SnipsNLU on nutrition dataset #16
Description
The rational of this task is to find a okayish recipe for NLU. Currently we employ Kaldi that cannot work with the open dictionary. We need to extend this solution to the cases when the dictionary is open, also when there is a repeating entities in the incoming sentence. The model task for this is a "nutrition task", so the chatbot can record consumed foods to later provide the calories.
Research the quality of Snips NLU https://snips-nlu.readthedocs.io/en/latest/ trained on our nutrition dataset https://huggingface.co/datasets/Spierocho/food_json_extract.
Conclude if the quality is enough to be usable
If so, implement Snips as a decider in BrainBox, with two endpoints train and decide. The nearest decider in terms of implementation is Resemblyzer.