[R] Faster union of polygons?

[Martin Morgan]

On 06/03/2010 04:54 PM, Remko Duursma wrote:
> Thanks for the tip - this cleans up the code a lot!
> Unfortunately, there is no gain in speed.

Playing a little bit dirty,

punion <-
    function(...)
{
    n <- nargs()
    if (0L == n) new("gpc.poly")
    else if (1L == n && is(..1, "gpc.poly")) ..1
    else {
        polygons <- list(...)
        if (!all(sapply(polygons, is, "gpc.poly")))
            stop("'...' must all be 'gpc.poly'")
        ## avoid method look-up
        to_numeric <- selectMethod(coerce, c("gpc.poly", "numeric"))
        vec <- to_numeric(polygons[[1]])
        for (p in polygons[-1]) {
            clip <- to_numeric(p)
            vec <- .Call("Rgpc_polygon_clip", vec, clip, 3,
                         PACKAGE="gpclib")
        }
        if (identical(vec, 0))
            new("gpc.poly")
        else
            as(vec, "gpc.poly")
    }
}

is about 4x faster on your example

> replicate(5, system.time(Reduce(union, leaves)))
            [,1]  [,2]  [,3]  [,4]  [,5]
user.self  1.272 1.272 1.272 1.268 1.268
sys.self   0.000 0.000 0.000 0.000 0.000
elapsed    1.271 1.272 1.272 1.273 1.281
user.child 0.000 0.000 0.000 0.000 0.000
sys.child  0.000 0.000 0.000 0.000 0.000
> replicate(5, system.time(do.call(punion, leaves)))
            [,1]  [,2]  [,3]  [,4]  [,5]
user.self  0.308 0.312 0.304 0.308 0.312
sys.self   0.004 0.000 0.004 0.004 0.000
elapsed    0.311 0.311 0.309 0.314 0.317
user.child 0.000 0.000 0.000 0.000 0.000
sys.child  0.000 0.000 0.000 0.000 0.000

Rprof suggests that most of the time is now in the C code

> Rprof("/tmp/leaves.Rprof")
> x <- replicate(5, system.time(do.call(punion, leaves)))
> Rprof(NULL)
> summaryRprof("/tmp/leaves.Rprof")
$by.self
                     self.time self.pct total.time total.pct
".Call"                   1.24     69.7       1.24      69.7
"gc"                      0.24     13.5       0.24      13.5
"FUN"                     0.08      4.5       1.78     100.0
[...SNIP...]

Martin

> 
> remko
> 
> 
> 
> On Thu, Jun 3, 2010 at 10:46 PM, nikhil kaza <nikhil.list at gmail.com> wrote:
>> Reduce might work. Not sure about the speed advantages though. It does
>> simplify code.
>>
>>  Unionall <- function(x) Reduce('union', x)
>> leaveout <- Unionall(leaves)
>>
>>
>> On Tue, Jun 1, 2010 at 9:53 PM, Remko Duursma <remkoduursma at gmail.com>
>> wrote:
>>>
>>> Dear R-helpers,
>>>
>>> thanks for yesterday's speeding-up tip. Here is my next query:
>>>
>>> I have lots of polygons (not necessarily convex ones, and they never
>>> have holes) given by x,y coordinates.
>>>
>>> I want to get the polygon that is the union of these polygons. This is
>>> my current method, but I am hoping there is a faster method (up to
>>> thousands of polygons, each with ca. 40 xy points).
>>>
>>> Example:
>>>
>>> library(gpclib)
>>>
>>> # A polygon
>>> leaf <- structure(c(0, 1, 12.9, 16.5, 18.8, 17, 16.8, 15.5, 12.1, 8.2,
>>> 6.3, 5, 2, 0, -1.5, -4.3, -6.6, -10.3, -14.8, -19.4, -22.2, -23.5,
>>> -22.2, -17.6, -7.8, 0, 0, -2.4, 2.8, 8.9, 19.9, 33.9, 34.8, 40.4,
>>> 49.7, 69.2, 77.4, 83.4, 91.4, 99, 92.8, 87.3, 81.2, 71.1, 57.6,
>>> 45.4, 39.2, 26, 15.6, 5.3, 0.6, 0), .Dim = c(26L, 2L), .Dimnames = list(
>>>    NULL, c("X", "Y")))
>>>
>>> # Lots of polygons:
>>> releaf <-
>>> function(leaf)cbind(leaf[,1]+rnorm(1,0,50),leaf[,2]+rnorm(1,0,50))
>>> leaves <- replicate(500, releaf(leaf), simplify=FALSE)
>>>
>>> # Make into gpc.poly class:
>>> leaves <- lapply(leaves, as, "gpc.poly")
>>>
>>> # Make union .....
>>> system.time({
>>> leavesoutline <- union(leaves[[1]], leaves[[2]])
>>> for(i in 3:length(leaves))leavesoutline <- union(leavesoutline,
>>> leaves[[i]])
>>> })
>>> # about 1sec here.
>>>
>>> # Check it:
>>> plot(leavesoutline)
>>>
>>>
>>>
>>> thanks!
>>>
>>> Remko
>>>
>>>
>>> -------------------------------------------------
>>> Remko Duursma
>>> Research Lecturer
>>>
>>> Centre for Plants and the Environment
>>> University of Western Sydney
>>> Hawkesbury Campus
>>> Richmond NSW 2753
>>>
>>> Dept of Biological Science
>>> Macquarie University
>>> North Ryde NSW 2109
>>> Australia
>>>
>>> Mobile: +61 (0)422 096908
>>> www.remkoduursma.com
>>>
>>> ______________________________________________
>>> 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.
>>
>>
> 
> ______________________________________________
> 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.


-- 
Martin Morgan
Computational Biology / Fred Hutchinson Cancer Research Center
1100 Fairview Ave. N.
PO Box 19024 Seattle, WA 98109

Location: Arnold Building M1 B861
Phone: (206) 667-2793