[R-sig-ME] lme4 troubles

[Ben Bolker]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 14-12-16 05:49 PM, Andrew C. Shaver wrote:
> Hi Ben!
> 
> I hope this note finds you well.

  [I'm fine, but cc'ing to r-sig-mixed-models at r-project.org]
> 
> I hate to trouble you, but I've having some troubles with the lme4 
> package that I'm hoping you might be able to help me quick
> resolve.
> 
> Using an earlier version of the package, I was running the
> following type of model without problem:
> 
> output <- glmer(Y ~ X + Z + (1 | Province) + (1 | Year), data =
> data, family = "binomial", verbose = TRUE, control =
> list(maxIter=1000))
> 
> I've since upgraded, and attempting to run this model now results
> in the following message:
> 
> Error: PIRLS step failed In addition: Warning message: In 
> checkArgs("glmer", sparseX, ...) : extra argument(s) ‘control’, 
> ‘verbose’ disregarded

  What version of lme4 are you now using (results of sessionInfo(),
or packageVersion("lme4") ?)

  Is there any chance you can provide a reproducible example?
Using one of the built-in examples from lme4 with the latest
(development) version of lme4 (although I think it should be identical
for these purposes to the CRAN version of lme4, 1.1-7):

library("lme4")
gm1 <- glmer(cbind(incidence, size - incidence) ~ period + (1 | herd),
                   data = cbpp, family = binomial,
                  verbose=TRUE, control=list(maxIter=1000))

gives

Warning in glmer(cbind(incidence, size - incidence) ~ period + (1 |
herd),  :
  Use control=glmerControl(..) instead of passing a list of class “list”
Error in (function (optimizer = c("bobyqa", "Nelder_Mead"),
restart_edge = FALSE,  :
  unused argument (maxIter = 1000)

If I use

gm1 <- glmer(cbind(incidence, size - incidence) ~ period + (1 | herd),
                   data = cbpp, family = binomial,
                  verbose=TRUE,
                 control=glmerControl(optCtrl=list(maxfun=20000)))

it works (I don't get the specified error) -- but of course it would
be likely to work with this well-behaved example.

> If I run something like: output <- glmer(Y ~ X + Z + (1 | Province)
> + (1 | Year), data = data, family = "binomial"), i get: "Error: 
> pwrssUpdate did not converge in 30 iterations". And if I attempt
> to increase the number of iterations by including 
> "control=glmerControl(optCtrl=list(maxfun=20000)", I get:
> 
> Error in checkArgs("glmer", sparseX, ...) : could not find
> function "glmerControl"

  That's surprising.

  FWIW the max number of PIRLS iterations is hard-coded in the current
version of lme4, although you can adjust the tolerance (tolPwrss).
However, usually when this goes wrong there's some other problem
that has to/can be fixed in another way.  Again, a reproducible example
would be really useful ...

> I am also no longer able to load the "blme" package. Before the 
> update, had no problems running, for instance, bglmer(). Now, when
> I attempt to load the package, I get the following messages:
> 
> Error : class "mer" is not exported by 'namespace:lme4' Error: 
> package/namespace load failed for ‘blme’

  I'm a little surprised/confused by this one.
  Make sure you're running a clean R session and (re)install lme4
and blme from scratch?

> Any hints on how to get back to running the model??
> 
> Thank you very much in advance!
> 
> Very best,
> 
> Andrew

> --- Andrew Shaver PhD Candidate Woodrow Wilson School Princeton 
> University scholar.princeton.edu/ashaver ---
> 
> 

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)

iQEbBAEBAgAGBQJUkMiRAAoJEOCV5YRblxUH0ooH+Iw/dnfWtjDwO97X8aILcB9E
8rcS1i4QrpN5rAEWa5SnuQDNYNCzduYMxpxMsfUu1uMreTq7hG7axxP0vC4rZJxL
onyg1wGY2JIzvwpjSHm7sEEyokFIz/EGyeruptCTdCGvLDBIVr2bwA0SXa+AGSpk
Ase7quPY6RBra7K4a+OJqsj0IZXqocf+HQIMAqdIfqdaBhsuAMXXAeaCRNbxcdnT
yQlYyHUVDH/uyfeRBhSaQn+wAu7guBYRUjcuyBRYP0bhd1RjSxMov8WkoRpef60S
b0en8Nb03e+cGjQaxpXEZv1hxnsQPPOVUqCC041qIjshi9UppZpxKq47eQij6w==
=Mu75
-----END PGP SIGNATURE-----