Sweave data-figure coupling
On Mon, Jan 05, 2009 at 01:52:00PM -0600, Sebastian P. Luque wrote:
<<binom-sim-fig, fig=TRUE, include=FALSE, height=3, echo=FALSE>>=
layout(matrix(1:2, ncol=2)); par(mar=c(5, 4, 2, 1), cex=0.75)
matplot(thetas, cbind(prior, lik, post), type="l", lty=c(2, 1, 1),
xlab="theta", ylab="probability density")
lik <- dbinom(60, 100, thetas)
lik.p <- prior * lik
post <- lik.p / sum(lik.p)
matplot(thetas, cbind(prior, lik, post), type="l", lty=c(2, 1, 1),
xlab="theta", ylab="probability density")
@
\includegraphics[width=\linewidth]{test-binom-sim-fig}
\end{document}
---<--------------------cut here---------------end--------------------->---
If the embedded chunks are evaluated directly in an R session, two
different panel data are produced by layout(), as expected. However,
when weaving and latexing the file produces, then the two panels show
the same data, corresponding to the last data set. This doesn't happen
if the data are much simpler, say using matplot() with simple sequences
of numbers. Any ideas what is causing this and how to get around it?
I was very confused, too when I first had this happen to me. The reason is that chunks with fig=TRUE are executed three times by default: once for the eps file, the pdf and the default device. Accordingly, whenever you modify data in such a code chunk, the results change with each round of execution. The solution to the problem is either - separating the manipulations from the plotting in two code chunks - or keeping the modified values in a separate variable instead of modifying the original. In your example you are modifying 'lik' and 'post'. Just call the modified versions likMod and postMod or something like this and your problem will go away. cu Philipp
Dr. Philipp Pagel Lehrstuhl f?r Genomorientierte Bioinformatik Technische Universit?t M?nchen Wissenschaftszentrum Weihenstephan 85350 Freising, Germany http://mips.gsf.de/staff/pagel