def test_top_k(k: int, preds: torch.Tensor, target: torch.Tensor, reduce: str, expected: torch.Tensor):
""" A simple test to check that top_k works as expected """
class_metric = StatScores(top_k=k, reduce=reduce, num_classes=3)
class_metric.update(preds, target)
assert torch.equal(class_metric.compute(), expected.T)
assert torch.equal(stat_scores(preds, target, top_k=k, reduce=reduce, num_classes=3), expected.T)
def test_v1_5_metric_classif_mix():
ConfusionMatrix.__init__._warned = False
with pytest.deprecated_call(match="It will be removed in v1.5.0"):
ConfusionMatrix(num_classes=1)
FBeta.__init__._warned = False
with pytest.deprecated_call(match="It will be removed in v1.5.0"):
FBeta(num_classes=1)
F1.__init__._warned = False
with pytest.deprecated_call(match="It will be removed in v1.5.0"):
F1(num_classes=1)
HammingDistance.__init__._warned = False
with pytest.deprecated_call(match="It will be removed in v1.5.0"):
HammingDistance()
StatScores.__init__._warned = False
with pytest.deprecated_call(match="It will be removed in v1.5.0"):
StatScores()
target = torch.tensor([1, 1, 0, 0])
preds = torch.tensor([0, 1, 0, 0])
confusion_matrix._warned = False
with pytest.deprecated_call(match="It will be removed in v1.5.0"):
assert torch.equal(
confusion_matrix(preds, target, num_classes=2).float(),
torch.tensor([[2.0, 0.0], [1.0, 1.0]]))
target = torch.tensor([0, 1, 2, 0, 1, 2])
preds = torch.tensor([0, 2, 1, 0, 0, 1])
fbeta._warned = False
with pytest.deprecated_call(match="It will be removed in v1.5.0"):
assert torch.allclose(fbeta(preds, target, num_classes=3, beta=0.5),
torch.tensor(0.3333),
atol=1e-4)
f1._warned = False
with pytest.deprecated_call(match="It will be removed in v1.5.0"):
assert torch.allclose(f1(preds, target, num_classes=3),
torch.tensor(0.3333),
atol=1e-4)
target = torch.tensor([[0, 1], [1, 1]])
preds = torch.tensor([[0, 1], [0, 1]])
hamming_distance._warned = False
with pytest.deprecated_call(match="It will be removed in v1.5.0"):
assert hamming_distance(preds, target) == torch.tensor(0.25)
preds = torch.tensor([1, 0, 2, 1])
target = torch.tensor([1, 1, 2, 0])
stat_scores._warned = False
with pytest.deprecated_call(match="It will be removed in v1.5.0"):
assert torch.equal(stat_scores(preds, target, reduce="micro"),
torch.tensor([2, 2, 6, 2, 4]))
def test_wrong_params(reduce, mdmc_reduce, num_classes, inputs, ignore_index):
"""Test a combination of parameters that are invalid and should raise an error.
This includes invalid ``reduce`` and ``mdmc_reduce`` parameter values, not setting
``num_classes`` when ``reduce='macro'`, not setting ``mdmc_reduce`` when inputs
are multi-dim multi-class``, setting ``ignore_index`` when inputs are binary, as well
as setting ``ignore_index`` to a value higher than the number of classes.
"""
with pytest.raises(ValueError):
stat_scores(
inputs.preds[0], inputs.target[0], reduce, mdmc_reduce, num_classes=num_classes, ignore_index=ignore_index
)
with pytest.raises(ValueError):
sts = StatScores(reduce=reduce, mdmc_reduce=mdmc_reduce, num_classes=num_classes, ignore_index=ignore_index)
sts(inputs.preds[0], inputs.target[0])
def test_wrong_threshold():
with pytest.raises(ValueError):
StatScores(threshold=1.5)