ModernTCN (ICLR 2024 Spotlight)

This is an official implementation of paper: ModernTCN: A Modern Pure Convolution Structure for General Time Series Analysis.

Our Paper

Donghao Luo and Xue Wang. ModernTCN: A Modern Pure Convolution Structure for General Time Series Analysis. In International Conference on Learning Representations, 2024. [Our paper in OpenReview].

We study the open question of how to better use convolution in time series analysis and we take a seldom-explored way in time series community to successfully bring convolution back to time series analysis.

As a pure convolution structure, our ModernTCN achieves the consistent state-of-the-art performance on five mainstream time series analysis tasks (long-term and short-term forecasting, imputation, classification and anomaly detection) while maintaining the efficiency advantage of convolution-based models, therefore providing a better balance of efficiency and performance.

ModernTCN Block

ModernTCN block design: ModernTCN block can achieve larger ERF and better capture the cross-variable dependency, therefore being more suitable for time series analysis.


Figure 1. ModernTCN block design.	Figure 2. Visualization of ERF.

Main Results

Main Results: Our ModernTCN achieves the consistent state-of-the-art performance on five mainstream time series analysis tasks with better efficiency. Block Design

Get Started

Install Python 3.7 and necessary dependencies.
```
pip install -r requirements.txt
```
Download data. You can obtain all datasets from [Times-series-library].
Long-term forecasting tasks.

We provide the long-term forecasting experiment coding in ./ModernTCN-Long-term-forecasting and experiment scripts can be found under the folder ./scripts. To run the code on ETTh2, just run the following command:

cd ./ModernTCN-Long-term-forecasting

sh ./scripts/ETTh2.sh

Short-term forecasting tasks.

We provide the short-term forecasting experiment coding in ./ModernTCN-short-term and experiment scripts can be found under the folder ./scripts. Please run the following command:

cd ./ModernTCN-short-term

sh ./scripts/M4.sh

Imputation tasks.

We provide the imputation experiment coding in ./ModernTCN-imputation and experiment scripts can be found under the folder ./scripts. To run the code on ETTh2, just run the following command:

cd ./ModernTCN-imputation

sh ./scripts/ETTh2.sh

Classification tasks.

We provide the classification experiment coding in ./ModernTCN-classification and experiment scripts can be found under the folder ./scripts. Please run the following command:

cd ./ModernTCN-classification

sh ./scripts/classification.sh

Anomaly detection tasks.

We provide the anomaly detection experiment coding in ./ModernTCN-detection and experiment scripts can be found under the folder ./scripts. To run the code on SWaT, just run the following command:

cd ./ModernTCN-detection

sh ./scripts/SWaT.sh

Contact

If you have any question or want to use the code, please contact ldh21@mails.tsinghua.edu.cn.

Citation

If you find this repo useful, please cite our paper.

@inproceedings{
donghao2024moderntcn,
title={Modern{TCN}: A Modern Pure Convolution Structure for General Time Series Analysis},
author={Luo donghao and wang xue},
booktitle={The Twelfth International Conference on Learning Representations},
year={2024},
url={https://openreview.net/forum?id=vpJMJerXHU}
}

Acknowledgement

We appreciate the following github repos a lot for their valuable code base or datasets:

https://github.com/ts-kim/RevIN

https://github.com/PatchTST/PatchTST

https://github.com/thuml/Time-Series-Library

https://github.com/facebookresearch/ConvNeXt

https://github.com/MegEngine/RepLKNet

luodhhh / ModernTCN

readme