# MSPT

**Repository Path**: btbuvislab/mspt

## Basic Information

- **Project Name**: MSPT
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2025-07-16
- **Last Updated**: 2025-07-18

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# **MSPT: A Framework of the Multi-Scale Patch Decomposition Transformer and Visual Analytics for Enhancing Time Series Forecasting

This is the [PyTorch](https://pytorch.org/) and [Lightning](https://lightning.ai/) implementation of our paper: ***MSPT: A Framework of the Multi-Scale Patch Decomposition Transformer and Visual Analytics for Enhancing Time Series Forecasting***. 

## Abstract
Transformer-based models have shown strong performance in time series forecasting, but often struggle to capture temporal patterns at multiple scales and lack interpretability. We propose the Multi-Scale Patch Decomposition Transformer (MSPT), a framework that integrates a forecasting model and a visual analytics system. The MSPT model introduces a multi-scale temporal patching mechanism into the Transformer architecture, enabling the decomposition of time series into scale-specific components. It combines intra-patch, inter-patch, and inter-channel attention to capture rich temporal dependencies and improve forecasting accuracy. To support interpretability, we develop MSPT-vis, a visual analytics system that reveals attention distributions, patching effects, and prediction errors. Qualitative evaluations on real-world datasets demonstrate its utility for analyzing model behavior and guiding parameter configuration. Extensive experiments on six public datasets show that MSPT outperforms three state-of-the-art baselines. The whole framework is available at: \url{https://gitee.com/btbuvislab/mspt}.


## Dependency Setup

* Create a conda virtual environment
  ```bash
  conda create -n MSPT python=3.10
  conda activate MSPT
  ```
* Install Python Packages
  ```bash
  pip install -r requirements.txt
  ```

## Run MSPT

Please use `python main.py` to run the experiments. Please use the `-h` or `--help` argument for details.


Example training commands:
* Run all benchmarks
    ```bash
    python main.py ltf

    # equivalent
    python main.py ltf --dataset all --pred_len all
    ```
* Run specific benchmarks
    ```bash
    python main.py ltf --dataset etth1 etth2 --pred_len 96 192 336




If you find this repo useful, please consider citing our paper:
```
@article{Hao2025MSPT,
      title={MSPT: A Framework of the Multi-Scale Patch Decomposition Transformer and Visual Analytics for Enhancing Time Series Forecasting}, 
      author={Jinlong Hao, Yi Chen, Cheng LV, Xianpeng Yuan, Haomiao Wang, Qinghui Zhang, Yu Dong},
      year={2025}
}
```