R help - Oct 2008 - histogram loses top row with alpha transparency under Windows

Hello all. 

Trying to use transparency for overlaid "histogram" plots I have come
across an interesting inconsistency, possibly a bug when running under
Windows. Originally noticed in R 2.7.1, it is still there in 2.8.0 beta.

library(lattice)
zz <- function(n,alpha)
{
  ranges <- NULL
  for(ds in 1:n){
    ranges <-
rbind(ranges,data.frame(confidence=c(0,100),dataset=as.character(ds),cor
rect=c(FALSE,TRUE)))
  }
  panel.colhist = function(x, group.number, col, ...) {
    panel.histogram(x, col=group.number+1, ...)
  }
  x <- histogram(~confidence|dataset,data=ranges,alpha=alpha,
 
panel=panel.superpose,panel.groups=panel.colhist,groups=correct)
  print(x)
}
zz(12,1) # works as expected, 12 identical plots
zz(12,0.5) # top row of plots has no bars at all, lower rows are as
expected
zz(1,1) # two bars fine
zz(1,0.5) # no bars at all

(I also find the default breaks slightly counter-intuitive here, as the
number of breaks seems to depend on the total size of the data, rather
than on the size within each plot - the bars meet in the zz(1,1) case
but are well separated in the zz(12,1) case.)

I am running on 64-bit Windows Vista Business SP1, session info is as
follows:

R version 2.8.0 beta (2008-10-07 r46631) 
i386-pc-mingw32 

locale:
LC_COLLATE=English_United Kingdom.1252;LC_CTYPE=English_United
Kingdom.1252;LC_MONETARY=English_United
Kingdom.1252;LC_NUMERIC=C;LC_TIME=English_United Kingdom.1252

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base


other attached packages:
[1] lattice_0.17-15

loaded via a namespace (and not attached):
[1] grid_2.8.0

It appears to work correctly under (Suse) linux, with the following
session characteristics:

R version 2.7.2 (2008-08-25) 
i686-pc-linux-gnu 

locale:
LC_CTYPE=en_GB.UTF-8;LC_NUMERIC=C;LC_TIME=en_GB.UTF-8;LC_COLLATE=en_GB.U
TF-8;LC_MONETARY=C;LC_MESSAGES=en_GB.UTF-8;LC_PAPER=en_GB.UTF-8;LC_NAMEC;LC_ADDRESS=C;LC_TELEPHONE=C;LC_MEASUREMENT=en_GB.UTF-8;LC_IDENTIFICATI
ON=C

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base


other attached packages:
[1] lattice_0.17-13

loaded via a namespace (and not attached):
[1] grid_2.7.2

                           sysname                            release 
                           "Linux"           
"2.6.22.18-0.2-default"
                           version                           nodename 
"#1 SMP 2008-06-09 13:53:20 +0200"                          
"node76"
                           machine                              login 
                            "i686"                        
"kponting"
                              user 
                        "kponting"



Keith Ponting
Aurix Ltd, Malvern WR14 3SZ  UK

k.ponting wrote:
> 
> Hello all. 
> 
> Trying to use transparency for overlaid "histogram" plots I have
come
> across an interesting inconsistency, possibly a bug when running under
> Windows. Originally noticed in R 2.7.1, it is still there in 2.8.0 beta.
> 
> library(lattice)
> zz <- function(n,alpha)
> {
>   ranges <- NULL
>   for(ds in 1:n){
>     ranges <-
> rbind(ranges,data.frame(confidence=c(0,100),dataset=as.character(ds),cor
> rect=c(FALSE,TRUE)))
>   }
>   panel.colhist = function(x, group.number, col, ...) {
>     panel.histogram(x, col=group.number+1, ...)
>   }
>   x <- histogram(~confidence|dataset,data=ranges,alpha=alpha,
>  
> panel=panel.superpose,panel.groups=panel.colhist,groups=correct)
>   print(x)
> }
> zz(12,1) # works as expected, 12 identical plots
> zz(12,0.5) # top row of plots has no bars at all, lower rows are as
> expected
> zz(1,1) # two bars fine
> zz(1,0.5) # no bars at all
> 
The rectangles being drawn extend higher than the top of the panel.  (Your y
axis ranges from 0 to 50, but the bars go up to 100.)

In the top row of plots, depending upon the shape of your device window, the
bars can extend beyond the range of the device window.  For some reason,
(take a look in panel.rect), when you specify alpha less than 1, this
prevents the bar being drawn.  

You need to add type="count" to the call to histogram, or rescale the
bar
heights.


-----
Regards,
Richie.

Mathematical Sciences Unit
HSL
-- 
View this message in context:
http://www.nabble.com/histogram-loses-top-row-with-alpha-transparency-under-Windows-tp19879687p19897501.html
Sent from the R help mailing list archive at Nabble.com.

Richard.Cotton wrote:
> The rectangles being drawn extend higher than the top of the panel.
(Your y
> axis ranges from 0 to 50, but the bars go up to 100.)
> 
Thankyou - I can also make the bars on the lower panels vanish by
tinkering with ylim.
> In the top row of plots, depending upon the shape of your device
window, the
> bars can extend beyond the range of the device window.  For some
reason,
> (take a look in panel.rect), when you specify alpha less than 1, this
> prevents the bar being drawn.  
The problem lies deeper. I can demonstrate the same effect with raw grid
calls (but cannot find a way into "grid.draw" to see what is going
on):

