Upper limit / max count for paraphrased questions under retrieval intents?

what will be the maximum count or the upper limit to paraphrased questions under the retrieval intents so that it perform best and also it shouldn’t got overfit… @here

@baval Hi! Do you mean a set of variations of questions with respect to answer in retrieval_intent?

1 Like


@baval There is I guess for training example for intent is min 2 or 3 I guess, but for particular intent as much as training examples you can provide it will be great for the bot to learn and predict the response. It’s a basic thumb rule in deep learning and NLU. I hope you got my point on this.

@baval see the Rasa GitHub link rasa-demo/faq.yml at main · RasaHQ/rasa-demo · GitHub they have provided so many sets of variations of questions a user can ask for a particular question. So, nothing to worry about. Keep in mind every user doesn’t have proper grammar knowledge.

I hope it will solve your doubt :slight_smile:

1 Like

@nik202 i appreciate your response but my concern was to know if there is any upper limit what is that, because at some point model will start to overfit.

@baval As, I told you there is no specific upper limit for the training example, till it will be relevant to the intent you creating for the particular question. If still you have doubt please visit this rasa github link for your reference: How many training examples should I feed to prepare good model · Issue #780 · RasaHQ/rasa · GitHub

@baval Do check this blog for 10 best practice to design training data 10 Best Practices for Designing NLU Training Data

I guess now it will solve your doubt :slight_smile:

1 Like

alright thank you man, i will see…