np.issubdtype does not work well with pandas

For example, below I am trying to use np.issubdtype to check if a pandas categorical is a subtype of a np.number. IMO this should definitely be false rather than throwing an error.

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-99-4ba76ca46d4a> in <module>()
     32 
---> 33 np.issubdtype(df.dtypes[1], np.number)

/Users/jhoward/homebrew/lib/python2.7/site-packages/numpy/core/numerictypes.pyc in issubdtype(arg1, arg2)
    753     """
    754     if issubclass_(arg2, generic):
--> 755         return issubclass(dtype(arg1).type, arg2)
    756     mro = dtype(arg2).type.mro()
    757     if len(mro) > 1:

TypeError: data type not understood

This issue should be fixed on numpy's side because external types should by default compare as false unless the external package developers have explicitly used the numpy type system.

If this is a reasonable update, I would be glad to fix this myself.

Start with np.issubdtype and the dtype handling shown in numpy/core/numerictypes.pyc, then inspect the current implementation and related tests. Reproduce the pandas categorical example and verify that comparing an unsupported external dtype returns false rather than raising TypeError.