R help - Apr 2008 - Why R is 200 times slower than Matlab ?

I am switching from Matlab to R, but I found that R is 200 times slower than
matlab.

Since I am newbie to R, I must be missing some important programming tips.

Please help me out on this.

Here is the function:
## make the full pair-wise permutation of a vector
## input_fc=c(1,2,3);
## output_fc=(
1 1 1 2 2 2 3 3 3
1 2 3 1 2 3 1 2 3
);

grw_permute = function(input_fc){

fc_vector = input_fc

index = 1

k = length(fc_vector)

fc_matrix = matrix(0,2,k^2)

for(i in 1:k){

for(j in 1:k){

fc_matrix[index]  =  fc_vector[i]

fc_matrix[index+1]  =  fc_vector[j]

index = index+2

}

}

return(fc_matrix)

}

For an input vector of size 300. It took R 2.17 seconds to run.

But the same code in matlab only needs 0.01 seconds to run.

Am I missing sth in R.. Is there a away to optimize.  ???

Thanks

-- 
Zhandong Liu

Genomics and Computational Biology
University of Pennsylvania

616 BRB II/III, 421 Curie Boulevard
University of Pennsylvania School of Medicine
Philadelphia, PA 19104-6160

	[[alternative HTML version deleted]]

Hi, ZD,
Your comment about speed is too general. Here is a benchmark
comparison among several languages and HTH.
http://www.sciviews.org/benchmark/index.html

On Wed, Apr 30, 2008 at 4:15 PM, Zhandong Liu
<zhandong at mail.med.upenn.edu> wrote:
> I am switching from Matlab to R, but I found that R is 200 times slower
than
>  matlab.
>
>  Since I am newbie to R, I must be missing some important programming tips.
>
>  Please help me out on this.
>
>  Here is the function:
>  ## make the full pair-wise permutation of a vector
>  ## input_fc=c(1,2,3);
>  ## output_fc=(
>  1 1 1 2 2 2 3 3 3
>  1 2 3 1 2 3 1 2 3
>  );
>
>  grw_permute = function(input_fc){
>
>  fc_vector = input_fc
>
>  index = 1
>
>  k = length(fc_vector)
>
>  fc_matrix = matrix(0,2,k^2)
>
>  for(i in 1:k){
>
>  for(j in 1:k){
>
>  fc_matrix[index]  =  fc_vector[i]
>
>  fc_matrix[index+1]  =  fc_vector[j]
>
>  index = index+2
>
>  }
>
>  }
>
>  return(fc_matrix)
>
>  }
>
>  For an input vector of size 300. It took R 2.17 seconds to run.
>
>  But the same code in matlab only needs 0.01 seconds to run.
>
>  Am I missing sth in R.. Is there a away to optimize.  ???
>
>  Thanks
>
>  --
>  Zhandong Liu
>
>  Genomics and Computational Biology
>  University of Pennsylvania
>
>  616 BRB II/III, 421 Curie Boulevard
>  University of Pennsylvania School of Medicine
>  Philadelphia, PA 19104-6160
>
>         [[alternative HTML version deleted]]
>
>  ______________________________________________
>  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.
>


-- 
==============================WenSui Liu
ChoicePoint Precision Marketing
Phone: 678-893-9457
Email : wensui.liu at choicepoint.com
Blog : statcompute.spaces.live.com

I would rather not comment on matlab (where is
your matlab code by the way?), but your function
could be simplified a bit:

grw.permute <- function(v) {
  cbind( rep(v, each=length(v)), rep(v, length(v)) )
}
> system.time(tmp <- f( 1:300))
   user  system elapsed 
  0.020   0.000   0.019 

This is on my quite busy 4 years old laptop....

Best,
Gabor

On Wed, Apr 30, 2008 at 04:15:46PM -0400, Zhandong Liu
wrote:
> I am switching from Matlab to R, but I found that R is 200 times slower
than
> matlab.
> 
> Since I am newbie to R, I must be missing some important programming tips.
> 
[...]

-- 
Csardi Gabor <csardi at rmki.kfki.hu>    UNIL DGM

But please consider that this benchmark is five years old, and i believe 
that R has changed quite a lot since version 1.9. 

Gabor

On Wed, Apr 30, 2008 at 04:21:51PM -0400, Wensui Liu
wrote:
> Hi, ZD,
> Your comment about speed is too general. Here is a benchmark
> comparison among several languages and HTH.
> http://www.sciviews.org/benchmark/index.html
> 
[...]

