The R Project SVN R

\documentclass[a4paper]{article}
%\VignetteIndexEntry{Rotated Viewports}
%\VignettePackage{grid}
\newcommand{\grid}{{\tt grid}}
\newcommand{\R}{{\tt R}}
\setlength{\parindent}{0in}
\setlength{\parskip}{.1in}
\setlength{\textwidth}{140mm}
\setlength{\oddsidemargin}{10mm}
\title{Rotated Viewports}
\author{Paul Murrell}

\begin{document}
\maketitle

<<echo=FALSE, results=hide>>=
library(grDevices)
library(graphics) # for boxplot
library(stats) # for rnorm
library(grid)
ps.options(pointsize=12)
options(width=60)

@
It is possible to specify an angle of rotation for a Grid viewport.
For example, the following code draws the example multipanel plot
at an angle of $15^\circ$.

<<fig=TRUE, results=hide>>=
pushViewport(viewport(h=.8, w=.8, angle=15))
grid.multipanel(newpage=FALSE)
popViewport()

@

\vspace{.5in}
A more complicated example is now developed.  First of all we
generate some data to plot; an $x$ and a $y$ with an obvious correlation.

<<complex1, results=hide>>=
x <- rnorm(50)
y <- x + rnorm(50, 1, 2)

@
\noindent
Next we generate some statistics from the data.

<<complex2, results=hide>>=
# We will extend the axes over the entire region so
# extrapolate scale from main data region
scale <- extendrange(r = range(x,y))
extscale <- c(min(scale), max(scale)+diff(scale)*1/3)

@
\noindent
Now generate a layout of regions:  a 3" by 3" region for a
scatterplot, inside a 4" by 4" region.

<<complex3, results=hide>>=
lay <- grid.layout(2, 2,
                   widths=unit(c(3, 1), "inches"),
                   heights=unit(c(1, 3), "inches"))
vp1 <- viewport(width=unit(4, "inches"), height=unit(4, "inches"),
                layout=lay,
                xscale=extscale, yscale=extscale)

@
\noindent
We draw a box around the outside and axes on the entire region.

<<complex4, results=hide>>=
grid.newpage()
pushViewport(vp1)
grid.rect()
grid.xaxis()
grid.text("Test", y=unit(-3, "lines"))
grid.yaxis()
grid.text("Retest", x=unit(-3, "lines"), rot=90)

@
\noindent
We draw points within the interior region.

<<complex5, results=hide>>=
vp2 <- viewport(layout.pos.row=2, layout.pos.col=1,
                xscale=scale, yscale=scale)
pushViewport(vp2)
grid.lines()
grid.points(x, y, gp=gpar(col="blue"))
popViewport()

@
\noindent
Now we use a rotated viewport to draw a boxplot which indicates
the distribution of the distances between the points in the scatterplot
and the line $y = x$\footnote{This may look like a large amount of
code, but that's mostly because its doing a boxplot by hand
rather than using a predefined high-level function.}.

The final output is shown on the last page.

<<complex6, results=hide>>=
diffs <- (y - x)
rdiffs <- range(diffs)
ddiffs <- diff(rdiffs)
bxp <- boxplot(diffs, plot=FALSE)
vp3 <- viewport(x=unit(3, "inches"),
                y=unit(3, "inches"),
                width=unit(.5, "inches"),
                # NOTE that the axis on the boxplot represents
                # actual (y - x) values BUT to make
                # the bits of the boxplot line
                # up with the data points we have to plot
                # (y - x)/sqrt(2)
                # Hence the sin(pi/4) below
                height=unit(ddiffs*sin(pi/4)/diff(scale)*3, "inches"),
                just=c("centre", "center"),
                angle=45,
            gp=gpar(col="red"),
                yscale=c(-ddiffs/2, ddiffs/2))
pushViewport(vp3)
left <- -.3
width <- .8
middle <- left + width/2
grid.rect(x=left, y=unit(bxp$conf[1,1], "native"),
          width=width, height=unit(diff(bxp$conf[,1]), "native"),
          just=c("left", "bottom"),
      gp=gpar(col=NULL, fill="orange"))
grid.rect(x=left, y=unit(bxp$stats[4,1], "native"),
          width=width, height=unit(diff(bxp$stats[4:3,1]), "native"),
          just=c("left", "bottom"))
grid.rect(x=left, y=unit(bxp$stats[3,1], "native"),
          width=width, height=unit(diff(bxp$stats[3:2,1]), "native"),
          just=c("left", "bottom"))
grid.lines(x=c(middle, middle), y=unit(bxp$stats[1:2,1], "native"))
grid.lines(x=c(middle, middle), y=unit(bxp$stats[4:5,1], "native"))
grid.lines(x=c(middle-.1, middle+.1), y=unit(bxp$stats[1,1], "native"))
grid.lines(x=c(middle-.1, middle+.1), y=unit(bxp$stats[5,1], "native"))
np <- length(bxp$out)
if (np > 0)
  grid.points(x=rep(middle, np), y=unit(bxp$out, "native"))
grid.yaxis(main=FALSE)
popViewport(2)

<<echo=FALSE, fig=TRUE, results=hide>>=
<<complex1>>
<<complex2>>
<<complex3>>
<<complex4>>
<<complex5>>
<<complex6>>

@
\section*{Problems}

\begin{enumerate}
\item Data symbols will not be affected by the angle of rotation.
For round data symbols this does not matter, but it will make
just about everything else look pretty odd.
\end{enumerate}

@
% Start a new page
% Not echoed, not evaluated
% ONLY here for checkVignettes so that all output doesn't
% end up on one enormous page

<<eval=FALSE, echo=FALSE>>=
grid.newpage()

@
\end{document}