17
Interactive Visualization in R

Adding interactivity to a visualization provides an additional mechanism through which data can be presented in an engaging, efficient, and communicative way. Interactions can allow users to effectively explore large data sets by panning and zooming through plots, or by hovering over specific plot geometry to gain additional details on demand.¹

¹Shneiderman, B. (1996). The eyes have it: A task by data type taxonomy for information visualizations. Proceedings of the. 1996 IEEE Symposium on Visual Languages (pp. 336–). Washington, DC: IEEE Computer Society. http://dl.acm.org/citation.cfm?id=832277.834354

While ggplot2 is the definitive, leading package for making static plots in R, there is not a comparably popular package for creating interactive visualizations. Thus this chapter briefly introduces three different packages for building such visualizations. Instead of offering an in-depth description (as with ggplot2), this chapter provides a high-level “tour” of these packages. The first two (Plotly and Bokeh) are able to add basic interactions to the plots you might make with ggplot2, while the third (Leaflet) is used to create interactive map visualizations. Picking among these (and other) packages depends on the type of interactions you want your visualization to provide, the ease of use, the clarity of the package documentation, and your aesthetic preferences. And because these open source projects are constantly evolving, you will need to reference their documentation to make the best use of these packages. Indeed, exploring these packages further is great practice in learning to use new R packages!

The first two sections demonstrate creating interactive plots of the iris data set, a canonical data set in the machine learning and visualization world in which a flower’s species is predicted using features of that flower. The data set is built into the R software, and is partially shown in Figure 17.1.

For example, you can use ggplot2 to create a static visualization of flower species in terms of the length of the petals and the sepals (the container for the buds), as shown in Figure 17.2:

Click here to view code image

# Create a static plot of the iris data set
ggplot(data = iris) +
  geom_point(mapping = aes(x = Sepal.Width, y = Petal.Width, color = Species))

The following sections show how to use the plotly and rbokeh packages to make this plot interactive. The third section of the chapter then explores interactive mapping with the leaflet package.

²Plotly: https://plot.ly/r/

Plotly is an external package (like dplyr or ggplot2), so you will need to install and load the package before you can use it:

Click here to view code image

install.packages("plotly") # once per machine
library("plotly")          # in each relevant script

This will make all of the plotting functions you will need available.

With the package loaded, there are two main ways to create interactive plots. First, you can take any plot created using ggplot2 and “wrap” it in a Plotly plot,³ thereby adding interactions to it. You do this by taking the plot returned by the ggplot() function and passing it into the ggplotly() function provided by the plotly package:

³Plotly ggplot2 library: https://plot.ly/ggplot2/ (be sure to check the navigation links in the menu on the left).

Click here to view code image

# Create (and store) a scatterplot of the `iris` data set using ggplot2
flower_plot <- ggplot(data = iris) +
  geom_point(mapping = aes(x = Sepal.Width, y = Petal.Width, color = Species))

# Make the plot interactive by passing it to Plotly's `ggplotly()` function
ggplotly(flower_plot)

This will render an interactive version of the iris plot! You can hover the mouse over any geometry element to see details about that data point, or you can click and drag in the plot area to zoom in on a cluster of points (see Figure 17.3).

When you move the mouse over a Plotly chart, you can see the suite of interaction types built into it through the menu that appears (see Figure 17.3). You can use these options to navigate and zoom into the data to explore it.

In addition to making ggplot plots interactive, you can use the Plotly API itself (e.g., calling its own functions) to build interactive graphics. For example, the following code will create an equivalent plot of the iris data set:

Click here to view code image

# Create an interactive plot of the iris data set using Plotly
plot_ly(
  data = iris,      # pass in the data to be visualized
  x = ~Sepal.Width, # use a formula to specify the column for the x-axis
  y = ~Petal.Width, # use a formula to specify the column for the y-axis
  color = ~Species, # use a formula to specify the color encoding
  type = "scatter", # specify the type of plot to create
  mode = "markers"  # determine the "drawing mode" for the scatter (points)
)

Plotly plots are created using the plot_ly() function, which is a sort of corollary to the ggplot() function. The plot_ly() function takes as arguments details about how the chart should be rendered. For example, in the preceding code, arguments are used to specify the data, the aesthetic mappings, and the plot type (that is, geometry). Aesthetic mappings are specified as formulas (using a tilde ~), indicating that the visual channel is a “function of” the data column. Also note that Plotly will try to “guess” values such as type and mode if they are left unspecified (and in which case it will print out a warning in the console).

