MTSCI: Multivariate Time Series Consistent Imputation

It's a pytorch implementation of paper "MTSCI: A Conditional Diffusion Model for Consistent Imputation in Incomplete Time Series" accepted in CIKM2024.

Requirements

pip install -r requirements.txt

Datasets

• ETT
• Weather
• METR-LA (you should unzip the METRLA.tar.gz tar -zxvf METRLA.tar.gz)

How to run

Train MTSCI from scratch

cd scripts/
bash ETT_point.sh
bash ETT_block.sh
bash Weather_point.sh
bash Weather_block.sh
bash METRLA_point.sh
bash METRLA_block.sh

Test MTSCI from trained model

First, set the scratch is False in scripts. Then, run these scripts.

bash ETT_point.sh
bash ETT_block.sh
bash Weather_point.sh
bash Weather_block.sh
bash METRLA_point.sh
bash METRLA_block.sh

Citation

If you find this repo useful, please cite our paper. Thanks for your attention.

@inproceedings{zhou2024mtsci,
  title={MTSCI: A Conditional Diffusion Model for Multivariate Time Series Consistent Imputation},
  author={Zhou, Jianping and Li, Junhao and Zheng, Guanjie and Wang, Xinbing and Zhou, Chenghu},
  booktitle={Proceedings of the 33rd ACM International Conference on Information and Knowledge Management},
  pages={3474--3483},
  year={2024}
}