Functions applied on .expanding() receive ndarrays rather than pandas objects

[max-sixty]

.apply(...) on window-based groupbys receive an ndarray rather than a Pandas object.
Is this intentional? Is it a performance issue?

It means that functions that work on Series can't necessarily be used on groupbys, which limits abstraction.

In [7]: pd.Series(range(10)).expanding().apply(lambda x: x.pow(2).sum())
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-7-970afaf538ff> in <module>()
----> 1 pd.Series(range(10)).expanding().apply(lambda x: x.pow(2).sum())

/Library/Frameworks/Python.framework/Versions/3.5/lib/python3.5/site-packages/pandas/core/window.py in apply(self, func, args, kwargs)
    929     @Appender(_shared_docs['apply'])
    930     def apply(self, func, args=(), kwargs={}):
--> 931         return super(Expanding, self).apply(func, args=args, kwargs=kwargs)
    932 
    933     @Substitution(name='expanding')

/Library/Frameworks/Python.framework/Versions/3.5/lib/python3.5/site-packages/pandas/core/window.py in apply(self, func, args, kwargs)
    547                                       kwargs)
    548 
--> 549         return self._apply(f, center=False)
    550 
    551     def sum(self, **kwargs):

/Library/Frameworks/Python.framework/Versions/3.5/lib/python3.5/site-packages/pandas/core/window.py in _apply(self, func, window, center, check_minp, how, **kwargs)
    487                 result = np.apply_along_axis(calc, self.axis, values)
    488             else:
--> 489                 result = calc(values)
    490 
    491             if center:

/Library/Frameworks/Python.framework/Versions/3.5/lib/python3.5/site-packages/pandas/core/window.py in calc(x)
    482 
    483                 def calc(x):
--> 484                     return func(x, window, min_periods=self.min_periods)
    485 
    486             if values.ndim > 1:

/Library/Frameworks/Python.framework/Versions/3.5/lib/python3.5/site-packages/pandas/core/window.py in f(arg, window, min_periods)
    545             minp = _use_window(min_periods, window)
    546             return algos.roll_generic(arg, window, minp, offset, func, args,
--> 547                                       kwargs)
    548 
    549         return self._apply(f, center=False)

pandas/algos.pyx in pandas.algos.roll_generic (pandas/algos.c:40613)()

<ipython-input-7-970afaf538ff> in <lambda>(x)
----> 1 pd.Series(range(10)).expanding().apply(lambda x: x.pow(2).sum())

AttributeError: 'numpy.ndarray' object has no attribute 'pow'

Expected (without the coercion):

In [9]: pd.Series(range(10)).expanding().apply(lambda x: pd.Series(x).pow(2).sum())
Out[9]: 
0      0.0
1      1.0
2      5.0
3     14.0
4     30.0
5     55.0
6     91.0
7    140.0
8    204.0
9    285.0
dtype: float64

Pandas 0.18

[jreback]

dupe of #5071 (and noted in the master issue). This is not hard to fix now that we are in the new structure.

[max-sixty]

Ah, mea culpa. I searched for numpy rather than ndarray