HTTP API : Adding a training data to an intent without sending the whole dataset

(SeyoS) #1


Is it possible to add a training data text to an intent, using HTTP API, without re-sending the whole dataset ? Ex : I have an intent with two training text, I want to add one more like this : { “rasa_nlu_data”: { “common_examples”: [ { “text”: “TEXT”, “intent”: “INTENT”, “entities”: [] } ] } } And generate a new model.

Is there a way to do so ?

Thanks for your help.

(David Axelrod) #2

Hi SeyoS!

Doesn’t look like it D: Looking through the Trainer.train method in the python api, it looks like each new training build up from nothing. Since a existing model isnt referenced, the weights from old data are not there so rasa has to “re learn” new weights from the same dataset.

It’s possible to retrain existing models, see sklearn’s docs on partial fits but i’d have to dig a bit deeper into the repo

(SeyoS) #3

I’m not sure to understand. What do you mean by ‘Doesn’t look like it’ ?

The idea is just to add a training data to the existing training data set, generate a new model and load it as soon as it is generated.