Indirect references in ML.NET using Dependency Injection

[eerhardt]

I've found 2 indirect references in our product that I'm not sure how to proceed on.

1. LearnerFeatureSelectionTransform depends on SDCA, but ML.Transforms doesn't have a reference to ML.StandardLearners.
2. The 'train' commands in ML.Data (CV, Train, TrainTest) all have a default Trainer of AveragedPerceptron. but AveragedPerceptron lives in the ML.StandardLearners assembly. And ML.StandardLearners references ML.Data, but this dependency is in the wrong order.

A couple of options:

1. Change these parameters to be 'user required', i.e. they don't have a default and fail if you didn't specify it.
2. Move some code around so the dependencies line up correctly. For example, we could move our 'train' commands out of ML.Data and into a higher assembly.
3. For (1) above, we could move LearnerFeatureSelectionTransform into ML.StandardLearners instead of being in ML.Transforms. Or we could add the dependency from ML.Transforms to ML.StandardLearners.
4. Continue to use Dependency Injection and just assume/hope the component is there at runtime.

From an internal conversation with @TomFinley, we are thinking of taking option (1) above - make these parameters required and not have a default.

Thoughts?

cc @TomFinley @Zruty0 @codemzs @Ivanidzo4ka

[codemzs]

Closing this issue as I believe this has been addressed with the stable API.