feat: update JSONParseEvaluator

donut/util.py

@@ -137,14 +137,13 @@ def __getitem__(self, idx: int) -> Tuple[torch.Tensor, torch.Tensor, torch.Tenso
 
 class JSONParseEvaluator:
     """
-    Calculate n-TED(Normalized Tree Edit Distance) based accuracy and  F1 accuracy score
+    Calculate n-TED(Normalized Tree Edit Distance) based accuracy and F1 accuracy score
     """
 
     @staticmethod
     def flatten(data: dict):
         """
         Convert Dictionary into Non-nested Dictionary
-
         Example:
             input(dict)
                 {
@@ -153,13 +152,15 @@ def flatten(data: dict):
                         {"name" : ["juice"], "count" : ["1"]},
                     ]
                 }
-            output(dict)
-                {
-                    "menu.name": ["cake", "juice"],
-                    "menu.count": ["2", "1"],
-                }
+            output(list)
+                [
+                    ("menu.name", "cake"),
+                    ("menu.count", "2"),
+                    ("menu.name", "juice"),
+                    ("menu.count", "1"),
+                ]
         """
-        flatten_data = defaultdict(list)
+        flatten_data = list()
 
         def _flatten(value, key=""):
             if type(value) is dict:
@@ -169,10 +170,10 @@ def _flatten(value, key=""):
                 for value_item in value:
                     _flatten(value_item, key)
             else:
-                flatten_data[key].append(value)
+                flatten_data.append((key, value))
 
         _flatten(data)
-        return dict(flatten_data)
+        return flatten_data
 
     @staticmethod
     def update_cost(label1: str, label2: str):
@@ -225,10 +226,11 @@ def normalize_dict(self, data: Union[Dict, List, Any]):
         elif isinstance(data, list):
             if all(isinstance(item, dict) for item in data):
                 new_data = []
-                for item in sorted(data, key=lambda x: str(sorted(x.items()))):
+                for item in data:
                     item = self.normalize_dict(item)
                     if item:
                         new_data.append(item)
+                new_data = sorted(new_data, key=lambda x: str(x.keys())+str(x.values()))
             else:
                 new_data = sorted([str(item) for item in data if type(item) in {str, int, float} and str(item)])
         else:
@@ -243,14 +245,14 @@ def cal_f1(self, preds: List[dict], answers: List[dict]):
         total_tp, total_fn_or_fp = 0, 0
         for pred, answer in zip(preds, answers):
             pred, answer = self.flatten(self.normalize_dict(pred)), self.flatten(self.normalize_dict(answer))
-            for pred_key, pred_values in pred.items():
-                for pred_value in pred_values:
-                    if pred_key in answer and pred_value in answer[pred_key]:
-                        answer[pred_key].remove(pred_value)
-                        total_tp += 1
-                    else:
-                        total_fn_or_fp += 1
-        return total_tp / (total_tp + (total_fn_or_fp) / 2)
+            for field in pred:
+                if field in answer:
+                    total_tp += 1
+                    answer.remove(field)
+                else:
+                    total_fn_or_fp += 1
+            total_fn_or_fp += len(answer)
+        return total_tp / (total_tp + total_fn_or_fp / 2)
 
     def construct_tree_from_dict(self, data: Union[Dict, List], node_name: str = None):
         """