enyandai / ganf

Offical implementation of "Graph-Augmented Normalizing Flows for Anomaly Detection of Multiple Time Series" (ICLR 2022)

GANF

Offical implementation of "Graph-Augmented Normalizing Flows for Anomaly Detection of Multiple Time Series" (ICLR 2022). [paper]

Requirements

torch==1.7.1

Overview

• ./models: This directory includes the code of GANF as well as basline methods.
• ./checkpoint: This directory stores the trained models. The trained models for the datasets SWaT and Metr-LA are given in ./checkpoint/eval.
• ./train_water.py and ./train_traffic.py: These programs are used to train GANF on the corresponding datasets.
• ./data: This directory is used to store the datasets.

Datasets

The paper uses three datasets for experiments:

• SWaT: This water system dataset can be requested from iTrust. We utilze the attack_v0 data in Dec/2015 for experimentation. You may need to first convert the file format to .csv to use our code. Then, use ./dataset.py to perform train/val/test split.
• Metr-LA: This traffic dataset does not include ground-truth outliers. It can be used for exploratory studies of density estimation. The dataset can be downloaded from this GitHub.
• PMU: This power grid dataset is proprietary and we are unable to offer it for public use.

Experiments

To train a GANF model on SWaT, run the bash script:

bash train_water.sh

The training log will be located at ./log as a reference to reproduce the results in the paper.

We also provide trained models in ./checkpoint/eval for evaluation. You can call:

python eval_water.py

To train a GANF model on Metr-LA, run:

python train_traffic.py

Citation

If you find this repo useful, please cite the paper. Thank you!

@inproceedings{
dai2022graphaugmented,
title={Graph-Augmented Normalizing Flows for Anomaly Detection of Multiple Time Series},
author={Enyan Dai and Jie Chen},
booktitle={International Conference on Learning Representations},
year={2022},
url={https://openreview.net/forum?id=45L_dgP48Vd}
}