Story not working/resuming after custom action executed using Followup Event (Story stuck before checkpoint )

Rasa Open Source
1

I have a story in which after asking some questions, feedback is being provided to the user. This I have done by using logic in custom action with Followup event.

My goal is after proving feedback to users, I want to suggest some opinions to the user. This I want to implement using the checkpoint.

Here is my custom action logic

class AssessorFeedback(Action):

    def name(self) -> Text:
        return "action_assessor_feedback"

    def run(self, dispatcher: CollectingDispatcher, tracker: Tracker, domain: Dict[Text, Any]) -> List[Dict[Text, Any]]:
        is_asking = tracker.get_slot('state') == 'questions'
        if is_asking:

            dispatcher.utter_message(text="Thank you for answering my questions 😁")
            if self.__user_needs_evaluation(tracker):
                score_resolution = ["This is some issue with your mind"]
            else:

                score_resolution = ["There is defintly something wrong with your mind."]
            print("this is the score resolution")
            print(score_resolution[0])

            dispatcher.utter_message(text=str(score_resolution[0]))
            return [SlotSet('state', 'feedback')]

        else:
            dispatcher.utter_message(text="It is ok. I hope to talk to you some other time!")

            return [SlotSet('state', 'feedback')]

The above feedback was executed using the following custom action :

class ActionAskQuestion(Action):

    def name(self) -> Text:
        return "action_ask_question"

    def run(self, dispatcher: CollectingDispatcher, tracker: Tracker, domain: Dict[Text, Any]) -> List[Dict[Text, Any]]:

        current_question = tracker.get_slot('question_id')
        if current_question == len(QUESTIONS):
            return [FollowupAction("action_assessor_feedback")]

My original story -

 stories:
    - story: new happy path
      steps:
      - intent: affirm
      - slot_was_set:
        - state: intro
      - action: action_start_questions
      - action: action_ask_question
      - intent: answer_question
      - slot_was_set:
          - state: questions
      - action: action_update_score
      - action: action_give_feedback
      - action: action_ask_question
      - slot_was_set:
        - frequency: null
      - slot_was_set:
          - state: feedback
      - checkpoint: ask_feedback


- story: happy path 2 genrated using RASA x
  steps:
  - intent: affirm
  - slot_was_set:
      - state: intro
  - action: action_start_questions
  - action: action_ask_question
  - slot_was_set:
    - question_id: 1
  - intent: answer_question
    entities:
    - frequency: low-medium
  - slot_was_set:
    - frequency: low-medium
  - action: action_update_score
  - slot_was_set:
    - score_questions: 1
  - action: action_give_feedback
  - action: action_ask_question
  - slot_was_set:
    - question_id: 2
  - intent: answer_question
    entities:
    - frequency: medium
  - slot_was_set:
    - frequency: medium
  - action: action_update_score
  - slot_was_set:
    - score_questions: 3
  - action: action_give_feedback
  - action: action_ask_question
  - slot_was_set:
    - frequency: null
  - slot_was_set:
    - state: feedback
  - checkpoint: ask_feedback

###############################################################################3
- story: user said yes to bunty
  steps:
  - checkpoint: ask_feedback
  - action: utter_bunty_request
  - intent: affirm
  - action: action_bunty_link
  - action: action_end_conversation
#################################################################

Now the issue the checkpoint never executed, it conversation only works till –

  - action: action_ask_question
  - slot_was_set:
    - frequency: null
  - slot_was_set:
    - state: feedback

In this part I got the bot feedback but the checkpoint conversation never works. Bot predicts action_default_fallback.

I also try by creating a story without checkpoints like this—

- story: happy path 2 when user accept bunty's conversation
  steps:
  - intent: affirm
  - slot_was_set:
      - state: intro
  - action: action_start_questions
  - action: action_ask_question
  - slot_was_set:
    - question_id: 1
  - intent: answer_question
    entities:
    - frequency: low-medium
  - slot_was_set:
    - frequency: low-medium
  - action: action_update_score
  - slot_was_set:
    - score_questions: 1
  - action: action_give_feedback
  - action: action_ask_question
  - slot_was_set:
    - question_id: 2
  - intent: answer_question
    entities:
    - frequency: medium
  - slot_was_set:
    - frequency: medium
  - action: action_update_score
  - slot_was_set:
    - score_questions: 3
  - action: action_give_feedback
  - action: action_ask_question
  - slot_was_set:
    - frequency: null
  - slot_was_set:
    - state: feedback
  - action: utter_bunty_request
  - intent: affirm
  - action: action_bunty_link
  - action: action_end_conversation

But the same issue, the story stucks after the bot’s feedback. Any help is appreciated. @souvikg10

2

I recommand switching off the core fallback behavior in your pipeline and check wether the stories work that you teached the bot. In order to prove that you should also switch off MemoizationPolicy. In the end your fallback threshold might just be to high. But despite these suggestions: Your action never sets the slot “frequency”, but it is set within the story. That already might cause the problem.

Also post your pipeline, please.

3

Thanks for the reply @harloc . I will try the suggestion and let you know.

Here is the pipeline

language: en
pipeline:
- name: WhitespaceTokenizer
- name: RegexFeaturizer
- name: LexicalSyntacticFeaturizer
- name: CountVectorsFeaturizer
- name: CountVectorsFeaturizer
  analyzer: char_wb
  min_ngram: 1
  max_ngram: 4
- name: DIETClassifier
  epochs: 100
  model_confidence: linear_norm
  constrain_similarities: true
- name: EntitySynonymMapper
- name: FallbackClassifier
  threshold: 0.7
policies:
- name: MemoizationPolicy
- name: TEDPolicy
  max_history: 5
  epochs: 100
  model_confidence: linear_norm
  constrain_similarities: true
- name: RulePolicy
4

I have corrected the slot setting and retrain with the modified pipeline, but still, checkpoints are not working.

image
image1244×404 41.6 KB

image
image545×908 60.5 KB

The bot is not calling checkpoint, i think it just call action_listen() after followup event. Any other idea how to debug. Any input from RASA core members. @amn41 Thanks