This is an updated version of the proposed TF-C model. 

1. Fixed buggs

2. Replaced the CNN blocks with 2-layer transformer as the encoder architecture

3. Cleaned the codes and added comments for easy understanding

4. Uploaded the weights of an exampled pretrained model. The model is pretrained on SleepEEG with only 1280 samples (a small subset that only takes ~1% samples of the whole dataset). 

5. For the downstream classifier, added a KNN classifier as an alternation of the MLP classifier. 

6. Will share some ideas of how to further improve the TFC model. 
