Labels mismatch when adding series with repeated index values in 0.19rc1

[mrocklin]

This showed up in the dask.dataframe test suite when testing against the 0.19 release candidate. I'm unsure if this was an intended change or not:

Pandas 0.18

In [1]: import pandas as pd

In [2]: a = pd.Series([1, 2], index=[1, 1])

In [3]: b = pd.Series([10, 10], index=[1, 2])

In [4]: a
Out[4]: 
1    1
1    2
dtype: int64

In [5]: b
Out[5]: 
1    10
2    10
dtype: int64

In [6]: a + b
Out[6]: 
1    11.0
1    12.0
2     NaN
dtype: float64

Pandas 0.19rc1

In [1]: import pandas as pd

In [2]: a = pd.Series([1, 2], index=[1, 1])

In [3]: b = pd.Series([10, 10], index=[1, 2])

In [4]: a
Out[4]: 
1    1
1    2
dtype: int64

In [5]: b
Out[5]: 
1    10
2    10
dtype: int64

In [6]: a + b
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-6-f96fb8f649b6> in <module>()
----> 1 a + b

/home/mrocklin/Software/anaconda/lib/python3.5/site-packages/pandas/core/ops.py in wrapper(left, right, name, na_op)
    671             return NotImplemented
    672 
--> 673         left, right = _align_method_SERIES(left, right)
    674 
    675         converted = _Op.get_op(left, right, name, na_op)

/home/mrocklin/Software/anaconda/lib/python3.5/site-packages/pandas/core/ops.py in _align_method_SERIES(left, right, align_asobject)
    613                                                 return_indexers=True)
    614             # if DatetimeIndex have different tz, convert to UTC
--> 615             left.index = index
    616             right.index = index
    617 

/home/mrocklin/Software/anaconda/lib/python3.5/site-packages/pandas/core/generic.py in __setattr__(self, name, value)
   2754         try:
   2755             object.__getattribute__(self, name)
-> 2756             return object.__setattr__(self, name, value)
   2757         except AttributeError:
   2758             pass

pandas/src/properties.pyx in pandas.lib.AxisProperty.__set__ (pandas/lib.c:44336)()

/home/mrocklin/Software/anaconda/lib/python3.5/site-packages/pandas/core/series.py in _set_axis(self, axis, labels, fastpath)
    306         object.__setattr__(self, '_index', labels)
    307         if not fastpath:
--> 308             self._data.set_axis(axis, labels)
    309 
    310     def _set_subtyp(self, is_all_dates):

/home/mrocklin/Software/anaconda/lib/python3.5/site-packages/pandas/core/internals.py in set_axis(self, axis, new_labels)
   2774             raise ValueError('Length mismatch: Expected axis has %d elements, '
   2775                              'new values have %d elements' %
-> 2776                              (old_len, new_len))
   2777 
   2778         self.axes[axis] = new_labels

ValueError: Length mismatch: Expected axis has 3 elements, new values have 5 elements

[jreback]

http://pandas-docs.github.io/pandas-docs-travis/whatsnew.html#series-operators-for-different-indexes

this may now not be well defined for duplicates

[mrocklin]

Alrighty. Thanks for the pointer.

[jorisvandenbossche]

I don't think this change was intentional. The whatsnew also says "Arithmetic operators align both index (no changes)."

[TomAugspurger]

@jorisvandenbossche agreed that it wasn't intentional, but does the old behavior make sense under the new equality rules (or the old ones for that matter)?

[jorisvandenbossche]

If we go from the "align both index" description (of course not sure if this is also how it is in the docs or is meant initially), I would expect this yes:

In [19]: a_temp, b_temp = a.align(b)

In [20]: a_temp
Out[20]: 
1    1.0
1    2.0
2    NaN
dtype: float64

In [21]: b_temp
Out[21]: 
1    10
1    10
2    10
dtype: int64

In [22]: a_temp + b_temp
Out[22]: 
1    11.0
1    12.0
2     NaN

[TomAugspurger]

Whoops, yeah obviously that's the expected output.

[jorisvandenbossche]

I wouldn't call it that "obvious" :-), but it does follow from the rules in some way