TST: Allow assert_numpy_array_equal to compare nan

[sinhrks]

Related to #10373. pandas assert_array_equal needs to be updated to perform nan comparison properly.

import numpy as np
import pandas as pd
import pandas.util.testing as pdt
import numpy.testing.utils as npt

npt.assert_array_equal(np.array([1, np.nan, 3]), np.array([1, np.nan, 3]))
# no AssertionError

pdt.assert_numpy_array_equal(np.array([1, np.nan, 3]), np.array([1, np.nan, 3]))
# AssertionError: [  1.  nan   3.] is not equal to [  1.  nan   3.].

[jreback]

no, this is the point of array_equivalent. This is on purpose essentially np.array_equal.

most/all of these numpy comparison functions are broken for object dtypes , we don't in general use them. All that said, I don't see the harm in fixing this.

[sinhrks]

Thanks. Based on impl, there is no benefit to use assert_numpy_array_equal except for speed?

How about merging with numpy_array_equivalent to avoid confusion, assuming testing time is not so much affected because Series and DataFrame comparison is mostly done by assert_almost_equal.

[jreback]

yes these should be the same

I think assert_numpy_array_equal is used much more

[sinhrks]

OK, let's deprecate assert_numpy_array_equivalent. assert_numpy_array_equal is good for typing:)