This is the code associated with the submission "AffinityTune: A Prompt-Tuning Framework for Few-Shot Anomaly Detection on Graphs".
conda install pytorch==1.12.0 torchvision==0.13.0 torchaudio==0.12.0 cudatoolkit=11.6 -c pytorch -c conda-forge
conda install -c dglteam/label/cu116 dgl
conda install scikit-learn
pip install pygod
For real-world datasets, they can be downloaded from https://github.com/pygod-team/data and placed in the "dataset" folder. Also, you can inject anomalies by executing "python inject_ano.py".
Run python pretrain.py to perform the first stage of the framework and obtain the trained GNN model.
Run python tune.py to perform prompt tuning and anomaly detection.