Removing one tableGrob when applied to a box plot with a facet_wrap

Question

I\'m using the code below to enrich a box plot with a summary table for categorical variable created on the x-axis.

# Libs
require(ggplot2); require(gridExtr         


        

Answer 1

This is just an illustration of the comment.

ggp <- ggplot(data = mtcars, aes(x = factor(cyl), y = qsec, fill = as.factor(gear))) +
  geom_boxplot() +
  scale_y_continuous(limits = quantile(mtcars$qsec, c(0.1, 0.9))) +
  scale_fill_tableau("gear",palette = "tableau10") +
  xlab("cyl") + ylab("qsec") +
  facet_wrap(~am) 

# this requires gridExtra 2.0.0
tt <- ttheme_default(core    = list(fg_params=list(cex = 0.7)),
                     colhead = list(fg_params=list(cex = 0.7)))
grid.newpage()
grid.draw(arrangeGrob(ggp))
grid.draw(grobTree(tableGrob(fun_dta_sum(var_sum = mtcars$qsec, group = mtcars$cyl, data = mtcars),
                             rows=NULL, theme=tt), 
                   vp=viewport(x=unit(0.20,"npc"),y=unit(0.20,"npc"))))

The point is that you really just need to tweak the x=... and y=... arguments to viewport(...). Using annotation_custom(...), even if you could hack the gTable to get rid of one of the grobs, you would still need to tweak the position (using xmin=... and ymin=...). This approach does not maintain the relative position when you shrink or enlarge the image, but neither does annotation_custom(...), so overall I don't really see this as any more difficult.

Answer 2

It would probably make sense to let annotation_custom access facetting info *; this trivial change seems to do the trick,

library(ggplot2)
library(grid)
library(gridExtra)

annotation_custom2 <- 
function (grob, xmin = -Inf, xmax = Inf, ymin = -Inf, ymax = Inf, data) 
{
  layer(data = data, stat = StatIdentity, position = PositionIdentity, 
        geom = ggplot2:::GeomCustomAnn,
        inherit.aes = TRUE, params = list(grob = grob, 
                                          xmin = xmin, xmax = xmax, 
                                          ymin = ymin, ymax = ymax))
}

p <- ggplot(mtcars) + geom_point(aes(mpg, wt)) + facet_wrap(~ cyl)


tg <- tableGrob(iris[1:2,1:2], rows=NULL)
# position the table within the annotation area
tg$vp=viewport(x=unit(0,"npc") + 0.5*sum(tg$widths),
               y=unit(0,"npc") + 0.5*sum(tg$heights))
# need to wrap in a gTree since annotation_custom overwrites the vp
g <- grobTree(tg)
p + annotation_custom2(g, data=data.frame(cyl=8))

Edit * hadley has a different view though, annotation is designed to appear in all panels. It's not clear to me how to produce the geom equivalent for this particular case, if possible.