I have a question relating to mvpart, which I hope you can answer.
We recently conducted a study using TBR. In our first study, we used
"regular" TBR in SPSS to model 1 dependent variable. Note we have a
relatively small data-set of 100 cases. In SPSS, we used a minimum change of
improvement smaller than 0.000001 as a stopping rule. Also, we chose the 1SE
"rule", set the maximum tree depth to 10 levels, and the minimum
number of
cases was set to 5 for parent nodes and 3 for child nodes. Now we would like
to proceed with fitting a multivariate tree. We only used pruning by the
way, no v-fold cross validation afterwards. Using the aforementioned
criteria in univariate analyses resulted in relatively large trees in SPSS,
but using mvpart with xv=1se, cp=0.000001,minsplit=5,minbucket=3 resulted in
a tree with only 1 or 2 splits. This makes us wonder what causes this
dramatic difference in the tree size produced by SPSS vs. mvpart. If I use
the "pick" option in mvpart I am able to "pick" the
SPSS-tree, but the X-val
Relative Error is quite large. The plot looks nothing like the one in the
paper by De'ath by the way, the Relative Error keeps increasing at first and
then levels off quite far above the Min+1SE line.
Thanks!
--
View this message in context:
http://r.789695.n4.nabble.com/mvpart-versus-SPSS-tp4583079p4583079.html
Sent from the R help mailing list archive at Nabble.com.
