Martin Maechler wrote:
Hallo Sebastian,
"SP" == Sebastian Pölsterl <s...@k-d-w.org>
on Sun, 14 Jun 2009 14:04:52 +0200 writes:
SP> Hello Martin,
SP> I plotting the silhouette of a clustering and storing it as png. When I
SP> try to store the image as png the bars are missing. The bars are plotted
SP> when I use x11 or postscript as device. In addition, it seems to work
SP> when I use a smaller matrix (e.g. ruspini).
SP> Would be great if you have look at this issue.
Hmm, I've been at a conference in Italy...
The silhouette plot only uses standard R plotting functions,
so any problem with it exposes problems in standard R
graphics.
--> Such a message should really go to R-help.
to which I CC now.
----------
library(cluster)
nmat <- matrix(rnorm(2500*300), ncol=300, nrow=2500)
rmat <- matrix(rchisq(1000, 300, 50), ncol=300, nrow=1000)
mat <- rbind(nmat, rmat)
pr <- pam(mat, 2)
sil <- silhouette(pr)
png("sil.png")
#postscript("sil.ps")
plot(sil)
dev.off()
----------
Anyway, I can confirm the "problem",
but of course, it has not much to do with the silhouette
function, but rather with the png() device which produces a
bitmap, and the lines you draw are too fine (in the bitmap
resolution) and so are "rounded to invisible".
You can reproduce the problem much more simply:
set.seed(1); x <- rlnorm(5000)
png("bar.png");barplot(x,col="gray",border=0,horiz=TRUE);dev.off()
system("eog bar.png &")
## which is also empty, and the completely analogue, replacing
## png [bitmap] with pdf [vector graphic]
pdf("bar.pdf");barplot(x,col="gray",border=0,horiz=TRUE);dev.off()
system("evince bar.pdf &")
## gives a very nice plot, into which you can zoom and see all details.
----------------
Now in principle you should be able to use png() with a much
higher resolution than the default one,
but replacing the above
png("bar.bng")
with
png("bar.bng", res = 1200)
did not help, as we now get the infamous
Error in plot.new() : figure margins too large
Other R-help readers will be able to make the png() example work
for such cases, where you need so many lines.
{but let's stick with barplot(*, border=0, *)}
Well, it's pretty hard, but one of the few cases where conversion from
some vector image to the bitmap produces fair results, hence I'd use
bitmap() with ghostscript to generate the png as in:
bitmap("sil.png", type="pnggray", res=300)
plot(sil)
dev.off()
Best,
Uwe
Regards,
Martin Maechler, ETH Zurich
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
