properly fix intent_name by removing it (#6962)

* properly fix intent_name by removing it

* fix training_data filtering

* return dissappeared nlu_examples

* fix creating message from events

* remove unused import

* fix a mess created by empty string

* pass only nlu examples

* fix is_core_message

* fix is_core_message

* fix pattern utils tests

* fix test_training_data

* fix test_regex_entity_extractor

* fix test_test

* use only NLU examples if needed

* fix test_bilou_utils

* increase timeout for pipeline test

* update examples for tests

* update check if training is possible

Co-authored-by: Tanja Bergmann <[email protected]>
Co-authored-by: Daksh <[email protected]>

data/examples/wit/demo-flights.json

@@ -34,6 +34,12 @@
     {
       "text" : "i'm looking for a flight from london to amsterdam next monday",
       "entities" : [
+        {
+          "entity" : "intent",
+          "value" : "\"flight_booking\"",
+          "start" : 0,
+          "end" : 42
+        },
         {
           "entity" : "location",
           "value" : "\"london\"",
@@ -59,6 +65,12 @@
     {
       "text" : "i want to fly to berlin",
       "entities" : [
+        {
+          "entity" : "intent",
+          "value" : "\"flight_booking\"",
+          "start" : 0,
+          "end" : 42
+        },
         {
           "entity" : "location",
           "value" : "\"berlin\"",
@@ -71,6 +83,12 @@
     {
       "text" : "i want to fly from london",
       "entities" : [
+        {
+          "entity" : "intent",
+          "value" : "\"flight_booking\"",
+          "start" : 0,
+          "end" : 42
+        },
         {
           "entity" : "location",
           "value" : "\"london\"",

data/test/wit_converted_to_rasa.json

@@ -28,7 +28,8 @@
         ]
       },
       {
-        "text": "i'm looking for a flight from london to amsterdam next monday", 
+        "text": "i'm looking for a flight from london to amsterdam next monday",
+        "intent": "flight_booking",
         "entities": [
           {
             "start": 30, 
@@ -53,7 +54,8 @@
         ]
       }, 
       {
-        "text": "i want to fly to berlin", 
+        "text": "i want to fly to berlin",
+        "intent": "flight_booking",
         "entities": [
           {
             "start": 17, 
@@ -65,7 +67,8 @@
         ]
       }, 
       {
-        "text": "i want to fly from london", 
+        "text": "i want to fly from london",
+        "intent": "flight_booking",
         "entities": [
           {
             "start": 19, 

rasa/nlu/test.py

@@ -1306,7 +1306,7 @@ def get_eval_data(
 
     should_eval_entities = is_entity_extractor_present(interpreter)
 
-    for example in tqdm(test_data.training_examples):
+    for example in tqdm(test_data.nlu_examples):
         result = interpreter.parse(example.get(TEXT), only_output_properties=False)
 
         if should_eval_intents:
@@ -1861,7 +1861,7 @@ def compare_nlu(
             _, train_included = train.train_test_split(percentage / 100)
             # only count for the first run and ignore the others
             if run == 0:
-                training_examples_per_run.append(len(train_included.training_examples))
+                training_examples_per_run.append(len(train_included.nlu_examples))
 
             model_output_path = os.path.join(run_path, percent_string)
             train_split_path = os.path.join(model_output_path, "train")

rasa/nlu/utils/bilou_utils.py

@@ -130,7 +130,7 @@ def build_tag_id_dict(
     distinct_tags = set(
         [
             tag_without_prefix(e)
-            for example in training_data.training_examples
+            for example in training_data.nlu_examples
             if example.get(bilou_key)
             for e in example.get(bilou_key)
         ]
@@ -157,7 +157,7 @@ def apply_bilou_schema(training_data: TrainingData) -> None:
     Args:
         training_data: the training data
     """
-    for message in training_data.training_examples:
+    for message in training_data.nlu_examples:
         entities = message.get(ENTITIES)
 
         if not entities:

rasa/shared/core/events.py

@@ -1039,7 +1039,7 @@ class ActionExecuted(Event):
 
     def __init__(
         self,
-        action_name: Text,
+        action_name: Optional[Text] = None,
         policy: Optional[Text] = None,
         confidence: Optional[float] = None,
         timestamp: Optional[float] = None,

rasa/shared/importers/importer.py

@@ -13,7 +13,7 @@
 from rasa.shared.core.training_data.structures import StoryGraph
 from rasa.shared.nlu.training_data.message import Message
 from rasa.shared.nlu.training_data.training_data import TrainingData
-from rasa.shared.nlu.constants import INTENT_NAME, TEXT
+from rasa.shared.nlu.constants import ENTITIES, ACTION_NAME
 from rasa.shared.importers.autoconfig import TrainingType
 from rasa.shared.core.domain import IS_RETRIEVAL_INTENT_KEY
 
