[R] Lattice: Superimposing histograms with different colors and transparency effects

[Yvonnick NOEL]

Dear users,

I would like to plot several histograms superimposed on the same panel
with different colors, with superimposed polygons appearing with
transparency effects. I also want estimated densities to appear on the
same plot. For several reasons, including that I like it, I want to use
the lattice package.

I have several questions regarding the use of the 'histogram' function
with a group structure.

I first thought that defining well-chosen values of alpha in trellis
parameters would give the transparency effets, but this is not the case:

library(lattice)

# Some data
x = c(rnorm(100), rnorm(100,2), rnorm(100,4))
grouping = gl(3,100,300)

# Trellis parameters
trellis.par.set("superpose.polygon"=list(alpha=rep(.5,3)))

histogram(~x ,groups=grouping, type = "density",
       panel = "panel.superpose",
       panel.groups = function(x,...) {
         panel.histogram(x,...)
         panel.mathdensity(dmath=dnorm,args =
list(mean=mean(x),sd=sd(x)),...)
       })

Besides transparency, I get no filling colors at all in the, even though
plot.polygon and superpose.polygon parameters are set. I clearly need to
define my own colors with alpha channel set:

mycolors = rgb(c(228, 55, 77), 
             c(26, 126, 175),
             c(28, 184, 74),alpha = 50,maxColorValue = 255)

... and include 'mycolors' as an explicit argument in the histogram
function:

histogram(~x ,groups=grouping, type = "density",ylim=c(0,.45),
       panel = "panel.superpose",col=mycolors,
       auto.key=list(space="right",rectangles=FALSE,col=mycolors),
       panel.groups = function(x,...) {
         panel.histogram(x,...)
         panel.mathdensity(dmath=dnorm,
args=list(mean=mean(x),sd=sd(x)),...)
       })

-> First question: Is it the only mean to get histogram bars filled, or
do I do something wrong in the use of trellis.arg.set ?

The problem with the previous approach is that the 'col' argument also
affects the density curves colors, for which I don't want transparency
effects. The 'col.lines' argument doesn't seem to change anything.
Removing the (...) arguments is not an interesting option, as it
suppresses some useful parameters for histograms (breaks, etc.).

-> Second question: How do I get superimposed density curves with colors
that differ from the bar colors (i.e. here: No transparency effects)?

-> Third question: How do I find nice (and common) ylim values for the
three histograms? I have set ylim=c(0,.45) above by hand, but I would
like to see this calibrated beforehand. Adding a prepanel function is
probably the way to go, but I am not sure how to manage this.

-> Fourth question: I would like the bar borders to have colors that
also vary from group to group, but unlike the 'col=' argument, adding a
'border=mycolors' argument in the histogram function call change colors
from bar to bar!

Thank you very much in advance.

Best wishes,

Yvonnick Noel, PhD.
University of Brittany, Rennes
France