def test_if_returns_expected_value_for_multiclass(self, images, target,
classes, result):
images = images.unsqueeze(dim=CHANNEL_DIM).repeat(1, classes, 1)
target = target.unsqueeze(dim=CHANNEL_DIM).repeat(1, classes, 1)
result = [result for _ in range(classes)]
assert np.all(
np.isclose(F.f_score(images, target, 1), result, atol=1.e-3))
def __call__(self, prediction: torch.Tensor,
target: torch.Tensor) -> torch.Tensor:
"""
Args:
prediction: Network output.
Dimensions - (Batch, Class, Depth, Height, Width)
target: Target values. The classes should be ont-hot encoded.
Dimensions - (Batch, Class, Depth, Height, Width)
Returns:
torch.Tensor: F Score for each class averaged
across the whole batch
"""
return F.f_score(prediction, target, self.beta, self.epsilon)
def test_if_returns_correct_number_of_classes(self, input_shape, result):
images = torch.ones(*input_shape)
target = torch.ones(*input_shape)
assert len(F.f_score(images, target, 1)) == result
def test_if_f2_returns_expected_values(self, images, target, result):
assert np.isclose(F.f_score(images.unsqueeze(dim=CHANNEL_DIM),
target.unsqueeze(dim=CHANNEL_DIM),
beta=2),
result,
atol=1.e-3)
def test_if_returns_0_for_worst_fit(self):
images = torch.zeros(*BATCH_DIMS)
target = torch.ones(*BATCH_DIMS)
assert np.isclose(F.f_score(images, target, beta=1), 0, atol=1.e-4)
def test_if_returns_1_for_perfect_fit(self):
images = torch.ones(*BATCH_DIMS)
target = torch.ones(*BATCH_DIMS)
assert np.isclose(F.f_score(images, target, beta=1), 1, atol=1.e-4)