def test_zero_variance(self):
input = torch.tensor([
[[0.0, 0.0], [0.0, 0.0]],
[[0.0, 0.0], [0.0, 0.0]],
[[0.0, 0.0], [0.0, 0.0]],
])
layer = Standardize(dims=[-1, -2])
output = layer(input)
assert not torch.isnan(
output).any(), "divided by zero because variance=0"
def test_basic_input(self):
input = torch.tensor([
[[0.0, 1.0], [1.0, 2.0]],
[[0.0, 1.0], [1.0, 2.0]],
[[0.0, 1.0], [1.0, 2.0]],
])
expected = torch.tensor([
[[-1.5, 0.0], [0.0, 1.5]],
[[-1.5, 0.0], [0.0, 1.5]],
[[-1.5, 0.0], [0.0, 1.5]],
])
layer = Standardize(dims=[-1, -2])
output = layer(input)
assert torch.allclose(output, expected)
def test_non_contiguous_dims(self):
input = torch.tensor([
[[0.0, 1.0], [1.0, 2.0]],
[[0.0, 1.0], [1.0, 2.0]],
[[0.0, 1.0], [1.0, 2.0]],
]).permute(1, 0, 2)
expected = torch.tensor([
[[-1.5, 0.0], [0.0, 1.5]],
[[-1.5, 0.0], [0.0, 1.5]],
[[-1.5, 0.0], [0.0, 1.5]],
]).permute(1, 0, 2)
layer = Standardize(dims=[0, 2])
output = layer(input)
assert torch.allclose(output, expected)
def test_repr(self):
layer = Standardize(dims=[-1, -2])
print(layer)