I would like to perform a regression model with random effects and a Cox model, both with many covariates (20000). Then I need to include some kind of regularization such as lasso.
But I haven’t found that option in MixedModels.jl nor Survival.jl.
How can I do it?
I think there is no solution for mixed models + lasso now. Instead Cox regression you can use Generalized Mixed Models (MixedModels.jl) with log link-function and Poisson distribution. May be this can help to get insight.
As far as I know, there is no frequentist package offer mixed models with regularized fixed effects. You can of course set up such models in one of the Bayesian packages like Turing.jl, but that requires a lot more effort. I have started working on implementing RegularizedMixedModels, but I haven’t had time to work on it and won’t have time in the next several months.
Hello! Could you kindly point me, what penalty functions use in likelihood calculation? I found this paper, but seems it work if
R is diagonal (in V = ZDZ’ + R). Or it doesn’t matter? And I can use as second penalty functions some function of
D, for example: ?