strandquistg / eeg-ssl

Self-supervised learning for EEG

Open Sourced Self-Supervised Representation Learning From Electroencephalography Signals

This implementation is based on the paper by Banville et al.

Directories

• experiments : Improvement experiments folder
• images : Folder for images for README and results
• model : Folder for model scripts
• preprocessing : Folder for preprocessing scripts
• ssl : Folder for SSL scripts

Installation and tutorial

Installation

$ python eeg_ssl.py data_folder T_pos_RP T_neg_RP T_pos_TS T_neg_TS

Inputs

• data_folder: a folder containing EEG .edf files
• T_pos_RP: an integer representing the positive limit for relative positioning.
• T_neg_RP: an integer representing the negative limit for relative positioning.
• T_pos_TS: an integer representing the positive limit for temporal shuffling.
• T_neg_TS: an integer representing the negative limit for temporal shuffling.

Outputs

• RP_dataset: pairs of 30 second normalized EEG time windows
• RP_labels:
  • +1 if the distance between the two windows is T_pos_RP
  • -1 if the distance between the two windows is T_neg_RP
• TS_dataset: triples of 30 second normalized EEG time windows
• TS_labels:
  • +1 if the distance between the two windows is T_pos_TS
  • -1 if the distance between the two windows is T_neg_TS

License

MIT