CLASS-L Archives

November 2002


Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
"Arthur J. Kendall" <[log in to unmask]>
Reply To:
Classification, clustering, and phylogeny estimation
Thu, 14 Nov 2002 16:36:30 -0500
text/plain (70 lines)
If Doug Carroll said there is a problem with the MDS in SPSS for your
kind of problem, then, of course, use something else for the final run.

If you use SPSS for the final analysis, set the memory as high as your
hardware and OS will allow. This will minimize execution time for the
final runs.  . Running your syntax in production mode after it is fully
debugged bypasses the need for the GUI.

Final execution time is usually a small fraction of total person time
and not that important a desideratum. Using a package  with proven track
record in human factors like SPSS can be very time-saving in the long
run. Use of syntax for one procedure easily generalizes to syntax for
others. Good quality assurance and documentation  in the preparation
phase of analysis are crucial.  Preparation of the data can easily be 80
or 90% of all the time in an analysis. With most software applications
you will have many attempts before getting the analysis done the way you
want it.  With SPSS you will probably need fewer attempts.  Even if you
use some other software for the final procedure, you would often find
SPSS very effective for the prep. Locating and downloading source code
can be time consuming.  Learning how to adapt and compile source
language code (FORTRAN, C++, IMSL) can take a lot of time.  It also
means that you have to have access to them.  It certainly took me much
more than a few hours to adopt SINDSCAL source code so that it had all
dimension arguments changed to parameters. After that it didn't take
long to adapt it to a particular analysis, but it still had to be compiled.

Hope this helps.

[log in to unmask]
Social Research Consultants
University Park, MD USA

[log in to unmask] wrote:

>MDS using SAS or SPSS will most likely be very slow. I have done MDS with SPSS and a matrix of your dimensions and 1 hour later it was still running.
>If performance is an issue, writting code will always run faster then the commercial stat packages. I use IMSL which has an MDS function and the runtime results were vastly superior. One reason is with SPSS, the GUI slows things done, their code does lots of memory allocation, slowing things down also.
>"Hand-Rolling" your own code with a proven MDS function (which are easy to find) will run much fast as you can just cache (store your matrix in memory) and let it fly.
>The net conclusion, unless it's a small sample size and I want to test my own programs... I compare my results with SPSS to affirm the results match. Direct programming is always much faster.
>Neil Gottlieb
>Profile Depot, Inc.
>you wrote:
>Dear Listmembers,
>I would like to do a MDS with a 400X400 square, symetric matrix of
>(dis)similaritys. Most MDS-Software is limited to a much smaller amount
>of variables though. I have been told, that SAS might be able to process
>the analyis I need, but this is quite inconvenient for me in terms of
>access to the software, hardware-requirements etc. Could anybody tell
>me, if there is a stand-alone program that can do MDS (preferably
>nonmetric) with a matrix of that size?
>Thanks a lot for your help,
>P.S: My Matrix is not suitable for factor analysis and clustering does
>not produce the results I need.