[R] Problem using with panel.average in Lattice package

[Deepayan Sarkar]

On Fri, Oct 2, 2009 at 8:03 PM, Mark Dalphin <mark.dalphin at peblnz.com> wrote:
> Hi,
>
> I'm having a problem getting the panel.average function to work as I
> expect it to in a lattice plot. I wish to draw lines between the
> averages of groups of y-values at specific x-values. I have created a
> dataset below which is similar to my real data. I also show an example
> of using panel.loess in place of panel.average; it performs in a
> manner similar to what I want panel.average to do except it shows a
> loess line rather than a straight line connecting the means of the
> groups.
>
> Please see my coded examples, below.
>
> Regards,
> Mark Dalphin
>
> =================================================================
> My system information:
>
>> library(lattice)
>> print(sessionInfo())
>
> R version 2.9.1 (2009-06-26) i686-pc-linux-gnu
> locale:
> LC_CTYPE=en_NZ.UTF-8;LC_NUMERIC=C;LC_TIME=en_NZ.UTF-8;LC_COLLATE=en_NZ.UTF-8;
> LC_MONETARY=C;LC_MESSAGES=en_NZ.UTF-8;LC_PAPER=en_NZ.UTF-8;LC_NAME=C;
> LC_ADDRESS=C;LC_TELEPHONE=C;LC_MEASUREMENT=en_NZ.UTF-8;LC_IDENTIFICATION=C
>
> attached base packages:
> [1] stats     graphics  grDevices utils     datasets  methods   base
> other attached packages:
> [1] lattice_0.17-25
>
> loaded via a namespace (and not attached):
> [1] grid_2.9.1  tools_2.9.1
>
> ##--------------------------------------------------------------
> ## This dataset is too complicated, but it does show the type of plot I
> want.
> ##
> ## Create a fake qPCR dataset: Eight 96-well plates over 4 days (2 per day),
> ## 2 genes per plate (multiplexed), and 4 "Hi" positive control and
> ## 4 "Lo" positive controls per plate.
> ## Create the experimental data; by rights it is all identical, expect for
> ## experimental errors with in days and between days.
> ## For this simulation, each gene will be given a base value.
> ## In qPCR the higher the "Ct" value, the lower the concentration.
> library(lattice)  # Add for ease of cut-n-paste of this code
> date  <- c('2009-09-07', '2009-09-08', '2009-09-10', '2009-09-14')
> probe <- c('Gene.A1', 'Gene.A2', 'Gene.B1', 'Gene.B2')
> conc  <- c('Lo', 'Hi')
> base.lo <- c(Gene.A1=29, Gene.A2=25, Gene.B1=28, Gene.B2=31)
> base.hi <- base.lo - 8
> day.err <- c(Day.1=0, Day.2=1, Day.3=1.5, Day.4=1.0)
>
> d <- data.frame()
> for(i in seq(along=date)) {
>   for(j in seq(along=probe)) {
>       for(k in seq(along=conc)) {
>            d <- rbind(d, data.frame(Date=rep(date[i], length=4),
>                                    Probe=rep(probe[j], length=4),
>                                    Conc=rep(conc[k], length=4),
>                                    Ct=rnorm(4, sd=0.5) + (k-1)*8 +
> base.hi[j] + day.err[i]
>                                    ))
>       }
>   }
> }
> d$Date <- as.POSIXct(d$Date)
>
> ##--------------------------------------------------
> ## Example 1
> ## Print with LOESS line showing the 'means' for the groups.
> ## This is close, but I don't want a loess line; I want straight lines
> ## between mean values.
> print(xyplot(Ct ~ Date|Probe, group=Conc, data=d,
>            panel="panel.superpose",
>  panel.groups=function(x, y, ...) {
>                panel.loess(x, y, ...)
>                panel.xyplot(x, y, ...)
>            },
>            auto.key=TRUE))
>
> ##--------------------------------------------------
> ## Example 2
> ## Parallel construction to the loess example, above.
> ## Note the loss of the lines. The 'horizontal' default
> ## is different between 'panel.loess' and 'panel.average'.
> print(xyplot(Ct ~ Date|Probe, group=Conc, data=d,
>            panel="panel.superpose",
>            panel.groups=function(x, y, ...) {
>                panel.average(x, y, horizontal=FALSE, ...)
>                panel.xyplot(x, y, ...)
>            },
>            auto.key=TRUE))

Unfortunately, the implicit type="p" argument in panel.superpose is
overriding the type="l" in panel.average (may be it should be
unmodifiable). So, you need

print(xyplot(Ct ~ Date|Probe, group=Conc, data=d,
             panel="panel.superpose",
             panel.groups=function(x, y, ..., type) {
                 panel.average(x, y, ..., type = "l", horizontal = FALSE)
                 panel.xyplot(x, y, ..., type = type)
             },
             auto.key=TRUE))

It's more common to use the syntactic sugar provided by panel.xyplot:

print(xyplot(Ct ~ Date|Probe, group=Conc, data=d,
             type = c("p", "a"),
             auto.key=TRUE))


>
> ##--------------------------------------------------
> ## Example 3
> ## Don't pass along the '...' to the panel.average. Now I
> ## get lines, but not matching colours to the points.
> print(xyplot(Ct ~ Date|Probe, group=Conc, data=d,
>            panel="panel.superpose",
>            panel.groups=function(x, y, ...) {
>                panel.average(x, y, horizontal=FALSE)
>                panel.xyplot(x, y, ...)
>            },
>            auto.key=TRUE))
>
> ##**********************************************************************
> Main question:
>
> I want to create a plot that looks like Example 3, but with the
> coloured lines of Example 1. Suggestions? I've looked in RSiteSearch()
> for both "panel.average" and "panel.linejoin" but found nothing addressing
> this.
>
> Side question:
>
> I also read the source code to panel.average, panel.loess and
> panel.superpose.  Which leads to a side question; how do I determine
> what parameters are being passed within '...'? I tried recreating my
> panel.groups function above as an explicit (non-anonymous) function
> and using debug() on it, but running formals() within that debugging session
> didn't seem to show me the huge list of arguments from
> panel.superpose, "do.call(panel.groups, args)", coming into my panel
> function. Is there some way to "see" these arguments?

I usually put in

str(list(...))

in the function. For interactive debugging, you could have

dots <- list(...)

and then debug.

-Deepayan