cols <- rainbow(2,alpha=0.5)
library(grid)
grid.newpage()
pushViewport(plotViewport(c(5,4,2,2)))
# draw viewport limits:
grid.rect() 
# VISIBLE: small projection outside viewport, transparent fill and edge
grid.rect(gp=gpar(fill=cols[2],col=cols[1]),height=unit(1.1,"npc"),width
=unit(0.18,"npc"),x=0.1)
# VISIBLE: small projection outside viewport, standard colours fill and
edge
grid.rect(gp=gpar(fill=2,col=3),height=unit(1.1,"npc"),width=unit(0.18,"
npc"),x=0.3)
# INVISIBLE: larger projection outside viewport, transparent colours
fill and edge
grid.rect(gp=gpar(fill=cols[2],col=cols[1]),height=unit(1.5,"npc"),width
=unit(0.18,"npc"),x=0.5)
# VISIBLE: larger projection outside viewport, standard colours fill and
edge
grid.rect(gp=gpar(fill=2,col=3),height=unit(1.5,"npc"),width=unit(0.18,"
npc"),x=0.7)
# MIXED: larger projection outside viewport, standard colour edge
visible, transparent fill not
grid.rect(gp=gpar(fill=cols[2],col=3),height=unit(1.5,"npc"),width=unit(
0.18,"npc"),x=0.9)

I can also show the same effect using standard graphics calls:

par(xpd=NA)
plot(c(1,2))
rect(1.01,1,1.19,2.2,border=cols[1],col=cols[2])
rect(1.21,1,1.39,2.2,border=3,col=2)
rect(1.41,1,1.59,4,border=cols[1],col=cols[2])
rect(1.61,1,1.79,4,border=3,col=2)
rect(1.81,1,1.99,4,border=3,col=cols[2])

Both of these examples have missing alpha transparency colours under
Windows but produce the expected filled rectangles under Linux.
>
> You need to add type="count" to the call to histogram, or rescale
the
bar
> heights.
My post was a rather simplified example - I am actually trying to make
visible a "small" distribution on the skirt of a large one, which
means
that scaling the bar heights is not sufficient. I have worked around the
problem by modifying panel.histogram to _clip_ the bar heights.

Keith

Keith Ponting
Aurix Ltd, Malvern WR14 3SZ  UK

I *presume* this is the windows() device in a current version of R, but 
you failed to tell us almost any of the information requested in the 
posting guide.

If so, you are seeing how Windows GDI copying operations work: if the 
region to be copied goes off-window, they clip the region completely. Your 
example works in R-devel, and if that survives a few weeks of testing, I 
might port the workaround to 2.8.1.  (A similar fix in 2.7.1 patched ended 
up with undesirable side effects in 2.7.2, so careful testing is needed.)

Implementing semi-transparent colours on windows() was hard, and it would 
be nice to see some appreciation of the effort involved.

On Fri, 10 Oct 2008, Keith Ponting wrote:
> Richard.Cotton wrote:
>
>> The rectangles being drawn extend higher than the top of the panel.
> (Your y
>> axis ranges from 0 to 50, but the bars go up to 100.)
>>
>
> Thankyou - I can also make the bars on the lower panels vanish by
> tinkering with ylim.
>
>> In the top row of plots, depending upon the shape of your device
> window, the
>> bars can extend beyond the range of the device window.  For some
> reason,
>> (take a look in panel.rect), when you specify alpha less than 1, this
>> prevents the bar being drawn.
>
> The problem lies deeper. I can demonstrate the same effect with raw grid
> calls (but cannot find a way into "grid.draw" to see what is
going on):
>
> cols <- rainbow(2,alpha=0.5)
> library(grid)
> grid.newpage()
> pushViewport(plotViewport(c(5,4,2,2)))
> # draw viewport limits:
> grid.rect()
> # VISIBLE: small projection outside viewport, transparent fill and edge
>
grid.rect(gp=gpar(fill=cols[2],col=cols[1]),height=unit(1.1,"npc"),width
> =unit(0.18,"npc"),x=0.1)
> # VISIBLE: small projection outside viewport, standard colours fill and
> edge
>
grid.rect(gp=gpar(fill=2,col=3),height=unit(1.1,"npc"),width=unit(0.18,"
> npc"),x=0.3)
> # INVISIBLE: larger projection outside viewport, transparent colours
> fill and edge
>
grid.rect(gp=gpar(fill=cols[2],col=cols[1]),height=unit(1.5,"npc"),width
> =unit(0.18,"npc"),x=0.5)
> # VISIBLE: larger projection outside viewport, standard colours fill and
> edge
>
grid.rect(gp=gpar(fill=2,col=3),height=unit(1.5,"npc"),width=unit(0.18,"
> npc"),x=0.7)
> # MIXED: larger projection outside viewport, standard colour edge
> visible, transparent fill not
>
grid.rect(gp=gpar(fill=cols[2],col=3),height=unit(1.5,"npc"),width=unit(
> 0.18,"npc"),x=0.9)
>
> I can also show the same effect using standard graphics calls:
>
> par(xpd=NA)
> plot(c(1,2))
> rect(1.01,1,1.19,2.2,border=cols[1],col=cols[2])
> rect(1.21,1,1.39,2.2,border=3,col=2)
> rect(1.41,1,1.59,4,border=cols[1],col=cols[2])
> rect(1.61,1,1.79,4,border=3,col=2)
> rect(1.81,1,1.99,4,border=3,col=cols[2])
>
> Both of these examples have missing alpha transparency colours under
> Windows but produce the expected filled rectangles under Linux.
>
>>
>> You need to add type="count" to the call to histogram, or
rescale the
> bar
>> heights.
>
> My post was a rather simplified example - I am actually trying to make
> visible a "small" distribution on the skirt of a large one, which
means
> that scaling the bar heights is not sufficient. I have worked around the
> problem by modifying panel.histogram to _clip_ the bar heights.
>
> Keith
>
> Keith Ponting
> Aurix Ltd, Malvern WR14 3SZ  UK
>
> ______________________________________________
> R-help at 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.
>
-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595