MPTSNet

MPTSNet is an implementation of the paper [MPTSNet: Integrating Multiscale Periodic Local Patterns and Global Dependencies for Multivariate Time Series Classification] (AAAI 2025).

🛠️ Setup

Repository

Clone the repository:

git clone [your-repo-url] && cd MPTSNet

Installation

Create a conda environment with all dependencies:

conda create --name mptsnet python=3.8
conda activate mptsnet
pip install -r requirements.txt

🚀 Usage

Quick Start

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

Model Configuration

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

Dataset Structure

Structure your dataset as follows:

dataset/General/
└── YOUR_DATASET_NAME
    ├── YOUR_DATASET_NAME_TRAIN.ts
    └── YOUR_DATASET_NAME_TEST.ts

📊 Key Features

• Automatic period detection using FFT
• Multi-scale local feature extraction through inception modules
• Attention-based global feature fusion
• Adaptive embedding dimensions
• Enhanced interpretability

📈 Results

Training results are saved in the results/ directory, including:

• Model checkpoints
• Training logs
• Accuracy metrics

🎓 Citation

If you use MPTSNet in your research, please cite our paper:

@misc{mu2025mptsnetintegratingmultiscaleperiodic,
      title={MPTSNet: Integrating Multiscale Periodic Local Patterns and Global Dependencies for Multivariate Time Series Classification}, 
      author={Yang Mu and Muhammad Shahzad and Xiao Xiang Zhu},
      year={2025},
      eprint={2503.05582},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2503.05582}, 
}

📝 License

This project is licensed under the MIT License.