BUG: .filter with unicode labels when can't encode

[griai]

Edit #10506 breaks if the DataFrame contains unicode column names with non-ASCII characters.

import pandas as pd
df = pd.DataFrame({u'a': [1, 2, 3], u'ä': [4, 5, 6]})
df.filter(regex=u'a')

throws me a

---------------------------------------------------------------------------
UnicodeEncodeError                        Traceback (most recent call last)
<ipython-input-10-9de5a19c260e> in <module>()
----> 1 df.filter(regex=u'a')

C:\Users\...\AppData\Local\Continuum\32bit\Anaconda\envs\test\lib\site-packages\pandas\core\generic.pyc in filter(self, items, like, regex, axis)
   2013             matcher = re.compile(regex)
   2014             return self.select(lambda x: matcher.search(str(x)) is not None,
-> 2015                                axis=axis_name)
   2016         else:
   2017             raise TypeError('Must pass either `items`, `like`, or `regex`')

C:\Users\...\AppData\Local\Continuum\32bit\Anaconda\envs\test\lib\site-packages\pandas\core\generic.pyc in select(self, crit, axis)
   1545         if len(axis_values) > 0:
   1546             new_axis = axis_values[
-> 1547                 np.asarray([bool(crit(label)) for label in axis_values])]
   1548         else:
   1549             new_axis = axis_values

C:\Users\...\AppData\Local\Continuum\32bit\Anaconda\envs\test\lib\site-packages\pandas\core\generic.pyc in <lambda>(x)
   2012         elif regex:
   2013             matcher = re.compile(regex)
-> 2014             return self.select(lambda x: matcher.search(str(x)) is not None,
   2015                                axis=axis_name)
   2016         else:

UnicodeEncodeError: 'ascii' codec can't encode character u'\xe4' in position 0: ordinal not in range(128)

[jreback]

xref #10384

yeah str(x) will try to encode, so probably easiest to either just catch this (and pass thru if it cannot encode), or just stringify integers (but then that leaves out things like float columns and such).

So I think the former is ok. want to do a PR?

would need to add some tests for other column label types as well

(e.g. the tests should loop thru all of the index types).

[griai]

I don't have an installed git environment at the moment. So I cannot do the Pull Request, unfortunately.
I would support the passing-through solution if the argument cannot be encoded, since it is the easiest and a pretty general fix (although this fallback mechanism might seem a bit intransparent).