fix for MetricCollection.update gives identical results

CHANGELOG.md

@@ -27,7 +27,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Fixed
 
--
+- Fixed `MetricCollection.update` gives identical results ([#2944](https://github.com/Lightning-AI/torchmetrics/issues/2944))
 
 
 ---

src/torchmetrics/collections.py

@@ -315,10 +315,12 @@ def _equal_metric_states(metric1: Metric, metric2: Metric) -> bool:
                 return False
 
             if isinstance(state1, Tensor) and isinstance(state2, Tensor):
-                return state1.shape == state2.shape and allclose(state1, state2)
+                if not (state1.shape == state2.shape and allclose(state1, state2)):
+                    return False
 
             if isinstance(state1, list) and isinstance(state2, list):
-                return all(s1.shape == s2.shape and allclose(s1, s2) for s1, s2 in zip(state1, state2))
+                if not (all(s1.shape == s2.shape and allclose(s1, s2) for s1, s2 in zip(state1, state2))):
+                    return False
 
         return True
 

tests/unittests/bases/test_collections.py

@@ -743,3 +743,35 @@ def compute(self):
     # Print the calculated metrics
     assert "my_prefix/accuracy/my_postfix" in res
     assert "my_prefix/precision/my_postfix" in res
+
+
+def test_collection_update():
+    """Test that metric collection updates metrics.
+
+    See issue: https://github.com/Lightning-AI/torchmetrics/issues/2916
+
+    """
+    from torchmetrics.text import BLEUScore
+
+    scores = MetricCollection({
+        "bleu-1": BLEUScore(1),
+        "bleu-2": BLEUScore(2),
+        "bleu-3": BLEUScore(3),
+        "bleu-4": BLEUScore(4),
+    })
+
+    preds = ["the cat is on the mat"]
+    target = [["there is a cat on the mat", "a cat is on the mat"]]
+
+    scores.update(preds, target)
+    actual = scores.compute()
+
+    expected = {
+        "bleu-1": torch.tensor(0.8333),
+        "bleu-2": torch.tensor(0.8165),
+        "bleu-3": torch.tensor(0.7937),
+        "bleu-4": torch.tensor(0.7598),
+    }
+
+    for k, v in expected.items():
+        torch.testing.assert_close(actual=actual.get(k), expected=v, rtol=1e-4, atol=1e-4)