Skip to content

Latest commit

 

History

History
104 lines (72 loc) · 2.21 KB

README.md

File metadata and controls

104 lines (72 loc) · 2.21 KB
Raw

MMA

This repository is the implementation of "MLP-Mixer based Masked Autoencoders Are Effective,Explainable and Robust for Time Series Anomaly Detection". We propose the MMA framework to achieve effective, explainable, and robust time series anomaly detection.

MMA model architecture

Main results

Results on multivariate datasets

Results on the univariate time series dataset: the KDD21 dataset

Case study

The detection results on the UCR dataset

Usage

Setup

Installation

conda create -n MMA python=3.9
conda activate MMA
pip install -r requirements.txt

Prepare datasets

Download datasets from this link: ano_dataset and put them in the ano_dataset folder.

├─ano_dataset
├───ASD
├───sate   
├───synthetic
└───UCR

Visualization of the datasets is provided at https://drive.google.com/drive/folders/1ZmOJ-lAN0FfgDr6unwsU2LLubdQovv1x?usp=sharing

Reproduce the effectiveness experiment results

Run our model

sh runners\run_asd.sh
sh runners\run_sate.sh
sh runners\run_synthetic.sh
sh runners\run_ucr.sh

Run baseline models

sh runners\run_asd_other.sh
sh runners\run_sate_other.sh
sh runners\run_synthetic_other.sh
sh runners\run_ucr_other.sh

For ablation experiments and parameter sensitivity analysis, please refer to the instructions in runners\run_asd.sh

Reproduce the explainability experiment results

sh runners\run_asd_ex.sh
sh runners\run_sate_ex.sh
sh runners\run_synthetic_ex.sh
sh runners\run_ucr_ex.sh

Reproduce the robustness experiment results

sh runners\run_asd_pollute.sh
sh runners\run_sate_pollute.sh
sh runners\run_synthetic_pollute.sh
sh runners\run_ucr_pollute.sh

Acknowledgement

We appreciate the following github repo very much for the valuable code base

https://github.com/PatchTST/PatchTST

https://github.com/ibm-granite/granite-tsfm