Are there plans to implement drop-in replacements for parts of sklearn.feature_selection?

[nirizr]

Some (like VarianceThreshold) may be pretty easy to do, others can be a little more complex and may require more thought.

I'm wondering if there's a plan to cover that as part of dask-ml?

[TomAugspurger]

Yes, some of those would definitely be in scope. Looking through the list, the Recursive ones probably can't be parallelized.

The various Select* ones should be doable, assuming the score functions can be implemented.

Do you have any interest in working on these? It can be done in bits and pieces. I suspect that you're correct that VarianceThreashold is the most straightforward.

[nirizr]

I expect to be needing some soon, at least VarianceThreshold to get us started. I was going to implement it directly in our source code but if its in-scope I would definitely prefer putting it here.

We're also using a slightly naive DT implementation on top of Dask, let me know if you will be interested that as well and I'll open a PR and we can continue discussing the details there.

[TomAugspurger]

Great!

What's DT stand for?

[nirizr]

sorry, I meant implementing a random forest based on parallelizing decision trees using dask.