What is the difference between the DIET Classifier and the TED policy? When I run “rasa test” I get a TEDPolicy_confusion_matrix.png and a DIETClassifier_confusion_matrix.png. The DIET Classifier matric looks more correct and reports the correct number (16) nlu examples I have but the TED Policy shows only 1s and 2s.How do I correct the TED policy confusion because adding nlu examples don’t seem to register. Am I missing something?
You should review the Rasa architecture and docs on pipeline components vs policies. Pipeline componets are part of the NLU which predicts the intent and entities. Policies are use to predict the next step in the dialogue.
You can read about the DIETClassifier here. It’s an NLU component which predicts intents and entities in a user utterance.
You can read about the TEDPolicy here. It predicts the next step in the dialogue.
Thank you for your response @stephens. I have read those but they didn’t help me fix the confusion matrix. Thank you for your clarification on the difference. If the TEDPolicy is just supposed to predict the next step in dialogue then why is rasa test using it to test entity extraction and giving me a TED Policy entity confusion matrix? Is it testing entities that influence conversation? Does creating more stories help the confusion because creating more nlu examples isn’t. Are you aware of the issue I linked in my reply?