This repository is an example of a simple Rasa project that shows how you can implement your own classifiers. The data was automatically generated by running:
rasa init
sklearn_classifier.py exemplifies implementing a simple SVM classifier.
transformer_classifier.py can take any Hugging Face transformer that has an AutoModelForSequenceClassification implementation and train it. We're using albert-base-v2 as an example because it is a small model trainable without the need for a GPU.
config.yml contains examples of how you would insert the two classifiers into the Rasa pipeline, and how you would pass parameters to them.
To train the custom rasa intent classification model:
rasa train nlu
To test the model:
rasa test nlu --nlu <path_to_test_data>
To interact with the model:
rasa shell