On Apr 11, 2011, at 23:53 , Joris Meys wrote:
> Based on a discussion on SO I ran some tests and found that converting
> to a factor is best done early in the process. Hence, I propose to
> rewrite the gl() function as :
>
> gl2 <- function(n, k, length = n * k, labels = 1:n, ordered = FALSE){
> rep(
> rep(
> factor(1:n,levels=1:n,labels=labels, ordered=ordered),rep.int(k,n)
> ),length.out=length
> )
> }
>
That's bizarre! You are relying on an optimization in rep.factor whereby it
replicates the internal codes and exploits that the result has the same
structure as the input. I.e., it just tacks on class and levels attributes
rather than call match() as factor() does internally.
However, you can do the same thing straight away:
> gl2
function (n, k, length = n * k, labels = 1:n, ordered = FALSE)
{
y <- rep(rep.int(1:n, rep.int(k, n)), length.out = length)
structure(y, levels=as.character(labels),
class=c(if(ordered)"ordered","factor"))
}
I get this to be a bit faster than your version, although with a smaller
speedup factor, which probably just indicates that match() is faster on this
machine.
> Some test results :
>
>> system.time(X1 <- gl(5,1e7))
> user system elapsed
> 29.21 0.30 29.58
>
>> system.time(X2 <- gl2(5,1e7))
> user system elapsed
> 1.87 0.45 2.37
>
>> all.equal(X1,X2)
> [1] TRUE
>
>> system.time(X1 <- gl(5,100,1e7))
> user system elapsed
> 5.98 0.05 6.05
>
>> system.time(X2 <- gl2(5,100,1e7))
> user system elapsed
> 0.21 0.03 0.25
>
>> all.equal(X1,X2)
> [1] TRUE
>
>> system.time(X1 <- gl(5,100,1e7,labels=letters[1:5]))
> user system elapsed
> 5.88 0.02 5.98
>
>> system.time(X2 <- gl2(5,100,1e7,labels=letters[1:5]))
> user system elapsed
> 0.20 0.05 0.25
>
>> all.equal(X1,X2)
> [1] TRUE
>
>> system.time(X1 <- gl(5,100,1e7,labels=letters[1:5],ordered=T))
> user system elapsed
> 5.82 0.03 5.89
>
>> system.time(X2 <- gl2(5,100,1e7,labels=letters[1:5],ordered=T))
> user system elapsed
> 0.22 0.04 0.25
>
>> all.equal(X1,X2)
> [1] TRUE
>
> reference to SO :
> http://stackoverflow.com/questions/5627264/how-can-i-efficiently-construct-a-very-long-factor-with-few-levels
>
> --
> Joris Meys
> Statistical consultant
>
> Ghent University
> Faculty of Bioscience Engineering
> Department of Applied mathematics, biometrics and process control
>
> tel : +32 9 264 59 87
> [email protected]
> -------------------------------
> Disclaimer : http://helpdesk.ugent.be/e-maildisclaimer.php
>
> ______________________________________________
> [email protected] mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
--
Peter Dalgaard
Center for Statistics, Copenhagen Business School
Solbjerg Plads 3, 2000 Frederiksberg, Denmark
Phone: (+45)38153501
Email: [email protected] Priv: [email protected]
______________________________________________
[email protected] mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
