topic Re: Lasso Logistic Regression in SAS Data Science

Lasso Logistic Regression

fuatengin — Sun, 28 Aug 2016 07:21:19 GMT

Hello,
I am trying to develop a regularized(LASSO) logistic regression model in SAS EM 14.1 but as far as I understand LASSO is available only for linear regression in SAS EM. I consider selecting variables with LASSO then put the resulting non zero variables as inputs into logistic regression. Does this approach work? Or are there any ways to come up with this issue?

Thanks for your help

Engin

Re: Lasso Logistic Regression

Ksharp — Sun, 28 Aug 2016 08:07:31 GMT

The LASSO method is in PROC GLMSELECT, Check whether it could be applied in logistic regression. And actually proc logistic have an option OFFSET= which can constraint parameter sum to one due to the fact that offset variable's parameter is always fixed as 1. E.X. if you want var1 has 0.2, var2 has 0.8 then put 0.2*var1+0.8*var2 into OFFSET= . More info check documentation of proc logistic.

Re: Lasso Logistic Regression

fuatengin — Sun, 28 Aug 2016 11:53:53 GMT

Hello, thanks for your reply but I plan to use SAS Miner Nodes not SAS STAT. Can I select inputs with LARS Node and use the nonzero features in Logistic Regression model?

Re: Lasso Logistic Regression

rayIII — Sun, 28 Aug 2016 23:32:15 GMT

Hi, check out the HP Regression Node. I think it will do what you want.

Another option ion is to preselect variables using the HP Variable Selection node.

Hope this helps.

Ray