[Rd] faster base::sequence

[Hervé Pagès]

Hi Romain,

FWIW I see at least 2 small differences in the way sequence_c()
behaves with respect to good old sequence(): zeros and names.

 > sequence(c(a=5, b=0, c=2))
a1 a2 a3 a4 a5 c1 c2
  1  2  3  4  5  1  2

sequence_c() ignores the names and doesn't support zeros in the input.

Cheers,
H.


On 11/28/2010 01:56 AM, Romain Francois wrote:
> Le 28/11/10 10:30, Prof Brian Ripley a écrit :
>> Is sequence used enough to warrant this? As the help page says
>>
>> Note that ‘sequence <- function(nvec) unlist(lapply(nvec,
>> seq_len))’ and it mainly exists in reverence to the very early
>> history of R.
>
> I don't know. Would it be used more if it were more efficient ?
>
>> I regard it as unsafe to assume that NA_INTEGER will always be negative,
>> and bear in mind that at some point not so far off R integers (or at
>> least lengths) will need to be more than 32-bit.
>
> sure. updated and dressed up as a patch.
>
> I've made it a .Call because I'm not really comfortable with .Internal,
> etc ...
>
> Do you mean that I should also use something else instead of "int" and
> "int*". Is there some future proof typedef or macro for the type
> associated with INTSXP ?
>
>
>> On Sun, 28 Nov 2010, Romain Francois wrote:
>>
>>> Hello,
>>>
>>> Based on yesterday's R-help thread (help: program efficiency), and
>>> following Bill's suggestions, it appeared that sequence:
>>>
>>>> sequence
>>> function (nvec)
>>> unlist(lapply(nvec, seq_len))
>>> <environment: namespace:base>
>>>
>>> could benefit from being written in C to avoid unnecessary memory
>>> allocations.
>>>
>>> I made this version using inline:
>>>
>>> require( inline )
>>> sequence_c <- local( {
>>> fx <- cfunction( signature( x = "integer"), '
>>> int n = length(x) ;
>>> int* px = INTEGER(x) ;
>>> int x_i, s = 0 ;
>>> /* error checking */
>>> for( int i=0; i<n; i++){
>>> x_i = px[i] ;
>>> /* this includes the check for NA */
>>> if( x_i <= 0 ) error( "needs non negative integer" ) ;
>>> s += x_i ;
>>> }
>>>
>>> SEXP res = PROTECT( allocVector( INTSXP, s ) ) ;
>>> int * p_res = INTEGER(res) ;
>>> for( int i=0; i<n; i++){
>>> x_i = px[i] ;
>>> for( int j=0; j<x_i; j++, p_res++)
>>> *p_res = j+1 ;
>>> }
>>> UNPROTECT(1) ;
>>> return res ;
>>> ' )
>>> function( nvec ){
>>> fx( as.integer(nvec) )
>>> }
>>> })
>>>
>>>
>>> And here are some timings:
>>>
>>>> x <- 1:10000
>>>> system.time( a <- sequence(x ) )
>>> utilisateur système écoulé
>>> 0.191 0.108 0.298
>>>> system.time( b <- sequence_c(x ) )
>>> utilisateur système écoulé
>>> 0.060 0.063 0.122
>>>> identical( a, b )
>>> [1] TRUE
>>>
>>>
>>>
>>>> system.time( for( i in 1:10000) sequence(1:10) )
>>> utilisateur système écoulé
>>> 0.119 0.000 0.119
>>>>
>>>> system.time( for( i in 1:10000) sequence_c(1:10) )
>>> utilisateur système écoulé
>>> 0.019 0.000 0.019
>>>
>>>
>>> I would write a proper patch if someone from R-core is willing to push
>>> it.
>>>
>>> Romain
>
>
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel


-- 
Hervé Pagès

Program in Computational Biology
Division of Public Health Sciences
Fred Hutchinson Cancer Research Center
1100 Fairview Ave. N, M2-B876
P.O. Box 19024
Seattle, WA 98109-1024

E-mail: hpages at fhcrc.org
Phone:  (206) 667-5791
Fax:    (206) 667-1319