BUG/inconsistency: IntervalIndex.get_loc gives a location for non-exact inputs

[topper-123]

This one is a bit complex to explain, but I'll do my best.

Currently IntervalIndex.get_indexer fails if the other index doesn't contain Interval only (there's also another bug, but let's keep it simple here).

The underlying issue is that IntervalIndex.get_indexer depends on IntervalIndex.get_loc which is ambigous for how it treats number inputs:

>> ii = pd.IntervalIndex.from_breaks([0,1,2,3])
>> ii.get_loc(pd.Interval(1, 2))
1  # ok
>> ii.get_loc(1)  # do we mean exactly 1, or if an interval contains the number 1?
1  # ambigous

The issue is that get_loc returns the location for both exact matches and inexact matches (i.e. if the number input is in an interval). For the purposes of get_indexer however, this behavious fails, as get_indexer needs get_loc to find exact matches only.

See #19021 (comment) for further discussion.

Solution

A solution could be adding a 'strict' option to the method parameter of IntervalIndex.get_loc.

This wasn't so difficult after all, and I've already made a PR on this, see #19353

[jorisvandenbossche]

cc @jschendel is this covered by the PR?

[jorisvandenbossche]

Ah, no, it is about scalars passed to get_loc, I don't think anything changed for that?