Rasa: Training using multiple training data files

[sankethpb]

Am a beginner with RASA Nlu, need some suggestions regarding below.
I want my bot to handle multiple intents. So, I want to train rasa_nlu with multiple intents. Is it required to create one training json with all intents or should I use separate training json for each instance? If I have to use separate file, I think rasa_nlu creates different models for each file. Is it possible to load multiple models at once per instance of rasa_nlu?

I use backend Spacy_scikitlearn.

Thanks in advance.

[tmbo]

@sankethpb feel free to use the gitter chat for these kinds of questions (https://gitter.im/golastmile/rasa_nlu).

If you want to classify multiple intents, you do not necessarily need multiple models. You can put all your training instances into one json file and use that to train the classifiers. The resulting model will be able to distinguish between the different intents (given enough training 😏 ).

Does that answer your question?

[sankethpb]

Thank you @tmbo , I will use gitter chat next time :)

Yes, it answers my question. I will start it using one file, but I think its going to be tough to maintain one json file as I keep adding new intents and extend existing intents. Will there be some improvements in this regard in future?

[tmbo]

Yeah, sure. So what you are basically looking for is a way to split the training data across multiple json files that are then used to train one model?

[tejuafonja]

@tom how can one do that? split the training data across multiple json
files that are then used to train one model

On Mon, Feb 27, 2017 at 10:45 AM, Tom Bocklisch [e-mail]
wrote:

Yeah, sure. So what you are basically looking for is a way to split the
training data across multiple json files that are then used to train one
model?

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
https://github.com/golastmile/rasa_nlu/issues/180#issuecomment-282673151,
or mute the thread
https://github.com/notifications/unsubscribe-auth/ARkacqguQ3Y5AUeTBrbLKEhbZRpSaDZzks5rgps1gaJpZM4ML9ek
.

[tmbo]

That's not implemented yet, just asking to clarify this is the feature you are looking for.

[sankethpb]

Yes, this is feature I am looking for. Thank you for clarifications.

[ram2408]

Is the enhancement done now? Like can we train the model using multiple files?