our manuscript, all baselines and their URLs are:

WWL (https://github.com/BorgwardtLab/WWL).

g-U-Nets (https://github.com/HongyangGao/Graph-U-Nets).

SAGPool (https://github.com/inyeoplee77/SAGPool).

DIFFPOOL (https://github.com/RexYing/diffpool).

GMT (https://github.com/JinheonBaek/GMT).

OCGIN (https://github.com/LingxiaoShawn/GLOD-Issues).

OCGTL (https://github.com/boschresearch/GraphLevel-AnomalyDetection).

GLocalKD (https://github.com/RongrongMa/GLocalKD).

iGAD (https://github.com/graph-level-anomalies/iGAD/tree/main).