(506f) Teaching Domain Experts Data Science: A Progress Report from a Purdue Initiative

Domain knowledge plays an underemphasized but nonetheless critical role in nearly all stages of successful applications of data science and machine learning. From data collection to application implementation—with curation, problem featurization, model design, and validation, among other ingredients, in between—domain knowledge informs countless decisions made by practitioners regarding what to collect, how to collect it, how to model it, and how to evaluate performance. In the context of chemical engineering education, this means that training opportunities need to exist that enable domain experts to learn the language of modern data science, including its limitations and pitfalls. In response to this, two years ago we developed a data science course for chemical engineers with the goal of training data fluent chemical engineers. In this presentation, I’ll cover what we have done, what has worked, and what has failed based on the last three offerings of our course. Among the salient observations is that a sizeable group of chemical engineering undergraduates are already eager for course offerings along these lines and are self-educating as they are able. We also observe that the domain experts in the course are quick to realize how certain modeling assumptions can lead to unphysical consequences. These are very hopeful signs. We also observe a gap between programming experience and data science needs for a large fraction of students that are otherwise enthusiastic about this topic. This is a more difficult gap to bridge in a single course, and motivates some suggestions at the end of the presentation on what training is required for chemical engineers to effectively work on data science problems.