def test_forward_inverse_are_consistent(self):
features = 100
batch_size = 50
inputs = torch.randn(batch_size, features)
transforms = [
orthogonal.HouseholderSequence(features=features,
num_transforms=num_transforms)
for num_transforms in [1, 2, 11, 12]
]
self.eps = 1e-5
for transform in transforms:
with self.subTest(transform=transform):
self.assert_forward_inverse_are_consistent(transform, inputs)
def test_matrix(self):
features = 100
for num_transforms in [1, 2, 11, 12]:
with self.subTest(num_transforms=num_transforms):
transform = orthogonal.HouseholderSequence(
features=features, num_transforms=num_transforms)
matrix = transform.matrix()
self.assert_tensor_is_good(matrix, [features, features])
self.eps = 1e-5
self.assertEqual(matrix @ matrix.t(),
torch.eye(features, features))
self.assertEqual(matrix.t() @ matrix,
torch.eye(features, features))
self.assertEqual(matrix.t(), torch.inverse(matrix))
det_ref = torch.tensor(1.0 if num_transforms %
2 == 0 else -1.0)
self.assertEqual(matrix.det(), det_ref)
def test_inverse(self):
features = 100
batch_size = 50
for num_transforms in [1, 2, 11, 12]:
with self.subTest(num_transforms=num_transforms):
transform = orthogonal.HouseholderSequence(
features=features, num_transforms=num_transforms)
matrix = transform.matrix()
inputs = torch.randn(batch_size, features)
outputs, logabsdet = transform.inverse(inputs)
self.assert_tensor_is_good(outputs, [batch_size, features])
self.assert_tensor_is_good(logabsdet, [batch_size])
self.eps = 1e-5
self.assertEqual(outputs, inputs @ matrix)
self.assertEqual(
logabsdet,
torchutils.logabsdet(matrix) * torch.ones(batch_size))
