Rasa nlu is not recogonizing the intent correctly

Hi All,

I want to learn rasa nlu I have created the project with only nlu.md file and config.yml file and train the nlu model with 2 intents without any stories.

I am using the pretrained_embeddings_spacy pipeline which is already trained but when I query the shell nlu with a different message that doesn’t belong to any intent then always trying to use the intent defined in nlu.md only.

if i give the car as input then also it using greet as intent.if pretrained_embeddings_spacy is already trained library why it is not able to detect the correct intent from the message. Please correct me if my understanding is not correct.


1- config.yml


2- nlu.md

Hi @piyush29programmer the “pretrained embeddings” part of the name means that it’s using pretrained word vectors, which means it already has a predefined representation of each word. What this does not mean is that it’s pretrained to classify intents. The intent classifier part still gets trained on only your training data. Therefore, it will take a representation of the sentence you enter and pick the most likely intent. In the case of “car”, the model is not very confident about it’s classification and so you could use core’s FallbackPolicy for example, to respond something like “Sorry, I didn’t understand you” to the user. Alternatively, you can create an out_of_scope intent with things you want the bot to understand, but which isn’t part of your bot