Discarded idea for the Learning from Networks course of University of Padova.
PCA.py contains the PCAConv layer and a network to test it. The code was updated up to the moment we decided to switch to SRGNN, so it may not work properly, since it is not finished.
REG.py contains our proposed regularization method which penalizes similarity among embeddings of the same layer. It was then adapted to the SRGNN model.