[R-pkgs] homals package and core loop

Jan de Leeuw deleeuw at stat.ucla.edu
Sat Sep 19 21:41:56 CEST 2009


The homals package

http://www.jstatsoft.org/v31/i04

will get a major programming overhaul. This will take some time,
but what's a few years on a 40-year project. Suggestions
from the audience are welcome.

homals() has a core loop over the m variables in which

1. tapply is used to compute category quantification (centroids)
2. category quantifications are then adjusted to satisfy the
   rank and level constraints

The idea is to replace 1 by least squares fitting of a B-spline and
to adjust 2 accordingly. The core will be re-programmed in C
and/or FORTRAN (sometimes using existing routines), and the loop over
variables will be parallelized using OpenMP or Grand Central Dispatch.
For ordinal variables this requires some alternative quadratic  
programming
routines to be linked in (which could then also become part of the
isotone package).

In addition the basic homals code will have object (individual) weights
(minor addition) and the prehom routines (from the jacobi package) will
be added. Because of the use of B splines the variables in homals will
no longer be all factors, some can now be numerical ("continuous").

Otherwise the package will keep its many analysis options (multi-set  
non-metric canonical analysis,
non-metric discriminant analysis, non-metric regression, multiple  
correspondence analysis,
non-metric principal components analysis, non-metric additive conjoint  
analysis)
and its many plot options.

====================================================================
     Jan de Leeuw, 11667 Steinhoff Rd, Frazier Park, CA 93225
     home 661-245-1725 mobile 661-231-5416 work 310-825-9550
     .mac: jdeleeuw +++  aim: deleeuwjan +++ skype: j_deleeuw
====================================================================
If I wanted you to understand it, I would have explained it  
better.                                                					
                                                   -- Johan Cruijff



More information about the R-packages mailing list