geom_boxplot() + scale_y_log10() bug

[ra81]

This bug cost me a full day of working.

Some sample of working code

abc <- adply(matrix(rnorm(100, mean = 20, sd = 5), ncol = 5), 2, quantile, c(0, .25, .5, .75, 1))
abc$X1  <- NULL
abc$X1  <- c(1:5)
abc[6, ]  <- c(10, 18, 20, 25, 1000, 6)

# working code
b <- ggplot(abc, aes(group=X1, x = X1, ymin = `0%`, lower = `25%`, middle = `50%`, upper = `75%`, ymax = `100%`))
 b + geom_boxplot(stat = "identity", aes(fill=X1)) + coord_trans(y="log10")

This all ok. Picture is ok.

# broken code
b <- ggplot(abc, aes(group=X1, x = X1, ymin = `0%`, lower = `25%`, middle = `50%`, upper = `75%`, ymax = `100%`))
d <- b + geom_boxplot(stat = "identity") + scale_y_log10()

ggplot makes ranges of y axe as c(0.8490745, 3.1024250).
max value from abc is 1000 = 10^3.
y range upper bound = 10^3 + small margin. So it is log scale.
BUT when data is plotted ggplot uses absolute values of abc data not log values.
So we have wrong picture

Problem like this was discussed here
#274
http://stackoverflow.com/questions/9502003/ggplot-scale-y-log10-issue

[hadley]

Here's a simpler reproducible example:

library(plyr)
df <- data.frame(
  x = c("a", "b"),
  y1 = c(10, 10),
  y2 = c(15, 18),
  y3 = c(20, 20),
  y4 = c(22, 25),
  y5 = c(30, 1000)
)

ggplot(df, aes(x = x, ymin = y1, lower = y2, middle = y3, upper = y4, ymax = y5)) + 
  geom_boxplot(stat = "identity") + 
  scale_y_log10()

[hadley]

Oh hmmm, it's because lower, middle and upper aren't transformed by the scale (while ymin and ymax are). This isn't usually a problem because the stat operates on the transformed values, so you'll only see this with stat identity and a transformed y axis.