October 5-9, 2014

Abstract

O5.1 Hackable User Interfaces in Astronomy

Christopher Beaumont (Harvard University)

Alyssa Goodman, Harvard University Perry Greenfield, Space Telescope Science Institute

Most tools for data visualization, exploration, and analysis fall into one of two categories: either they are primarily driven by writing code, or by interacting with a graphical user interface. These two modes have complementary strengths, but neither is ideal for the entire data analysis lifecycle -- GUIs are better suited for highly exploratory tasks, while code-driven workflows are more expressive, precise, and scalable. Very few tools handle both graphical and code-based workflows equally well, nor do they facilitate easily switching between different tools. This imposes a burden on the researcher, who pays a cognitive penalty associated with juggling data and ideas between several tools. This "interface between interfaces" deserves more attention. How might we build hybrid tools that better incorporate the best features of GUI-based and code-based analysis? How can we make switching between different workflows more fluid? What features should such a "Hackable User Interface" possess? In this talk I'll present Glue (glueviz.org), a data exploration tool that attempts to implement such a hybrid interface. Glue's primary design goal is to make it easy to inter-compare multidimensional datasets spread across several files, by building linked-view visualizations of those files. It is written in Python, and aims to make it easy for researchers to integrate Python-based analysis code into the GUI -- without getting bogged down in writing user interface code. I'll discuss Glue's philosophy and design, demonstrate the kinds of problems it is best able to solve, and discuss the lessons we've learned so far about building Hackable User Interfaces for data exploration.

Mode of presentation: Invited Speaker (need to be confirmed by the SOC)

Applicable ADASS XXIV theme category: Visualisation for astronomy