Rasa not following the story order

Rasa Open Source
1

Hi, I am new to RASA, I wanted to create an onboarding flow for users where yes answer Yes or No to proceed further, It was working as Intended until I trained the model again, the actions are not being executed as per the story. So the reply from rasa stops after action: utter_intro2, it does not execute utter_ask2. I even tried to fiddle around with config.yml by changing the policies max_history: 10 and enabled MemoizationPolicy. Here’s my code of stories below.

- story: intro path positive
  steps:
  - intent: intro
  - action: utter_intro
  - action: utter_ask1
  - intent: affirm
  - action: utter_intro2
  - action: utter_ask2
  - intent: affirm
  - action: utter_video
  - action: utter_ask_conditions
  - action: utter_ask_symptoms
  - action: utter_ask_medications
  - action: utter_ask_sync
  - action: utter_unpaid_progress
  - action: utter_unpaid_end

- story: intro path negative 1
  steps:
  - intent: intro
  - action: utter_intro
  - action: utter_ask1
  - intent: deny
  - action: utter_goodbye

- story: intro path negative 2
  steps:
  - intent: intro
  - action: utter_intro
  - action: utter_ask1
  - intent: affirm
  - action: utter_intro2
  - action: utter_ask2
  - intent: deny
  - action: utter_goodbye

Please let me know if you need more info

2

I’ve come across this problem a year ago and increasing max_history and training Core for more epochs fixed it for me but I think there might be a better answer. (mine is based on experience)

3

Hey @prathik-s, welcome to the forum! What you’re describing is very interesting. What do you see as the output around the critical point (utter_intro2) when you try to have this kind of conversation using rasa shell --debug?

Additionally, to enable others to reproduce this weird behaviour, could you share your domain and config files? And what version of Rasa and Python are you using?

4

@prathik-s : Can you please provide the solution to this problem if you got any? I am also facing same issue where my story order is not executed in the sequence which is mentioned in story.yml file. After every training, the behavior is changing.

5

Rules are equivalent to code; Stories are data used to train a machine learning model. With the latter it is not always obvious why you get a particular prediction nor to debug it.

Imagine an image classification model predicted a cat instead of a dog. Rather than edit the images your solution would likely involve adding more data and retraining. Same applies here. Use rasa interactive to correct errors and add new stories until it behaves as you expect.