Review

• A data visualisation consists of data symbols, guides, and labels.

• We produce data symbols by mapping data values to the visual features of a shape.

• Which visual feature we choose depends on:

  • Whether we have quantitative or qualitative data.
  • The accuracy and capacity of the visual feature.
  • The question we are interested in answering.

Review

• The effectiveness of a data visualisation also depends on coordination between different elements.

  • Contrast.
  • Repetition.
  • Alignment.
  • Proximity.

• There may also be a justification for elements that do not map directly to data values.

Customisation

• With {ggplot2} we are able to specify a high-level description of a data visualisation.

  • A lot of the details are filled in for us.
  • It can be difficult to control all of the details.

• This topic explores a different approach to producing a data visualisation.

  • We control all of the details.
  • We have to specify all of the details ourselves.

• There are some things that are hard to do with {ggplot2}.

Customisation

• The {grid} package

• Integrating {grid} with {ggplot2}

• Integrating {ggplot2} with {grid}

• With the {ggplot2} package, we describe a data visualisation in terms of higher-level concepts, such as mappings from data to aesthetics.

  ggplot(crimeGenderTotal) +
      geom_col(aes(x=prop, y="", fill=gender))

  

• With the {grid} package, we just draw shapes.

  library(grid)

  grid.newpage()

  grid.rect(gp=gpar(col=NA, fill="grey90"))

  

• grid.rect() draws rectangles

• gpar() controls graphical settings like (border) colour and fill colour.

  grid.rect(gp=gpar(col=NA, fill="grey90"))

  

• grid.segments() draws line segments.

• gpar(lwd) controls line width.

  grid.rect(gp=gpar(col=NA, fill="grey90"))
  grid.segments(0, .5, 1, .5, gp=gpar(col="white", lwd=.5))
  grid.segments(0:4/4, 0, 0:4/4, 1, gp=gpar(col="white", lwd=2))

  

• Positions are proportions of the page (by default)
  (0 = left/bottom, 1 = right/top)

• These are "npc" (normalised parent) coordinates.

  grid.rect(gp=gpar(col=NA, fill="grey90"))
  grid.segments(0, .5, 1, .5, gp=gpar(col="white", lwd=.5))
  grid.segments(0:4/4, 0, 0:4/4, 1, gp=gpar(col="white", lwd=2))

  

• Shapes can be based on data!

  widths <- crimeGenderTotal$prop
  widths

  [1] 0.2995225 0.7004775

  fills <- scales::hue_pal()(2)
  fills

  [1] "#F8766D" "#00BFC4"

• Shapes can be based on data!

  grid.rect(gp=gpar(col=NA, fill="grey90"))
  grid.segments(0, .5, 1, .5, gp=gpar(col="white", lwd=.5))
  grid.segments(0:4/4, 0, 0:4/4, 1, gp=gpar(col="white", lwd=2))
  grid.rect(c(1, 0), width=widths, height=.8, 
            hjust=c(1, 0), gp=gpar(col=NA, fill=fills))

  

• Drawing is always relative to a viewport.

  vp <- viewport(width=.8, height=.6)

  

• Drawing is always relative to a viewport.

  pushViewport(vp)
  grid.rect(gp=gpar(col=NA, fill="grey90"))
  grid.segments(0, .5, 1, .5, gp=gpar(col="white", lwd=.5))
  grid.segments(0:4/4, 0, 0:4/4, 1, gp=gpar(col="white", lwd=2))
  grid.rect(c(1, 0), width=widths, height=.8, 
            hjust=c(1, 0), gp=gpar(col=NA, fill=fills))
  popViewport()

  

• Drawing is always relative to a coordinate system.

• vp(xscale, yscale) define "native" coordinates.

• "npc" coordinates are also still available.

  vp <- viewport(width=.8, height=.6, xscale=c(-.05, 1.05))

  

• Drawing is always relative to a coordinate system.

• default.units selects the default coordinates.

  grid.segments(0, .5, 1, .5, gp=gpar(col="white"))
  grid.segments(0:4/4, 0, 0:4/4, 1, gp=gpar(col="white"), 
                default.units="native")

• Drawing is always relative to a coordinate system.

  pushViewport(vp)
  grid.rect(gp=gpar(col=NA, fill="grey90"))
  grid.segments(0, .5, 1, .5, gp=gpar(col="white"))
  grid.segments(0:4/4, 0, 0:4/4, 1, gp=gpar(col="white"), 
                default.units="native")
  grid.rect(c(1, 0), width=crimeGenderTotal$prop, height=.8, 
            hjust=c(1, 0), gp=gpar(col=NA, fill=fills), 
            default.units="native")

  

• unit() associates values with a coordinate system.

