RoCA: Robust Contrastive One-class Time Series Anomaly Detection with Contaminated Data

This repository provides the implementation of the RoCA: Robust Contrastive One-class Time Series Anomaly Detection with Contaminated Data method, called RoCA below.

Abstract

The accumulation of time-series signals and the absence of labels make time-series Anomaly Detection (AD) a self-supervised task of deep learning. Most methods based on normality assumptions face the following two challenges: (1) a single assumption could hardly characterize the whole normality or lead to some deviation, e.g. Contrastive Learning (CL) methods distance negative pairs, many of which consist of both normal samples, thus reducing the AD performance. (2) their basic assumption is that the training data is uncontaminated (free of anomalies), which is unrealistic in practice. When the proportion of contaminated data increases, the performance will be affected to varying degrees. This paper proposes a novel approach for time series anomaly detection based on multiple normality assumptions. We fuse the assumptions of one-class classification, contrastive learning, and autoencoder in a single learning process to characterize a complete so-called normality. On the other hand, we introduce the idea of outlier exposure in the latent space, which helps exclude the influence of abnormal samples and utilize the contained anomaly knowledge.

Citation

Link to our paper here. If you use this code for your research, please cite our paper:

We will add citation information later.

Installation

This code is based on Python 3.8, all requirements are written in requirements.txt. Additionally, we should install saleforce-merlion v1.1.1 and ts_dataset as Merlion suggested.

pip install salesforce-merlion==1.1.1
pip install -r requirements.txt

Dataset

We acknowledge the dataset's contributors, including AIOps, UCR, SWaT, and WADI. This repository already includes Merlion's data loading package ts_datasets.

AIOps (KPI, IOpsCompetition) and UCR.

1. AIOps Link: https://github.com/NetManAIOps/KPI-Anomaly-Detection
2. UCR Link: https://wu.renjie.im/research/anomaly-benchmarks-are-flawed/ and https://www.cs.ucr.edu/~eamonn/time_series_data_2018/UCR_TimeSeriesAnomalyDatasets2021.zip
3. Download and unzip the data in data/iops_competition and data/ucr respectively. e.g. For AIOps, download phase2.zip and unzip the data/iops_competition/phase2.zip before running the program.

SWaT and WADI.

1. For SWaT and WADI, you need to apply by their official tutorial. Link: https://itrust.sutd.edu.sg/itrust-labs_datasets/dataset_info/
2. Because multiple versions of these two datasets exist, we used their newer versions: SWaT.SWaT.A2_Dec2015, version 0 and WADI.A2_19Nov2019.
3. Download and unzip the data in data/swat and data/wadi respectively. Then run the swat_preprocessing() and wadi_preprocessing() functions in dataloader/data_preprocessing.py for preprocessing.

Repository Structure

conf

This directory contains experiment parameters for all models on AIOps (as IOpsCompetition in the code), UCR, SWaT, and WADI datasets.

models

Source code of CutAddPaste model.

results

Directory where the experiment result is saved.

RoCA Usage

# RoCA Method (dataset_name: IOpsCompetition, UCR, SWaT, WADI)
python coca+.py --selected_dataset <dataset_name> --device cuda --seed 2

Baselines

Anomaly Transformer(AnoTrans, AOT), AOC, RandomScore(RAS), NCAD, LSTMED, OC_SVM, IF, SR, RRCF, SVDD, DAMP, TS_AD(TCC)

We reiterate that in addition to our method, the source code of other baselines is based on the GitHub source code provided by their papers. For reproducibility, we changed the source code of their models as little as possible. We are grateful for the work on these papers.

We consult the GitHub source code of the paper corresponding to the baseline and then reproduce it. For baselines that use the same datasets as ours, we use their own recommended hyperparameters. For different datasets, we use the same hyperparameter optimization method Grid Search as our model to find the optimal hyperparameters.

Acknowledgements

Part of the code, especially the baseline code, is based on the following source code.

1. Anomaly Transformer(AOT)
2. AOC
3. Deep-SVDD-PyTorch
4. TS-TCC
5. DAMP and DAMP-python
6. LSTM_ED, SR, and IF are reproduced based on saleforce-merlion
7. RRCF
8. Metrics:affiliation-metrics