Interactive and Static Statistical Graphics: Bridge to Integration

Abstract

There are plenty of graphical packages in R which play an important role in building graphics for data analysis, either static graphics (e.g., `graphics`, `grid`, `ggplot2`) or interactive graphics (e.g., `loon`, `shiny`). Each of them has certain strengths and weaknesses. Typically, analysts only use one graphical system at a time during data analysis. However, it may not be sufficient in some circumstances. To better achieve goals, analysts sometimes need more than one graphical packages. For example, an analyst aims to use interactive plots to uncover patterns of interest in data exploration, in which case, a web-based app or an animation could better deliver the analysis dynamically in the presentation. Unfortunately, due to the dissimilarity of the design, data analysis using multiple graphical systems could be too complicated to accomplish. To simplify the process, the idea of ``bridge'' is introduced. A bridge is a peer to peer transformation and works as a connection to map elements (i.e., visual display or visual structure) from one graphical system to another. Usually, the difficulty level of building a bridge mainly depends on how well the abstraction level can be matched.

In this thesis, we mainly focus on four packages. The graphical system `loon` provides interactive visualization toolkit for data exploration. The package `ggplot2` offers tools to extend the flexibility of drawing static plots in data analysis based upon a grammar of graphics. The package `grid` is a core graphical system in R, providing low-level, general purpose graphics functions. The package `shiny` provides interactive web applications in R.

To integrate the strengths of each, three bridges are introduced: bridge `loon.ggplot` is to transform a `loon` widget to a `ggplot` object, or backwards; bridge `loonGrob` is to turn a `loon` widget to a static `grid` graphic; bridge `loon.shiny` is to render a `loon` widget into a `shiny` web app. In addition, a new package `loon.tourr` is also discussed. Even though it is not a bridge, it could be useful to help find interesting lower projections from a high dimensional subspace in an interactive way.