Using numpy.testing functions with unittest

Question

I am using numpy.testing.assert_almost_equal in a unittest environment - but I am not sure what the right way to combine numpy and unittest is.

My first approach was to use assertTrue from unittest in combination with a is None comparison like so:

from unittest import TestCase
import numpy as np

class TestPredict(TestCase):
    def test_succeeding(self):
        self.assertTrue(
            np.testing.assert_almost_equal(1, 0.9999999999999) is None
        )
    def test_failing(self):
        self.assertTrue(
            np.testing.assert_almost_equal(1, 0.9) is None
        )

This gives the correct test results, but it is a bit hacky and it bloats the test code.

A simpler approach is the following:

from unittest import TestCase
import numpy as np

class TestPredict(TestCase):
    def test_succeeding(self):
        np.testing.assert_almost_equal(1, 0.9999999999999)
    def test_failing(self):
        np.testing.assert_almost_equal(1, 0.9)

This code also returns the correct test statistics as the above, but it is much more readable. The only downside I see with this is that pylint complains about the "R0201 method could be a function" message. Can this become an issue?

PS: I checked multiple posts here on SO that seemed related but didn't answer my specific question about the integration of unittest and numpy testing. (e.g. https://stackoverflow.com/a/4319870/6018688 talks bout catching Exceptions inside unittests. This seems bo be wrong or just simply an overkill.)

user_12480730 · Answer

If you are in a unittest enviroment, your second try is perfectly ok. If you don't want the pylint warnings, you can make static functions from the methods:

from unittest import TestCase
import numpy as np

class TestPredict(TestCase):
    @staticmethod
    def test_succeeding():
        np.testing.assert_almost_equal(1, 0.9999999999999)

    @staticmethod
    def test_failing():
        np.testing.assert_almost_equal(1, 0.9)

Note that pylint is basically just warning about the unused self parameter - this would not cause any issue except from the warning itself.

If you can use pytest instead, the code gets even cleaner, because you don't have to derive from a test case class:

import numpy as np

def test_succeeding():
    np.testing.assert_almost_equal(1, 0.9999999999999)

def test_failing():
    np.testing.assert_almost_equal(1, 0.9)

Using numpy.testing functions with unittest

1 Answers