Re: [R] Matrix oriented computing

From: Prof Brian Ripley <ripley_at_stats.ox.ac.uk>
Date: Sat 27 Aug 2005 - 01:55:55 EST

Try profiling. Doing this many times to get an overview, e.g. for sapply with df=1:1000:

    %       self        %       total
  self     seconds    total    seconds    name
  98.26      6.78     98.26      6.78     "FUN"
   0.58      0.04      0.58      0.04     "unlist"
   0.29      0.02      0.87      0.06     "as.vector"
   0.29      0.02      0.58      0.04     "names<-"
   0.29      0.02      0.29      0.02     "names<-.default"
   0.29      0.02      0.29      0.02     "names"

so almost all the time is in qchisq.

On Fri, 26 Aug 2005, Marc Schwartz (via MN) wrote:

> On Fri, 2005-08-26 at 15:25 +0200, Peter Dalgaard wrote:

>> Marc Schwartz <MSchwartz@mn.rr.com> writes:
>>
>>> x <- c(0.005, 0.010, 0.025, 0.05, 0.1, 0.5, 0.9,
>>>        0.95, 0.975, 0.99, 0.995)
>>>
>>> df <- c(1:100)
>>>
>>> mat <- sapply(x, qchisq, df)
>>>
>>>> dim(mat)
>>> [1] 100  11
>>>
>>>> str(mat)
>>>  num [1:100, 1:11] 3.93e-05 1.00e-02 7.17e-02 2.07e-01 4.12e-01 ...
>>
>> outer() is perhaps a more natural first try... It does give the
>> transpose of the sapply approach, though.
>>
>> round(t(outer(x,df,qchisq)),2)
>>
>> should be close. You should likely add dimnames.
>
>
>

> What I find interesting, is that I would have intuitively expected
> outer() to be faster than sapply(). However:
>
>
>>  system.time(mat <- sapply(x, qchisq, df), gcFirst = TRUE)

> [1] 0.01 0.00 0.01 0.00 0.00
>
>>  system.time(mat1 <- round(t(outer(x, df, qchisq)), 2),

> gcFirst = TRUE)
> [1] 0.01 0.00 0.01 0.00 0.00
>

> # No round() or t() to test for overhead
>>  system.time(mat2 <- outer(x, df, qchisq), gcFirst = TRUE)

> [1] 0.01 0.00 0.02 0.00 0.00
>
>

> # Bear in mind the round() on mat1 above
>> all.equal(mat, mat1)

> [1] "Mean relative difference: 4.905485e-05"
>
>> all.equal(mat, t(mat2))

> [1] TRUE
>
>

> Even when increasing the size of 'df' to 1:1000:
>
>
>>  system.time(mat <- sapply(x, qchisq, df), gcFirst = TRUE)

> [1] 0.16 0.01 0.16 0.00 0.00
>
>>  system.time(mat1 <- round(t(outer(x, df, qchisq)), 2), gcFirst =

> TRUE)
> [1] 0.16 0.00 0.18 0.00 0.00
>
>>  # No round() or t() to test for overhead
>>  system.time(mat2 <- outer(x, df, qchisq), gcFirst = TRUE)

> [1] 0.16 0.01 0.17 0.00 0.00
>
>
>

> It also seems that, at least in this case, t() and round() do not add
> much overhead.

Definitely not for such small matrices.

-- 
Brian D. Ripley,                  ripley@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

______________________________________________
R-help@stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
Received on Sat Aug 27 02:08:08 2005

This archive was generated by hypermail 2.1.8 : Fri 03 Mar 2006 - 03:39:56 EST