shiny and rpanel - a quick comparison

Shiny is a package from RStudio that lets you produce interactive web pages. You build a page with some control widgets and a handler that does something dependent on the value of those widgets. You can build your interface programmatically or create a boilerplate html page that gets filled in by control and output widgets.

A conceptually similar pattern is implemented by the rpanel package, but this uses the tcltk toolkit. A panel is created, control widgets added, and callbacks on the controls can run R code to, for example, update a plot.

qq plot example

Here's the rpanel version:

require(rpanel)
# box-cox transform
bc.fn <- function(y, lambda) {
    if (abs(lambda) < 0.001) 
        z <- log(y) else z <- (y^lambda - 1)/lambda
}
# qqplot of transformed data
qq.draw <- function(panel) {
    z <- bc.fn(panel$y, panel$lambda)
    qqnorm(z, main = paste("lambda =", round(panel$lambda, 2)))
    panel
}
# create a new panel with some initial data
panel <- rp.control(y = exp(rnorm(50)), lambda = 1)
# add a slider that calls qq.draw on change
rp.slider(panel, lambda, -2, 2, qq.draw)

Run these functions and you should see a slider and a graphics window. Move the slider to modify the plot.
Note that this might not work too well under RStudio because of the way the embedded RStudio graphics device captures output.

And here is the shiny version, which comes in two files living in their own folder. First qqplot/ui.R:

library(shiny)
# this defines our page layout
shinyUI(pageWithSidebar(
  headerPanel("qqplot example"),
  sidebarPanel(
  # a slider called 'lambda':
    sliderInput("lambda", 
                "Lambda value", 
                min = -2, 
                max = 2,
                step=0.01,
                value = 0)
  ),
  mainPanel(
    # the main panel is the plotted output from qqplot:
    plotOutput("qqPlot")
  )
))

and qqplot/server.R:

library(shiny)
shinyServer(function(input, output) {
    # b-c transform
    bc.fn <- function(y, lambda) {
        if (abs(lambda) < 0.001) 
            z <- log(y) else z <- (y^lambda - 1)/lambda
    }
    # initial data
    y = exp(rnorm(50))
    # here's the qqplot method:
    output$qqPlot <- reactivePlot(function() {
        z <- bc.fn(y, input$lambda)
        qqnorm(z, main = paste("lambda =", round(input$lambda, 2)))
    })
})

With that done, launch the app with:

runApp("qqplot")

that should open up the page in your web browser. Hit break, stop, or control-C to quit.

Notes

The rpanel plot updates as you drag the slider, whereas shiny updates only when you let go of the slider.

I find that when I hit Control-C and break a running shiny app, then my tcltk windows go all unresponsive until I quit R and start again. Threading issues? This is on Linux. I've always had problems with tcltk widgets going unresponsive on me, or ending up unkillable.

The shiny UI looks, well, “shiny”, but the rpanel interface looks a bit old and not very exciting (if you can get excited by user interfaces).

Using the tkrplot package, you can build integrated rpanel packages with controls and plots in the same window. Without it, you are stuck with separate graphics and control windows.

Which should I use?

How do I know?! Shiny looks better, but I do like the update on drag of rpanel - it gives you much better feedback as you control the plot. Maybe this can be done in shiny with some additional work.

I don't really like the two-file method of shiny. Looking at the code I see the files just get sourced in, so conceivably it could be possible to run shiny apps just by specifying the shinyServer and shinyUI functions - but shiny monitors the server.R and ui.R file for changes and updates the application, which is quite nice.

So there's the basic existential dilemma. Choice. I can even throw some more things into the mix if you want - there' RServe, or RApache with gWidgetsWWW and probably many many more. I'm sure we can all agree that the days of needing Java and Apache Tomcat to deploy R applications to the web are now over (http://sysbio.mrc-bsu.cam.ac.uk/Rwui/tutorial/quick_tour.html).

I might try and implement some more of the rpanel examples in shiny shortly. Or why don't you have a go, and publish your works here?