Merge branch 'master' into sa-sdr

CHANGELOG.md

@@ -26,6 +26,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Fixed
 
+- Fixed the use of `max_fpr` in `AUROC` metric when only one class is present ([#1895](https://github.com/Lightning-AI/torchmetrics/pull/1895))
+
+
 - Fixed bug related to empty predictions for `IntersectionOverUnion` metric ([#1892](https://github.com/Lightning-AI/torchmetrics/pull/1892))
 
 

requirements/doctest.txt

@@ -3,4 +3,4 @@
 
 pytest >=6.0.0, <7.5.0
 pytest-doctestplus >=0.9.0, <=0.13.0
-pytest-rerunfailures >=10.0, <12.0
+pytest-rerunfailures >=10.0, <13.0

requirements/test.txt

@@ -5,7 +5,7 @@ coverage ==7.2.7
 pytest ==7.4.0
 pytest-cov ==4.1.0
 pytest-doctestplus ==0.13.0
-pytest-rerunfailures ==11.1.2
+pytest-rerunfailures ==12.0
 pytest-timeout ==2.1.0
 phmdoctest ==1.4.0
 

requirements/visual.txt

@@ -1,5 +1,5 @@
 # NOTE: the upper bound for the package version is only set for CI stability, and it is dropped while installing this package
 #  in case you want to preserve/enforce restrictions on the latest compatible version, add "strict" as an in-line comment
 
-matplotlib >=3.2.0, <=3.7.1
+matplotlib >=3.2.0, <=3.7.2
 SciencePlots >= 2.0.0, <= 2.1.0

src/torchmetrics/functional/classification/auroc.py

@@ -86,7 +86,7 @@ def _binary_auroc_compute(
     pos_label: int = 1,
 ) -> Tensor:
     fpr, tpr, _ = _binary_roc_compute(state, thresholds, pos_label)
-    if max_fpr is None or max_fpr == 1:
+    if max_fpr is None or max_fpr == 1 or fpr.sum() == 0 or tpr.sum() == 0:
         return _auc_compute_without_check(fpr, tpr, 1.0)
 
     _device = fpr.device if isinstance(fpr, Tensor) else fpr[0].device

tests/unittests/classification/test_auroc.py

@@ -393,3 +393,17 @@ def test_valid_input_thresholds(metric, thresholds):
     with pytest.warns(None) as record:
         metric(thresholds=thresholds)
     assert len(record) == 0
+
+
+@pytest.mark.parametrize("max_fpr", [None, 0.8, 0.5])
+def test_corner_case_max_fpr(max_fpr):
+    """Check that metric returns 0 when one class is missing and `max_fpr` is set."""
+    preds = torch.tensor([0.1, 0.2, 0.3, 0.4])
+    target = torch.tensor([0, 0, 0, 0])
+    metric = BinaryAUROC(max_fpr=max_fpr)
+    assert metric(preds, target) == 0.0
+
+    preds = torch.tensor([0.5, 0.6, 0.7, 0.8])
+    target = torch.tensor([1, 1, 1, 1])
+    metric = BinaryAUROC(max_fpr=max_fpr)
+    assert metric(preds, target) == 0.0