Need help on convergence issue when fitting zero-inflated Poisson with random coefficients using gamlss
Since the proximal problem looks like (guessing from the error messages) a call to the nlminb optimizer within a call to lme is running out of iterations before converging, I'd try to see if there is a way to set the number of parameters higher. Unfortunately, I don't know how to do this off the top of my head (a brief review of the gamlss documentation didn't get me anywhere, and I haven't had time to dig through the code to see if/how this is possible) ... There are other packages that are capable of fitting mixed ZIP models (e.g. https://journal.r-project.org/archive/2017/RJ-2017-066/index.html shows examples using glmmTMB, inla, MCMCglmm, gamlss, ...) in case you can't solve the problem with gamlss (and you aren't stuck using gamlss for other reasons) cheers Ben Bolker
On Wed, Jan 22, 2020 at 2:01 PM Xia Li <odditylee at gmail.com> wrote:
Hello,
I have trouble getting convergence when fitting ZIP with (individual level)
random coefficients, the function that I used:
m1 <- gamlss(y ~ re(fixed = ~ treatment, random = ~ treatment|unit_id),
family = ZIP, data = dat)
Specifically I wanted to include global or population level treatment
effect, and individual random effect nested under treatment.
I always got the following error messages:
Error in lme.formula(fixed = fix.formula, data = Data, random = random, :
nlminb problem, convergence error code = 1
message = iteration limit reached without convergence (10)
Is there anything that I can debug from? Changing the control parameters? I
tried different algorithms like method = CG() but seems it did not help
much.
Looking for help. Thanks!
--
Best,
Xia
[[alternative HTML version deleted]]
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models