The tools we use to investigate data affect the way that we do science -- we are constantly (if subtly) drawn towards questions that are easily answered by our analytical and software tools, and comparatively discouraged from research directions that are less-well matched to these tools. A crucial skill that most scientists learn early in their careers is how to identify the most fruitful scientific questions, given the current state of analysis techniques. However, some of us are drawn to to the problem from the other side -- how might we build new tools to answer different classes of questions, and how would this shape scientific research?
This talk will explore that question, by sketching some general limitations to our current tools. In particular, we currently live with a strong dichotomy between code-driven and interactive workflows. While GUI-based tools offer more fluidity for common tasks, code-driven workflows offer greater expressiveness and reproducibility. These are complementary strengths, but switching between different workflows is extremely cumbersome. Inspired by this, we have been building Glue (http://glueviz.org) as a "hackable user interface" to bridge this gap. Glue lets users easily build interactive, linked-view visualizations to drill down into high-dimensional datasets. At the same time, it is designed to be easily and deeply scriptable, so that users can customize Glue for their particular data needs. I'll demonstrate these features and discuss Glue's design philosophy, to suggest how future tools might better empower researchers.
Chris Beaumont is a Senior Software Engineer at the Harvard-Smithsonian Center for Astrophysics, where he build tools to visualize and analyze multidimensional datasets. His main focus is currently Glue, which is being supported by NASA and the Space Telescope Science Institute to analyze future data from the James Webb Space Telescope. Prior to this, Chris was a graduate student at the University of Hawaii's Institute for Astronomy, where he received a PhD in 2013.