The method currently in the repository (90% complete) is the one in https://www.nature.com/articles/s41598-019-51269-8 . This method suffers from the fact that they use 20-second epochs instead of 4-second epochs, which may reduce the classification slightly (since in all likelihood the probability of transition withining from wake to sleep or sleep to wake in a 20 second interval is rare, albeit rarer in a 4 second interval). Therefore, I decided to try to implement https://arxiv.org/pdf/1809.08443.pdf which actually uses 2-EEG channels (unlike the nature article which only used 1 and much more training data) and also 4-second epochs so theoretically it should be much more straightforward to implement.