def test_2d_tensor(self):
a = np.array([[0.0, 1.0], [1.0, 0.0], [1.0, 1.0]])
b = np.roll(a, 1, axis=0)
c = (a - b)**2
c = np.sqrt(np.clip(np.sum(c, axis=1), 1e-6, 10))
expected = torch.exp(-5.0 * torch.tensor(c)).sum()
a, b = torch.tensor(a), torch.tensor(b)
actual = compute_disambiguation(a, b)
assert torch.isclose(expected.float(), actual.float())
def test_with_zero_1d(self):
a, b = torch.tensor([[0.0]]), torch.tensor([[0.0]])
expected = torch.exp(-5.0 * torch.tensor([[1e-3]]))
actual = compute_disambiguation(a, b)
assert torch.isclose(expected.float(), actual.float())
def test_large_value(self):
a, b = torch.tensor([[0.0]]), torch.tensor([[100.0]])
expected = torch.exp(-5.0 * torch.tensor([[np.sqrt(10.0)]]))
actual = compute_disambiguation(a, b)
assert torch.isclose(expected.float(), actual.float())
def test_with_constant_difference(self):
a, b = torch.tensor([[0.0]]), torch.tensor([[2.0]])
expected = torch.exp(-5.0 * torch.tensor([[2.0]]))
actual = compute_disambiguation(a, b)
assert torch.isclose(expected.float(), actual.float())