Iterating over groupby fails with period-indexed dataframe

[goyodiaz]

I have found this while working with period-indexed data frames on Windows XP 32 bits:

import numpy as np
import pandas as pd
index = pd.period_range(start='1999-01', periods=5, freq='M')
s1 = pd.Series(np.random.rand(len(index)), index=index)
s2 = pd.Series(np.random.rand(len(index)), index=index)
series = [('s1', s1), ('s2',s2)]
df = pd.DataFrame.from_items(series)
grouped = df.groupby(df.index.month)
list(grouped)

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "D:\Python27\lib\site-packages\pandas\core\groupby.py", line 595, in get_iterator
    for key, (i, group) in izip(keys, splitter):
  File "D:\Python27\lib\site-packages\pandas\core\groupby.py", line 2214, in __iter__
    sdata = self._get_sorted_data()
  File "D:\Python27\lib\site-packages\pandas\core\groupby.py", line 2231, in _get_sorted_data
    return self.data.take(self.sort_idx, axis=self.axis)
  File "D:\Python27\lib\site-packages\pandas\core\frame.py", line 2891, in take
    new_index = self.index.take(indices)
  File "D:\Python27\lib\site-packages\pandas\tseries\period.py", line 1110, in take
    taken = self.values.take(indices, axis=axis)
TypeError: Cannot cast array data from dtype('int64') to dtype('int32') according to the rule 'safe'

This happens with the last stable version 0.11.0. However iterating over grouped.s1 or grouped.s2 just works.

[cpcloud]

See the docs for numpy.ndarray.astype for why this is happening. Let d1 and d2 be two dtypes. array([1,2], d1).astype(d2, cast='safe') will not work if d1.itemsize > d2.itemsize because 'safe' implies that you are not okay with losing information when casting from d1 to d2. You are on a 32-bit system and it sounds like there is no 64-bit integer emulation on your OS.

[jreback]

@cpcloud this might be a bug I have to look
period index is backed by int64, by there are conversions necessary at times to platform int
(which should be transparent)

will take a look

[cpcloud]

@jreback I was just about to ask if this is a numpy or pandas issue :)

[jreback]

closed by #3591