Fallback is triggered not correctly?

Rasa Open Source
1

Maybe it is due to a underlying mechansim which I did not see. I use core 10.4 with tensorflow emebeding (13.3). I customize the intent for a specific condition if it has intent None and is just one word in the parse function of model.py of NLU. But it seems that the fallback policy doesn not notice anything about this change? Although in Debug mode it says that it received the message with intent as I wanted and the tracker is also updated with this new intent. Why is therefore the fallback policy triggered?

When and where exatcly is the fallback triggered with intent None?

2

Can you share the debug logs ?

3

Here and I studied the policy ensemble code,it seems that fallback is predicted not due to to tresholds but but as normal? Although I have this story few times in my stroy file…

DEBUG:rasa_core.processor:Received user message 'bl a scdf' with intent '{'name': 'Cost', 'confidence': 0.99}' and entities '[]'
parsedata
{'intent': {'name': 'Cost', 'confidence': 0.99}, 'entities': [], 'intent_ranking': [], 'text': 'bl a scdf'}
DEBUG:rasa_core.processor:Logged UserUtterance - tracker now has 24 events
DEBUG:rasa_core.processor:Current slot values:
 some slots...
DEBUG:rasa_core.policies.memoization:Current tracker state [{'prev_action': 1.0, 'intent_Cost': 1.0, 'entity_Entity': 1.0}, {'prev_action_listen': 1.0}]
DEBUG:rasa_core.policies.memoization:There is no memorised next action
DEBUG:rasa_core.policies.ensemble:Predicted next action using policy_2_FallbackPolicy
DEBUG:rasa_core.policies.ensemble:Predicted next action 'action_default_fallback' with prob 1.00.
4

0.10.4 might be using a different version of the fallback policy. Try upgrading to 0.11

Though, i’m not sure i understand your issue exactly. from your debug log there it might be that Core predicted the next action with a low confidence

5

What kind of? I hesitate to update since it is completly new with the custom action as server. And my projetc is right now working good with 10.4.

Are there some major changes in standard policy made?

6

I’d recommend upgrading anyways, given we always add new features. I can’t remember now what kind of changes were made anymore, you can check the changelog to see if anything was changed.