Rasa-core: 'action_default_fallback' triggered for valid input

Rasa Open Source
1

I have created bot using rasa-core working fine for below story.

User: Hi
BOT: Hi, How are you?
User: fine
BOT: How can I help you?
User: what is the weather today?
BOT: In which location do you want to know?
User: London
Bot: It's cloudy with 22 degrees during day time.
User: Thanks
Bot: Thank you

I have an issue if I am trying to give valid input after invalid input triggers ‘action_default_action’.

User: Hi
BOT: Hi, How are you?
User: fine
BOT: How can I help you?
User: what is the weather today?
BOT: In which location do you want to know?
User: Thanks
Bot: Sorry, I can't understand (default fallback action given in policy)
User: London
Bot: Sorry, I can't understand (default fallback action given in policy)

How can I make a bot to understand that input is valid for previous question?

2

hi @madhanaravind - welcome to the forum.

Can you please run the bot with rasa shell --debug ? that should show why the fallback is triggered

3

Hi @amn41, I got below response

2020-01-22 19:32:47 DEBUG    rasa_core.policies.memoization  - Current tracker state [{'intent_thanks': 1.0, 'prev_action_listen': 1.0}, {'prev_utter_location_name': 1.0, 'intent_thanks': 1.0}, {'intent_greet': 1.0, 'prev_action_listen': 1.0}]
2020-01-22 19:32:47 DEBUG    rasa_core.policies.memoization  - There is no memorised next action
2020-01-22 19:32:47 DEBUG    rasa_core.policies.form_policy  - There is no active form
2020-01-22 19:32:47 DEBUG    rasa_core.policies.ensemble  - Predicted next action using policy_2_FallbackPolicy
2020-01-22 19:32:47 DEBUG    rasa_core.processor  - Predicted next action 'utter_default' with prob 1.00.
2020-01-22 19:32:47 DEBUG    rasa_core.processor  - Action 'utter_default' ended with events '[]'
2020-01-22 19:32:47 DEBUG    rasa_core.processor  - Bot utterance 'BotUttered(text: I am not sure what you're aiming for, data: {
  "elements": null,
  "buttons": null,
  "attachment": null
})'

Please let me know the feedback.Thanks

4

from the debug you can see that

There is no memorised next action

this means that the conversation you are testing is not present in your training data (up to max_history)

I would suggest adding the augmemo policy or embedding policy to your config.yml so that Rasa can figure out what to do

5

@amn41 Thanks. Working fine now.