rename multiindex bug in DataFrame

[joeb1415]

related / same as #4160

from datetime import datetime, time
from pandas import date_range, Series, DataFrame

time_ind = date_range(datetime(2013, 1, 1), datetime(2013, 1, 10), freq='1min')

s = Series(1, index=time_ind)
s_g = s.groupby([lambda x: x.hour, lambda x: x.minute]).sum()
s_g = s_g.rename(lambda (h, m): time(h, m, 0))

df = DataFrame(1, index=time_ind, columns=['a', 'b'])
df_g = df.groupby([lambda x: x.hour, lambda x: x.minute]).sum()
df_g = df_g.rename(lambda (h, m): time(h, m, 0))

Works fine on Series, but throws error on DataFrame:

Traceback (most recent call last):

File "<ipython-input-25-4e51131ae70d>", line 2, in <module>
  df_g = df_g.rename(lambda (h, m): time(h, m, 0))

File "C:\Python27\lib\site-packages\pandas\core\frame.py", line 3539, in rename
  result._rename_index_inplace(index_f)

File "C:\Python27\lib\site-packages\pandas\core\frame.py", line 3548, in _rename_index_inplace
  self._data = self._data.rename_axis(mapper, axis=1)

File "C:\Python27\lib\site-packages\pandas\core\internals.py", line 1734, in rename_axis
  new_axis = MultiIndex.from_tuples([tuple(mapper(y) for y in x) for x in index], names=index.names)

File "C:\Python27\lib\site-packages\pandas\core\internals.py", line 1734, in <genexpr>
  new_axis = MultiIndex.from_tuples([tuple(mapper(y) for y in x) for x in index], names=index.names)

File "<ipython-input-25-4e51131ae70d>", line 2, in <lambda>
  df_g = df_g.rename(lambda (h, m): time(h, m, 0))

TypeError: 'long' object is not iterable

[jreback]

do it like this

In [32]: df_g.index = map(lambda (h, m): time(h, m, 0),df_g.index)

In [33]: df_g
Out[33]: 
<class 'pandas.core.frame.DataFrame'>
Index: 1440 entries, 00:00:00 to 23:59:00
Data columns (total 2 columns):
a    1440  non-null values
b    1440  non-null values
dtypes: int64(2)

[joeb1415]

This works. Thanks.
Any reason DataFrame.rename doesn't / shouldn't work too ?

[jreback]

its a bug sort of of; using objects as column labels is frowned upon, but sometimes works depending on the repr. Pandas really doesn't have a straight time index, so I understand what you are doing; going to mark it as a feature request.

Its a bug because I think the time is somehow misinterpreted as a multi-index tuple.....

[jreback]

@wesm reason to have a TimeIndex?

[wesm]

might make sense. at some point we'll need to unify the column/index array business (i.e. if you want to call datetimeindex methods on a datetime64 column, you have to convert it first)

[jreback]

same as #4160