Lattice problems: lattice objects coming from JAGS, but device can't be set

I ran JAGS with runjags in R and I开发者_JAVA技巧 got a giant list back (named results for this example).

Whenever I access results$density, two lattice plots (one for each parameter) pop up in the default quartz device.

I need to combine these with par(mfrow=c(2, 1)) or with a similar approach, and send them to the pdf device.

Nothing I tried is working. Any ideas?

I've tried dev.print, pdf() with dev.off(), etc. with no luck.

Here's a way to ditch the "V1" panels by manipulation of the Trellis structure:

p1 <- results$density$c
p2 <- results$density$m

p1$layout <- c(1,1)
p1$index.cond[[1]] <- 1   # remove second index
p1$condlevels[[1]] <- "c"   # remove "V1"
class(p1) <- "trellis"   # overwrite class "plotindpages"

p2$layout <- c(1,1)
p2$index.cond[[1]] <- 1   # remove second index
p2$condlevels[[1]] <- "m"   # remove "V1"
class(p2) <- "trellis"   # overwrite class "plotindpages"

library(grid)
layout <- grid.layout(2, 1, heights=unit(c(1, 1), c("null", "null")))
grid.newpage()
pushViewport(viewport(layout=layout))
pushViewport(viewport(layout.pos.row=1))
print(p1, newpage=FALSE)
popViewport()
pushViewport(viewport(layout.pos.row=2))
print(p2, newpage=FALSE)
popViewport()
popViewport()

pot of c.trellis() result http://img142.imageshack.us/img142/3272/ctrellisa.png

The easiest way to combine the plots is to use the results stored in results$mcmc:

# prepare data, see source code of "run.jags"
thinned.mcmc <- combine.mcmc(list(results$mcmc),
                             collapse.chains=FALSE,
                             return.samples=1000)
print(densityplot(thinned.mcmc[,c(1,2)], layout=c(1,2),
                  ylab="Density", xlab="Value"))

For instance, for the included example from run.jags, check the structure of the list using

sink("results_str.txt")
str(results$density)
sink()

Then you will see components named layout. The layout for the two plots of each variable can be set using

results$density$m$layout <- c(1,2)
print(results$density$m)

The plots for different parameters can be combined using the c.trellis method from the latticeExtra package.

class(results$density$m) <- "trellis"   # overwrite class "plotindpages"
class(results$density$c) <- "trellis"   # overwrite class "plotindpages"
library("latticeExtra")
update(c(results$density$m, results$density$c), layout=c(2,2))

output of c.trellis http://img88.imageshack.us/img88/6481/ctrellis.png

Another approach is to use grid viewports:

library("grid")
results$density$m$layout <- c(2,1)
results$density$c$layout <- c(2,1)
class(results$density$m) <- "trellis"
class(results$density$c) <- "trellis"
layout <- grid.layout(2, 1, heights=unit(c(1, 1), c("null", "null")))
grid.newpage()
pushViewport(viewport(layout=layout))
pushViewport(viewport(layout.pos.row=1))
print(results$density$m, newpage=FALSE)
popViewport()
pushViewport(viewport(layout.pos.row=2))
print(results$density$c, newpage=FALSE)
popViewport()
popViewport()

grid output http://img88.imageshack.us/img88/5967/grida.png