When I used the glmmlasso for a linear mixed model (gaussian), I got a warning message:
1: In log((1 - mu)[y == 0]) : NaNs produced
2: In nlminb(1e-16, taylor.opt, y = y, X = Z_alles, fixef = Delta_start[1:lin], ... :
NA/NaN function evaluation
I looked into the code of glmmlasso and in the function taylor.opt, it defines the loglik as
loglik<--(sum(log(mu[y==1]))+sum(log((1-mu)[y==0])))+0.5*t(delta[(length(fixef)+1):length(delta)])%*%P%*%delta[(length(fixef)+1):length(delta)]
And in my data I have some value of y==0 and the calculated mu are greater than 100, and then I got the warnings above.
Does anyone knows why it needs (sum(log(mu[y==1]))+sum(log((1-mu)[y==0]))) even for gaussian case?
