This repository supplements our paper "DTAAD: Dual Tcn-Attention Networks for Anomaly Detection in Multivariate Time Series Data" accepted in arXiv. For more information, please see: https://arxiv.org/abs/2302.10753. This is a refactored version of the code used for results in the paper for ease of use. Follow the below steps to replicate each cell in the results table.

This code needs Python-3.7 or higher.

$ git clone https://github.com/Yu-Lingrui/DTAAD  # clone
$ cd DTTAD
$ pip3 install torch==1.11.0+cu113 torchvision==0.12.0+cu113 torchaudio==0.11.0 --extra-index-url https://download.pytorch.org/whl/cu113
$ pip3 install -r requirements.txt

Preprocess all datasets using the command

$ python3 preprocess.py SMAP MSL SWaT WADI SMD MSDS UCR MBA NAB

Distribution rights to some datasets may not be available. Check the readme files in the ./data/ folder for more details. If you want to ignore a dataset, remove it from the above command to ensure that the preprocessing does not fail.

To run a model on a dataset, run the following command:

$ python3 main.py --model <model> --dataset <dataset> --retrain

where <model> can be either of 'DTAAD', 'GDN', 'MAD_GAN', 'MTAD_GAT', 'MSCRED', 'USAD', 'OmniAnomaly', 'LSTM_AD', and dataset can be one of 'SMAP', 'MSL', 'SWaT', 'WADI', 'SMD', 'MSDS', 'MBA', 'UCR' and 'NAB. To train with 20% data, use the following command

python3 main.py --model <model> --dataset <dataset> --retrain --less

You can use the parameters in src/params.json to set values in src/constants.py for each file.

For ablation studies, use the following models: 'DTAAD_Tcn_Local', 'DTAAD_Tcn_Global', 'DTAAD_Callback', 'DTAAD_Transformer'.

The output will provide anomaly detection and diagnosis scores and training time. For example:

$ python3 main.py --model DTAAD --dataset SMAP --retrain 
Using backend: pytorch
Creating new model: DTAAD
Training DTAAD on SMAP
Epoch 0,        L1 = 0.14307714637902505
Epoch 1,        L1 = 0.028436464400400804
Epoch 2,        L1 = 0.02235450599727518
Epoch 3,        L1 = 0.02094299886140656
Epoch 4,        L1 = 0.019440197744748654
100%|███████████████████████████████████████████████████████████████████| 5/5 [00:04<00:00,  1.57it/s]
Training time:     4.1132 s
Testing DTAAD on SMAP
{'FN': 0,
 'FP': 167,
 'Hit@100%': 1.0,
 'Hit@150%': 1.0,
 'NDCG@100%': 1.0,
 'NDCG@150%': 1.0,
 'TN': 7590,
 'TP': 748,
 'f1': 0.9022741135930116,
 'precision': 0.8219863298635374,
 'recall': 0.9999999866310163,
 'threshold': 0.23806398726921338}

All outputs can be run multiple times to ensure statistical significance.

Our paper is available in the Proceedings of arXiv: https://arxiv.org/pdf/2302.10753.pdf. If you use this work, please cite using the following bibtex entry.

@article{yu2023dtaad,
  title={DTAAD: Dual Tcn-Attention Networks for Anomaly Detection in Multivariate Time Series Data},
  author={Yu, Lingrui},
  journal={arXiv preprint arXiv:2302.10753},
  year={2023}
}

BSD-3-Clause.
Copyright (c) 2022, Lingrui Yu, Shreshth Tuli.
All rights reserved.

See License file for more details.