Predicted action not following the story

Hello, I have the following two stories. When I train on them individually the action gets predicted, however, when I train using both stories at the same time the action for Story 5 is not predicted correctly.

Story 5

  • utter_greet
  • expressions.greet
  • utter_ask_any_questions**

Story 24

  • utter_greet
  • expressions.greet
  • utter_after_hours
  • action_listen
  • utter_apology_for_unresponsiveness

I am using the following configuration: policies:

  • name: “KerasPolicy”

    epochs: 100

    max_history: 3

    augmentation_factor: 0

    batch_size: 10 validation_split: 0.0 random_seed: 0

  • name: “MemoizationPolicy”

    max_history: 3

  • name: “FallbackPolicy”

    nlu_threshold: 0.7

    core_threshold: 0.5

    fallback_action_name: “action_default_fallback”

These stories are conflicting – you tell your model that with the exact same context, it’s supposed to do something different, which doesn’t make sense. Where is the point where the model is supposed to change its path? It looks like there is some sort of after hours thing here, so you should have a slot that gets set depending on the time, e.g. slot{"after_hours": True} to drive your story one way or another.

Thanks for your help!