def test_input_types(self, dtype):
torch.random.manual_seed(42)
coords = torch.rand(1000, 3).to(dtype)
output = center(coords)
assert output.is_floating_point()
if coords.is_floating_point():
assert coords.dtype == output.dtype
def test_center_basic(self, strategy):
torch.random.manual_seed(42)
coords = torch.randint(5, 11, (1000, 3)).float()
output = center(coords, strategy=strategy)
if strategy == "minmax":
assert torch.allclose(output.max(), torch.tensor([2.5]))
assert torch.allclose(output.min(), torch.tensor([-2.5]))
else:
assert torch.allclose(output.mean(dim=0), torch.tensor([0.0, 0.0, 0.0]), atol=1e-4)
def test_inplace(self, inplace):
torch.random.manual_seed(42)
coords = torch.rand(1000, 3)
coords_copy = coords.clone()
output = center(coords, inplace=inplace)
if inplace:
assert not torch.allclose(coords, coords_copy)
assert output is coords
else:
assert torch.allclose(coords, coords_copy)
assert not torch.allclose(coords, output)
def test_grad(self):
torch.random.manual_seed(42)
coords = torch.rand(1000, 3, requires_grad=True)
output = center(coords)
output.sum().backward()
def test_center_xy_only(self):
torch.random.manual_seed(42)
coords = torch.randint(5, 11, (1000, 3)).float()
output1 = center(coords[..., :2])
output2 = center(coords)
assert torch.allclose(output1, output2[..., :2])