When I test a bot using response-selector in rasa shell using the --debug option, I can see the main intent (chitchat or faq) in my case being identified with the degree of certainty but I have no information on the identification (name, degree of certainty, …) of sub-intents. Is there a way to see this information?
Hi @marcco411 what version of Rasa are you using?
This applies for Rasa 2.x:
You can implicitly see the name in the response that is logged e.g.
2021-02-22 13:49:16 DEBUG rasa.core.processor - Action 'utter_chitchat' ended with events '[BotUttered('Blue!', {"elements": null, "quick_replies": null, "buttons": null, "attachment": null, "image": null, "custom": null}, {"template_name": "utter_chitchat/colour"}, 1613998156.13262)]'.
You can also see the details of the response selector predictions when running rasa shell nlu
e.g.
"response_selector": {
"all_retrieval_intents": [
"chitchat"
],
"default": {
"response": {
"id": -3174625603948499783,
"response_templates": [
{
"text": "Blue!"
}
],
"confidence": 0.8385758996009827,
"intent_response_key": "chitchat/colour",
"template_name": "utter_chitchat/colour"
},
"ranking": [
{
"id": -3174625603948499783,
"confidence": 0.8385758996009827,
"intent_response_key": "chitchat/colour"
},
{
"id": 4474014727561821834,
"confidence": -0.9020220637321472,
"intent_response_key": "chitchat/language"
}
]
}
}
This information would also be available inside a custom action, in tracker.latest_message
Hi, so the response_selector only work with the retrieval intent (chitchat, FAQs) isn’t it? What if I don’t use the chitchat, FAQs but still want it to respond to by using the response utter_, what should I do. I don’t want to use chitchat and FAQs because my intent numbers is about 500 and it will decrease the accuracy of the retrieval intent if I put too much intents in one retrieval intent (faq). If there’s no way, should I create more intents to solve this? Thank you so much for any help!
Hi, I have tried to divide my big faq retrieval_intent into small retrieval intents (about 10) and start training them. Each ResponseSelector’s epochs is just 10. But it keep stopping before the training finish. This is a picture of the problem:
Can someone help me with this? Thank you!!I got this one solved by decreasing the epochs of each Response Selector (from 10 to 5).