This article (12 page PDF) identifies the challenges and pain points of running things like data analytics in things like Jupyter Notebooks. It's good to see this sort of article, because while some people (including myself) have touted this sort of approach as a way to generate dynamic and interactive learning resources, the actuality is that they are far from usable for the average person. In fact, after reading this article, we see that they are far from usable for the average data scientist. But the research quite clearly identifies where these problems are, and will serve as a valuable guide not only for notebook designers but for learning tool developers generally. For a short summary of the paper, see this blog post.