Hey there !
I didn’t found something clear about featurization of NLU as you did for CORE. I would like to learn more about how you extract feature from the training file (list of sentence for each intent).And how you chose it ?
I try to understand what is wrong with my interpreter to rearrange my training file. I understand that rasa NLU uses the Word2Vect from Spacy which take the sum of each word vector. But you can also use the technique of Bag of words or N-grams.
Next you use a SVM for the classification, why ?