def setUp(self):
self.shape = [2, 3, 4]
self.batch_size = 10
self.transforms = [
nl.PiecewiseLinearCDF(self.shape),
nl.PiecewiseQuadraticCDF(self.shape),
nl.PiecewiseCubicCDF(self.shape),
nl.PiecewiseRationalQuadraticCDF(self.shape),
]
def test_forward_inverse_are_consistent(self):
shape = [2, 3, 4]
batch_size = 10
transforms = [
nl.PiecewiseLinearCDF(shape, tails="linear"),
nl.PiecewiseQuadraticCDF(shape, tails="linear"),
nl.PiecewiseCubicCDF(shape, tails="linear"),
nl.PiecewiseRationalQuadraticCDF(shape, tails="linear"),
]
for transform in transforms:
with self.subTest(transform=transform):
inputs = 3 * torch.randn(batch_size, *shape)
self.eps = 1e-4
self.assert_forward_inverse_are_consistent(transform, inputs)