Series.drop() with boolean index: inconsistent behaviour

[urraca]

The behaviour below occurs in versions '0.15.0rc1-21-g32c5016' and '0.14.1'.

When the label passed to the drop method of a Series is not in the index:
(a) if the index is not all booleans then an error is raised
(b) if the index is all booleans and has length > 1 then no error is raised; the original series is returned
(c) if the index is all booleans and has length == 1 then an error is raised

I propose that:

1. the difference between the behaviour in (a) and (b) should be documented
2. the behaviour in (c) should be changed to match that in (b)

Examples of current behaviour:
(a)

>>> pd.Series([1, 1], index=['a', True]).drop(False)
ValueError: labels [False] not contained in axis

(b)

>>> pd.Series([1, 1], index=[True, True]).drop(False)
True    1
True    1
dtype: int64

(c)

>>> pd.Series([1], index=[True]).drop(False)
ValueError: labels [False] not contained in axis

[jreback]

feel free to submit a PR for 1) docs

why would c) be incorrect? A value NOT in the axis by definition raises

as an aside, using a fully boolean index by definition almost always has duplicates which makes it a bit tricky to work with it.

[jreback]

#6599 is an attempt to consolidate the drop behavior, feel free to pick this up / comment as appropriate as well.

[urraca]

Thanks for your comments.

My key point is that (b) and (c) are inconsistent in a way users would not expect.

Personally, I find it convenient that errors are not raised in (b) but I suppose that it is most important that the behaviour be consistent.

To change the behaviour so that (b) raises an error would have the advantage of making the behaviour consistent across all index types and lengths, perhaps making further documentation unnecessary.

[jreback]

sorry, you are right. I think b) should raise, that looks like a bug (as I said before, just be careful with bool indexes).

[jreback]

@urraca yes this stems from the fact that b) has a duplicate index, while c) does not. Have to think about this.

[wesm]

This behavior is not present in 0.23.2