Disentangled Parameter-Efficient Linear Model for Long-Term Time Series Forecasting (DASFAA 2026)

April 21, 2026 ยท View on GitHub

arXiv DOI license

The official implementation of paper "Disentangled Parameter-Efficient Linear Model for Long-Term Time Series Forecasting" (DASFAA 2026)

Requirements

We recommend using the latest versions of dependencies. However, you can refer to the environment.yml file to set up the same environment as we used.

Dataset

All datasets are stored as CSV files and compressed in GZ format. Please place the datasets in the ./dataset directory.

  • For the M5 dataset, we recommend downloading it from M5-methods and preprocessing it using preprocessing/M5.py.
  • For other datasets, we recommend downloading them from Autoformer.

Usage

All experiments can be reproduced using the scripts/DiPE.sh script.

Citation

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

@misc{zhao2026disentangledparameterefficientlinearmodel,
      title={Disentangled Parameter-Efficient Linear Model for Long-Term Time Series Forecasting}, 
      author={Yuang Zhao and Tianyu Li and Jiadong Chen and Shenrong Ye and Fuxin Jiang and Xiaofeng Gao},
      year={2026},
      eprint={2411.17257},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2411.17257}, 
}

License

This repo is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments