|
Abstract:
|
Multipanel display systems like Trellis Display have proven to be very effective tools for visualizing complex data sets in detail. In Trellis Displays, data are broken into subsets and a visualization method is applied to each subset, resulting in a set of panels that are displayed in a grid, resembling a garden trellis. When the size of the data becomes very large, it is often the case that there are too many panels for the analyst to consume at one time. A simple idea put forth by John Tukey is to compute "cognostics", metrics that help bring different, interesting sets of panels in a display to the analyst's attention and allow the analyst to interactively sort and filter the panels. Cognostics can include statistical summaries, descriptive variables, goodness-of-fit metrics, etc. In this talk, I will discuss ideas regarding default cognostic choices based on what is being plotted and will discuss an implementation of a large-scale multipanel display system with cognostics called Trelliscope.
|
