Debugging Rasa systems

A chatbot is a complex software system. Therefore, we need to design and configure Rasa projects carefully. It is pretty common for developers to get different kinds of bugs when building Rasa-based chatbots. In general, those bugs can be of two types: one is that the prediction results are not as expected; another is that there is a code error in the Rasa system, and the bot cannot run normally. We will cover both types of bugs in the following subsections.

Wrong prediction of results

Two problems may cause the wrong prediction of results. It can be that the Natural Language Understanding (NLU) module makes the wrong prediction on user intent and entities, or it can be that a policy makes the wrong prediction on the next action. It is crucial to first make sure which of these problems is causing the wrong predictions.

Fortunately, most of the commands in Rasa have the debug function. Developers can turn on the debugging option to obtain critical internal...