Inconsistent behaviour of rename for MultiIndexed DataFrames and Series

[alefnula]

If I try to rename a MultiIndexed DataFrame, everything works correctly:

In [3]: pd.DataFrame([11,21,31],
            index=pd.MultiIndex.from_tuples([("A",x) for x in ["a","B","c"]])).rename(str.lower)
Out[3]: 
      0
a a  11
  b  21
  c  31

But if I do the same thing with the Series, it doesn't work:

In [4]: pd.Series([11,21,31],
            index=pd.MultiIndex.from_tuples([("A",x) for x in ["a","B","c"]])).rename(str.lower)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-4-aa25b537c1df> in <module>()
----> 1 pd.Series([11,21,31],index=pd.MultiIndex.from_tuples([("A",x) for x in ["a","B","c"]])).rename(str.lower)

/home/viktor/pandas/pandas/core/series.py in rename(self, mapper, inplace)
   3187         result = self if inplace else self.copy()
   3188         result.index = Index([mapper_f(x)
-> 3189                              for x in self.index], name=self.index.name)
   3190 
   3191         if not inplace:

TypeError: descriptor 'lower' requires a 'str' object but received a 'tuple'

This behavior is observed in 0.12 and in the master branch.

[cpcloud]

And here's the SO question.

looks like it should be called on every element of the tuple similar to frames. not sure why it's not

[jreback]

yep..think this has always been there, adding to the post series2 cleaning; another inconsistency between series/frame....should be fixable (though these are not always trivial as series jumped thru hoops sometimes to do some of these types of things)

[cpcloud]

maybe we should call it "aftermath"? 😄

[cancan101]

Another small observation: DataFrame take an index kwarg whereas Series does not.

[cpcloud]

That's not true. Do Series?? in IPython. It's possible that there's an inconsistency in the documentation somewhere though.

[jreback]

@cancan101 example?

[cancan101]

This seems to indicate that Series.rename does not take index: http://pandas.pydata.org/pandas-docs/stable/generated/pandas.Series.rename.html#pandas.Series.rename

whereas DataFrame.rename does: http://pandas.pydata.org/pandas-docs/stable/generated/pandas.DataFrame.rename.html#pandas.DataFrame.rename

[jreback]

I suppose instead of the mapper argument it could/should be called index? (for consistency)

[cancan101]

Yes. Although for backward compatibility I would imagine Series.rename will need to take both and assert that only one is set.

[jreback]

ok....marking as an API issue (and a bug), will look at some point

[jtratner]

it probably just needs to special case MultiIndex, since if you look at the
code snippet, it's assuming that it's an Index instead.

[jreback]

Series.rename just needs to combine with DataFrame.rename/Panel.rename and be put in core/generic.py pretty straightforward (I put it on the todo for after series2)