BUG: unstacking with object columns defaults to None as a fill value

[jreback]

In [41]: df = DataFrame({'A' : [1,1,2], 'B' : list('aba'), 'C' : ['foo', np.nan, 'bar']}).set_index(['A','B'])

In [42]: df.unstack('B')
Out[42]: 
     C      
B    a     b
A           
1  foo   NaN
2  bar  None

In [43]: df.unstack('B', fill_value=np.nan)
Out[43]: 
     C     
B    a    b
A          
1  foo  NaN
2  bar  NaN

Introduced is this PR: #10246

you can make a case to do this as we DO support None as a fill value, but almost always even with object columns we use np.nan, unless very explicity set by the user.

[jreback]

cc @amcpherson

[amcpherson]

So the preferred functionality is?:

In [42]: df.unstack('B')
Out[42]: 
     C      
B    a     b
A           
1  foo  NaN
2  bar  NaN

default to None for object columns seems reasonable to me, but I can fix if np.nan is preferred.

[jreback]

yeah we just use nan even for object columns
unless the user is very explicit about it

[mattayes]

Hey folks, I'm happy to pick this up if no one else is working on it. I'm fairly confident with what needs to change (thanks for linking to the original PR, @jreback).