• There are "mm" and "in" coordinates available.

• "npc" and "native" are still also available.

  twoMM <- unit(2, "mm")

• unit() associates values with a coordinate system.

  pushViewport(vp)
  grid.rect(gp=gpar(col=NA, fill="grey90"))
  grid.segments(0, .5, 1, .5, gp=gpar(col="white"))
  grid.segments(0:4/4, 0, 0:4/4, 1, gp=gpar(col="white"), 
                default.units="native")
  grid.rect(c(1, 0), width=crimeGenderTotal$prop, height=.8, 
            hjust=c(1, 0), gp=gpar(col=NA, fill=fills), 
            default.units="native")
  grid.segments(0:4/4, 0, 0:4/4, -twoMM, default.units="native")

  

• grid.text() draws text.

  pushViewport(vp)
  grid.rect(gp=gpar(col=NA, fill="grey90"))
  grid.segments(0, .5, 1, .5, gp=gpar(col="white"))
  grid.segments(0:4/4, 0, 0:4/4, 1, gp=gpar(col="white"), 
                default.units="native")
  grid.rect(c(1, 0), width=crimeGenderTotal$prop, height=.8, 
            hjust=c(1, 0), gp=gpar(col=NA, fill=fills), 
            default.units="native")
  grid.segments(0:4/4, 0, 0:4/4, -twoMM, default.units="native")
  grid.text(0:4/4, unit(0:4/4, "native"), unit(-1, "lines"))

  

{grid} Shapes

gpar() Settings

{grid} units

• {grid} means more work, but you get complete control.

• {grid} means more work, but you get complete control.

  

• {grid} means more work, but you get complete control.

  

• {grid} means more work, but you get complete control.

• {grid} means more work, but you get complete control.

Cats and Dogs

• The pets data frame contains an estimate of the number of pet cats and dogs globally.

  pets <- data.frame(pet=c("dog", "cat"), count=c(471, 373))
  pets

    pet count
  1 dog   471
  2 cat   373

Cats and Dogs

ggplot(pets) +
    geom_col(aes(x=count, y=pet))

Cats and Dogs

• The raster images cat and dog depict silhouettes of a cat and a dog.

  grid.rect(gp=gpar(col=NA, fill="grey"))
  grid.raster(catImg, x=1/3, height=1/4)
  grid.raster(dogImg, x=2/3, height=1/4)

  

{grid} Grobs

• Instead of drawing {grid} output directly, we can create a grob (graphical objects), but draw nothing, and then draw the grob later.

  catGrob <- rasterGrob(catImg, x=1/3, height=1/4)
  dogGrob <- rasterGrob(dogImg, x=2/3, height=1/4)

{grid} Grobs

grid.rect(gp=gpar(col=NA, fill="grey"))
grid.draw(catGrob)
grid.draw(dogGrob)

{grid} Grobs

• We can combine grobs into a single gTree (grob Tree).

  petGrob <- grobTree(catGrob, dogGrob)

  grid.rect(gp=gpar(col=NA, fill="grey"))
  grid.draw(petGrob)

  

The {gggrid} package

• The {gggrid} package allows us to mix {grid} drawing with a {ggplot2} plot.

  library(gggrid)

• The grid_panel() function draws a grob on a {ggplot2} plot.

  ggplot(pets) +
      geom_col(aes(x=count, y=pet)) +
      grid_panel(petGrob)

  

• The grobs are drawn within a viewport corresponding to the {ggplot2} panel.

  catGrob2 <- rasterGrob(catImg, x=.2, y=1/4, height=1/4)
  dogGrob2 <- rasterGrob(dogImg, x=.2, y=3/4, height=1/4)

  ggplot(pets) +
      geom_col(aes(x=count, y=pet)) +
      grid_panel(catGrob2) +
      grid_panel(dogGrob2)

  

• grid_panel() also accepts a function that calculates what to draw.

• The function is called with data and coords (transformed data) for the {ggplot2} plot panel.

• The function returns a grob or gTree for {ggplot2} to draw.

  demo <- function(data, coords) {
      print(data)
      print(coords)
      rectGrob()
  }

• grid_panel() also accepts a function that calculates what to draw.

• The function is called with data and coords (transformed data) for the {ggplot2} plot panel.

  ggplot(pets) +
      geom_col(aes(x=count, y=pet)) +
      grid_panel(demo, aes(x=count, y=pet))

      x y PANEL group
  1 471 2     1     2
  2 373 1     1     1
            x         y PANEL group
  1 0.9545455 0.7272727     1     2
  2 0.7653928 0.2727273     1     1

• grid_panel() also accepts a function that calculates what to draw.

