In this project, I aim at applying transfer learning on EEG in order to show significant differences between Parkinson's Disease patients and normal control and build classification model. Dr. X WH (though not obtained the degree yet) is working together with me on this topic. He is focusing on data augmentation based on GAN. The dataset we are using is from Pred+ct. We decided to select channels around Cz (Pz, Cz, Fz, P3, C3, F3, P4, C4, F4) as we believe this region is greatly related to cognitive functions.
Why hdf5 files?
Because when I tried to save them as .mat files, the size of the file was strangely huge. I didn't spend time finding the reason, and hdf5 is a second choice as a bridge between Matlab and Python.
The disappointing thing is that there is always some trivia in the final year thus I cannot fully focus on the project. This week I tried to build models and the dataloader. Dataloader is always something bloated. In the past, I built it and ended up with too many parameters. This time I tried to make it as light as possible but it seems that I failed to make differences.