NumPy scalars are not collections.Hashable on Python 3

[shoyer]

xref: pandas-dev/pandas#9276

For example:

In [1]: import numpy as np

In [2]: import collections

In [3]: isinstance(3.14, collections.Hashable)
Out[3]: True

In [4]: isinstance(np.float64(3.14), collections.Hashable)
Out[4]: False

On Python 2.7, the last output here is True (which is the correct value)

[sylviamic]

This peculiarity seems to be limited in scope - I only see it with np.float(64) in my short tests.

In [4]: isinstance(np.float32(3.14), collections.Hashable)
Out[4]: True

In [5]: isinstance(np.float64(3.14), collections.Hashable)
Out[5]: False

In [6]: isinstance(np.float128(3.14), collections.Hashable)
Out[6]: True

In [7]: isinstance(np.float(3.14), collections.Hashable)
Out[7]: True

In [8]: isinstance(np.int(3.14), collections.Hashable)
Out[8]: True

I'm not sure where these types are implemented, but I think this afternoon I'm going to poke around the code and look to see what differences there are between the different float types.

[mattip]

Cannot reproduce on linux64, numpy 1.13, numpy 1.16, python 3.6.6.

[mattip]

Closing. Please reopen or open a new issue if the issue can be reproduced