DEPR: Deprecate NDFrame.filter

[topper-123]

• xref Rename NDFrame.filter to .select? #26642
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

I propose to deprecate (DataFrame|Series).filter. The method's functionality is already in the .locmethod, except the regex functionality, for which I've made a proposal adding regex functionality to .loc in #27363.

[jreback]

ok, we need to decide on #27363 first

[jbrockmendel]

@topper-123 can you rebase

[jreback]

can you rebase again.

[topper-123]

I've rebased. As you noted, this depends on #27363.

[jbrockmendel]

@topper-123 im trying to clear the PR backlog; can this be closed pending #27363?

[WillAyd]

Knowing the outstanding issue is around regex capability, maybe we should just deprecate all keywords from filter for now that aren't regex?

[topper-123]

That sounds like a stepping stone to deprecating. Wouldn't it make more sense then to just deprecate this method before 1.0 is released?

[WillAyd]

I don't think we'd want to outright deprecate without a replacement for the regex unless we instruct users to do df[(re.search(x) for x in df.columns)]

[topper-123]

Oh yes, I meant choose another way to select by regex at the same time, sorry.

[jreback]

can you merge master and will look again

[topper-123]

I've rebased.

I do think an alternative should be settled on for the regex use case and I've had two proposals that were not accepted. So I'm not sure where this is floating currently?

[WillAyd]

Thanks for the PR! But I think agreement on approach here is still outstanding. If something you want to drive forward can revisit discussion in issue first