R help - Jul 2010 - Feature selection via glmnet package (LASSO)

Hello,

   I'm trying to select features of cetain numbers(like 100 out of 1000) via
   LASSO, based on multinomial model, however, it seems the glmnet package
   provides a very sparse estimation of coefficients(most of coefficients are
   0), which selects very few number of variables, like only 10, based on my
   easy dataset.

   I try to connect the choice of lambda to the selecting numbers, but it seems
   the choice is close to 0 and very sensitive.

   Can someone help me or point me to another package to do that?

   Thanks.


   Lai

Hello Lai!

You can try the elastic net which is a mixture of lasso and ridge
regression. Setting the parameter alpha to less than one will provide you
with more coefficients different from zero. I am not sure about the R
implementation. You have to search for it on your own.

Johannes
-- 
View this message in context:
http://r.789695.n4.nabble.com/Feature-selection-via-glmnet-package-LASSO-tp2308635p2532271.html
Sent from the R help mailing list archive at Nabble.com.