-- 
Csardi Gabor <csardi at rmki.kfki.hu>    UNIL DGM

Zhandong Liu wrote:
> I am switching from Matlab to R, but I found that R is 200 times slower
than
> matlab.
>
> Since I am newbie to R, I must be missing some important programming tips.
>
> Please help me out on this.
>
> Here is the function:
> ## make the full pair-wise permutation of a vector
> ## input_fc=c(1,2,3);
> ## output_fc=(
> 1 1 1 2 2 2 3 3 3
> 1 2 3 1 2 3 1 2 3
> );
>
> grw_permute = function(input_fc){
>
> fc_vector = input_fc
>
> index = 1
>
> k = length(fc_vector)
>
> fc_matrix = matrix(0,2,k^2)
>
> for(i in 1:k){
>
> for(j in 1:k){
>
> fc_matrix[index]  =  fc_vector[i]
>
> fc_matrix[index+1]  =  fc_vector[j]
>
> index = index+2
>
> }
>
> }
>
> return(fc_matrix)
>
> }
>
> For an input vector of size 300. It took R 2.17 seconds to run.
>
> But the same code in matlab only needs 0.01 seconds to run.
>
> Am I missing sth in R.. Is there a away to optimize.  ???
>
> Thanks
>
>   
This is pretty characteristic. With R, you really don't want nested 
loops doing single-element accessing (if you have better things to do 
with 2.16 seconds of our life). You will usually find that this sort of 
problem is handled either using vectorized operations at a higher level, 
or pushed into C code which is dynamically loaded. For the particular 
problem, notice that the same result is obtained with

 > system.time(rbind(rep(1:300,300),rep(1:300,each=300)))
   user  system elapsed
  0.041   0.006   0.050

or even (OK, so it's transposed)

 > system.time(expand.grid(1:300,1:300))
   user  system elapsed
  0.027   0.011   0.040


-- 
   O__  ---- Peter Dalgaard             ?ster Farimagsgade 5, Entr.B
  c/ /'_ --- Dept. of Biostatistics     PO Box 2099, 1014 Cph. K
 (*) \(*) -- University of Copenhagen   Denmark      Ph:  (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)              FAX: (+45) 35327907

On Thu, 01 May 2008, Zhandong Liu wrote:
> I am switching from Matlab to R, but I found that R is 200 times slower
> than matlab.
>
> Since I am newbie to R, I must be missing some important programming tips.
>
> Please help me out on this.
>
> Here is the function:
> ## make the full pair-wise permutation of a vector
> ## input_fc=c(1,2,3);
> ## output_fc=(
> 1 1 1 2 2 2 3 3 3
> 1 2 3 1 2 3 1 2 3
> );
>
> grw_permute = function(input_fc){
>
> fc_vector = input_fc
>
> index = 1
>
> k = length(fc_vector)
>
> fc_matrix = matrix(0,2,k^2)
>
> for(i in 1:k){
>
> for(j in 1:k){
>
> fc_matrix[index]  =  fc_vector[i]
>
> fc_matrix[index+1]  =  fc_vector[j]
>
> index = index+2
>
> }
>
> }
>
> return(fc_matrix)
>
> }
>
> For an input vector of size 300. It took R 2.17 seconds to run.
>
> But the same code in matlab only needs 0.01 seconds to run.
I am not a MATLAB user, but I suspect it wasn't "the same code"
that produced
an answer in MATLAB, but you don't provide your MATLAB code, nor do you 
specify what version of R, of MATLAB, or what hardware and OS you are using.

I get {NetBSD, R version 2.6.0 (2007-10-03), Core 2 Duo, 3.x
GHz}:
> input_fc <- sample(1:600)
> unix.time(a1 <- grw_permute(input_fc))
   user  system elapsed
  3.279  -0.001   3.280
> unix.time({n <- length(input_fc); a2 <- matrix(c(rep(input_fc,
each=n),
rep(input_fc, n)), 2, n*n, byrow = T)})
   user  system elapsed
  0.019   0.020   0.040
> all.equal(a1, a2)
[1] TRUE
>               
A sample of length 300 took less than 1 second using your grw_permute() (so 
your OS may be making a difference as well).
>
> Am I missing sth in R.. Is there a away to optimize.  ???
>
Yes. Loops are not efficient in R.
> Thanks
HTH,
Ray Brownrigg

You just have to use the right functions: is this fast enough
> system.time(x <- expand.grid(1:300, 1:300))
   user  system elapsed
   0.00    0.01    0.01



