Replace pytest's paremeterization with parameterized #1157
Conversation
| torch.testing.assert_allclose(norm_tensor, expected_norm_tensor, atol=1e-5, rtol=1e-5) | ||
| class TestComplexNorm(common_utils.TorchaudioTestCase): | ||
|
|
||
| @parameterized.expand([ |
There was a problem hiding this comment.
I suppose this is not quite right. list(itertools.product(...) might be the right way to go about this. However that would require importing import itertools.
@mthrok which would be the best way to go about this?.
There was a problem hiding this comment.
Yeah, that's right. That's the pattern I use everywhere. itertools is a standard library so there is no problem.
Maybe it's good time to add a helper decorator for @partameterized.expand(list(itertools.product pattern, but let's not bother at this moment.
|
Following occurrences will be substituted (
|
| (torch.randn(1, 2, 1025, 400, 2), 1), | ||
| (torch.randn(1025, 400, 2), 2) | ||
| ]) | ||
| def test_complex_norm(self, complex_tensor, power): |
There was a problem hiding this comment.
Performing torch.randn in decorator makes it difficult to control the randomness, so instead can you parameterize the shape?
def test_complex_norm(self, shape, power):
torch.manual_seed(0)
complex_tensor = torch.randn(*shape)
mthrok
left a comment
mthrok
left a comment
There was a problem hiding this comment.
Looks good. Made some suggestions.
| complex_tensor = torch.randn(*shape) | ||
| expected_norm_tensor = complex_tensor.pow(2).sum(-1).pow(power / 2) | ||
| norm_tensor = F.complex_norm(complex_tensor, power) | ||
| torch.testing.assert_allclose(norm_tensor, expected_norm_tensor, atol=1e-5, rtol=1e-5) |
There was a problem hiding this comment.
Can you replace torch.testing.assert_allclose with self.assertEqual?
the use of torch.testing.assert_allclose is discouraged now.
|
|
||
| class TestDB_to_amplitude(common_utils.TorchaudioTestCase): | ||
| def test_DB_to_amplitude(self): | ||
| torch.random.manual_seed(42) |
There was a problem hiding this comment.
Let's leave this test untouched. #1113 is modifying this test suite.
| complex_stretch = complex_specgrams_stretch[index].numpy() | ||
| complex_stretch = complex_stretch[..., 0] + 1j * complex_stretch[..., 1] | ||
|
|
||
| assert np.allclose(complex_stretch, expected_complex_stretch, atol=1e-5) |
There was a problem hiding this comment.
Can you use self.assertEqual here? You might need to wrap the expected using torch.from_numpy.
|
Thanks! |
3 participants
Made Changes to
fixes: #1156