I have taken 3456 proteins (used for 'Deepcon covariance') and tried the implementation of ResPRE on my CNN model of architecture depth = 32.
I got an accuracy of 74.7 for covariance calculations, also similar 74.6 for precision calculations. I am not sure why I am not able to get better prediction(as per the claim of the ResPRE paper).
I used the default option for weight calculation while using the 'calNF_ly'.
a. Is there a specific normalization value that gives high prediction?
b. ResPRE specifically mentions that precision values were higher than covariance values when tested on DNCON2 protein dataset. Can you throw more insight in to this?