MPTSNet is an implementation of the paper [MPTSNet: Integrating Multiscale Periodic Local Patterns and Global Dependencies for Multivariate Time Series Classification] (AAAI 2025).
Clone the repository:
git clone https://github.com/MUYang99/MPTSNet.git && cd MPTSNet
Create a conda environment with all dependencies:
conda create --name mptsnet python=3.8
conda activate mptsnet
pip install -r requirements.txt
The project supports various UEA time series classification datasets from Time Series Classification Repository. Place your dataset in the dataset/General/
and dataset/UEA/
directory. For quick access to all datasets used in the paper, you can visit this link.
Train MPTSNet:
python train.py
Evaluate MPTSNet:
python eval.py
The model automatically adapts its parameters based on input dimensions:
- Embedding dimensions are scaled based on input channels
- Periodic patterns are automatically detected using FFT
- Early stopping and learning rate scheduling are implemented for optimal training
Structure your dataset as follows:
dataset/General/
└── YOUR_DATASET_NAME
├── YOUR_DATASET_NAME_TRAIN.ts
└── YOUR_DATASET_NAME_TEST.ts
Training results are saved in the results/
directory, including:
- Model checkpoints
- Training logs
- Accuracy metrics
If you find it's useful, please cite our paper:
@inproceedings{mu2025mptsnet,
title={MPTSNet: Integrating Multiscale Periodic Local Patterns and Global Dependencies for Multivariate Time Series Classification},
author={Mu, Yang and Shahzad, Muhammad and Zhu, Xiao Xiang},
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
volume={39},
number={18},
pages={19572--19580},
year={2025}
}
This project is licensed under the MIT License.