For a complete list of options available to the plot_ly() function, see the official documentation.⁴ It’s often easiest to learn to make Plotly charts by working from one of the many examples.⁵ We suggest that you find an example that is close to what you want to produce, and then read that code and modify it to fit your particular use case.

⁴Plotly: R Figure Reference: https://plot.ly/r/reference/

⁵Plotly: Basic Charts example gallery: https://plot.ly/r/#basic-charts

In addition to using the plot_ly() function to specify how the data will be rendered, you can add other chart options, such as titles and axes labels. These are specified using the layout() function, which is conceptually similar to the labs() and theme() functions from ggplot2. Plotly’s layout() function takes as an argument a Plotly chart (e.g., one returned by the plot_ly() function), and then modifies that object to produce a chart with a different layout. Most commonly, this is done by piping the Plotly chart into the layout() function:

Click here to view code image

# Create a plot, then pipe that plot into the `layout()` function to modify it
# (Example adapted from the Plotly documentation)
plot_ly(
  data = iris,      # pass in the data to be visualized
  x = ~Sepal.Width, # use a formula to specify the column for the x-axis
  y = ~Petal.Width, # use a formula to specify the column for the y-axis
  color = ~Species, # use a formula to specify the color encoding
  type = "scatter", # specify the type of plot to create
  mode = "markers"  # determine the "drawing mode" for the scatter (points)
) %>%
  layout(
    title = "Iris Data Set Visualization",                  # plot title
    xaxis = list(title = "Sepal Width", ticksuffix = "cm"), # axis label + format
    yaxis = list(title = "Petal Width", ticksuffix = "cm")  # axis label + format
  )

The chart created by this code is shown in Figure 17.4. The xaxis and yaxis arguments expect lists of axis properties, allowing you to control many aspects of each axis (such as the title and the ticksuffix to put after each numeric value in the axis). You can read about the structure and options to the other arguments in the API documentation.⁶

⁶Plotly layout: https://plot.ly/r/reference/#layout

⁷Bokeh: http://bokeh.pydata.org

⁸rbokeh, R Interface for Bokeh: http://hafen.github.io/rbokeh/

As with other packages, you will need to install and load the rbokeh package before you can use it. At the time of this writing, the version of rbokeh on CRAN (what is installed with install.packages()) gives warnings—but not errors!—for R version 3.4; installing a development version from the package’s maintainer Ryan Hafen fixes this problem.

Click here to view code image

# Use `install_github()` to install the version of a package on GitHub
# (often newer)
devtools::install_github("hafen/rbokeh") # once per machine
library("rbokeh") # in each relevant script

You create a new plot with Bokeh by calling the figure() function (which is a corollary to the ggplot() and plot_ly() functions). The figure() function will create a new plotting area, to which you add layers of plot elements such as plot geometry. Similar to when using geometries in ggplot2, each layer is created with a different function—all of which start with the ly_ prefix. These layer functions take as a first argument the plot region created with figure(), so in practice they are “added” to a plot through piping rather than through the addition operator.

