Original authors have uploaded their code here https://github.com/vlawhern/arl-eegmodels

EEGNet

PyTorch implementation of EEGNet: A Compact Convolutional Network for EEG-based Brain-Computer Interfaces

Requirements

• Python 2
• Dataset of your own choice, works well with BCI Competition 3 Dataset 2.
• Pytorch 0.2+
• Jupyter notebook

Usage

• GPU - Just shift+enter everything.
• No GPU - Remove all .cuda(0) before running.

Notes

• I found ELU to work inferior, would not recommend. Linear units work better than ReLU as well.
• I found that ELU/Linear/ReLU are similar in performance.

Results

• BCI Competition 3 Dataset 2 - Fmeasure (0.402)

Credits

• Original paper - https://arxiv.org/abs/1611.08024
• PyTorch documentation.

Hope this helped you. Raise an issue if you spot errors or contact sriram@ucsd.edu.