Series.drop regression and unexpected behaviour

[ruidc]

On 0.12 stable this works:

>>> s = pandas.Series([1,1,2],index=["one","one","two"])
>>> s.drop(["one"])
two    2
dtype: int64

but a singleton silently fails:

>>> s.drop("one")
one    1
one    1
two    2
dtype: int64

except when index is unique:

>>> s = pandas.Series([1,2],index=["one","two"])
>>> s.drop("one")
two    2
dtype: int64

according to the docs, the input to drop should be array like, so i'm not sure this should be permissible.

On head, with unique index, it matches, but with non-unique:

>>> pandas.__version__
'0.12.0-892-g6830600'
>>> s = pandas.Series([1,1,2],index=["one","one","two"])
>>> s2 = s.drop(["one"])
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "pandas/core/generic.py", line 1133, in drop
    return self.ix[tuple(slicer)]
  File "pandas/core/indexing.py", line 53, in __getitem__
    return self._getitem_tuple(key)
  File "pandas/core/indexing.py", line 558, in _getitem_tuple
    return self._multi_take(tup)
  File "pandas/core/indexing.py", line 598, in _multi_take
    raise self._exception
KeyError

[jreback]

thanks for the report! fixed up

[ruidc]

Wow, that was quick! Thanks
But if singletons are meant to be supported, shouldn't the docs say so? and is it not a bit ambiguous?

e.g.:

>>> s = pandas.Series([1,1,2,3],index=["one","one","two", "e"])
>>> s.drop("egg")

[jtratner]

That's how duck typing works in Python - I guess we could add a scalar check, but I think that will get complicated with Index objects that gave tuples as labels.

[jreback]

I think this IS consistent with the doc-string (array-like) and actual
maybe i'll add a test though

In [1]: s = Series(range(3),index=list('abc'))

In [2]: s
Out[2]: 
a    0
b    1
c    2
dtype: int64

In [3]: s.drop('a')
Out[3]: 
b    1
c    2
dtype: int64

In [4]: s.drop('ab')
ValueError: labels ['ab'] not contained in axis

[ruidc]

hmm, I'd still think that if this is intentional (which it now clearly seems to be) then the docs for Series.drop() are misleading in stating that the labels argument is "array-like". To me that implies that the elements of the array will be dropped, rather than scalars which could be iterables, are treated as array elements - in short, shouldn't his behaviour for strings be mentioned?

[jreback]

I think the convieence of using a single value that you just want to drop outweight the possibility of really meaning a list-like.

Maybe doc-string could say list-like (a string is not list-like, but suppose it could be construed as array-like).

i'll do a doc-update...