1) Use  model.matrix() / model.frame to create 'x' in the examples

2) Consider using *sparse* model matrices (for 'x') -- in the functions!

3) -- apply supclust() to
   library(caret)
   data(segmentationData)
  and compare with the prediction there,
  --> ~/R/Meetings-Kurse-etc/2011-Warwick/Tutorials/Kuhn_caret/user_2011_caret.r

4) Note: tests/wilma-ex.R  fit4 -- gives quite(?) different results for
   	 32 bit and 64 bit.
   What does that mean?  [overparametrized - "random" results ?]