• The function returns a grob or gTree for {ggplot2} to draw.

  petFun <- function(data, coords) {
      grobTree(rasterGrob(catImg, y=coords$y[2], height=1/4,
                          x=unit(coords$x[2], "npc") - unit(1, "cm"),
                          just="right"),
               rasterGrob(dogImg, y=coords$y[1], height=1/4,
                          x=unit(coords$x[1], "npc") - unit(1, "cm"),
                          just="right"))
  }

The {gggrid} package

ggplot(pets) +
    geom_col(aes(x=count, y=pet)) +
    grid_panel(petFun, aes(x=count, y=pet))

• {gggrid} means greater control, at the cost of more effort.

• A {ggplot2} plot is “just” {grid} drawing in the end.

  ggplot(crimeGenderTotal) +
      geom_col(aes(x=prop, y="", fill=gender))

  

• A {ggplot2} plot is “just” {grid} drawing in the end.

  ggplot(crimeGenderTotal) +
      geom_col(aes(x=prop, y="", fill=gender))

  grid.force()
  grid.ls()

  layout
    background.1-13-16-1
      plot.background..rect.462
    panel.9-7-9-7
      panel-1.gTree.428
        grill.gTree.426
          panel.background..rect.419
          panel.grid.minor.x..polyline.421
          panel.grid.major.y..polyline.423
          panel.grid.major.x..polyline.425
        NULL
        geom_rect.rect.415
        NULL
        panel.border..zeroGrob.416

• We can draw a {ggplot2} plot within a {grid} viewport.

  gg <- 
      ggplot(crimeGenderTotal) +
          geom_col(aes(x=prop, y="", fill=gender))

  pushViewport(viewport(y=0, height=3/4, just="bottom"))
  print(gg, newpage=FALSE)

  

• This allows us to draw other content alongside.

  pushViewport(viewport(y=0, height=3/4, just="bottom"))
  print(gg, newpage=FALSE)
  popViewport()
  pushViewport(viewport(y=3/4, height=1/4, just="bottom", clip=TRUE))
  grid.rect(gp=gpar(lwd=3))
  grid.raster(policeLogo, x=.1, just="left", height=.8)

  

Arranging components

• The crimeYear data frame contains the number of offenders per year.

  

Arranging components

• We can see changes more easily by directly visualising changes.

  

The {patchwork} Package

• We can answer more questions if we have both views of the data.

  library(patchwork)

  gg1 <- ggplot(crimeYear) +
      geom_line(aes(x=year, y=total)) +
      scale_x_continuous(limits=c(2010, 2022), 
                         expand=expansion(c(0, 0)),
                         breaks=seq(2010, 2020, 2))
  gg2 <- ggplot(crimeYear) +
      geom_col(aes(x=year, y=change)) +
      scale_x_continuous(limits=c(2010, 2022), 
                         expand=expansion(c(0, 0)),
                         breaks=seq(2010, 2020, 2))

The {patchwork} Package

gg1 + gg2

Summary

• We can produce a data visualisation with complete control over all of the details with {grid}
  • Draw basic shapes.
  • Control graphical parameters.
  • Draw within viewports.
  • Associate locations and dimensions with units.

• {ggplot2} is built on top of {grid}
  • We can add {grid} drawing to {ggplot2} with {gggrid}.
  • We can draw {ggplot2} within {grid} viewports.

Exercise

• Can you use {grid} to draw any of these diagrams?

  

Exercise

• The male() function creates a male symbol grob.

  male <- function(x=.5, y=.5) {
      vp <- viewport(x, y, width=unit(6, "mm"), height=unit(6, "mm"),
                     gp=gpar(col="white"))
      grobTree(segmentsGrob(.5, .5, 1, 1),
               segmentsGrob(1, 1, 2/3, 1),
               segmentsGrob(1, 1, 1, 2/3),
               circleGrob(1/3, 1/3, r=1/3, gp=gpar(fill="skyblue")),
               vp=vp)
  }

  grid.rect(gp=gpar(col=NA, fill="grey40"))
  grid.draw(male())

  

Exercise

• The female() function creates a female symbol grob.

  female <- function(x=.5, y=.5) {
      vp <- viewport(x, y, width=unit(6, "mm"), height=unit(6, "mm"),
                     gp=gpar(col="white"))
      grobTree(segmentsGrob(2/3, .5, 2/3, -1/3),
               segmentsGrob(1/3, 0, 1, 0),
               circleGrob(2/3, 2/3, r=1/3, gp=gpar(fill="pink")),
               vp=vp)
  }

  grid.rect(gp=gpar(col=NA, fill="grey40"))
  grid.draw(female())

  

Exercise

• Can you use the male() and female() functions (and the crimeGenderTotal data frame) to produce the data visualisation below?