Harmonisation of classification algorithms

Apparently, the same algorithms are implemented in both bob.pad.face and bob.pad.voice. For instance, we have VideoLRPadAlgorithm in face and LogRegrAlgorithm in voice.

I totally understand that they are a bit different (feature normalization in one case, PCA in the other for instance), but at the core, both are supposed to do Logistic Regression on some features useful for PAD. It would then be nice to have a base class in bob.pad.base. (and then you're free to do whatever you want in a derived class if you need to).

The same goes for GMM ...

@pkorshunov @amohammadi @onikisins @ageorge @andre.anjos @sbhatta : what do you think ? Who's willing / available to take care of this ? (I can, starting on March 5th)

Thanks