inheritance of aesthetics in ggplot2 0.9.3 & the behavior of annotation_custom

Question

Following up on a recent question of mine , this one is a bit different and illustrates the problem more fully using simpler examples. Below are two data sets and three functions. The first one draws some points and a circle as expected:

library("ggplot2")
library("grid")

td1 <- data.frame(x = rnorm(10), y = rnorm(10))

tf1 <- function(df) { # works as expected
    p <- ggplot(aes(x = x, y = y), data = df)
    p <- p + geom_point(color = "red")
    p <- p + annotation_custom(circleGrob())
    print(p)
}

tf1(td1)

This next one seems to ask for the exact sample plot but the code is slightly different. It does not give an error but does not draw the circle:

tf2 <- function(df) { # circle isn't draw, but no error either
    p <- ggplot()
    p <- p + geom_point(data = df, aes(x = x, y = y), color = "red")        
    p <- p + annotation_custom(circleGrob())
    print(p)
    }

tf2(td1)

Finally, this one involves a more complex aesthetic and gives an empty layer when you try to create the circle:

td3 <- data.frame(r = c(rnorm(5, 5, 1.5), rnorm(5, 8, 2)),
    f1 = c(rep("L", 5), rep("H", 5)), f2 = rep(c("A", "B"), 5))

tf3 <- function(df) {
    p <- ggplot()
    p <- p + geom_point(data = df, 
        aes(x = f1, y = r, color = f2, group = f2))     
#   p <- p + annotation_custom(circleGrob()) # comment out and it works
    print(p)
    }

tf3(td3)

Now, I suspect the problem here is not the code but my failure to grasp the inner workings of ggplot2. I could sure use an explanation of why the circle is not drawn in the 2nd case and why the layer is empty in the third case. I looked at the code for annotation_custom and it has a hard-wired inherit.aes = TRUE which I think is the problem. I don't see why this function needs any aesthetic at all (see the docs on it). I did try several ways to override it and set inherit.aes = FALSE but I was unable to fully penetrate the namespace and make it stick. I tried to example the objects created by ggplot2 but these proto objects are nested very deeply and hard to decipher.

Answer 1

To answer this :

"I don't see why this function needs any aesthetic at all".

In fact annotation_custom need x and y aes to scale its grob, and to use after the native units. Basically it did this :

  x_rng <- range(df$x, na.rm = TRUE)                            ## ranges of x :aes x
  y_rng <- range(df$y, na.rm = TRUE)                            ## ranges of y :aes y
  vp <- viewport(x = mean(x_rng), y = mean(y_rng),              ##  create a viewport
                 width = diff(x_rng), height = diff(y_rng),
                 just = c("center","center"))
  dd <- editGrob(grod =circleGrob(), vp = vp)                  ##plot the grob in this vp 

To illustrate this I add a grob to a dummy plot used as a scale for my grob. The first is a big scale and the second is a small one.

base.big   <- ggplot(aes(x = x1, y = y1), data = data.frame(x1=1:100,y1=1:100))
base.small <- ggplot(aes(x = x1, y = y1), data = data.frame(x1=1:20,y1=1:1))

I define my grob, see I use the native scales for xmin,xmax,ymin,ymax

annot <- annotation_custom(grob = circleGrob(),  xmin = 0, 
                                                 xmax = 20, 
                                                 ymin = 0, 
                                                 ymax = 1)

Now see the scales difference( small point / big circle ) between ( base.big +annot ) and ( base.small + annot ).

library(gridExtra)
grid.arrange(base.big+annot,
             base.small+annot)