Towards stabilisation of the core API

[ablaom]

The purpose of this "FYI" issue is to articulate a number of personal priorities for stabilizing the core API in MLJBase - a preliminary triage for MLJBase 1.0, if you will.

I am assigning myself to these tasks. I fear if they are not addressed, then they willl cause much pain should my involvement in the project wane.

I will not be responding to any comments posted here. Please provide any feedback by joining the discussion in one of the separate threads referenced below, or open a new topic-specific issue, thanks.

• Enable model implementations to support sample weights. Actually, the interface part is done, but machines and their methods are now aware of this yet. (Add sample-weight interface point? #177)

• Make output_scitype an "inspectable" trait of supervised learners, to support those learners that transform as well as predict (Can a supervised model also implement the transform method. MLJModels.jl#117)

• Enable model implementations to support on-line learning (Integrating online and active learning models #60)

• Reorganize the MLJ stack to enhance scalability (Re-organizing the MLJ stack #317)

• Tweak the machine interface to facilitate serialisation/deserialization of hyperparameters and learned parameters Saving and loading models #138

[ablaom]

Closing in favour of #, the remaining item on my list. (Probably need a new list 😄 )