Is Mamba Effective for Time Series Forecasting?

:loudspeaker: Latest Updates

• 2024 Dec-12 : Accepted by Neurocomputing [Neurocomputing].
• 2024 Apr-27 : Updated our paper (v3). [arXiv] [PDF].
• 2024 Apr-02 : Updated our paper and released the code. You can refer to [arXiv] for more details.

S-Mamba

Contributions :trophy:

• We propose S-Mamba, a Mamba-based model for time series forecasting, which delegates the extraction of inter-variate correlations and temporal dependencies to a bidirectional Mamba block and a Feed-Forward network.
• We evaluate the performance of S-Mamba, which not only has low GPU memory required and short time for forecasts but also maintains superior performance compared to the representative and state-of-the-art models.
• We conduct extensive experiments to further delve deeper into Mamba's potential in time series forecasting tasks.

🌟 Getting Start

🛠️ Installation

pip install -r requirements.txt

📦 Datasets

The datasets can be obtained from here.

🚀 Train and evaluate

# ECL
bash ./scripts/multivariate_forecasting/ECL/S_Mamba.sh
# Exchange
bash ./scripts/multivariate_forecasting/Exchange/S_Mamba.sh
# Traffic
bash ./scripts/multivariate_forecasting/Traffic/S_Mamba.sh
# Weather
bash ./scripts/multivariate_forecasting/Weather/S_Mamba.sh
# Solar-Energy
bash ./scripts/multivariate_forecasting/SolarEnergy/S_Mamba.sh
# PEMS
bash ./scripts/multivariate_forecasting/PEMS/S_Mamba_03.sh
bash ./scripts/multivariate_forecasting/PEMS/S_Mamba_04.sh
bash ./scripts/multivariate_forecasting/PEMS/S_Mamba_07.sh
bash ./scripts/multivariate_forecasting/PEMS/S_Mamba_08.sh
# ETT
bash ./scripts/multivariate_forecasting/ETT/S_Mamba_ETTm1.sh
bash ./scripts/multivariate_forecasting/ETT/S_Mamba_ETTm2.sh
bash ./scripts/multivariate_forecasting/ETT/S_Mamba_ETTh1.sh
bash ./scripts/multivariate_forecasting/ETT/S_Mamba_ETTh2.sh

:pray: Acknowledgement

We are grateful for the following awesome projects when implementing S-Mamba:

• iTransformer
• Mamba

🤝 Join the Collaboration

We warmly welcome your participation! Whether you have ideas for improvements, feature additions, or bug fixes, feel free to open an issue or submit a pull request.

Contributors:

📖 Citation

If you find our work useful in your research, please consider citing us:

@article{wang2024mamba,
  title={Is mamba effective for time series forecasting?},
  author={Wang, Zihan and Kong, Fanheng and Feng, Shi and Wang, Ming and Yang, Xiaocui and Zhao, Han and Wang, Daling and Zhang, Yifei},
  journal={Neurocomputing},
  pages={129178},
  year={2024},
  publisher={Elsevier}
}