OS: macOS Sierra (10.12.5)
Python version: 2.7.12
numpy version: 1.13.0

Passing np.array_equal() two apparently identical arrays containing NaN values returns False. See the following example replicating this issue:

>>> a = np.array([1, 2, np.nan])
>>> np.array_equal(a, a)
False

It would make sense to me if this returned True, though I understand that the following comparison returns False by definition:

>>> np.nan == np.nan
False

I presume this is related to changes in np.equal and np.not_equal. If this is the desired behaviour, I wonder if it might be more helpful for the element-wise comparison to raise an exception on encountering NaN comparisons, as this has caused unit test failures for me and took a while to track down.