R help - Aug 2010 - lattice::xyplot() with one factor for points and another for lines

Hi:

In lattice, how does one handle separate graphical behavior for two
different factors? In the xyplot below, the objective is to use the levels
of one factor to distinguish corresponding shapes and colors, and the levels
of the other factor to perform level-wise loess smooths.

# Illustrative data:
d <- data.frame(time = rep(1:8, each = 6), val = rnorm(48),
                gp1 = factor(rep(1:6, 8)),
                gp2 = factor(rep(rep(c('A', 'B'), each = 3),
8)))

Based on the code from the Lattice book, p. 160, I set up the following:

mypch <- 1:6
mycol <- 1:6

with(d,
xyplot(val ~ time,
       panel = function(x, y, ..., groups, subscripts) {
                 pch <- mypch[gp1[subscripts]]
                 col <- mycol[gp1[subscripts]]
                 grp <- gp2
                 panel.xyplot(x, y, pch = pch, col = col)
                 panel.loess(x, y, groups = grp, lty = 1,
                             col.line = c('blue', 'red'))
          }
      )  )

As stated in the book, the with() wrapper allows one to use variable names
within the panel function. I was hoping to get away with using both groups
and subscripts in the panel function and fake my way through, but no luck. I
get 90% of what I want: the points are plotted correctly, but only one loess
line shows up instead of two. I put groups before the ellipsis in the
argument list, defined groups = gp2 in the function call and a couple of
other things, but got the same result.

I also tried to put the groups argument at the level of the xyplot() call,
but as expected, it plots only the points and skips the loess curves
altogether (but does tell you something about how the panel function
behaves):

# Vain, desperate, naive attempt:
with(d,
xyplot(val ~ time, groups = gp2, type = c('p', 'smooth'),
        col.lines = c('red', 'blue'), lty = 1,
       panel = function(x, y, ..., groups = gp2, subscripts) {
                 pch <- mypch[gp1[subscripts]]
                 col <- mycol[gp1[subscripts]]
                 panel.xyplot(x, y, pch = pch, col = col)
          } )   )

Adding panel.loess(x, y, groups = groups, ...) leaves the same single line.

What's the correct mantra? I'm probably missing something fairly
obvious,
but it escapes me.  [And before Hadley chimes in, I already have a working
version in ggplot2, but the person I'm trying to help wants it in lattice,
too.]

TIA for your assistance,
Dennis

	[[alternative HTML version deleted]]

On Sat, Aug 21, 2010 at 5:58 AM, Dennis Murphy <djmuser at gmail.com>
wrote:
> Hi:
>
> In lattice, how does one handle separate graphical behavior for two
> different factors? In the xyplot below, the objective is to use the levels
> of one factor to distinguish corresponding shapes and colors, and the
levels
> of the other factor to perform level-wise loess smooths.
>
> # Illustrative data:
> d <- data.frame(time = rep(1:8, each = 6), val = rnorm(48),
> ? ? ? ? ? ? ? ?gp1 = factor(rep(1:6, 8)),
> ? ? ? ? ? ? ? ?gp2 = factor(rep(rep(c('A', 'B'), each = 3),
8)))
>
> Based on the code from the Lattice book, p. 160, I set up the following:
>
> mypch <- 1:6
> mycol <- 1:6
>
> with(d,
> xyplot(val ~ time,
> ? ? ? panel = function(x, y, ..., groups, subscripts) {
> ? ? ? ? ? ? ? ? pch <- mypch[gp1[subscripts]]
> ? ? ? ? ? ? ? ? col <- mycol[gp1[subscripts]]
> ? ? ? ? ? ? ? ? grp <- gp2
> ? ? ? ? ? ? ? ? panel.xyplot(x, y, pch = pch, col = col)
> ? ? ? ? ? ? ? ? panel.loess(x, y, groups = grp, lty = 1,
> ? ? ? ? ? ? ? ? ? ? ? ? ? ? col.line = c('blue', 'red'))
> ? ? ? ? ?}
> ? ? ?) ?)
>
> As stated in the book, the with() wrapper allows one to use variable names
> within the panel function. I was hoping to get away with using both groups
> and subscripts in the panel function and fake my way through, but no luck.
I
> get 90% of what I want: the points are plotted correctly, but only one
loess
> line shows up instead of two. I put groups before the ellipsis in the
> argument list, defined groups = gp2 in the function call and a couple of
> other things, but got the same result.
You are on the right track, except that panel.loess is not
"groups-aware".

I would do something like

with(d,
     xyplot(val ~ time,
            pch = mypch, col = mycol,
            lty = 1, col.line = c('blue', 'red'),
            subscripts = TRUE,
            panel = function(x, y, ..., groups) {
                panel.superpose(x, y, ..., groups = gp1, panel = panel.points)
                panel.superpose(x, y, ..., groups = gp2, panel = panel.loess)
            }))

The first call inside the panel function could be simpler, but this
makes the intention clearer.

-Deepayan