@@ -533,18 +533,23 @@ def _unique_events_from_stories(
 
 
 def _messages_from_user_utterance(event: UserUttered) -> Message:
-    return Message(data={TEXT: event.text, INTENT_NAME: event.intent_name})
+    # sub state correctly encodes intent vs text
+    data = event.as_sub_state()
+    # sub state stores entities differently
+    if data.get(ENTITIES) and event.entities:
+        data[ENTITIES] = event.entities
+
+    return Message(data=data)
 
 
 def _messages_from_action(event: ActionExecuted) -> Message:
-    return Message.build_from_action(
-        action_name=event.action_name, action_text=event.action_text or ""
-    )
+    # sub state correctly encodes action_name vs action_text
+    return Message(data=event.as_sub_state())
 
 
 def _additional_training_data_from_default_actions() -> TrainingData:
     additional_messages_from_default_actions = [
-        Message.build_from_action(action_name=action_name)
+        Message(data={ACTION_NAME: action_name})
         for action_name in rasa.shared.core.constants.DEFAULT_ACTION_NAMES
     ]
 

rasa/shared/nlu/constants.py

@@ -4,7 +4,6 @@
 INTENT_RESPONSE_KEY = "intent_response_key"
 ACTION_TEXT = "action_text"
 ACTION_NAME = "action_name"
-INTENT_NAME = "intent_name"
 INTENT_NAME_KEY = "name"
 METADATA = "metadata"
 METADATA_INTENT = "intent"

rasa/shared/nlu/training_data/message.py

@@ -19,7 +19,6 @@
     FEATURE_TYPE_SEQUENCE,
     ACTION_TEXT,
     ACTION_NAME,
-    INTENT_NAME,
 )
 
 if typing.TYPE_CHECKING:
@@ -132,24 +131,6 @@ def build(
             # pytype: enable=unsupported-operands
         return cls(data, **kwargs)
 
-    @classmethod
-    def build_from_action(
-        cls,
-        action_text: Optional[Text] = "",
-        action_name: Optional[Text] = "",
-        **kwargs: Any,
-    ) -> "Message":
-        """
-        Build a `Message` from `ActionExecuted` data.
-        Args:
-            action_text: text of a bot's utterance
-            action_name: name of an action executed
-        Returns:
-            Message
-        """
-        action_data = {ACTION_TEXT: action_text, ACTION_NAME: action_name}
-        return cls(data=action_data, **kwargs)
-
     def get_full_intent(self) -> Text:
         """Get intent as it appears in training data"""
 
@@ -330,9 +311,8 @@ def is_core_message(self) -> bool:
         Returns:
             True, if message is a core message, false otherwise.
         """
-        return (
-            self.data.get(ACTION_NAME) is not None
-            or self.data.get(INTENT_NAME) is not None
+        return bool(
+            self.data.get(ACTION_NAME)
             or self.data.get(ACTION_TEXT)
             or (
                 (self.data.get(INTENT) or self.data.get(RESPONSE))

rasa/shared/nlu/training_data/training_data.py

@@ -21,7 +21,7 @@
     ENTITIES,
     TEXT,
     ACTION_NAME,
-    INTENT_NAME,
+    ACTION_TEXT,
 )
 from rasa.shared.nlu.training_data.message import Message
 from rasa.shared.nlu.training_data import util
@@ -130,21 +130,21 @@ def sanitize_examples(examples: List[Message]) -> List[Message]:
 
         return list(OrderedDict.fromkeys(examples))
 
+    @lazy_property
+    def nlu_examples(self) -> List[Message]:
+        return [ex for ex in self.training_examples if not ex.is_core_message()]
+
     @lazy_property
     def intent_examples(self) -> List[Message]:
-        return [ex for ex in self.training_examples if ex.get(INTENT)]
+        return [ex for ex in self.nlu_examples if ex.get(INTENT)]
 
     @lazy_property
     def response_examples(self) -> List[Message]:
-        return [ex for ex in self.training_examples if ex.get(INTENT_RESPONSE_KEY)]
+        return [ex for ex in self.nlu_examples if ex.get(INTENT_RESPONSE_KEY)]
 
     @lazy_property
     def entity_examples(self) -> List[Message]:
-        return [ex for ex in self.training_examples if ex.get(ENTITIES)]
-
-    @lazy_property
-    def nlu_examples(self) -> List[Message]:
-        return [ex for ex in self.training_examples if not ex.is_core_message()]
+        return [ex for ex in self.nlu_examples if ex.get(ENTITIES)]
 
     @lazy_property
     def intents(self) -> Set[Text]:
@@ -577,38 +577,23 @@ def is_empty(self) -> bool:
         """Checks if any training data was loaded."""
 
         lists_to_check = [
-            self._training_examples_without_empty_e2e_examples(),
+            self.training_examples,
             self.entity_synonyms,
             self.regex_features,
             self.lookup_tables,
         ]
         return not any([len(lst) > 0 for lst in lists_to_check])
 
-    def without_empty_e2e_examples(self) -> "TrainingData":
-        """Removes training data examples from intent labels and action names which
-        were added for end-to-end training.
-
-        Returns:
-            Itself but without training examples which don't have a text or intent.
-        """
-        training_examples = copy.deepcopy(self.training_examples)
-        entity_synonyms = self.entity_synonyms.copy()
-        regex_features = copy.deepcopy(self.regex_features)
-        lookup_tables = copy.deepcopy(self.lookup_tables)
-        responses = copy.deepcopy(self.responses)
-        copied = TrainingData(
-            training_examples, entity_synonyms, regex_features, lookup_tables, responses
-        )
-        copied.training_examples = self._training_examples_without_empty_e2e_examples()
-
-        return copied
+    def can_train_nlu_model(self) -> bool:
+        """Checks if any NLU training data was loaded."""
 
