Correct me if I’m wrong, Look like Rasa Interactive learning only update the model after we save new data (generated during interactive conversations) and retain the model offline.
Does the model is retrained on the entire data (old + new) or just on new data only?
Does it apply any kind of imitation learning (learning from user corrections) or just purely learning by looking at new labeled examples?
It would be great if there’s detailed documentation explains how Rasa core interactive learning work.
The model is not updated on the fly. After the interactive learning finished, the newly annotated data is written down. You need to re-train your model to improve it with the new data. The data should be merged, so by default the model will be trained on the old + new data. We do not learn anything during the interactive learning itself. Just the re-training improves the model.
Does that clarify your questions? Let me know if you want to know anything else.