Skip to content

glmmadmb help

1 message · Ben Bolker

#
[cc'ing to r-sig-mixed-models]

  I had some trouble reconstructing your results, as I had difficulty
reading the file, but I did eventually succeed.  I tried it with both
lme4::glmer and glmmADMB::glmmadmb.   glmer finished after 100 seconds
(note that you also have values of "" for GENDER_; unless you know you
have claims for which an indeterminate GENDER_ is acceptable, you should
double-check your data cleaning process!)

  glmmadmb ran for 94 minutes and then quit.  It *is* quite possible to
do such large problems with AD Model Builder (the foundation of
glmmADMB), but at present glmmadmb is built for flexibility rather than
performance -- there are some tricks for speeding up random effects that
haven't been applied.  See e.g. the 'wildflowers' or 'owls' projects at
https://groups.nceas.ucsb.edu/non-linear-modeling/projects .

=====================
library(data.table)
y <- as.data.frame(fread("Article.txt"))

library(lme4)
t1 <- system.time(tt <- glmer(Claims~GENDER_+(1|T), data=y,
family="poisson",
            verbose=1))
## t1: 100 seconds elapsed

library(glmmADMB)
y$T <- factor(y$T)
t2 <- system.time(tt2 <- glmmadmb(Claims~GENDER_+(1|T),
                                  data=y, family="poisson",
                                  extra.args="-ndi 500000"))
On 13-09-29 11:50 AM, Younes Mouatassim wrote: