Supporting Independent Learning and Rapid Experimentation in Data Science

Abstract

Data scientists iteratively discover insights from data or test pre-existing insights with data. This is called sensemaking and insights derived from sensemaking facilitate decisions. Given the potential consequences of making decisions this way, the insights derived from data are often validated in collaboration with domain experts and other data scientists. Data scientists use literate programming environments like computational notebooks to enable collaborative validation. Notebooks can capture a relatively complete picture of the data science workflow, encapsulating code, intermediate results, documentation, and visualizations. Notebook authors can use these components to form a computational narrative---a storytelling device to effectively communicate the sensemaking results and process. The literature nature of these documents also allow them to be excellent mediums for teaching data science practices. However, previous literature has found notebook authors have difficulty translating their sensemaking iterations to singular narratives in a linear document like Jupyter Notebook. Notebook tutorials suffer from the same limitations and fail to convey the best practices in sensemaking. To overcome these issues, we develop a meta-analysis approach to automatically track sensemaking and sensemaking iterations in computational notebooks. We can use similar meta-analysis approaches to automatically communicate sensemaking in extensible, visual, and interactive environments in the absence of traditional computational narratives. Finally, we demonstrate a specific interactive environment that can aid independent learning and rapid experimentation.