Docs reorganization about dask-ml

[TomAugspurger]

The separation between single machine and distributed learning seems odd to me.

Agreed. I'm not sure why I found that distinction so relevant a few months ago.

Your proposed TOC sounds reasonable. For now, I think Pipelines would make a good home for things in preprocessing/.

I've also found it pleasant recently to start sections with the API relevant for that section.

That could get a bit boring for scikit-learn style estimators, since it'll (almost) always be.fit, .transform, etc :) Perhaps I'm misunderstanding though.

[TomAugspurger]

And if you're busy, I'll have time to work on this later today or tomorrow, as I wait around for pandas release things to finish.

[mrocklin]

That could get a bit boring for scikit-learn style estimators, since it'll (almost) always be.fit, .transform, etc :) Perhaps I'm misunderstanding though.

Yeah, to be clear I'm saying that we just add the .. autosummary: line at the top of each relevant section. When you go to the Generalized Linear Models section you see LinearRegression, LogisticRegression, PoissonRegression, _GLM at the top of the section. If you care to you can click on those to be taken to the full API docs on another page.