The code for our paper "Generative Cooperative Network for Unsupervised Outlier Detection".
Our project is implemented using the PyTorch framework. The following are the necessary dependencies:
- python==3.11.5
- pytorch==2.1.0
- CUDA==12.1
- numpy==1.26.0
- pandas==2.1.1
- matplotlib==3.8.0
- scipy==1.11.3
- sklearn==1.3.2
We provide 13 datasets in ./data, as referenced in our paper. Parameters for these datasets are detailed in parameters.txt. Below is an example illustrating the training process for our models (ACE & GCN-UOD) on the SpamBase dataset. By default, the results, including model weights, AUC scores, inference times, and training history, are stored in ./results.
-
Train ACE to describle a preliminary boundary that separates significant outliers from normal data:
python main.py --model ACE --data SpamBase --lr_E 1e-5 --epochs 600 --alpha 0.05 -
Copy the weights of ACE to
./weights:cp path/to/weights ./weights/ACE_Spambase.pt -
Train GCN-UOD to describle a more resonable boundary:
python main.py --model GCN-UOD --data SpamBase --lr_E 1e-4 --lr_G 1e-4 --epochs 500 --drop_one