P3Net: Systematic Deep Learning Model Selection for P300-Based Brain-Computer Interfaces

This repository contains the source code used in our research paper titled "A Systematic Deep Learning Model Selection for P300-Based Brain-Computer Interfaces". The study explores the feasibility of conducting systematic model selection combined with mainstream deep learning architectures to construct accurate classifiers for decoding P300 event-related potentials.

Features

• Diverse Model Architectures: Implementation of various deep learning models, including CNNs, LSTMs, CNN-LSTM hybrids, EEGNet, and ShallowConvNet.
• Comprehensive Evaluation: Assessment of 232 CNNs (4 datasets × 58 structures), 36 LSTMs (4 datasets × 9 structures), and 320 CNN-LSTM models (4 datasets × 80 structures) of varying complexity.
• Subject-Specific and Pooled Training: Support for both subject-specific and pooled data training approaches.
• Reproducibility: Scripts and utilities to reproduce the results presented in the study.

Installation

1. Clone the repository:

   git clone https://github.com/berdakh/P3Net.git
   cd P3Net

2. Create a virtual environment (optional but recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install required packages:

   pip install -r requirements.txt

Note: Ensure that PyTorch is installed, as it's central to the functionalities provided.

Usage

The repository includes various scripts for training and evaluating different model architectures:

• Data Loading and Utilities:

  • nu_data_loader.py: Functions for loading and preprocessing EEG data.
  • nu_train_utils.py: Utility functions for training models.([github.com][1])

• Model Definitions:

  • nu_models.py: Definitions of CNN, LSTM, CNN-LSTM, EEGNet, and ShallowConvNet architectures.([github.com][1])

• Training Scripts:

  • train_CNN_pooled.py / train_CNN_subject_specific.py: Training CNN models on pooled or subject-specific data.
  • train_LSTM_pooled.py / train_LSTM_subject_specific.py: Training LSTM models on pooled or subject-specific data.
  • train_CNNLSTM_pooled.py / train_CNNLSTM_subject_specific.py: Training CNN-LSTM hybrid models.
  • train_EEGNET_subject_specific.py: Training EEGNet models on subject-specific data.
  • train_ShallowConvNet_subject_specific.py: Training ShallowConvNet models on subject-specific data.

To execute a training script, navigate to the repository directory and run:

python train_CNN_pooled.py

Replace train_CNN_pooled.py with the desired training script.

Datasets

The study utilizes four publicly available EEG datasets for evaluating model performance. Please refer to the respective dataset sources for access and usage guidelines.

Contribution

Contributions are welcome! If you have suggestions, bug reports, or enhancements, please open an issue or submit a pull request.

License

This project is open-source and available under the MIT License.

Acknowledgments

This repository was developed by Berdakh Abibullaev, focusing on systematic deep learning model selection for P300-based brain-computer interfaces.

---

For detailed explanations and methodologies, refer to the research paper associated with this repository.

If you need further assistance or have specific questions about any script or functionality, feel free to ask!