Thanks for taking the time! In the meanwhile I checked PDL::Transform and 
arrived at the same conclusion…

I know Maggies's module -- PDL::Stats::Kmeans is the step in my chain 
(clustering on the reduced matrix). However PCA is not the same a PCoA… If 
don't think that exists under PDL or more generally for perl. I have a little 
hope with R, since Rserve can quite speed up things 
(http://www.rforge.net/Rserve/), and someone just started a pelr lint 
(http://search.cpan.org/~djunkim/Statistics-RserveClient-0.01/lib/Statistics/RserveClient.pm).
 We'll see…


Le 31 janv. 2013 à 22:27, Derek Lamb <[email protected]> a écrit :

> OK, with this information I can tell you that PDL::Transform is probably not 
> what you want--that does coordinate transforms (I was interpreting your 
> question as in, "can I scale things (images, coordinates, etc) in multiple 
> dimensions").  Thank you for the MDS wikipedia link.
> 
> PCA might be a better search term.  Look at Maggie's PDL::Stats module, which 
> contains PDL::Stats::GLM.  That includes some PCA routines.
> 
> I don't know that GSL does PCA explicitly, but it should handle the 
> eigenvalue or SVD portions of PCA just fine.  There are some PDL::GSL:: 
> modules but I don't think they include the eigen calculations.
> 
> PDL::Slatec has the 'eigsys' function which may also be helpful.
> 
> PDL::MatrixOps has the eigens.  These last two I found by doing 
> 
> $pdldoc -a eigen
> 
> from the command line after PDL was installed.
> 
> There is a general R/Splus <--> Perl package here: 
> http://www.omegahat.org/RSPerl/ but it has not been updated in many years, 
> and I have never used it.
> 
> best,
> Derek
> 
> 
> On Jan 31, 2013, at 1:52 PM, Jean Véronis wrote:
> 
>> Thanks Derek, I am going to check PDL::Transform right away.
>> 
>> My need is to reduce dimensionality on large distance matrices. For this is 
>> use multidimensional scaling  
>> (http://en.wikipedia.org/wiki/Multidimensional_scaling) by calling R from 
>> perl (cmdscale function : 
>> http://stat.ethz.ch/R-manual/R-patched/library/stats/html/cmdscale.html).
>> 
>> It's perfectly fine, expect in terms of performance, since at the moment I 
>> didn't find any better way than use Statistics:R, which basically passes 
>> data through stdin/stdout. When it comes to huge matrices, it is extremely 
>> inefficient. Hence my question.
>> 
>> I tried to google "pdl multidimensional scaling", "gsl multidimensional 
>> scaling", "perl multidimensional scaling", and all possible variants 
>> involving "pcinipal coordinates analysis", "PCoA", with no success.
>> 
>> Many thanks, again.
>> --j
>> 
>> Le 31 janv. 2013 à 21:41, Derek Lamb <[email protected]> a écrit :
>> 
>>> Hi Jean,
>>> 
>>> The best hint I can provide based on the extremely limited information you 
>>> have given is PDL::Transform.
>>> 
>>> You might get a better response if you describe what you are trying to do 
>>> in more than one sentence, what you have tried, whether you are new to PDL 
>>> or Perl or are already using one or both, and what documentation you have 
>>> looked at.
>>> 
>>> best,
>>> Derek
>>> 
>>> On Jan 31, 2013, at 1:24 PM, Jean Véronis wrote:
>>> 
>>>> Hi all,
>>>> 
>>>> I am looking for a multidimensional scaling package.
>>>> Any hint would be appreciated.
>>>> 
>>>> Many thanks
>>>> --j
>>> 
>> 
> 

_______________________________________________
Perldl mailing list
[email protected]
http://mailman.jach.hawaii.edu/mailman/listinfo/perldl

Reply via email to