Anthony Blaom, PhD

Of course, this comment does not address the other "fly in the ointment" which is tables, requiring trait dispatch.

Well, no, I'm not suggesting we change the *definition* of scitype for arrays (see [Property 3](https://github.com/JuliaAI/ScientificTypesBase.jl#what-is-provided-here)) - only the implementation. According to the definition, we will need to look inside...

Sorry, I guess this one fell under the radar. Just skimmed your comment but here's a quick reply, which hopefully addressed your point: In general, because one is resampling to...

As explained above, the best we can expect here is for user-specified holdout train/test pairs to work in addition to `Holdout` resampling strategy. [This PR](https://github.com/alan-turing-institute/MLJBase.jl/pull/559) resolves this (also in the...

I'm inclined to go with option 2, which is more user-friendly. The other issue ought to be solved on the tables interface side, in my opinion.

`all-zero` looks like the simplest. One question for `category` is how to handle `missing` values that appear for a feature that did not have`missing` values in training (`fit`). Here's a...

No, rather it's the same as the current behaviour, except instead of `missing`s, use zeros. You don't need to spawn an extra column in this case: ``` julia> X =...

Yes, great catch, that's a bug: https://github.com/JuliaAI/MLJModels.jl/issues/467 Are you willing an able to make a PR with a test?

Done. You have an invitation to accept.

@tlienart You are right. DBSCAN is not like KMeans clustering. I stand corrected. However, I do wonder if the sk-learn way of conceptualising this class of clustering problems is the...