np.isnan() == False, but np.isnan() is not False

问题:

As far as I understand it, == checks for equality of value, and is checks for identity of structure behind value (as, say === in some other languages).

Given that, I don't understand the following:

np.isnan(30) == False
Out[19]: 
True
np.isnan(30) is False
Out[20]: 
False

It appears not to be the case with other identity checks:

(5 == 4) == False
Out[22]: 
True
(5 == 4) is False
Out[23]: 
True

It appears as if np.isnan() returns False as a value, but not as identity. Why is that the case?

回答1:

numpy.isnan() returns a compatible type object:

>>> import numpy
>>> type(numpy.isnan(0))
<class 'numpy.bool_'>

This is a custom boolean that can be stored efficiently in numpy arrays, see Numpy's Data Types documentation. The numpy.isnan() function can also operate on arrays, producing another array with results:

>>> numpy.isnan(numpy.array([1, 2]))
array([False, False], dtype=bool)

where again the dtype is the Numpy boolean object.

Python makes no guarantees that boolean operations must always return a singleton boolean value. You should never test for is True or is False anyway. Use numpy.isnan() output directly in boolean operations, use not to test for false values:

if numpy.isnan(foo):

and

if not numpy.isnan(bar):

回答2:

np.isnan(30) returns np.False_ which has a different identity from False; don't rely on this though.

>>> import numpy as np
>>> np.isnan(30) is np.False_
True
>>> np.False_ is False
False
>>>