error with groupby and transform while using datetime64[ns] as part of a key

[MiloszD]

from pandas.util.testing import choice
from pandas import DataFrame, date_range
from numpy.random import randn

n = 8
dates = date_range('1/1/2000', periods=n, freq='1T')
colors = choice(['red', 'green'], size=n)

df = DataFrame({'value': randn(n), 'date': dates, 'color': colors})

zscore = lambda x: (x - x.mean()) / x.std()

df
Out[2]:
color date value
0 red 2000-01-01 00:00:00 -0.382902
1 green 2000-01-01 00:01:00 0.991723
2 green 2000-01-01 00:02:00 -0.442476
3 green 2000-01-01 00:03:00 -0.776372
4 red 2000-01-01 00:04:00 -1.372511
5 red 2000-01-01 00:05:00 -0.276006
6 red 2000-01-01 00:06:00 -0.182162
7 red 2000-01-01 00:07:00 -0.232967

[8 rows x 3 columns]

df.dtypes
Out[3]:
color object
date datetime64[ns]
value float64
dtype: object

df.groupby(['color','date'])['value'].transform(zscore)

---

KeyError Traceback (most recent call last)
in ()
----> 1 df.groupby(['color','date'])['value'].transform(zscore)

C:\Users\MD\Anaconda\lib\site-packages\pandas-0.13.1.dev-py2.7-win-amd64.egg\pandas\core\groupby.pyc in transform(self, func, _args, *_kwargs)
1845 # this needs to be an ndarray
1846 result = Series(result)
-> 1847 result.iloc[self._get_index(name)] = res
1848 result = result.values
1849

C:\Users\MD\Anaconda\lib\site-packages\pandas-0.13.1.dev-py2.7-win-amd64.egg\pandas\core\groupby.pyc in _get_index(self, name)
265 """ safe get index """
266 try:
--> 267 return self.indices[name]
268 except:
269 if isinstance(name, Timestamp):

KeyError: ('green', Timestamp('2000-01-01 00:01:00', tz=None))

[jreback]

this is a dup of #3794

[jreback]

you can do a nested group or resample followed by a groupby as a work around