fix(similarity): handle edge case of unknown symbol class

Author: AmitMY

signwriting_evaluation/metrics/similarity.py

@@ -60,6 +60,9 @@ def calculate_distance(self, hyp: SignSymbol, ref: SignSymbol) -> float:
         hyp_class = self.get_shape_class_index(hyp_veq)
         ref_class = self.get_shape_class_index(ref_veq)
 
+        if hyp_class is None or ref_class is None:
+            return self.max_distance
+
         hyp_veq = self.weight_vector(hyp_veq)
         ref_veq = self.weight_vector(ref_veq)
         distance = (dis.euclidean(hyp_veq, ref_veq) +

signwriting_evaluation/metrics/test_similarity.py

@@ -65,6 +65,15 @@ def test_score_swu(self):
         self.assertIsInstance(score, float)  # Check if the score is a float
         self.assertAlmostEqual(score, 0.5509574768254414)
 
+    def test_unknown_symbol_class_returns_zero_score(self):
+        # Test that symbols with shapes outside defined class ranges are handled gracefully
+        # When a symbol's shape doesn't match any defined symbol class ranges, the metric
+        # should return maximum distance (resulting in zero similarity score) rather than crashing
+        hypothesis = "M530x538S38c00508x462"  # S38c00 has shape 0x38c, outside all defined ranges
+        reference = "M530x538S10000508x462"   # S10000 has shape 0x100, in hands_shapes range
+        score = self.metric.score(hypothesis, reference)
+        self.assertEqual(score, 0)
+
 
 if __name__ == '__main__':
     unittest.main()