fix ted training e2e entities when none are given

changelog/8194.bugfix.md

@@ -0,0 +1,2 @@
+Fix `TEDPolicy` training e2e entities when no entities are present in the stories
+but there are entities in the domain.

rasa/core/policies/ted_policy.py

@@ -393,14 +393,26 @@ def _create_label_data(
 
         return label_data, encoded_all_labels
 
+    @staticmethod
+    def _should_extract_entities(
+        entity_tags: List[List[Dict[Text, List["Features"]]]]
+    ) -> bool:
+        for turns_tags in entity_tags:
+            for turn_tags in turns_tags:
+                # if turn_tags are empty or all entity tag indices are `0`
+                # it means that all the inputs only contain NO_ENTITY_TAG
+                if turn_tags and np.any(turn_tags[ENTITY_TAGS][0].features):
+                    return True
+        return False
+
     def _create_data_for_entities(
         self, entity_tags: Optional[List[List[Dict[Text, List["Features"]]]]]
     ) -> Optional[Data]:
         if not self.config[ENTITY_RECOGNITION]:
             return
 
         # check that there are real entity tags
-        if entity_tags and any([any(turn_tags) for turn_tags in entity_tags]):
+        if entity_tags and self._should_extract_entities(entity_tags):
             entity_tags_data, _ = convert_to_data_format(entity_tags)
             return entity_tags_data
 

tests/core/policies/test_ted_policy.py

@@ -35,7 +35,6 @@
     VALUE_RELATIVE_ATTENTION,
     MODEL_CONFIDENCE,
     COSINE,
-    INNER,
     AUTO,
     LINEAR_NORM,
 )
@@ -93,7 +92,7 @@ def test_train_model_checkpointing(self, tmp_path: Path):
 
     def create_policy(
         self, featurizer: Optional[TrackerFeaturizer], priority: int
-    ) -> Policy:
+    ) -> TEDPolicy:
         return TEDPolicy(featurizer=featurizer, priority=priority)
 
     def test_similarity_type(self, trained_policy: TEDPolicy):