fix: evaluation

chaine/optimization/metrics.py

@@ -108,17 +108,18 @@ def evaluate_predictions(true: list[list[str]], pred: list[list[str]]) -> dict[s
         true_labels = [l.removeprefix("B-").removeprefix("I-") for l in true_labels]
         predicted_labels = [l.removeprefix("B-").removeprefix("I-") for l in predicted_labels]
 
+        if len(true_labels) != len(predicted_labels):
+            raise ValueError(f"Different lengths: '{true_labels}' vs. '{predicted_labels}'")
+
         for true_label, predicted_label in zip(true_labels, predicted_labels):
-            if true_label == "O" and true_label == predicted_label:
-                counts["tn"] += 1
-            elif true_label == "O" and true_label != predicted_label:
-                counts["fp"] += 1
-            elif true_label != "O" and true_label == predicted_label:
+            if true_label != "O" and predicted_label == true_label:
                 counts["tp"] += 1
-            elif true_label != "O" and predicted_label == "O":
-                counts["fn"] += 1
-            elif true_label != "O" and true_label != predicted_label:
+            if predicted_label != "O" and predicted_label != true_label:
                 counts["fp"] += 1
+            if true_label == "O" and predicted_label == "O":
+                counts["tn"] += 1
+            if true_label != "O" and predicted_label == "O":
+                counts["fn"] += 1
 
     # calculate precision, recall and f1 score
     return {