This is a companion to the NLU concept guide, which is a detailed outline of the NLU module. Here we’ll talk about usage issues specific to the Python client library.
As mentioned in the concept guide, every NLU model will have three files:
The path to the directory containing these files is passed as the
model_dir argument on initialization of the NLU.
As mentioned in the Getting Started guide, initializing the Spokestack NLU is just like other Spokestack components.
from spokestack.nlu.tflite import TFLiteNLU nlu = TFLiteNLU("model_dir")
NLU results are returned in a wrapper that allows for easy access of the properties. The
Result class contains:
utterance: unmodified transcript received by the NLU
intent: classified user intent
confidence: model confidence in intent classification
slots: slots tagged by the model