Save plot with a given aspect ratio

You can use grid functions to calculate the full size of the ggplot grob, but there are (edit: at least) two caveats:

  • an extra device window will open, to do the unit conversion

  • the plot panel size will be 0 by default, as it is meant to be calculated on-the-fly according to the device (viewport) it lives in, not the opposite.

That being said, the following function attempts to open a device that fits the ggplot exactly,

library(ggplot2)
library(grid)

sizeit <- function(p, panel.size = 2, default.ar=1){

  gb <- ggplot_build(p)
  # first check if theme sets an aspect ratio
  ar <- gb$plot$coordinates$ratio

  # second possibility: aspect ratio is set by the coordinates, which results in 
  # the use of 'null' units for the gtable layout. let's find out
  g <- ggplot_gtable(gb)
  nullw <- sapply(g$widths, attr, "unit")
  nullh <- sapply(g$heights, attr, "unit")

  # ugly hack to extract the aspect ratio from these weird units
  if(any(nullw == "null"))
    ar <- unlist(g$widths[nullw == "null"]) / unlist(g$heights[nullh == "null"])

  if(is.null(ar)) # if the aspect ratio wasn't specified by the plot
       ar <- default.ar

  # ensure that panel.size is always the larger dimension
  if(ar <= 1 ) panel.size <- panel.size / ar

  g$fullwidth <- convertWidth(sum(g$widths), "in", valueOnly=TRUE) + 
    panel.size
  g$fullheight <- convertHeight(sum(g$heights), "in", valueOnly=TRUE) + 
    panel.size / ar

  class(g) <- c("sizedgrob", class(g))
  g
}


print.sizedgrob <- function(x){
  # note: dev.new doesn't seem to respect those parameters
  # when called from Rstudio; in this case it 
  # may be replaced by x11 or quartz or ...
  dev.new(width=x$fullwidth, height=x$fullheight)
  grid.draw(x)
}


p1 <- ggplot(mtcars, aes(x = wt, y = mpg)) + geom_point() + coord_fixed() +
  theme(plot.background = element_rect(colour = "red"))

p2 <- p1 + aes(x = mpg, y = wt)

# need for an explicit dummy device open, otherwise it's a bit off
# for no apparent reason that I can understand
dev.new() 

sizeit(p1, 0.1)

enter image description here

sizeit(p2, 2)

enter image description here

Leave a Comment