-    def _training_examples_without_empty_e2e_examples(self) -> List[Message]:
-        return [
-            example
-            for example in self.training_examples
-            if not example.get(ACTION_NAME) and not example.get(INTENT_NAME)
+        lists_to_check = [
+            self.nlu_examples,
+            self.entity_synonyms,
+            self.regex_features,
+            self.lookup_tables,
         ]
+        return not any([len(lst) > 0 for lst in lists_to_check])
 
 
 def list_to_str(lst: List[Text], delim: Text = ", ", quote: Text = "'") -> Text:

rasa/train.py

@@ -157,7 +157,7 @@ async def _train_async_internal(
         file_importer.get_stories(), file_importer.get_nlu_data()
     )
 
-    if stories.is_empty() and nlu_data.is_empty():
+    if stories.is_empty() and nlu_data.can_train_nlu_model():
         print_error(
             "No training data given. Please provide stories and NLU data in "
             "order to train a Rasa model using the '--data' argument."
@@ -174,7 +174,7 @@ async def _train_async_internal(
             additional_arguments=nlu_additional_arguments,
         )
 
-    if nlu_data.is_empty():
+    if nlu_data.can_train_nlu_model():
         print_warning("No NLU data present. Just a Rasa Core model will be trained.")
         return await _train_core_with_validated_data(
             file_importer,
@@ -495,7 +495,7 @@ async def _train_nlu_async(
     )
 
     training_data = await file_importer.get_nlu_data()
-    if training_data.is_empty():
+    if training_data.can_train_nlu_model():
         print_error(
             f"Path '{nlu_data}' doesn't contain valid NLU data in it. "
             f"Please verify the data format. "

tests/nlu/extractors/test_regex_entity_extractor.py

@@ -4,7 +4,7 @@
 
 from rasa.shared.nlu.training_data.training_data import TrainingData
 from rasa.shared.nlu.training_data.message import Message
-from rasa.shared.nlu.constants import ENTITIES, TEXT
+from rasa.shared.nlu.constants import ENTITIES, TEXT, INTENT
 from rasa.nlu.extractors.regex_entity_extractor import RegexEntityExtractor
 
 
@@ -86,12 +86,17 @@ def test_process(
     training_data.lookup_tables = lookup
     training_data.training_examples = [
         Message(
-            data={TEXT: "Hi Max!", "entities": [{"entity": "person", "value": "Max"}]}
+            data={
+                TEXT: "Hi Max!",
+                INTENT: "greet",
+                ENTITIES: [{"entity": "person", "value": "Max"}],
+            }
         ),
         Message(
             data={
                 TEXT: "I live in Berlin",
-                "entities": [{"entity": "city", "value": "Berlin"}],
+                INTENT: "inform",
+                ENTITIES: [{"entity": "city", "value": "Berlin"}],
             }
         ),
     ]
@@ -165,12 +170,17 @@ def test_lowercase(
     training_data.lookup_tables = lookup
     training_data.training_examples = [
         Message(
-            data={TEXT: "Hi Max!", "entities": [{"entity": "person", "value": "Max"}]}
+            data={
+                TEXT: "Hi Max!",
+                INTENT: "greet",
+                ENTITIES: [{"entity": "person", "value": "Max"}],
+            }
         ),
         Message(
             data={
                 TEXT: "I live in Berlin",
-                "entities": [{"entity": "city", "value": "Berlin"}],
+                INTENT: "inform",
+                ENTITIES: [{"entity": "city", "value": "Berlin"}],
             }
         ),
     ]
@@ -184,18 +194,23 @@ def test_lowercase(
 
 
 def test_do_not_overwrite_any_entities():
-    message = Message(data={TEXT: "Max lives in Berlin."})
+    message = Message(data={TEXT: "Max lives in Berlin.", INTENT: "infrom"})
     message.set(ENTITIES, [{"entity": "person", "value": "Max", "start": 0, "end": 3}])
 
     training_data = TrainingData()
     training_data.training_examples = [
         Message(
-            data={TEXT: "Hi Max!", "entities": [{"entity": "person", "value": "Max"}]}
+            data={
+                TEXT: "Hi Max!",
+                INTENT: "greet",
+                ENTITIES: [{"entity": "person", "value": "Max"}],
+            }
         ),
         Message(
             data={
                 TEXT: "I live in Berlin",
-                "entities": [{"entity": "city", "value": "Berlin"}],
+                INTENT: "inform",
+                ENTITIES: [{"entity": "city", "value": "Berlin"}],
             }
         ),
     ]