EEGDiffWave

Diffusion for EEG

Version 2.0: training for deeper network

1. to run new experiemnts: use new_main.py and eegWave-new.ipynb
   
2. To run the code, change the following paths in the configure.json file:
   a) data_path field of the trainset_config should be your bci-comp-iv2a directory, which can be downloaded through this link: http://bnci-horizon-2020.eu/database/data-sets
   
3. Change the configure.json file in the new_main.py or the eegWave-new.ipynb depends on your choice of the network to train:
   a) configure-same.json: use the same number of channels as the eeg data in the residual blocks: 0.8M parameters
   b) configure-deep.json: use 128 channels in the residual blocks: 5.9M parameters
   c) change the name (line 186 in the new_main.py)
   

Version 1.0: training for unconditional EEG generation

1. main.py shares the exact same content as the jupyter notebook file (eegWave.ipynb)
   
2. To run the code, change the following paths:
   a) line 122 in main.py: data_path should be your bci-comp-iv2a directory, which can be downloaded through this link: http://bnci-horizon-2020.eu/database/data-sets
   b) check points and logs are saved in the directory defined in the configure.json , if you want to continue training the same network, do not change anything in that json file.
   
3. To run on multiple GPUs, read the description of the train function in main.py. You should provide several parameters.