ENH: warn Series.interpolate(method='index') with unsorted index

[immaxchen]

Series.interpolate(method='index') use numpy.interp under the hood, which, from numpy doc, Does not check that the x-coordinate sequence xp is increasing. If xp is not increasing, the results are nonsense. (link)
Here adds a warning to warn the user about this.

• closes raise warning when calling .interpolate(method='index') when Index is not sorted with ascending=True #21037
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

[WillAyd]

Can this be fixed instead of producing a warning?

[immaxchen]

Yes it can, in the cost of introducing dependency on scipy. (like other interpolate methods)

[WillAyd]

We can't use pd.where(..., df.index) to fill from the index?

[immaxchen]

You mean we sort it, interpolate, then fill it back? this may leads to undefined behavior for duplicated indexes, better leaves the responsibility to scipy IMHO. 😂

[jreback]

You mean we sort it, interpolate, then fill it back? this may leads to undefined behavior for duplicated indexes, better leaves the responsibility to scipy IMHO. 😂

then would just raise rather than a warning.

[immaxchen]

So the preferred approach would be stick to numpy.interp but we do the sort for it right? No problem. I will close this one and do it in a new branch. Thanks for your suggestions.

[jreback]

i think either we sort then reorder to get the original
or raise