Fallback error

Rasa Open Source
1

Community, I have a problem with my bot, when I send a message the fallback returns and the “confidence” has 0.9919366836547852. This is my config

language: en
pipeline: supervised_embeddings

policies:
  - name: MemoizationPolicy
  - name: KerasPolicy
  - name: FallbackPolicy
    nlu_threshold: 0.3
    core_threshold: 0.3
    fallback_action_name: "action_default_fallback"
2

Hi @Cesarcuna! Does this happen for all messages sent to your assistant?

3

Hi @tyd!, not only some.

4

Can you please share which messages are as well as your nlu.md?

5

This my nlu.md nlu.md (51.8 KB) And my stories.md stories.md (21.6 KB)

6

At the moment he doesn’t answer me correctly, he answers them all with the default.

image
image1524×657 53.6 KB

7 
2020-02-18 23:08:00 DEBUG    rasa.core.tracker_store  - Recreating tracker from sender id 'POSTMAN1'
2020-02-18 23:08:00 DEBUG    rasa.core.processor  - Received user message 'Where can I find the difference between a responsible party and an owner?' with intent '{'name': 'owner_and_responsible_party', 'confidence': 0.9986124038696289}' and entities '[]'
2020-02-18 23:08:00 DEBUG    rasa.core.processor  - Logged UserUtterance - tracker now has 14 events
2020-02-18 23:08:00 DEBUG    rasa.core.policies.memoization  - Current tracker state [{'intent_time_of_an_inspection': 1.0, 'prev_action_listen': 1.0}, {'intent_time_of_an_inspection': 1.0, 'prev_utter_inspection_time_response': 1.0}, {'intent_time_of_an_inspection': 1.0, 'prev_action_listen': 1.0}, {'intent_time_of_an_inspection': 1.0, 'prev_utter_inspection_time_response': 1.0}, {'prev_action_listen': 1.0, 'intent_owner_and_responsible_party': 1.0}]
2020-02-18 23:08:00 DEBUG    rasa.core.policies.memoization  - There is no memorised next action
2020-02-18 23:08:00 DEBUG    rasa.core.policies.fallback  - NLU confidence threshold met, confidence of fallback action set to core threshold (0.3).
2020-02-18 23:08:00 DEBUG    rasa.core.policies.ensemble  - Predicted next action using policy_2_FallbackPolicy
2020-02-18 23:08:00 DEBUG    rasa.core.processor  - Predicted next action 'action_default_fallback' with confidence 0.30.
2020-02-18 23:08:00 DEBUG    rasa.core.processor  - Action 'action_default_fallback' ended with events '['BotUttered(text: Sorry, I do not have an answer for this question for now. I will report this question to my team so they can teach me how to answer it properly in the future. You can try  the Most Asked Questions feature to see what others are asking., data: {"elements": null, "quick_replies": null, "buttons": null, "attachment": null, "image": null, "custom": null}, metadata: {})', 'UserUtteranceReverted()']'
2020-02-18 23:08:00 DEBUG    rasa.core.policies.memoization  - Current tracker state [{'intent_time_of_an_inspection': 1.0, 'prev_utter_inspection_time_response': 1.0}, {'intent_time_of_an_inspection': 1.0, 'prev_action_listen': 1.0}, {'intent_time_of_an_inspection': 1.0, 'prev_utter_inspection_time_response': 1.0}, {'intent_time_of_an_inspection': 1.0, 'prev_action_listen': 1.0}, {'intent_time_of_an_inspection': 1.0, 'prev_utter_inspection_time_response': 1.0}]
2020-02-18 23:08:00 DEBUG    rasa.core.policies.memoization  - There is no memorised next action
2020-02-18 23:08:00 DEBUG    rasa.core.policies.fallback  - NLU confidence threshold met, confidence of fallback action set to core threshold (0.3).
2020-02-18 23:08:00 DEBUG    rasa.core.policies.ensemble  - Predicted next action using policy_1_KerasPolicy
2020-02-18 23:08:00 DEBUG    rasa.core.processor  - Predicted next action 'action_listen' with confidence 1.00.
2020-02-18 23:08:00 DEBUG    rasa.core.processor  - Action 'action_listen' ended with events '[]'
2020-02-18 23:08:00 DEBUG    rasa.core.tracker_store  - Tracker with sender_id 'POSTMAN1' stored to database
2020-02-18 23:08:00 DEBUG    rasa.core.lock_store  - Deleted lock for conversation 'POSTMAN1'.
8

@Cesarcuna The fallback action triggers if your NLU confidence is below the threshold in your config (0.3) or if none of the Core dialogue policies predict an action with confidence higher than the threshold in your config (0.3). In the examples, you gave it looks like your other dialogue policies are not predicting a next action with a confidence higher than your core_threshold, causing the fallback policy to execute.

A quick look at your stories shows me that each one of this is only includes one conversation turn. If you want users to be able to have conversations greater than one turn, I would add stories that represent those conversations. I recommend using Rasa X to collect real conversations, so you learn to handle the ways people actually talk to your assistant and include those stories. I also might checkout retrieval actions if a lot of those are FAQs

9

You will have an example chit-chat bot to do mine please, I have several types of response in my bot. And taking advantage of the help, you could give me an example bot to request personal data of a person with my bot.