[R] Howto? plot legend with no line behind the points

[Duncan Murdoch]

On 11/21/2005 10:33 AM, Martin Maechler wrote:
>>>>>> "Jan" == Jan Verbesselt <Jan.Verbesselt at biw.kuleuven.be>
>>>>>>     on Mon, 21 Nov 2005 12:28:35 +0100 writes:
> 
>     Jan> Hi R-help,
>     Jan> We are using R 2.2 on Win XP and have a detail question on how the
>     Jan> legend can be optimised.
> 
>     Jan> We use the following;
>     -> plot(,type="b",...)
>     Jan> The lines in the plot do not cross the points. How can we obtain the
>     Jan> same effect in the legend? (points without a line through them..)
> 
> not directly, however you can trick it, see below
> 
>     Jan> We tried setting the pt.bg to white but this did not help.
> 
> I think it does help some.. read on
> 
>     Jan> See script below.
> 
>      par(mar=c(5,5,2,4) +.1) # dit om extra text in Y-as te plaatsen
>      plot(ts.X, type="b", col=1, pch=19,ylim=c(-0.4,0.2),ylab=c("X"))
>      legend.txt <- c("X","Y")
>      # Define how the legend looks like, place it on the right location
>      legend("topright", legend.txt,col=1,lty=1, pch=c(19,1),bty="n", pt.bg=1)
>      par(new=T)
>      plot(ts.Y, type="b", lty=1, col=1, pch=1, ylab="",
>      xlab="",yaxt="n",ylim=c(0.1,0.7))
>      axis(4)
>      mtext(side=4, line=3, "Y", cex=1)
> 
> 
> Unfortunately you don't give a reproducible example {we don't
> have your ts.X and ts.Y}, so I make up a simpler version of
> the above:
> 
> ## A bivariate time-series
> set.seed(1)
> xy <-  cbind(x = ts(cumsum(rnorm(47))), y = ts(cumsum(rt(47, df=3))))
> 
> plot(xy, plot.type = "single", type = "b", pch = 21, col=1:2, bg = "light blue")
> legend("topright", c("x","y"), col=1:2, lty="47", 
>        pch = 21, bty="n", pt.bg="light blue")
> 
> 
> I'm using pch=21 and I've used pt.bg just to show it's effect
> here; of course you can play with these as well.
> 
> And the real trick was to "play with" 'lty'  using the nice  "on/off"
> convention ("47": a dash of length 4; a break of length 7).
> 
> Now of course, we also might accept patches for improving
> legend (source at 
>  https://svn.R-project.org/R/trunk/src/library/graphics/R/legend.R ).
> 
> Naturally, I think it's the argument  'merge' that
> could be extended to allow something analogous to plot type = "b".

Are there any examples with "merge = FALSE" that look good?  It seems to 
me that it would be a cleaner change to add a new parameter "type" that 
did what it does in plot(), and then eventually drop merge.  If there 
really are examples with merge = FALSE that people want to keep, then we 
could keep it, but I still think adding "type" would be better than 
fiddling with merge.

Of course, this suggestion isn't trivial to implement for "type = 'b'", 
where we'd want

    -- * --

in the legend, but I think it could be done with some sort of trickery 
involving 3 points with "type = 'c'" overlaid with one point with "type 
= 'p'".

Duncan Murdoch