For example, the following code shows how to recreate the iris visualization using Bokeh (shown in Figure 17.5):

Click here to view code image

# Create an interactive plot of the iris data set using Bokeh
figure(
  data = iris,                          # data for the figure
  title = "Iris Data Set Visualization" # title for the figure
) %>%
  ly_points(
    Sepal.Width,    # column for the x-axis (without quotes!)
    Petal.Width,    # column for the y-axis (without quotes!)
    color = Species # column for the color encoding (without quotes!)
  ) %>%
  x_axis(
    label = "Sepal Width",       # label for the axis
    number_formatter = "printf", # formatter for each axis tick
    format = "%s cm",            # specify the desired tick labeling
  ) %>%
  y_axis(
    label = "Petal Width",       # label for the axis
    number_formatter = "printf", # formatter for each axis tick
    format = "%s cm",            # specify the desired tick labeling
  )

The code for adding layers is reminiscent of how geometries act as layers in ggplot2. Bokeh even supports non-standard evaluation (referring to column names without quotes) just like ggplot2—as opposed to Plotly’s reliance on formulas. However, formatting the axis tick marks is more verbose with Bokeh (and is not particularly clear in the documentation).

The plot that is generated by Bokeh (Figure 17.5) is quite similar to the version generated by Plotly (Figure 17.4) in terms of general layout, and offers a comparable set of interaction utilities through a toolbar to the right of the chart. Thus you might choose between these packages based on which coding style you prefer, as well as any other aesthetic or interactive design choices of the packages.

⁹Leaflet: https://leafletjs.com

¹⁰Leaflet for R: https://rstudio.github.io/leaflet/

As with other packages, you will need to install and load the leaflet package before you can use it:

Click here to view code image

install.packages("leaflet") # once per machine
library("leaflet")          # in each relevant script

You can create a new Leaflet map by calling the leaflet() function. Just as calling ggplot() will create a blank canvas for constructing a plot, the leaflet() function will create a blank canvas on which you can build a map. Similar to the other visualization packages, Leaflet maps are then constructed by adding (via pipes) a series of layers with different visual elements to constitute the image—including map tiles, markers, lines, and polygons.

The most important layer to add when creating a Leaflet map are the map tiles, which are added with the addTiles() function. Map tiles are a series of small square images, each of which shows a single piece of a map. These tiles can then be placed next to each other (like tiles on a bathroom floor) to form the full image of the map to show. Map tiles power mapping applications like Leaflet and Google Maps, enabling them to show a map of the entire world at a wide variety of levels of zoom (from street level to continent level); which tiles will be rendered depends on what region and zoom level the user is looking at. As you interactively navigate through the map (e.g., panning to the side or zooming in or out), Leaflet will automatically load and show the appropriate tiles to display the desired map!

There are many different sources of map tiles that you can use in your maps, each of which has its own appearance and included information (e.g., rivers, streets, and buildings). By default, Leaflet will use tiles from OpenStreetMap,¹¹ an open source set of map tiles. OpenStreetMap provides a number of different tile sets; you can choose which to use by passing in the name of the tile set (or a URL schema for the tiles) to the addTiles() function. But you can also choose to use another map tile provider¹² depending on your aesthetic preferences and desired information. You do this by instead using the addProviderTiles() function (again passing in the name of the tile set). For example, the following code creates a basic map (Figure 17.6) using map tiles from the Carto¹³ service. Note the use of the setView() function to specify where to center the map (including the “zoom level”).

¹¹OpenStreetMap map data service: https://www.openstreetmap.org

¹²Leaflet-providers preview http://leaflet-extras.github.io/leaflet-providers/preview/

¹³Carto map data service: https://carto.com

Click here to view code image

# Create a new map and add a layer of map tiles from CartoDB
leaflet() %>%
  addProviderTiles("CartoDB.Positron") %>%
  setView(lng = -122.3321, lat = 47.6062, zoom = 10) # center the map on Seattle

The rendered map will be interactive in the sense that you can drag and scroll to pan and zoom—just as with other online mapping services!

After rendering a basic map with a chosen set of map tiles, you can add further layers to the map to show more information. For instance, you can add a layer of shapes or markers to help answer questions about events that occur at specific geographic locations. To do this, you will need to pass the data to map into the leaflet() function call as the data argument (i.e., leaflet(data = SOME_DATA_FRAME)). You can then use the addCircles() function to add a layer of circles to the map (similar to adding a geometry in ggplot2). This function will take as arguments the data columns to map to the circle’s location aesthetics, specified as formulas (with a ~).

