Answer seems to be choosen randomly when using entities

Hi all, I’m getting very strange answers with my Bot, it seems like stories are not well taken into account.

Here is my config

pipeline: supervised_embeddings language: fr policies:

  • name: MemoizationPolicy max_history: 1
  • name: KerasPolicy
  • name: MappingPolicy
  • name: FallbackPolicy nlu_threshold: 0.3 core_threshold: 0.1 fallback_action_name: action_default_fallback

In the stories I have for example :

rervation voiture

  • reservation{“reservable”:“voiture”}
    • utter_deplacements


  • reservation{“reservable”:“navette”}
    • utter_reservation_navette

And for example with interactive learning here is the result :

New Story

  • reservation{“reservable”:“voiture”}
    • utter_reservation_salle

Which should trigger utter_deplacements and not utter_reservation_salle as the training stories specify it. And every time I have training data with a different answer depending on the entity the answer is never the good one (not the one specified in the training data) and seems to be choosen randomly (even with twice the same example). And when I manage to make it choose the right answer for one entity, it does not work anymore with others.

What should I do ? Is there a configuration misunderstanding on my side ? Thanks a lot for any feedback

Is reservable a categorical slot? If not, rasa only cares if it has a value or not.

1 Like

Oh… That’s right I totally forgot to use slots… anyway thanks a lot for your help !