Universal representation learning for multivariate time series using the instance‐level and cluster‐level supervised contrastive learning

This repository contains the PyToch implementation of the SupCon-TS as described in the paper.

Datasets Preparation

Download all the UEA Multivariate time series archive datasets from [here] (http://www.timeseriesclassification.com/)

The data structure should look like this:

SupCon-TSC
├── ...
├── data
│   ├── BasicMotion
│   ├── Cricket

Overview

This work has proposed supervised contrastive learning for time series classifcation (SupCon-TSC). This model is based on the instance-level and cluster-level supervised contrastive learning approaches to learn the discriminative and universal representation for the multivariate time series dataset

Train

To train the model, please run the command below:

python main.py --lr1 0.001 --lr2 0.001 --batch_size_emb 10 --batch_size_cl 5 --Epoch 100 --name_dataset BasicMotions

Acknowledgments

This project is based on the following open-source projects. We thank their authors for making the source code publicly available.

1. SupContrast (link)
2. tsai (link)

Citation

Moradinasab, N., Sharma, S., Bar-Yoseph, R., Radom-Aizik, S., C Bilchick, K., M Cooper, D., Weltman, A. and Brown, D.E., 2024. Universal representation learning for multivariate time series using the instance-level and cluster-level supervised contrastive learning. Data Mining and Knowledge Discovery, pp.1-27.