There is the code of our paper

We adopt three widely-used benchmarks in time series anomaly task to evaluate our network. We have placed the pre-processed datasets in folder dataset/pocessed.

The NeurIPS-TS Benchmark mentioned in our paper can be found at:

https://github.com/datamllab/tods/tree/benchmark/benchmark/synthetic

Install Python >= 3.8, PyTorch >= 1.9 with CUDA.

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
pip install -r requirements.txt

You can train the network by running the file train.py, for example:

python train.py --dataset=SWaT --TA_replacing=0.5 --num_FD_perturbation_positions=15

After training, you will get the model parameters in the file logs, which is the default path.

If you need to evaluate the network, please select your dataset, trained model and corresponding parameters. Then run the file evaluate.py, for example:

python evaluate.py --dataset=SWaT --model_dict=logs\Example_SWaT\Example_SWaT_model.pt --parameters_dict=logs\Example_SWaT\state\Example_SWaT.pt --window_sliding=16 

If you want to save the evaluations results to the path and file you specify, please run as following:

python evaluate.py --matrics_outfile=path\file_name.txt --dataset=SWaT --model_dict=logs\Example_SWaT\Example_SWaT_model.pt --parameters_dict=logs\Example_SWaT\state\Example_SWaT.pt

We appreciate the following github repos a lot for their valuable code:

https://github.com/Jhryu30/AnomalyBERT