Hello,
I encounter a problem when using the "gl1ce" function from "lasso2" package.
First, I tried something like:
etastart <- log(mean(y))
fit.gl1ce.log <- gl1ce(y~., data = xy, family = gaussian(log))
and nearly all the coefficients in "fit.gl1ce.log" were zero (except the
intercept and one predictor out of 25), and no warnings were produced.
This is not what I expected, since the "glm" function with "gaussian(log))"
could fit a model using all the 25 predictors, and when using the " fit.l1ce"
function:
fit.l1ce <- l1ce(log(y)~., data = xy)
ca. 33% or the coefficients were set to zero, which is OK.
The problem with having too many zero-coefficients persists even when using the
"gl1ce" with identity link:
fit.gl1ce.id <- gl1ce(y ~ . , data = xy, family = gaussian())
fit.gl1ce.id <- gl1ce(log(y) ~ . , data = xy, family = gaussian())
I wonder what should I do more for getting a reasonable model with the "gl1ce"
function.
Thanks,
Liviu
-------------------------------------------------------------------
Liviu Theodor Ene, PhD
Researcher - Forest Inventory
Norwegian University of Life Sciences
Department of Ecology and
Natural Resource Management
Pb. 5003, 1500 - INA, 1432, Ås, Norway
Tel: +47 - 64 96 57 66
Fax: +47 - 64 96 58 01
www.umb.no
www.forestinventory.no
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
