Tidy up the model subpackages

🚀 Refactoring

The structure in pl_bolts.models is not user-friendly due to:

• an inconsistent structure, for example:

  from pl_bolts.models.vision import SemSegment  # This works fine.
  from pl_bolts.models.vision import ImageGPT    # This doesn't work,
  from pl_bolts.models import ImageGPT           # but this works.
  

• several ways to import a certain class, for example: <- (related: #423)

  from pl_bolts.models import LogisticRegression
  from pl_bolts.models.regression import LogisticRegression
  from pl_bolts.models.regression.logistic_regression import LogisticRegression
  

Pitch

Two options I have in mind to make it consistent are:

1. Make all models available right under pl_bolts.models

   pl_bolts.models.VAE
   pl_bolts.models.AE
   pl_bolts.models.ImageGPT
   pl_bolts.models.PixelCNN
   pl_bolts.models.SwAV
   pl_bolts.models.SimCLR
   ...
   

2. Split models into domains, e.g. vision, self_supervised… (which I prefer)

   pl_bolts.models.autoencoders.VAE
   pl_bolts.models.autoencoders.AE
   pl_bolts.models.vision.ImageGPT
   pl_bolts.models.vision.PixelCNN
   pl_bolts.models.self_supervised.SwAV
   pl_bolts.models.self_supervised.SimCLR
   ...
   

@Borda @ananyahjha93 @annikabrundyn @oke-aditya I’d love to hear your opinions on if it should be as is or another interface :]

[oke-aditya]

Haha so true, it is not regression a task. But maybe put it under pl_bolts.models.ml ? Where we could have simple ml models.

[Borda]

I am for some refactoring in this direction, just I would recommend stabilizing the state e have right now as 0.3 and do this refactor in 0.4 which can be even next week :]