Fallback NLU threshold is not being respected

Rasa Open Source
1

I have this configration for my fallback policy:

  - name: FallbackPolicy
    fallback_action_name: 'utter_default'
    core_threshold: 0.75
    nlu_threshold: 0.5

Pretty standard one, the thing that is happening is when I parse a message with the interpreter, I am getting an intent detected below that threshold: Example:

User input --> hay información sobre internet?
{'entities': [],
 'intent': {'confidence': 0.1366065889596939,
            'name': 'citizensinfo.online_request_neededdoc'},
 'intent_ranking': [{'confidence': 0.1366065889596939,
                     'name': 'citizensinfo.online_request_neededdoc'},
                    {'confidence': 0.10718376189470291,
                     'name': 'greetings.hello'},
                    {'confidence': 0.09952027350664139,
                     'name': 'tourism.acommodation_beds'},
                    ...
],
 'text': 'hay información sobre internet?'}

Then the answer predicted is utter_default, which I guess is triggering from NLU and not Core because there is a story with the intent detected, I wonder how can I get the default answer defined in class RasaNLUHttpInterpreter(NaturalLanguageInterpreter) line 209: default_return = { "intent": {"name": "", "confidence": 0.0}, "entities": [], "text": "", } script /rasa/core/interpreter.py I need this because I am saving conversations in a database and intent name is really relevant to calculate statistics.

Thanks in advance

Best regards

2

Based on what you’ve shown above, your fallback policy should be invoked since the intent ranking of .13 falls below the 0.5 NLU threshold.

Please provide the rasa log with --debug for this intent with the NLU and core messages.