How to draw arrows across panels of multi-panel plot? - r

Question

Suppose we have a multi-panel plot in R, created by using layout(). I would like to draw an arrow from a specified point in one panel to a specified point in another panel. Thus, the arrow goes across panels of the layout. The starting point of the arrow is specified in the coordinates of its panel, and the end point of the arrow is specified in the coordinates of the destination panel.
As a minimal example, consider this:
layout( matrix( 1:2 , nrow=2 ) )
plot( x=c(1,2) , y=c(1,2) , main="Plot 1" )
plot( x=c(10,20) , y=c(10,20) , main="Plot 2" )
# I want to make an arrow
# from point c(x=1.2,y=1.2) in Plot 1
# to point c(x=18,y=18) in Plot 2
I've searched for methods to accomplish this, but haven't found anything. Thank you for solutions or pointers.

Answer 1

Update
(I'm keeping the previous answer below this, but this more-programmatic way is better given your comments.)
The trick is knowing how to convert from "user" coordinates to the coordinates of the overarching device. This can be done with grconvertX and *Y. I've made some sloppy helper functions here, though they are barely necessary.
user2ndc <- function(x, y) {
list(x = grconvertX(x, 'user', 'ndc'),
y = grconvertY(y, 'user', 'ndc'))
}
ndc2user <- function(x, y) {
list(x = grconvertX(x, 'ndc', 'user'),
y = grconvertY(y, 'ndc', 'user'))
}
For the sake of keeping magic-constants out of the code, I'll predefine your points-of-interest:
pointfrom <- list(x = 1.2, y = 1.2)
pointto <- list(x = 18, y = 18)
It's important that the conversion from 'user' to 'ndc' happen while the plot is still current; once you switch from plot 1 to 2, the coordinates change.
layout( matrix( 1:2 , nrow=2 ) )
Plot 1.
plot( x=c(1,2) , y=c(1,2) , main="Plot 1" )
points(y~x, data=pointfrom, pch=16, col='red')
ndcfrom <- with(pointfrom, user2ndc(x, y))
Plot 2.
plot( x=c(10,20) , y=c(10,20) , main="Plot 2" )
points(y~x, data=pointto, pch=16, col='red')
ndcto <- with(pointto, user2ndc(x, y))
As I did before (far below here), I remap the region on which the next plotting commands will take place. Under the hood, layout is doing things like this. (Some neat tricks can be done with par(fig=..., new=T), including overlaying one plot in, around, or barely-overlapping another.)
par(fig=c(0:1,0:1), new=TRUE)
plot.new()
newpoints <- ndc2user(c(ndcfrom$x, ndcto$x), c(ndcfrom$y, ndcto$y))
with(newpoints, arrows(x[1], y[1], x[2], y[2], col='green', lwd=2))
I might have been able to avoid the ndc2user conversino from ndc back to current user points, but that's playing with margins and axis-expansion and things like that, so I opted not to.
It is possible that the translated points may be outside of the user-points region of this last overlaid plot, in which case they may be masked. To fix this, add xpd=NA to arrows (or in a par(xpd=NA) before it).
Generalized
Okay, so imagine you want to be able to determine the coordinates of any drawing after layout completion. There's a more complex implementation that currently supports what you're asking for. the only requirement is that you call NDC$add() after every (meaningful) plot. For example:
NDC$reset()
layout(matrix(1:4, nrow=2))
plot(1)
NDC$add()
plot(11)
NDC$add()
plot(21)
NDC$add()
plot(31)
NDC$add()
with(NDC$convert(1:4, c(1,1,1,1), c(1,11,21,31)), {
arrows(x[1], y[1], x[2], y[2], xpd=NA, col='red')
arrows(x[2], y[2], x[3], y[3], xpd=NA, col='blue')
arrows(x[3], y[3], x[4], y[4], xpd=NA, col='green')
})
Source can be found here: https://gist.github.com/r2evans/8a8ba8fff060bade13bf21e89f0616c5
Previous Answer
One way is to use par(fig=...,new=TRUE), but it does not preserve the coordinates you e
layout(matrix(1:4,nr=2))
plot(1)
plot(1)
plot(1)
plot(1)
par(fig=c(0,1,0,1),new=TRUE)
plot.new()
lines(c(0.25,0.75),c(0.25,0.75),col='blue',lwd=2)
Since you may be more likely to use this if you have better (non-arbitrary) control over the ends of the points, here's a trick to allow you more control over the points. If I use this, connectiong the top-left point with the bottom-right point:
p <- locator(2)
str(p)
# List of 2
# $ x: num [1:2] 0.181 0.819
# $ y: num [1:2] 0.9738 0.0265
and then in place of lines above I use this:
with(p, arrows(x[1], y[1], x[2], y[2], col='green', lwd=2))
I get
(This picture and the values in p demonstrate how the coordinates are different. When using par(fig=...,new=T);plot.new();, the coordinates return to
par('usr')
# [1] -0.04 1.04 -0.04 1.04
There might be trickery to try to workaround this (such as if you need to automate this step), but it likely will be non-trivial (and not robust).