mwshiny: Creating a Multi-Window Shiny App

Load Libraries and Datasets

As in any typical Shiny app, you start by loading any necessary libraries and data that you want to be globally accessed. I’ll start by doing that with visualization libraries and the iris dataset.

# load libraries
# note that attaching mwshiny also attaches Shiny
library(mwshiny) # our multiwindow app
library(ggplot2) # cool visualizations
library(datasets) # contains the iris dataset

Let’s load our iris data and get an idea of what it looks like.

data(iris) # load iris data
summary(iris) # just to get a look at what we're dealing with here

##   Sepal.Length    Sepal.Width     Petal.Length    Petal.Width   
##  Min.   :4.300   Min.   :2.000   Min.   :1.000   Min.   :0.100  
##  1st Qu.:5.100   1st Qu.:2.800   1st Qu.:1.600   1st Qu.:0.300  
##  Median :5.800   Median :3.000   Median :4.350   Median :1.300  
##  Mean   :5.843   Mean   :3.057   Mean   :3.758   Mean   :1.199  
##  3rd Qu.:6.400   3rd Qu.:3.300   3rd Qu.:5.100   3rd Qu.:1.800  
##  Max.   :7.900   Max.   :4.400   Max.   :6.900   Max.   :2.500  
##        Species  
##  setosa    :50  
##  versicolor:50  
##  virginica :50  
##                 
##                 
## 

Set Up UI for Each Window

Now that I’ve loaded my data and packages, I need to set up what my windows are going to look like. For this mwshiny app, I’m going to have one window be my “controller”, where I change all my inputs, and the other one be a window where my output is going to be. My output is going to be scatter plots of the data, colored by species.

# named list of ui pages that are the contain the title and content of each of my windows
ui_win <- list()

# first we add what we want to see in the controller to the list
ui_win[["Controller"]] <- fluidPage(
  titlePanel("Iris Dataset Explorer: Controller"),
  sidebarLayout(
    sidebarPanel(
      # choose what goes on the x axis
      selectInput(inputId = "x_axis",
                  label = "What would you like to see on the x axis?",
                  choices = colnames(iris)[colnames(iris)!="Species"]),
      # choose what goes on the y axis
      selectInput(inputId = "y_axis",
                  label = "What would you like to see on the y axis?",
                  choices = colnames(iris)[colnames(iris)!="Species"]),
      # choose which groups you want to see
      checkboxGroupInput(inputId = "spec",
                         label = "Which species would you like to see?",
                         choices = as.character(unique(iris$Species)),
                         selected = as.character(unique(iris$Species))),
      # only build the scatter plot when this is clicked
      actionButton(inputId = "go",
                   label = "Build!")
    ),
    # just an empty main panel
    mainPanel()
  )
)

# then we add what we want to see in the scatter section
ui_win[["Scatter"]] <- fluidPage(
  titlePanel("Iris Dataset Explorer: Scatter"),
  plotOutput(outputId = "iris_scatter")
)

Note that each of ui_win’s names correspond to what I want the name of my window to be (i.e., the UI I want for the controller window is first, then the scatter UI is second in the list). Further, window titles cannot contain spaces or have any duplicates.

Set Up the Calculations

Now that my UI is all set up, I want to set up a derived dataframe that I’ll use in my output, created when I click my “Build!” button. Note that this isn’t explicitly necessary to do if you would prefer to do such things in render functions – but if you’re using any observe()-type capabilities, this is the variable to do it in.

serv_calc is a list of functions of the form function(calc, sess), where any calculations based on input are done. calc is a reactive list containing all the input values with the same names and can be accessed in the same way (calc$x or calc[[x]]). Any variables that you’ve made that you want to pass to input should be added to calc with a name, which is a named list of reactive variables. This is done in the traditional manner, i.e. calc[[“y”]] <- y. sess is the traditional Shiny session variable. Note that these functions follow the traditional Shiny restrictions (reactive variables can only be used in a reactive context, etc.)!

Further, if you want render objects later based on an observe()-type change, assign the result that you will be using in the render to calc. We can see an example of that interaction below, where we use calc[[“sub.df”]] in our render, which changes as the result of our observeEvent().

It’s recommended for clarity that you make one function entry for each observe()-type function you’re using, but it’s not explicitly necessary.

# setting up the list of calculations I want to do
serv_calc <- list()

# I only want to build a scatterplot when I click my build button, so my list will be of length 1
serv_calc[[1]] <- function(calc, sess){
  # this is going to activate any time I press "build!"
  observeEvent(calc$go, {
    # create our data frame for visualizing
    sub.df <- data.frame("x" = iris[iris$Species %in% calc$spec,calc$x_axis],
                         "y" = iris[iris$Species %in% calc$spec,calc$y_axis],
                         "species"=iris[iris$Species %in% calc$spec,"Species"])
    
    # add this to calc, since we want to use this in our rendering
    calc[["sub.df"]] <- sub.df
  })
}

Set Up the Output

Now that I’ve calculated based on input and set that up to work only when I click “build!”, I want to actually render my plot for output.

serv_out is a named list of functions, where each function in the list corresponds to the output that it renders. Functions are of the form function(calc, sess), where calc is the named list which contains traditional Shiny input and the calculated variables that we added to previously, and sess is the traditional Shiny session variable. Entries in both lists can be accessed in a normal manner (i.e. calc$x or calc[[“y”]]). These functions also follow the standard Shiny restrictions.

# set up our serv_out list
serv_out <- list()

# we're just rendering our scatter plot based on the iris dataset
# note the name is the same as the outputid
serv_out[["iris_scatter"]] <- function(calc, sess){
  renderPlot({
    # we add this check to make sure our plot doesn't try to render before we've ever pressed "Build!"
    if (!is.null(calc$sub.df)){
      # build scatterplot
      ggplot(calc$sub.df, aes(x, y, color = factor(species)))+
        geom_point()+ # make scatter
        ggtitle("Iris Comparisons")+ # add title
        xlab(calc$x_axis)+ # change x axis label
        ylab(calc$y_axis)+ # change y axis label
        labs(color="species")+ # change legend label
        NULL
    }
  })
}

Run Multi-Window Shiny App!

Now that we have all our pieces, we can just plug these in and run our app!

If you are using an older version of mwshiny (v1.1.0 or v1.1.0), JS/CSS dependencies may be required, as well as different syntax. For more information on this previous version and an example using visNetwork to find and specify dependencies, please look at our other vignette “Specifying Package JS and CSS Dependencies with mwshiny v1.0.0/v1.1.0”.

#run!
mwsApp(ui_win, serv_calc, serv_out)

Note that multi-windowed apps work better when run in an external browser. Since I can’t show a multi-window Shiny app in a vignette form, here are some images of the final result.

When you initially run mwsApp(), you’ll first be greeted by the selection window, where links to your separate windows are displayed:

If I click on each of my two windows (opening these windows in new tabs), I see the inital states of my controller and scatter windows:

Once I change some inputs and click “build!”, I can see that my scatter window updates accordingly:

It’s as simple as that! Now that you know make a multi-window Shiny app, feel free to make some of your own!

Further Examples

To present this work at useR! 2019, I created a repository of some additional multi-window shiny apps, which you can find here. The examples demonstrate the following:

1. Multiple Monitors at a Workstation (Population Dynamics): folder “pop”
2. Controller Driving External Monitor (Cultural Exploration through Art): folder “art”
3. Alternative Visualization Structures through the Rensselaer Campfire (Circadian Rhythm Functionality): folder “health”

Hopefully these help a little as well. Enjoy your multi-window shiny apps!