R help - Nov 2004 - Unexpected results from sort function when partial and index are used

Hi,

Consider the following example:

sort(10:1, partial=3)
## 1  2  3  7  6  5  4  8  9 10

sort(10:1, index=T)
## $x: 1  2  3  4  5  6  7  8  9 10
## $ix: 10  9  8  7  6  5  4  3  2  1

sort(10:1, partial=3, index=T)
##  1  2  3  7  6  5  4  8  9 10

The first 2 calls gave expected returns; however, the third one did not
returned an index as requested. I could not find anything about it in
http://stat.ethz.ch/R-manual/R-patched/library/base/html/sort.html
<http://stat.ethz.ch/R-manual/R-patched/library/base/html/sort.html> , so
it
seems to be an "undocumented feature". 

Does any body know how to "convince" sort to return index of partially
sorted array?

Thanks

Jarek
=====================================\====                 
 Jarek Tuszynski, PhD.                               o / \ 
 Science Applications International Corporation  <\__,|  
 (703) 676-4192                        ">  \
 Jaroslaw.W.Tuszynski@saic.com                   `    \



	[[alternative HTML version deleted]]

Tuszynski, Jaroslaw W. wrote:
> Hi,
> 
> Consider the following example:
> 
> sort(10:1, partial=3)
> ## 1  2  3  7  6  5  4  8  9 10
> 
> sort(10:1, index=T)
> ## $x: 1  2  3  4  5  6  7  8  9 10
> ## $ix: 10  9  8  7  6  5  4  3  2  1
> 
> sort(10:1, partial=3, index=T)
> ##  1  2  3  7  6  5  4  8  9 10
> 
> The first 2 calls gave expected returns; however, the third one did not
> returned an index as requested. I could not find anything about it in
> http://stat.ethz.ch/R-manual/R-patched/library/base/html/sort.html
> <http://stat.ethz.ch/R-manual/R-patched/library/base/html/sort.html>
, so it
> seems to be an "undocumented feature". 
> 
> Does any body know how to "convince" sort to return index of
partially
> sorted array?
> 
> Thanks
> 
> Jarek
Jarek,

Looking at the code for sort, we see the following:

     if (!is.null(partial)) {
         if (!all(is.finite(partial)))
             stop("non-finite `partial'")
         y <- .Internal(psort(x, partial))
     } else {
        # other sort code
     }

so index.return is ignored if partial is provided. To get the index you 
can use ?match:

z <- rnorm(10)
x <- sort(z, partial = 3)
ix <- match(z, x)

Hopefully, I used ?match correctly. Please verify on your own.

--sundar

On Wed, 3 Nov 2004, Tuszynski, Jaroslaw W. wrote:
> Hi,
> 
> Consider the following example:
> 
> sort(10:1, partial=3)
> ## 1  2  3  7  6  5  4  8  9 10
> 
> sort(10:1, index=T)
> ## $x: 1  2  3  4  5  6  7  8  9 10
> ## $ix: 10  9  8  7  6  5  4  3  2  1
> 
> sort(10:1, partial=3, index=T)
> ##  1  2  3  7  6  5  4  8  9 10
> 
> The first 2 calls gave expected returns; however, the third one did not
> returned an index as requested. I could not find anything about it in
> http://stat.ethz.ch/R-manual/R-patched/library/base/html/sort.html
> <http://stat.ethz.ch/R-manual/R-patched/library/base/html/sort.html>
, so it
> seems to be an "undocumented feature". 
> 
> Does any body know how to "convince" sort to return index of
partially
> sorted array?
You cannot.  There is no underlying code to do so, and the person who 
added 'index.return' forgot this case.  I was against having it at all
--
we have sort.list for that purpose.  I've updated the documentation.

Sundar's match() solution will not work if there are duplicate values.  
If you need the index, just do a full sort -- partial sorting is only 
implemented for efficiency reasons, and nowadays full sorting is fast 
enough even on massive vectors.

-- 
Brian D. Ripley,                  ripley at 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