[Rd] an Update on the "Woods" package--classification and constrained L1 regression for binary response

Izmirlian, Grant (NIH/NCI) [E] izmirlig at mail.nih.gov
Fri Dec 9 21:14:13 CET 2005


Hello R-devel:

This is an update on my R package, "woods" that does bagged classification trees
using data structures in C. Most of the comments of my earlier post still
apply, with some additions (noted *)

    (i) fits a single classification tree to dataset (R function CT)
   (ii) basic functionality of Random Forest, e.g. bagged trees with choices
        about sample size, with/without replacement, size of (random) subset
        of covariates drawn when nodes are split.  Result contains the oob votes,
        and a matrix representing the forest structure.
 *(iii) for each element of the sample, discovers all unique paths from a root node to 
        a terminal node as a sequence of splits on covariates and uses these to fit
        a lasso regresssion to the binary response using a full c-implementation of
        the Turlach lasso2 function gl1ce.

It is now available at http://mysite.verizon.net/izmirlian/woods_1.00.tar.gz


Grant Izmirlian
NCI



More information about the R-devel mailing list