topic How to penalize subset of parameters in Statistical Procedures

How to penalize subset of parameters

dschmidt — Tue, 27 Oct 2020 13:41:14 GMT

Forgive me if a dumb question:

Can you penalize (L1 or L2 penalty) a subset of predictors in any SAS procedures?

I'm looking at:

2 categorical variables

1 group of non-mutually exclusive dummy variables (list of a few hundred medical conditions)

Can I penalize the latter group of dummy variables without penalizing the categorical variables?

Re: How to penalize subset of parameters

StatDave — Tue, 27 Oct 2020 14:28:37 GMT

NLMIXED is a procedure that can fit a wide variety of model and can be used to apply penalization including L1 (LASSO), L2 or a combination (elastic net) regularization. See this note which discusses and illustrates this. In particular, see Example 4 which shows how a subset of the model parameters can be penalized.

Re: How to penalize subset of parameters

dschmidt — Tue, 27 Oct 2020 14:34:12 GMT

Thanks!