Almost as soon as you start using R you hear about Sweave (probably because it is included with the base install of R) which allows you to embed R code within LaTeX documents. However, I placed learning it on the back-burner for about a year so I could just spend time getting my head around R (although how I’ve managed to avoid LaTeX until now, considering my background and the amount of technical reports I’ve written over the years, is another matter).

Then, as soon as you start using Sweave you come across the issue of plot fonts and sizes not matching the main report (personally I can live with the different font style, but the different sizes is annoying). And so that’s when you hear about pgfSweave.

However, I was using Lattice graphics built-in capability to produce multiple pages of plots (see ‘layout’):

The number of pages is by default set to as many as is required to plot all the panels, and so rarely needs to be specified. However, in certain situations the default calculation may be incorrect, and in that case the number of pages needs to be specified explicitly.

In Sweave, the way you would use one code chunk to produce multiple figures is to use a tex chunk rather than a fig:

%Traditional Sweave Loop
%Sweave file extract

<<eval=TRUE, echo=FALSE, results=hide>>=
#Cleanup - delete previous files
file.remove(list.files(pattern = glob2rx("ybygroup*.png")))
#No need to use a loop here, lattice graphics will produce the necessary number of pngs
png(file="ybygroup%03d.png", width=1000, height=2000)
xyplot(Y ~ Date | Grouping, datatable, layout=c(1,8), ylim=c(-100,100), xlim=as.Date(c("2011-01-01", "2011-07-01")) )
#This echos when using PNGs, so put in it's own block and hide.
dev.off()
@


<<results=tex, echo=FALSE>>=
files <- list.files(pattern = glob2rx("ybygroup*.png"))
for(file in files){
    cat("\\begin{figure}")
    cat("\\begin{center}")
    cat("\\includegraphics{", file, "}\n\n", sep="")
    cat("\\end{center}")
    cat("\\end{figure}")
}
@

Link to gist

I still haven’t fully got my head round pgfSweave and Tikz, but for individual plots you indicate the use of Tikz within the figure:

%Reference for pgfSweave / Tikz option
%Sweave file extract

\begin{figure}
<<fig=TRUE, tikz=T>>=
print(
	xyplot(Y ~ Date | Grouping, datatable[levels(datatable$Grouping)[1:8]], layout=c(1,8), ylim=c(-100,100), xlim=as.Date(c("2011-01-01", "2011-07-01")) )
)
@
\end{figure}

But I wanted to be able to produce all plots via a loop as per Sweave, rather than have to manually set up enough fig chunks so I came up with the following:

%Tikz loop
%Sweave file extract

\usepackage{tikz}
%...

<<results=tex, echo=FALSE>>=
#Need to know how many levels
nl <- length(levels(datatable$Grouping))
#then how many per graph
ng = 8
#Can then loop

for (i in 1:ceiling(nl/ng)) {
    cat("\\begin{figure}\n")
    cat("\\begin{center}\n")
    tikz(console=TRUE)
    start=(i-1)*ng+1
    end=i*ng
    print(
        xyplot(Y ~ Date | Grouping, datatable[levels(datatable$Grouping)[start:end]], layout=c(1,ng), ylim=c(-100,100), xlim=as.Date(c("2011-01-01", "2011-07-01")) )
	)
    dev.off()
    cat("\\end{center}\n")
    cat("\\end{figure}\n")
}

@

All of this looks and seems amazingly obvious as I write it up now, but it wasn’t at the time. Especially because I came across this weird bug (which, thinking about it now, might not be a bug, rather a nuance of the data.table package - I need to investigate further):

works:

#Extract of R code
#This works as expected

#Need to know how many levels
nl <- length(levels(datatable$Grouping))
#then how many per graph
ng = 8
#Can then loop

for (i in 1:ceiling(nl/ng)) {
    start=(i-1)*ng+1
    end=i*ng
    print(
        xyplot(Y ~ Date | Grouping, datatable[levels(datatable$Grouping)[start:end]], layout=c(1,ng), ylim=c(-100,100), xlim=as.Date(c("2011-01-01", "2011-07-01")) )
	)
}

Link to gist

doesn’t work:

#Extract of R code
#This doesn't work properly

#Need to know how many levels
nl <- length(levels(datatable$Grouping))
#then how many per graph
ng = 8
#Can then loop

for (i in 1:ceiling(nl/ng)) {
    print(
        xyplot(Y ~ Date | Grouping, datatable[levels(datatable$Grouping)[(i-1)*ng+1:(i*ng)]], layout=c(1,ng), ylim=c(-100,100), xlim=as.Date(c("2011-01-01", "2011-07-01")) )
	)
}

#Instead of using the start and end variables within the loop, the values are calculated directly
#as applied to the datatable. For some reason, from the 2nd loop onwards, `i*ng` seems to be 
#evaluated as `(i+1)*ng` which means Lattice automatically produces two pages of graphs, instead
#of just the one page per loop that it should. 

#This is probably just a data.table nuance actually, not a bug.

Since this just uses Tikz within a tex chunk, in theory pgfSweave isn’t actually required to process this and just plain Sweave would do, in practice however, since pgfSweave can externalise the graphics by default, it will avoid text memory capacity issues with a large number of loops, where Sweave will run into them.

When I got the Tikz solution working, I actually decided to go back to the plain Sweave way of doing things as for me it was much faster (producing Lattice xyplots, especially with lots of data points, is slow via Tikz) and using the \includegraphics approach, the pngs scale better and you don’t run into overfull \hbox issues. You just need to tweak the text sizes for the plots to look ok.

Since this, I’ve also come across a post which uses a latex loop (which I thought must be possible, but wasn’t clever enough to work out), but then also goes on to mention Brew which looks interesting… and easier.