View Composition

Combining multiple data displays

Vega-lite and Altair have several methods to help combine different views of these data.

Layering: Combine several displays on the same set of axes
Horizontal Concatenation: Position several displays next to each other, horizontally.
Vertical Concatenation: Position several displays next to each other, vertically
Faceting: Create panels showing the same view of different subsets of the data
Repeating: Create panels that differ in the encoding of one or more channels

For the first three methods, Altair has functions that take in two chart objects. Those chart objects can either be a simple chart made by alt$Chart, or a compound chart made by any of the methods above. Altair and this R package also provides operators for these methods:

Operation	Function Syntax	Operator Syntax
Layering	`alt$layer(chart1, chart2)`	`chart1 + chart2`
Horizontal Concatenation	`alt$hconcat(chart1, chart2)`	`chart1 \| chart2`
Vertical Concatenation	`alt$vconcat(chart1, chart2)`	`chart1 & chart2`

In contrast, faceting and repeating are methods for Chart objects, so if you had created a chart called chart1 you would call them using chart1$:

Operation	Function Syntax
Faceting	`chart1$facet(...)`
Repeating	chart1$`repeat`(...)

Note that for “repeat”, backticks are required around the word repeat. This is because repeat is a reserved word in R!

Layered charts

Layering involves combining multiple displays on the same set of axes. Two chart objections can be layered using either the + operator or the alt$layer function. A common use case is to add more than one mark for the same set of encodings. For example, you could combine points and a line for time-series data. Here is an example using the “beaver1” data, which has body temperature for a beaver over time:

library("altair")

base <- 
  alt$Chart(beaver1)$
  encode(
    x = alt$X("time"),
    y = alt$Y("temp", scale = alt$Scale(zero = FALSE))
  )

scatter_plot <- base$mark_point()
line_plot <- base$mark_line()

scatter_plot + line_plot

This layered plot could also have been made using:

alt$layer(scatter_plot, line_plot)

Passing the result of that function to vegalite would give the same output as using +.

Horizontal concatenation

We will now make a similar plot for another beaver, using the beaver2 dataset. We will use horizontal concatenation to juxtapose those two plots, using the | operator:

base2 <- 
  alt$Chart(beaver2)$
  encode(
    x = alt$X("time"),
    y = alt$Y("temp", scale = alt$Scale(zero = FALSE))
  )

scatter_plot2 <- base2$mark_point()
line_plot2 <- base2$mark_line()

combined_plot <- 
  (scatter_plot + line_plot)$
  properties(title = "Beaver 1", width = 200)

combined_plot2 <- 
  (scatter_plot2 + line_plot2)$
  properties(title = "Beaver 2", width = 200)

combined_plot | combined_plot2

This plot could also have been made using:

alt$hconcat(combined_plot, combined_plot2)

Rendering the result of that function would give the same output as using |.

Vertical concatenation

Those two plots could alternatively have been concatenated vertically using &:

combined_plot <- combined_plot$properties(height = 200)
combined_plot2 <- combined_plot2$properties(height = 200)

combined_plot & combined_plot2

This plot could also have been made using:

alt$vconcat(combined_plot, combined_plot2)

Passing the result of that function to vegalite would give the same output as using &.

Combining vertical and horizontal concatenation

Compound plots can be combined to create more elaborate composed views. Here is an example of combining two plots for each beaver into a compound plot with four panels:

hist_plot <- 
  alt$Chart(beaver1)$
  mark_bar()$
  encode(
    x = alt$X("temp:Q", bin = TRUE),
    y = alt$Y("count():Q")
  )$
  properties(width = 200, height = 200)

hist_plot2 <- 
  alt$Chart(beaver2)$
  mark_bar()$
  encode(
    x = alt$X("temp:Q", bin = TRUE),
    y = alt$Y("count():Q")
  )$
  properties(width = 200, height = 200)

(combined_plot | hist_plot) & (combined_plot2 | hist_plot2)

Note the use of parentheses to make sure the order of operations is correct.

Faceting will likely be familiar to R users comfortable with ggplot2. Faceting creates sub-panels for different subsets of the data. Faceting is specified in similar ways to encoding, using $facet() instead of $encode(). Here is an example of faceting with the mtcars data set, with the rows split by cylinders and the columns by gears:

plot <- 
  alt$Chart(mtcars)$
  mark_point()$
  encode(
    x = "mpg:Q",
    y = "hp:Q"
  )$
  properties(height = 150, width = 150)
  
plot$facet(row = "cyl:N", column = "gear:N")

Repeating

Repeating is a little different than faceting. Repeating creates panels based on different encodings, with the same data in each panel. To create a repeating plot, you specify the encodings with alt$`repeat`("column") or alt$`repeat`("row") instead of an actual column variable. Then the encodings to use for the column or row are specified when calling $`repeat`() on the chart object. Note the use of the backticks around “repeat”; these backticks are necessary because repeat is a reserved word in R.

Here is an example of creating plots with different x and y axes for the mtcars dataset:

plot2 <- 
  alt$Chart(mtcars)$
  mark_point()$
  encode(
    alt$X(alt$`repeat`("column"), type="quantitative"),
    alt$Y(alt$`repeat`("row"), type="quantitative")
  )$
  properties(height = 150, width = 150)$
  `repeat`(
    row = c("mpg", "qsec"),
    column = c("hp", "wt")
  )

plot2

More information

For more information on view composition in Vega-lite and Altair, check out the Vega-lite or Altair documentation about the subject!

Alicia Schep

2023-09-04