Series.fillna doesn't work with series input

[miketkelly]

Test code:

import pandas as pd

print pd.__version__

s1 = pd.Series([np.nan])
s2 = pd.Series([1])

s1.fillna(s2)

Results:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-10-28706fea97bb> in <module>()
      6 s2 = pd.Series([1])
      7 
----> 8 s1.fillna(s2)

/usr/local/anaconda/lib/python2.7/site-packages/pandas-0.13.0rc1_64_gceec8bf-py2.7-linux-x86_64.egg/pandas/core/generic.pyc in fillna(self, value, method, axis, inplace, limit, downcast)
   1917                         continue
   1918                     obj = result[k]
-> 1919                     obj.fillna(v, inplace=True)
   1920                 return result
   1921             else:

AttributeError: 'numpy.float64' object has no attribute 'fillna'

0.13.0rc1-64-gceec8bf

[jreback]

yep..should prob work

[jreback]

@mtkni see PR #5705

pls lmk what you think about the 'other' cases (which were not tested in 0.12, but are possible I guess)

doesn't seem to me that passing more than 1 fillna value makes sense for a Series/dict

[miketkelly]

My use case is more like the following, which requires multiple values. It worked in 0.12, so I got used to it.

s1 = pd.Series([1, 2, 3, 4, 5, 6])

# may produce missing values
s2 = do_some_calc(s1)

# fill missing values with an alternate calculation
s2 = s2.fillna(s1 - s1.mean())

[jreback]

ok so u want to match on the index and fill if it's na
ok easy enough then

that's effectively combine_first so will do exactly that

[miketkelly]

Now that I look at it, what I was doing only worked because my indexes where the same.

>>> s1 = pd.Series({'a': 1, 'b': 2})
>>> s1 = pd.Series([0, 1, 2], list('abc'))
>>> s1
a    0
b    1
c    2
dtype: int64
>>> s2 = pd.Series([0, np.nan, 2], list('bac')) 
>>> s2
b     0
a   NaN
c     2
dtype: float64
>>> s2.fillna(s1)
b    0
a    1
c    2
dtype: float64

A problem with combine_first approach is that it might expand the index.

[jreback]

It seems your example above from 0.12 is wrong. This is makes sense?

In [1]: s1 = Series([0, 1, 2], list('abc'))

In [2]: s1
Out[2]: 
a    0
b    1
c    2
dtype: int64

In [3]: s2 = Series([0, np.nan, 2], list('bac'))

In [4]: s2
Out[4]: 
b     0
a   NaN
c     2
dtype: float64

In [5]: s2.fillna(s1)
Out[5]: 
b    0
a    0
c    2
dtype: float64

[miketkelly]

Yes, that is the desired outcome.

[jreback]

ok gr8....