R help - Jan 2008 - Overlay plots from different data sets using the Lattice package

#After spending the entire day working on this question, I have
decided to reach out for support:

#I am trying to overlay a densityplot from one data set over a
histogram of another, if I were to plot the two individually, they
would look like:

# data frame construction

data.frame.A <- data.frame(rnorm(12*8), c(rep("one", 4),
rep("two",
4), rep("three", 4)), c("red", "orange",
"yellow", "green"))
names(data.frame.A) <- c("vals", "factor.1",
"factor.2")
data.frame.B <- data.frame(rnorm(12*15), c(rep("one", 4),
rep("two",
4), rep("three", 4)), c("red", "orange",
"yellow", "green"))
names(data.frame.B) <- names(data.frame.A)

# stand alone plots
histogram(~ vals|factor.1*factor.2, data.frame.A, type = "density")
densityplot(~ vals | factor.1*factor.2, data.frame.B, plot.points = F)

# I found http://www.ats.ucla.edu/STAT/R/faq/overlay_trellis.htm

xyplot(write~read | ses.f, hsb2,
  panel=function(x, y, subscripts){
	panel.xyplot(x, y, pch=16)
	panel.lmline(x, y, lty=4)
	panel.xyplot(hsb2$science[subscripts], hsb2$math[subscripts], pch=3)
	panel.lmline(hsb2$science[subscripts], hsb2$math[subscripts])
}, as.table=T, subscripts=T)

# but it seems that the overlay only works in cases where the data
comes from the same data.frame and has the same subscripting.

# I have attempted a variety of incarnations of:

histogram(~ vals|factor.1*factor.2, data.frame.A, type = "density",
  panel=function(x, ...){
	panel.histogram(x,...)
    panel.densityplot(~ vals | factor.1*factor.2, data.frame.B,
plot.points = F) })

# which obviously doesn't work.

# I'm not sure if the problem lies in my poor understanding of writing
functions or my poor understanding of Lattice. Is it indeed possible?

# Thank you for your time.

# Erin

On Jan 14, 2008 7:30 PM, Erin Steiner <steiner.erin at gmail.com>
wrote:
> #After spending the entire day working on this question, I have
> decided to reach out for support:
>
> #I am trying to overlay a densityplot from one data set over a
> histogram of another, if I were to plot the two individually, they
> would look like:
>
> # data frame construction
>
> data.frame.A <- data.frame(rnorm(12*8), c(rep("one", 4),
rep("two",
> 4), rep("three", 4)), c("red", "orange",
"yellow", "green"))
> names(data.frame.A) <- c("vals", "factor.1",
"factor.2")
> data.frame.B <- data.frame(rnorm(12*15), c(rep("one", 4),
rep("two",
> 4), rep("three", 4)), c("red", "orange",
"yellow", "green"))
> names(data.frame.B) <- names(data.frame.A)
>
> # stand alone plots
> histogram(~ vals|factor.1*factor.2, data.frame.A, type =
"density")
> densityplot(~ vals | factor.1*factor.2, data.frame.B, plot.points = F)
It isn't lattice, but this is pretty easy to do with ggplot2:

install.packages("ggplot2")
library(ggplot2)

qplot(vals, ..density.., data = data.frame.A, geom="histogram", facets
= factor.1 ~ factor.2, binwidth = 1) + geom_density(data=data.frame.B)

In ggplot2, every "layer" on the plot can have a different dataset.
You can find out more about ggplot2 at http://had.co.nz/ggplot2/, but
I'm still working on the documentation for these more advanced
features.

Hadley


-- 
http://had.co.nz/

On 1/14/08, Erin Steiner <steiner.erin at gmail.com>
wrote:
> #After spending the entire day working on this question, I have
> decided to reach out for support:
>
> #I am trying to overlay a densityplot from one data set over a
> histogram of another, if I were to plot the two individually, they
> would look like:
>
> # data frame construction
>
> data.frame.A <- data.frame(rnorm(12*8), c(rep("one", 4),
rep("two",
> 4), rep("three", 4)), c("red", "orange",
"yellow", "green"))
> names(data.frame.A) <- c("vals", "factor.1",
"factor.2")
> data.frame.B <- data.frame(rnorm(12*15), c(rep("one", 4),
rep("two",
> 4), rep("three", 4)), c("red", "orange",
"yellow", "green"))
> names(data.frame.B) <- names(data.frame.A)
The first step would be to combine the two data sources:

df.comb <- make.groups(data.frame.A, data.frame.B)

I would then just overlay two density plots:

densityplot(~vals | factor.1 * factor.2, df.comb,
            groups = which, plot.points=FALSE,
            auto.key = TRUE)

but you could do a histogram and a densityplot too:

histogram(~vals | factor.1 * factor.2, df.comb,
          type = "density",
          groups = which,
          panel = panel.superpose,
          panel.groups = function(x, group.number, col, ...) {
              if (group.number == 1)
                  panel.histogram(x, ...)
              else
                  panel.densityplot(x, ..., plot.points = FALSE)
          })

-Deepayan

hadley wickham <h.wickham <at> gmail.com> writes:
> qplot(vals, ..density.., data = data.frame.A, geom="histogram",
facets
> = factor.1 ~ factor.2, binwidth = 1) + geom_density(data=data.frame.B)
uhh.. how should I understand "y (?)=..density.."

Dieter

On Jan 15, 2008 12:27 PM, Dieter Menne <dieter.menne at menne-biomed.de>
wrote:
> hadley wickham <h.wickham <at> gmail.com> writes:
>
> > qplot(vals, ..density.., data = data.frame.A,
geom="histogram", facets
> > = factor.1 ~ factor.2, binwidth = 1) + geom_density(data=data.frame.B)
>
> uhh.. how should I understand "y (?)=..density.."
Yes, that does require a little explanation.  The .. tells ggplot the
variable (density) is not in the original dataset, but is instead
produce by the statistical transformation (stat_bin for the histogram
and stat_density for the density plot).  By default the histogram uses
..count.. (the raw count) while density uses ..density.., so y ..density.. tells
them both to use the same scale (you could also use
y = ..count.. but count on a density is plot is a little less common)

Hadley


-- 
http://had.co.nz/