BUG: nanops.nanmedian doesn't work when axis=None

[toddrjen]

The default value for the axis argument for nanops.nanmedia is None. However, the function raises an exception when using this value:

>>> from pandas.core import nanops
>>> import bottleneck
>>> import numpy as np
>>> nanops._USE_BOTTLENECK = False
>>> 
>>> val = np.random.randn(13, 11, 7)
>>> 
>>> bottleneck.nanmedian(val, axis=None).ndim
0
>>> nanops.nanmedian(val, axis=None)
TypeError: unsupported operand type(s) for +=: 'NoneType' and 'int'
>>> nanops.nanmedian(val)
TypeError: unsupported operand type(s) for +=: 'NoneType' and 'int'

[jreback]

couple of points

• their are really NO tests for nanops explicity (a few use it in test_frame/test_common but very limited)
• need a bottleneck testing decorator to set/unset the _USE_BOTTLENECK switch for testing (e.g. see test_frame/test_sem
• all of the nan* functions should have assert axis is not None (which we are not testing), as its supposed to ALWAYS be passed in

would welcome tests/fixes for these

[toddrjen]

I have fixes for all of these, and an nanops test suite, but I thought you said I needed to raise issues for the problems I found before I submit the pull request.

[jreback]

ahh gr8. yes I wanted issues for these because then you can reference them (conceptually you could put it in the top of the PR as well. but almost always better to create an issue then to close it with a PR. (makes finding it easier at the end of the day)

[jreback]

if this is the only once that doesn't handle axis=None then its a bug, but I still am puzzled how it gets passed that. I think that itself is a problem (that's why I am suggesting putting in asserts)