Connecting R with D3 for dynamic graphics, to explore multivariate data with tours

The tourr package in R has several algorithms and displays for showing multivariate data as a sequence of low-dimensional projections. It can display as a movie but has no capacity for interaction, such as stop/go, change tour type, drop/add variables. The tourrGui package provides these sorts of controls, but the interface is programmed with the dated RGtk2 package. This work explores using custom messages to pass data from R to D3 for viewing, using the Shiny framework. This is an approach that can be generally used for creating all sorts of interactive graphics.

Introduction

Did you know you can run any javascript you like in a Shiny application and you can pass whatever you want including JSON back and forth? This massively widens the scope of what you can do with Shiny, and generating a tour of multivariate data with this approach is a really good example of what is possible.

The tour algorithm (Asimov 1985) is a way of systematically generating and displaying projections of high-dimensional spaces in order for the viewer to examine the multivariate distribution of data. It can do this either randomly, or by picking projections judged interesting according to some criterion or index function. The tourr package (Wickham et al. 2011) provides the computing and display in R to make several types of tours: grand, guided, little and local. The projection dimension can be chosen between one and the number of variables in the data. The display, though, has no capacity for interaction. The viewer can watch the tour like a movie, but not pause it and restart, or change tour type, or number of variables.

These interactive controls were provided with the tourrGui package (Huang et al. 2012), with was programmed with the RGtk2 package (Lawrence and Temple Lang 2010). This is not the toolkit of choice today, and has been superceded with primarily web-capable tools, like Shiny (Chang et al. 2017). To display dynamic graphics though, is not straight-forward. This paper explains how to use D3 (Bostock et al. 2011) as the display engine in a Shiny graphical user interface (GUI), using custom message passing between server and client.

Creating a tour, with the tourr package

The tourr package (Wickham et al. 2011) is an R implementation of the tour algorithms discussed in (Cook et al. 2007). It includes methods for geodesic interpolation and basis generation, as well as an implementation of the simulated annealing algorithm to optimise projection pursuit indices for the guided tour. The tour can be displayed directly in the R graphics device, for example, the code below generates a 1D density tour. Figure 1 shows snapshots.

library(tourr)
# quartz() # to display on a Mac; X11() # For windows; The Rstudio graphics
# device is not advised
animate_dist(flea[, 1:6], center = TRUE)