The GGobi data pipeline

• Main Author: Michael Lawrence
• Other Authors: The Pennsylvania State University CiteSeerX Archives
• Format: Text
• Language: English
• Subjects: Orca
• Online Access: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.513.6927; http://www.ggobi.org/docs/pipeline-design.pdf

A data pipeline transforms data, through a series of stages, into visualizations. Buja et al. [1] introduce the use of the pipeline design pattern for statistical data visualization. Figure 1 gives an overview of the design. They argue that an interactive and dynamic visualization requires real-time data processing. Their suggested pipeline is constituted by seven stages. The pipeline begins with the raw data. The second stage performs non-linear transformations on the data, if requested. The next stage standardizes the variables, and the following stage randomizes variables, so that they may serve as a graphical permutation test. After randomization comes the projection engine, which reduces the dimensionality of the data, either by selecting variables for the axes or projecting multiple variables onto a lower dimensional space, as in tours. The viewporting stage is next; it decides the visible range and scale of the data, which may be specified by a user through pan, zoom and scale controls. The final stage is the actual plot. Thus, the pipeline incorporates many of the common needs of dynamic statistical graphics: transformation, standardization, permutation, projection, and viewporting. The Orca project [2] extends the Buja et al. pipeline to support multiple linked views. The linked views adhere to a model view controller (MVC) pattern, in that user manipulation, such as brushing, changes values in the underlying OrcaAppearance model, and these changes are then propagated to any