Request

LUIS has an endpoint to batch add Labeled Utterances (https://westus.dev.cognitive.microsoft.com/docs/services/5890b47c39e2bb17b84a55ff/operations/5890b47c39e2bb052c5b9c09). LUIS uses the same data format to allow batch testing via their UI (https://docs.microsoft.com/en-us/azure/cognitive-services/LUIS/luis-how-to-batch-test)

It would be convenient to generate out Train and Test data in the LUIS format. LUIS supports synonyms via Phrase Lists and List Entities, but these are separate endpoints which would require different file outputs and therefore would be far more complicated to support.

Data

LUIS's data format is almost identical to the core of RASA's

RASA schema:

rasa_nlu_data: object
    regex_features: []
    entity_synonyms: []
    common_examples: []
        text: string
        intent: string
        entities: []
            end: int
            entity: string
            start: int
            value: string

LUIS schema:

(base) []
    Text: string
    Intent: string
    Entities: []
        Entity: string
        StartPos: int
        EndPos: int

Current thoughts on implementation

After taking a look at how the RASA adapter is implemented it seems like a pretty simple adjustment to make it output LUIS's format instead. I'm not sure how I would go about hooking that up into the rest of the system however