I have a modeling task where I have quite a few potential predictors (11), but I need a simpler model and don't have a theoretical justification for simpler combinations of predictors. I'm interested in using a regularized horseshoe prior (e.g. rstanarm::hs) for regularization, but it isn't supported in glmmfields. Is it reasonable to you implement such a prior? An alternative approach to variable selection I'm interested in is projection predictive feature selection (i.e. projpred r package), but I imagine that is a bit more difficult to implement.
Hello all,
I have a modeling task where I have quite a few potential predictors (11), but I need a simpler model and don't have a theoretical justification for simpler combinations of predictors. I'm interested in using a regularized horseshoe prior (e.g. rstanarm::hs) for regularization, but it isn't supported in glmmfields. Is it reasonable to you implement such a prior? An alternative approach to variable selection I'm interested in is projection predictive feature selection (i.e. projpred r package), but I imagine that is a bit more difficult to implement.
Thanks, Connor