Code implementation for : Graph Neural Network-Based Anomaly Detection in Multivariate Time Series(AAAI'21)
# run after installing correct Pytorch package
bash install.sh
Run to check if the environment is ready
bash run.sh cpu msl
# or with gpu
bash run.sh <gpu_id> msl # e.g. bash run.sh 1 msl
We use part of msl dataset(refer to telemanom) as demo example.
# put your dataset under data/ directory with the same structure shown in the data/msl/
data
|-msl
| |-list.txt # the feature names, one feature per line
| |-train.csv # training data
| |-test.csv # test data
|-your_dataset
| |-list.txt
| |-train.csv
| |-test.csv
| ...
# using gpu
bash run.sh <gpu_id> <dataset>
# or using cpu
bash run.sh cpu <dataset>
You can change running parameters in the run.sh.
SWaT and WADI datasets can be requested from iTrust
If you find this repo or our work useful for your research, please consider citing the paper
@inproceedings{deng2021graph,
title={Graph neural network-based anomaly detection in multivariate time series},
author={Deng, Ailin and Hooi, Bryan},
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
volume={35},
number={5},
pages={4027--4035},
year={2021}
}