R help - Apr 2011 - Override col.lines and col.symbol in panel.xyplot with type='b'

Dear useRs,

I have a longitudinal experiment with several treatment groups, ~20 subjects per
group, ~6 timepoints and a continuous dependent variable.  I have been
successfully been using lattice::xyplot with this data.  However, I have been
stumped with a particular application of it.

I would like to use xyplot on my data, broken into treatment groups with the
groups argument, using type='b' to show subjectwise longitudinal data. 
So far so good, I have done this many times.  But now I wish to show the same
data but having the color of the lines and symbols overridden in some arbitrary
way, yet not without changing anything else about the plot, in particular the
structure/topology of the plot from using the groups argument and
type='b'.

This requires using a panel function of some sort.  I have come to think I will
need to use a function with a 'subscripts' argument as the panel
function, which then itself calls panel.xyplot() and uses its arguments col.line
and col.symbol.  The closest example I could find is on page 73 of Sarkar's
UseR! Lattice book, where the subscripts argument indexed the data within each
grouped subplot, and it was used as an index for a user-generated vector of
colors.  This seems like what I want to do.

But I could not get this paradigm to work.  Here is a simple example using
trivial data:

> unlist(R.Version())
                                      platform                                  
arch
                             "i386-pc-mingw32"                        
"i386"
                                            os                                  
system
                                     "mingw32"                        
"i386, mingw32"
                                        status                                  
major
                                     "Patched"                        
"2"
                                         minor                                  
year
                                        "12.2"                        
"2011"
                                         month                                  
day
                                          "03"                        
"18"
                                       svn rev                                  
language
                                       "55383"                        
"R"
                                version.string
"R version 2.12.2 Patched (2011-03-18 r55383)"
> search()
[1] ".GlobalEnv"        "package:stats"    
"package:graphics"
[4] "package:grDevices" "package:utils"    
"package:datasets"
[7] "package:methods"   "Autoloads"        
"package:base"
> require(lattice)
Loading required package: lattice
> set.seed(388659262)
> dat <- data.frame(Panel=rep(c('A','B'), each=4),
+                   ID=factor(rep(letters[1:4], each=2)),
+                   X=rep(c(0,1), times=4),
+                   Y=runif(8)
+                  )

##### now for the arbitrary colors.  Let's highlight one subject red, the
rest black
> dat$Color <- with(dat, ifelse(Panel=='A' & ID ==
'a', 2, 1))
> dat
  Panel ID X         Y Color
1     A  a 0 0.1138821     2
2     A  a 1 0.7361403     2
3     A  b 0 0.3304683     1
4     A  b 1 0.5866701     1
5     B  c 0 0.8819857     1
6     B  c 1 0.7329025     1
7     B  d 0 0.5000357     1
8     B  d 1 0.6365438     1

#####  The following standard plot is fine.
#####  Each subject is colored differently,
#####  I believe recycling through the colors from either
#####  trellis.par.get("superpose.symbol")$col or
#####  trellis.par.get("superpose.line")$col,
#####  but be default they are the same anyway
> xyplot(Y ~ X | Panel, data=dat, groups=ID, type='b',
+        scales=list(x=list(at=c(0,1),labels=c(0,1))))

#####  But for example, this following attempt to grab the
#####  corresponding values of dat$Color do not have
#####  my intended effect.  There are now three "subjects"
#####  plotted per group, each groups' line colors are the same,
#####  and the symbol colors are nearly the same as the line colors....but not
exactly!
> xyplot(Y ~ X | Panel, data=dat, groups=ID, type='b',
+        scales=list(x=list(at=c(0,1),labels=c(0,1))),
+        panel=function(..., groups, subscripts) panel.xyplot(...,
col.symbol=dat$Color[subscripts],
+                                                                               
col.line=dat$Color[subscripts])
+        )

At one point in my efforts I was actually able to get the symbol colors correct,
but the line colors were (to me) incomprehensibly wrong.  But alas I have not be
able to reproduce that to show here.

\begin{ignorant speculation alert}
I suspect that having (in the example) 8 points but only 4 lines causes
undesired recycling somewhere.
\end{speculation}

Any assistance as to how to properly use the panel functions (or any other
approach short of abandoning lattice graphics) to override the coloring would be
greatly appreciated.

John

John  Szumiloski,  Ph.D.
Senior Biometrician
Biometrics Research
WP53B-120
Merck Research Laboratories
P.O. Box 0004
West Point, PA 19486-0004
USA
(215) 652-7346 (PH)
(215) 993-1835 (FAX)
john<dot>szumiloski<at>merck<dot>com
___________________________________________________
These opinions are my own and do not necessarily reflect that of
Merck & Co., Inc.





Notice:  This e-mail message, together with any attachme...{{dropped:14}}

On Mon, Apr 11, 2011 at 9:27 PM, Szumiloski, John
<john_szumiloski at merck.com> wrote:
> Dear useRs,
>
> I have a longitudinal experiment with several treatment groups, ~20
subjects per group, ~6 timepoints and a continuous dependent variable. ?I have
been successfully been using lattice::xyplot with this data. ?However, I have
been stumped with a particular application of it.
>
> I would like to use xyplot on my data, broken into treatment groups with
the groups argument, using type='b' to show subjectwise longitudinal
data. ?So far so good, I have done this many times. ?But now I wish to show the
same data but having the color of the lines and symbols overridden in some
arbitrary way, yet not without changing anything else about the plot, in
particular the structure/topology of the plot from using the groups argument and
type='b'.
>
> This requires using a panel function of some sort.
No, it does not. You are making things unnecessarily complicated.

When you use 'groups', colors are associated with levels of the
grouping variable, not individual rows of your data. So:

set.seed(388659262)
dat <- data.frame(Panel=rep(c('A','B'), each=4),
                  ID=factor(rep(letters[1:4], each=2)),
                  X=rep(c(0,1), times=4),
                  Y=runif(8))

##### now for the arbitrary colors.  Let's highlight one subject red,
the rest black

myColor <- ifelse(levels(dat$ID) == "a", 2, 1)

xyplot(Y ~ X | Panel, data=dat, groups=ID,
       type='b', col = myColor,
       scales=list(x=list(at=c(0,1),labels=c(0,1))))

-Deepayan