s-news
[Top] [All Lists]

Re: [S] Canonical Correlation Analysis

To: Alexander Gershunov <gershuno@mombasa.ucsd.edu>
Subject: Re: [S] Canonical Correlation Analysis
From: Don McKenzie <dmck@silvae.cfr.washington.edu>
Date: Tue, 19 Oct 1999 12:19:32 -0700 (PDT)
Cc: s-news@wubios.wustl.edu, sasha@ucsd.edu
In-reply-to: <199910191751.AA27546@mombasa.ucsd.edu>
Sender: owner-s-news@wubios.wustl.edu
The CCA algorithm requires you to invert the covariance matrices of 
X and Y.  This will (should) bomb in any software package because if 
p,q, > n the matrices are singular.  Methods like CCA and factor analysis
won't work if you have more variables than observations.

(I post this to the group also in case I am out to lunch here for some
reason.  If so, enlightenment appreciated, as always.)

_______________________________________________________________________

                           DON MCKENZIE

                        Research Ecologist
              College of Forest Resources, Box 352100
                      University of Washington 
                        Seattle, WA 98195

                            206.543.9138                              
                            206.616.4015
                       dmck@u.washington.edu
     http://silvae.cfr.washington.edu/people/dmck/home.html

_______________________________________________________________________
-

On Tue, 19 Oct 1999, Alexander Gershunov wrote:

> 
> Dearest Splusers:
> 
> Does anyone have an efficient Splus CCA function that works for matrices with 
> large numbers of variables? By large, I mean larger than the number of 
> observations. The canned Splus version, cancor(), cannot(;) deal with 
> matrices 
> that have more variables than observations, i.e. ncol > nrow. Such matrices, 
> say 
> X, an n X p matrix and Y, an n X q matrix, where n is the number of 
> observations 
> (rows) and p and q are the numbers of variabes (columns) such that p,q > n, 
> are 
> dealt with in such a way by cancor() that the linear combinations of the 
> original variables ($xcoef and $ycoef) that are supposed to maximize 
> correlation 
> between the two sets of variables X and Y and be vectors of length p and q, 
> respectively, instead have length (n-1). I just thought to check with this 
> illustrious group before attempting to write my own CCA function. 
> 
> Thankfully,
> 
> Sasha
> -------------------------------------------
>       Alexander Gershunov
>       Climate Research Division
>       Scripps Institution of Oceanography
>       UC San Diego
>       La Jolla, CA 92093-0224
> 
> Phone:        (858) 534-8418; Fax: (858) 534-8561; email: sasha@ucsd.edu
> 
> 
> -----------------------------------------------------------------------
> This message was distributed by s-news@wubios.wustl.edu.  To unsubscribe
> send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
> message:  unsubscribe s-news
> 

-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu.  To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message:  unsubscribe s-news

<Prev in Thread] Current Thread [Next in Thread>