Organizer:

Leland Wilkinson, leland@spss.com

Speakers

8:15 a.m.
Linking Interactive Graphics for Exploring Many Types of Data
Graham Wills, Bell Laboratories (Lucent Technology)
Keywords: interactive graphics linking exploration

One common metaphor for exploring massive data sets is that of mining for information. Panning for a "nugget" of knowledge in a wide expanse of dross is the common image brought up by the phrase "Data Mining", but in fact the most common methods used for examining large data sets do not discard the bulk of the data and focus on a few examples; they process the entire data and extract knowledge from consideration of the data in its entirety. A better metaphor might be taken from cutting diamonds; an uncut stone has great potential, but it needs to be transformed to be valuable; little is discarded and the purpose is to see the majority in the best possible light.

Similarly, an analyst looks at a data set and decides as what form the data is best described (transaction, customer description, scientific, relational..) and must then create views of the data so as to reveal different properties. These views must interact with each other and build so as to reveal as much of the intrinsic value of the data as possible.

In this talk, I will concentrate on this aspect of building different views of a massive data base which interact and complement each other. Showing a scatterplot, histogram, bar chart or parallel axes plot view of some variables comprises one set of views that illuminate the data. Statistical models provide a second set, and other data mining models such as neural net or decision trees provide a third set of views. I will discuss one mechanism that has been successful for facilitating view interaction and show some examples of the process in action.

8:45 a.m.
Concrete Advantages of Reusable Statistical Graphics Components: Visual Programming, Distributed Processing, and Custom Interactive Controls
Dan Rope, Dan Rope Consulting
Keywords: Statistical Graphics, Software Components, JavaBeans

Truly re-usable software components have always been an industry "Holy Grail," and will likely continue to be for some time. Nonetheless, component architectures have evolved enough at this point to provide some very useful and promising tools and to chart a clear course for the future.

The demand for this capability has been greatly enhanced by the web--in that to take full advantage of a software network, it is desirable to provide customized interactive software wrapped around the content. The implications for statistical graphics are immense. With proper componentization, many of the visual tools that statisticians have been using for years can be customized and packaged into special-purpose graph applications that focus on the data at hand. In this context, the graph is the application.

This is one of the major goals of the GPL (Graphics Production Library). We will demonstrate how GPL takes advantage of the JavaBeans architecture to provide software development components for constructing graph applications using any Rapid Application Development (RAD) tool that supports JavaBeans.

9:15 a.m.
Graphboard: A User Interface Component for Building Quantitative Graphics
Leland Wilkinson, SPSS Inc. and Northwestern University
Keywords: components, statistical graphics

A graphboard is a whiteboard for constructing statistical graphics. It is a collection of components that implement user specification controls, a panel for drawing, and widgets for exploration. Statistical graphics programs have been traditionally classified by application: production, reporting, or exploratory tools. A graphboard is based on a different premise: construction and explora- tion of a statistical graphic are part of a common process.

The graphboard environment regularizes several classes of user actions. Data access is provided through scroll lists of variables and typing tools. Graph construction is through drag-and-drop into a panel that contains live target zones. These target zones demarcate important regions of a graph frame - the area devoted to graphic display. Dragging a variable to a margin of a frame, for example, assigns it to the nearest appropriate axis. Dragging a graphic representation element (such as a line or bar) into the center of a frame instantiates a particular chart type.

All actions are reversible. A representation element or variable can be removed by dragging it out of the frame area. Exploratory gestures are recognized and implemented in simple ways. The result is a predictable environment for graphic construction and analysis that does away with dialog boxes, menus, and other sequential or fixed interface methods.