I use rasa interactive to setup the stories. However, when try to run again in rasa shell, It cannot predict correctly:
- There is no memorised next action
- Predicted next action using policy_1_FallbackPolicy
My story: it helps insurance agent to get quotation of insurance product.(The form is shortened)
interactive_story_1
- query_available_insurance_type
- action_check_available_insurance_type
- inform{“insurance_type”: “domestic helper”}
- slot{“insurance_type”: “domestic helper”}
- action_check_available_insurance_plan
- inform{“insurance_plan_id”: “a1”}
- slot{“insurance_plan_id”: “a1”}
- utter_provide_info_for_quotation
- action_process_form (form is activated using this action)
- followup{“name”: “form_a1”}
- form{“name”: “form_a1”}
- form: form_a1
- slot{“requested_slot”: “intermediary_code”}
- form: goodbye
- form: form_a1
- slot{“intermediary_code”: “1234567890”}
- form: followup{“name”: “action_listen”}
- form{“name”: null}
- slot{“requested_slot”: null}
- affirm (the prediction fails here)
- utter_wait
- action_auto_quotation_and_order
Since I will have lots of forms, I use an action(action_process_form) to activate the corresponding form instead of putting all the possibilities in stories.
When the form submit, it dispatches a message to ask user if the user wants a quotation. When I input ‘yes’, it correctly regonized it as intent ‘affirm’ but fail to predict the action and resolve to fallback instead. Please advise how to setup the story.
config.yml
language: en
pipeline: supervised_embeddings
policies:
- name: FormPolicy
- name: FallbackPolicy
- name: MemoizationPolicy
- name: MappingPolicy
- name: KerasPolicy