Click here to view code image

# Create a data frame of locations to add as a layer of circles to the map
locations <- data.frame(
  label = c("University of Washington", "Seattle Central College"),
  latitude = c(47.6553, 47.6163),
  longitude = c(-122.3035, -122.3216)
)

# Create the map of Seattle, specifying the data to use and a layer of circles
leaflet(data = locations) %>% # specify the data you want to add as a layer
  addProviderTiles("CartoDB.Positron") %>%
  setView(lng = -122.3321, lat = 47.6062, zoom = 11) %>% # focus on Seattle
  addCircles(
    lat = ~latitude,   # a formula specifying the column to use for latitude
    lng = ~ longitude, # a formula specifying the column to use for longitude
    popup = ~label,    # a formula specifying the information to pop up
    radius = 500,      # radius for the circles, in meters
    stroke = FALSE     # remove the outline from each circle
  )

The preceding code also adds interactivity to the map by providing popups—information that pops up on click and remains displayed—as shown in Figure 17.7. Because these popups appear when users are interacting with the circle elements you created, they are specified as another argument to the addCircles() function—that is, as a value of the formula for which column to map to the popup. Alternatively, you can cause labels to appear on hover by passing in the label argument instead of popup.

¹⁴City of Seattle Land use permits: https://data.seattle.gov/Permitting/Building-Permits/76t5-zqzr

¹⁵Interactive visualization in action: https://github.com/programming-for-data-science/in-action/tree/master/interactive-vis

First, the data needs to be loaded into R and filtered down to the subset of data of interest (new buildings since 2010):

Click here to view code image

# Load data downloaded from
# https://data.seattle.gov/Permitting/Building-Permits/76t5-zqzr
all_permits <- read.csv("data/Building_Permits.csv", stringsAsFactors = FALSE)

# Filter for permits for new buildings issued in 2010 or later
new_buildings <- all_permits %>%
  filter(
    PermitTypeDesc == "New",
    PermitClass != "N/A",
    as.Date(all_permits$IssuedDate) >= as.Date("2010-01-01") # filter by date
  )

Before mapping these points, you may want to get a higher-level view of the data. For example, you could aggregate the data to show the number of permits issued per year. This will again involve a bit of data wrangling, which is often the most time-consuming part of visualization:

Click here to view code image

# Create a new column storing the year the permit was issued
new_buildings <- new_buildings %>%
  mutate(year = substr(IssuedDate, 1, 4)) # extract the year

# Calculate the number of permits issued by year
by_year <- new_buildings %>%
  group_by(year) %>%
  count()

# Use plotly to create an interactive visualization of the data
plot_ly(
  data = by_year, # data frame to show
  x = ~year,      # variable for the x-axis, specified as a formula
  y = ~n,         # variable for the y-axis, specified as a formula
  type = "bar",   # create a chart of type "bar" -- a bar chart
  alpha = .7,     # adjust the opacity of the bars
  hovertext = "y" # show the y-value when hovering over a bar
) %>%
  layout(
    title = "Number of new building permits per year in Seattle",
    xaxis = list(title = "Year"),
    yaxis = list(title = "Number of Permits")
  )

The preceding code produces the bar chart shown in Figure 17.9. Keep in mind that the data was downloaded before the summer of 2018, so the observed downward trend is an artifact of when the visualization was created!

After understanding this high-level view of the data, you likely want to know where buildings are being constructed. To do so, you can take the previous map of Seattle and add an additional layer of circles on top of the tiles (one for each building constructed) using the addCircles() function:

Click here to view code image

# Create a Leaflet map, adding map tiles and circle markers
leaflet(data = new_buildings) %>%
  addProviderTiles("CartoDB.Positron") %>%
  setView(lng = -122.3321, lat = 47.6062, zoom = 10) %>%
  addCircles(
    lat = ~Latitude,     # specify the column for `lat` as a formula
    lng = ~Longitude,    # specify the column for `lng` as a formula
    stroke = FALSE,      # remove border from each circle
    popup = ~Description # show the description in a popup
  )

The results of this code are shown in Figure 17.10—it’s a lot of new buildings. And because the map is interactive, you can click on each one to get more details!