On Wed, Apr 30, 2008 at 4:15 PM, Zhandong Liu
<zhandong at mail.med.upenn.edu> wrote:
> I am switching from Matlab to R, but I found that R is 200 times slower
than
> matlab.
>
> Since I am newbie to R, I must be missing some important programming tips.
>
> Please help me out on this.
>
> Here is the function:
> ## make the full pair-wise permutation of a vector
> ## input_fc=c(1,2,3);
> ## output_fc=(
> 1 1 1 2 2 2 3 3 3
> 1 2 3 1 2 3 1 2 3
> );
>
> grw_permute = function(input_fc){
>
> fc_vector = input_fc
>
> index = 1
>
> k = length(fc_vector)
>
> fc_matrix = matrix(0,2,k^2)
>
> for(i in 1:k){
>
> for(j in 1:k){
>
> fc_matrix[index]  =  fc_vector[i]
>
> fc_matrix[index+1]  =  fc_vector[j]
>
> index = index+2
>
> }
>
> }
>
> return(fc_matrix)
>
> }
>
> For an input vector of size 300. It took R 2.17 seconds to run.
>
> But the same code in matlab only needs 0.01 seconds to run.
>
> Am I missing sth in R.. Is there a away to optimize.  ???
>
> Thanks
>
> --
> Zhandong Liu
>
> Genomics and Computational Biology
> University of Pennsylvania
>
> 616 BRB II/III, 421 Curie Boulevard
> University of Pennsylvania School of Medicine
> Philadelphia, PA 19104-6160
>
>        [[alternative HTML version deleted]]
>
> ______________________________________________
> 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.
>


-- 
Jim Holtman
Cincinnati, OH
+1 513 646 9390

What is the problem you are trying to solve?

Zhandong Liu wrote:
> I am switching from Matlab to R, but I found that R is 200 times slower
than
> matlab.
> 
> Since I am newbie to R, I must be missing some important programming tips.
The most important tip I would give you is to use the vectorized nature 
of R whenever possible.  This helps avoid messy indexing and 'for'
loops.

Look at the following 3 functions.  Yours, Gabor's, and my own (which I 
was about to post when I saw Gabor's nice solution, and is basically the 
same).

Also see the system timings after the definitions.

grw_permute <- function(input_fc){
   fc_vector <- input_fc
   index <- 1
   k <- length(fc_vector)
   fc_matrix <- matrix(0, 2, k^2)

   for(i in 1:k){
     for(j in 1:k){
       fc_matrix[index]  <-  fc_vector[i]
       fc_matrix[index+1]  <-  fc_vector[j]
       index <- index + 2
     }
   }
   return(fc_matrix)
}

grw.permute2 <- function(v) {
   cbind( rep(v, each=length(v)), rep(v, length(v)) )
}


grw_permute3 <- function(input_fc) {
   matrix(c(rep(input_fc, each = length(input_fc)),
            rep.int(input_fc, times = length(input_fc))),
          nrow = 2, byrow = TRUE)
}

 > system.time(p1 <- grw_permute(1:300))
    user  system elapsed
   1.548   0.064   2.341
 > system.time(p2 <- grw_permute2(1:300))
    user  system elapsed
   0.009   0.001   0.010
 > system.time(p3 <- grw_permute3(1:300))
    user  system elapsed
   0.008   0.002   0.010


Erik Iverson

Aside from optiming your code by making use of R functions
that use C underneath as much as possible the big difference
between R and Matlab is Matlab's just-in-time compilation of
code.  When that was introduced in Matlab huge speedups of
Matlab programs were noticeable.

For R, there is a new package on CRAN, jit, that
aims to provide similar speedups.

On Wed, Apr 30, 2008 at 4:15 PM, Zhandong Liu
<zhandong at mail.med.upenn.edu> wrote:
> I am switching from Matlab to R, but I found that R is 200 times slower
than
> matlab.
>
> Since I am newbie to R, I must be missing some important programming tips.
>
> Please help me out on this.
>
> Here is the function:
> ## make the full pair-wise permutation of a vector
> ## input_fc=c(1,2,3);
> ## output_fc=(
> 1 1 1 2 2 2 3 3 3
> 1 2 3 1 2 3 1 2 3
> );
>
> grw_permute = function(input_fc){
>
> fc_vector = input_fc
>
> index = 1
>
> k = length(fc_vector)
>
> fc_matrix = matrix(0,2,k^2)
>
> for(i in 1:k){
>
> for(j in 1:k){
>
> fc_matrix[index]  =  fc_vector[i]
>
> fc_matrix[index+1]  =  fc_vector[j]
>
> index = index+2
>
> }
>
> }
>
> return(fc_matrix)
>
> }
>
> For an input vector of size 300. It took R 2.17 seconds to run.
>
> But the same code in matlab only needs 0.01 seconds to run.
>
> Am I missing sth in R.. Is there a away to optimize.  ???
>
> Thanks
>
> --
> Zhandong Liu
>
> Genomics and Computational Biology
> University of Pennsylvania
>
> 616 BRB II/III, 421 Curie Boulevard
> University of Pennsylvania School of Medicine
> Philadelphia, PA 19104-6160
>
>        [[alternative HTML version deleted]]
>
> ______________________________________________
> 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.
>