[R] scalassoc package

Jan de Leeuw deleeuw at stat.ucla.edu
Tue Apr 18 09:29:28 CEST 2006

Many new things at


The scalassoc package, which fits exponential distance association  
to indicator matrices, it now at version 1.0.0. It seems to be robust  
can analyze large examples easily. It is a major improvement (in  
speed and
robustness) over the distassoc package, which is at the same site.

scalassoc does something neat (if you like that sort of thing). It  
the changing configurations to a plotwindow for your default device,  
the iterations as a movie. But, if you have ffmpeg installed in your  
then it also has the option to write the iterations to a quicktime movie
file. Currently this creates a lot of intermediate jpeg's (although  
it cleans up
after itself). It may be possible to use ffmpeg to stream them directly
into a movie file.

Just to give you an idea, the data are in the n x k_j indicator  
matrices G_j, where g_{ijl}=1
if object i is in category (level) l of variable j. The log- 
likelihood we maximize is
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pastedGraphic.pdf
Type: application/pdf
Size: 29990 bytes
Desc: not available
Url : https://stat.ethz.ch/pipermail/r-help/attachments/20060418/6b1b37c4/attachment-0003.pdf 
-------------- next part --------------

Variable j has k_j levels, and after we are done we can make a Voronoi
diagram (using the deldir package) of the k_j points y_{jl}. Maximizing
the likelihood means trying to make sure each of the x_i is in the  
Voronoi cell, i.e. the Voronoi cell corresponding with the category  
of variable j that i was
in (i.e. for which g_{ijl}=1). This generalizes multidimensional IRT
models, the choice models used for voting data in political science,
the Goodman-Haberman-Gilula-Ritov distance association models, the
Luce-Shepard choice model, and so on, to multivariate/multicategory  

It is part of the "Gifi Goes Logistic" project.
The algorithm is based on majorization, starting with multiple  
analysis, and each iteration does one step of a truncated SVD (with a
different target in each iteration). The movies show the movement of X
from one iteration to the next.

Jan de Leeuw; Distinguished Professor and Chair, UCLA Department of  
Editor: Journal of Multivariate Analysis, Journal of Statistical  
US mail: 8125 Math Sciences Bldg, Box 951554, Los Angeles, CA 90095-1554
phone (310)-825-9550;  fax (310)-206-5658;  email: deleeuw at stat.ucla.edu
.mac: jdeleeuw ++++++  aim: deleeuwjan ++++++ skype: j_deleeuw
homepages: http://gifi.stat.ucla.edu ++++++ http://www.cuddyvalley.org
           No matter where you go, there you are. --- Buckaroo Banzai

More information about the R-help mailing list