GraphSleepNet: Adaptive Spatial-Temporal Graph Convolutional Networks for Sleep Stage Classification

These are source code and experimental setup for the MASS SS3 database.

GraphSleepNet: Adaptive Spatial-Temporal Graph Convolutional Networks for Sleep Stage Classification. (IJCAI 2020)

@inproceedings{ijcai2020-184,
  title     = {GraphSleepNet: Adaptive Spatial-Temporal Graph Convolutional Networks for Sleep Stage Classification},
  author    = {Jia, Ziyu and Lin, Youfang and Wang, Jing and Zhou, Ronghao and Ning, Xiaojun and He, Yuanlai and Zhao, Yaoshuai},
  booktitle = {Proceedings of the Twenty-Ninth International Joint Conference on
               Artificial Intelligence, {IJCAI-20}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},             
  pages     = {1324--1330},
  year      = {2020},
  month     = {7},
  doi       = {10.24963/ijcai.2020/184},
  url       = {https://doi.org/10.24963/ijcai.2020/184},
}

We evaluate our model on the Montreal Archive of Sleep Studies (MASS)-SS3 dataset. The Montreal Archive of Sleep Studies (MASS) is an open-access and collaborative database of laboratory-based polysomnography (PSG) recordings. Information on how to obtain it can be found here.

• Python 3.6
• Tensorflow 1.12.0
• Keras 2.2.4
• numpy 1.15.4
• scipy 1.1.0
• scikit-learn 0.21.3

• Data preparation

  Extract DE features and make data package.

  For more details, please refer to preprocess.

• Configuration

  Write the config file in the format of the example.

  • We provide a sample config file in /config/SS3.config

• Network training and testing

  Run python train.py with -c and -g parameters.

  • -c: The configuration file.
  • -g: The number of the GPU to use. E.g., 0, 1,3. Set this to -1 if only CPU is used.

  python train.py -c ./config/SS3.config -g -1