SST

The SST (State Space Transformer) code for the paper "SST: Multi-Scale Hybrid Mamba-Transformer Experts for Long-Short Range Time Series Forecasting".

Contributions

• We propose to decompose time series into global patterns and local variations according to ranges. We identify that global patterns as the focus of long range and local variations should be captured in short range.
• To effectively capture long-term patterns and short-term variations, we leverage the patching to create coarser PTS in long range and finer PTS in short range. Moreover, we introduce a new metric to precisely quantify the resolution of PTS.
• We propose a novel hybrid Mamba-Transformer experts architecture SST, with Mamba as a global patterns expert in long range, and LWT as a local variations expert in short range. A long-short router is designed to adaptively integrate the global patterns and local variations. With Mamba and LWT, SST is highly scalable with linear complexity O(L) on time series length L.

Getting Started

Environment

• python 3.10.13
• torch 1.12.1+cu116
• mamba-ssm 1.2.0.post1
• numpy 1.26.4
• transformers 4.38.2

The installation of mamba-ssm package can refer to https://github.com/state-spaces/mamba.

Run

To run SST on various dataset, run corrrsponidng dataset .sh files in the scripts folder.

For exmaple, run SST on the Weather dataset: ./weather.sh

Dataset

You can download all the datasets from the "Autoformer" project. Creatae a dataset folder in the current directory and put the downloaded datasets into dataset folder.

Acknowledgement

We would like to greatly thank the following awesome projects:

Mamba (https://github.com/state-spaces/mamba)

PatchTST (https://github.com/yuqinie98/PatchTST)

LTSF-Linear (https://github.com/cure-lab/LTSF-Linear)

Autoformer (https://github.com/thuml/Autoformer)

Cite

If you find this repository useful for your work, please consider citing the paper as follows (bib format from arxiv):

@misc{xu2024sstmultiscalehybridmambatransformer,
      title={SST: Multi-Scale Hybrid Mamba-Transformer Experts for Long-Short Range Time Series Forecasting}, 
      author={Xiongxiao Xu and Canyu Chen and Yueqing Liang and Baixiang Huang and Guangji Bai and Liang Zhao and Kai Shu},
      year={2024},
      eprint={2404.14757},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2404.14757}, 
}