While this visualization shows all of the new construction, it leaves unanswered the question of who benefits and who suffers as a result of this change. You would need to do further research into the number of affordable housing units being built, and the impact on low-income and homeless communities. As you may discover, building at such a rapid pace often has a detrimental effect on housing security in a city.

As with ggplot2, the visual attributes of each shape or marker (such as the size or color) can also be driven by data. For example, you could use information about the permit classification (i.e., if the permit is for a home versus a commercial building) to color the individual circles. To effectively map this (categorical) data to a set of colors in Leaflet, you can use the colorFactor() function. This function is a lot like a scale in ggplot2, in that it returns a specific mapping to use:

Click here to view code image

# Construct a function that returns a color based on the PermitClass column
# Colors are taken from the ColorBrewer Set3 palette
palette_fn <- colorFactor(palette = "Set3", domain = new_buildings$PermitClass)

The colorFactor() function returns a new function (here called palette_fn()) that maps from a set of data values (here the unique values from the PermitClass column) to a set of colors—it performs an aesthetic mapping. You can use this function to specify how the circles on the map should be rendered (as with ggplot2 geometries, further arguments can be used to customize the shape rendering):

Click here to view code image

# Modify the `addCircles()` method to specify color using `palette_fn()`
addCircles(
  lat = ~Latitude, # specify the column for `lat` as a formula
  lng = ~Longitude, # specify the column for `lng` as a formula
  stroke = FALSE, # remove border from each circle
  popup = ~Description, # show the description in a popup
  color = ~palette_fn(PermitClass) # a "function of" the palette mapping
)

To make these colors meaningful, you will need to add a legend to your map. As you might have expected, you can do this by adding another layer with a legend in it, specifying the color scale, values, and other attributes:

Click here to view code image

# Add a legend layer in the "bottomright" of the map
addLegend(
  position = "bottomright",
  title = "New Buildings in Seattle",
  pal = palette_fn, # the color palette described by the legend
  values = ~PermitClass, # the data values described by the legend
  opacity = 1
)

Putting it together, the following code generates the interactive map displayed in Figure 17.11.

Click here to view code image

# Create a Leaflet map of new building construction by category
leaflet(data = new_buildings) %>%
  addProviderTiles("CartoDB.Positron") %>%
  setView(lng = -122.3321, lat = 47.6062, zoom = 10) %>%
  addCircles(
    lat = ~Latitude, # specify the column for `lat` as a formula
    lng = ~Longitude, # specify the column for `lng` as a formula
    stroke = FALSE, # remove border from each circle
    popup = ~Description, # show the description in a popup
    color = ~palette_fn(PermitClass), # a "function of" the palette mapping
    radius = 20,
    fillOpacity = 0.5
  ) %>%
  addLegend(
    position = "bottomright",
    title = "New Buildings in Seattle",
    pal = palette_fn, # the palette to label
    values = ~PermitClass, # the values to label
    opacity = 1
  )

In summary, packages for developing interactive visualizations (whether plots or maps) use the same general concepts as ggplot2, but with their own preferred syntax for specifying plot options and customizations. As you choose among these (and other) packages for making visualizations, consider the style of code you prefer to use, the trade-off of customizability versus ease of use, and the visual design choices of each package. There are dozens (if not hundreds) of other packages available and more created every day; exploring and learning these packages is an excellent way to expand your programming and data science skills.

That said, when you are exploring new packages, be careful about using code that is poorly documented or not widely used—such packages may have internal errors, memory leaks, or even security flaws that haven’t been noticed or addressed yet. It’s a good idea to view the package code on GitHub, where you can check the popularity by looking at the number of stars (similar to “likes”) and forks for the project, as well as how actively and recently new commits have been made to the code. Such research and consideration are vital when choosing one of the many packages for building interactive visualizations—or doing any other kind of work—with R.

For practice building interactive visualizations, see the set of accompanying book exercises.¹⁶

¹⁶Interactive visualization exercises: https://github.com/programming-for-data-science/chapter-17-exercises