Debate: `apply_curation` should warn when `sparsity_overlap` fails

[chrishalcrow]

Hello, a debate!

We use apply_curation in our spike sorting pipeline after using automated curation tools to find possible unit merges. At the moment, if there isn't enough sparsity overlap between the units, apply_curation throws an error. This is annoying.

I'd prefer if the non-overlappying units simply weren't merged, and an warning was shown. Then I'd still get a curated analyzer out of apply_curation with most of the merges completed.

Thoughts?

[samuelgarcia]

I think I like the raise but we can have a flag to control this.
Merging non overlapping is very suspicious.

[chrishalcrow]

No no, I think we should not merge non-overlapping. But it shouldn't raise an error, it should just not do it. Image you have 50 merges and one fails the overlap threshold. Currently, it just errors and that's it. Instead, I think it should merge the 49 good ones, and skip the bad one.

[alejoe91]

I'm ok with this. We could have a failure_mode flag raise/skip to control the behavior. Fine to have skip by default.