BUG: Error raised accessing tuple value in Series when index contains duplicates

[rhshadrach]

E.g.

s = pd.Series([1, 1], index=[(1, 1), (1, 1)])
s[(1, 1)]

raises KeyError: (1, 1). Within pandas._libs.index.IndexEngine._get_loc_duplicates, we're using ndarray.searchsorted which interprets the tuple as an array-like of values to search for rather than a single tuple.

[phofl]

Similar problems with loc. Index does not have to be non-unique there

s = pd.Series([1, 1], index=[(1, 2), (1, 1)])

s.loc[(1, 2)]

pandas.core.indexing.IndexingError: Too many indexers

Tuple support is not very good here.

[GYHHAHA]

This is conflicting with the multiindex selector which uses tuple as the label for different levels.

>>>s = pd.Series([1, 1], index=pd.Index([(1, 2), (1, 1)]))
>>>s.loc[(1,2)]
1

Ambiguous when using tuple index.

[rhshadrach]

@GYHHAHA what is ambiguous? When Series has a multi-index, a tuple consists of labels for different levels. When Series doesn't have a multi-index, a tuple is a single label. Or is there some other case(s)?

[GYHHAHA]

>>>s = pd.Series([1,2], index=pd.Index([((1,1),1),((1,2),2)]))
>>>s.loc[(1,1)]
1    1
dtype: int64

When tuple index is the zero level of multiindex.

But it seems still reasonable when applying to a non-multiindex.

[rhshadrach]

Ack, I see. A similar example:

s = pd.Series([1, 2], index=pd.Index([((1, 1), 1), (1, 1)]))
print(s.loc[(1, 1)])

one might expect to get 2 rather than the actual output of 1. In any case, oddities in the case of nested tuples should not imply that we shouldn't make improvements on supporting tuples.