The Jupyter notebooks are reall really nice stuff. Currently there is a huge lack of this kind of base analytics for NLU in the Open Source world.

I tried to do something similiar some weeks ago, but implemented it with Node.js instead of Python:

https://medium.com/@floriantreml/tutorial-benchmark-your-chatbot-on-watson-dialogflow-wit-ai-and-more-92885b4fbd48