fix(corpus_score): get max score from all references

signwriting_evaluation/metrics/base.py

@@ -13,9 +13,13 @@ def __call__(self, hypothesis: str, reference: str) -> float:
     def score(self, hypothesis: str, reference: str) -> float:
         raise NotImplementedError
 
+    def score_max(self, hypothesis: str, references: list[str]) -> float:
+        all_scores = self.score_all([hypothesis], references)
+        return max(max(scores) for scores in all_scores)
+
     def corpus_score(self, hypotheses: list[str], references: list[list[str]]) -> float:
         # Default implementation: average over sentence scores
-        return sum(self.score(h, r) for h, r in zip(hypotheses, references)) / len(references)
+        return sum(self.score_max(h, r) for h, r in zip(hypotheses, references)) / len(references)
 
     def score_all(self, hypotheses: list[str], references: list[str]) -> list[list[float]]:
         # Default implementation: call the score function for each hypothesis-reference pair

signwriting_evaluation/metrics/test_similarity.py

@@ -27,6 +27,14 @@ def test_different_shape(self):
         self.assertIsInstance(score, float)
         self.assertAlmostEqual(score, 0.8326259781509948)
 
+    def test_corpus_score(self):
+        hypothesis = "M530x538S17600508x462S15a11493x494S20e00488x510S22f03469x517"
+        good_reference = "M530x538S17600508x462S12a11493x494S20e00488x510S22f13469x517"
+        bad_reference = "M530x538S17600508x462"
+        score = self.metric.corpus_score([hypothesis], [[good_reference, bad_reference]])
+        self.assertIsInstance(score, float)
+        self.assertAlmostEqual(score, 0.8326259781509948)
+
 
 if __name__ == '__main__':
     unittest.main()