[R-sig-ME] Crash fitting lmer in lme4_0.999375-4

[Douglas Bates]

On Mon, Mar 3, 2008 at 10:16 AM, Douglas Bates <bates at stat.wisc.edu> wrote:
> On Mon, Mar 3, 2008 at 6:18 AM, Nick Isaac <njbisaac at googlemail.com> wrote:
>  > I just installed lme4_0.999375-5 and the problem has disappeared. Thanks!
>
>  Hmm - interesting.  When I try the example using that version of the
>  lme4 package it segfaults.
>
>  I'm debugging now.  Thanks for the example.  I'm pretty sure it has to
>  do with the use of the weights and that should help to isolate the
>  problem.

There were indeed a couple of dumb programming mistakes that caused
segfaults when fitting a linear mixed model with weights.  After
having fixed those I can get your example to run but the results
(enclosed) don't look correct.  I think the problem is that I am using
the penalized, weighted residual sum of squares to evaluate the
conditional estimate of the common scale parameter (i.e. sigma)
without taking into account the sum of the weights.

I already have a FIXME comment in the code that I should evaluate and
store sigma in a different way so I will make that change and check on
the way that sigma should be estimated at the same time.



>  > On 20/02/2008, Douglas Bates <bates at stat.wisc.edu> wrote:
>  > > Thanks for the report Nick and thanks for including the data so that
>  > > we can reproduce the problem.
>  > >
>  > > I don't think I will have time to look at it until next week.  I am
>  > > travelling for the rest of the week (and, naturally, still need to
>  > > finish my presentation).
>  > >
>  > >
>  > > On Feb 20, 2008 11:18 AM, Nick Isaac <njbisaac at googlemail.com> wrote:
>  > > > Dear all,
>  > > >
>  > > > I just installed lme4_0.999375-4 and am sad to report unexpected
>  > behaviour
>  > > > the first time I ran lmer.
>  > > >
>  > > > > sessionInfo()
>  > > > R version 2.6.2 (2008-02-08)
>  > > > i386-apple-darwin8.10.1
>  > > >
>  > > > locale:
>  > > > en_GB.UTF-8/en_GB.UTF-8/en_GB.UTF-8/C/en_GB.UTF-8/en_GB.UTF-8
>  > > >
>  > > > attached base packages:
>  > > > [1] stats     graphics  grDevices utils     datasets  methods   base
>  > > >
>  > > > other attached packages:
>  > > > [1] lme4_0.999375-4   Matrix_0.999375-4 lattice_0.17-4
>  > > >
>  > > > loaded via a namespace (and not attached):
>  > > > [1] grid_2.6.2
>  > > >
>  > > >
>  > > >
>  > > > The dataset is attached. A call to:
>  > > >
>  > > > data<-read.table("ESW_080220.txt",sep="\t",header=T)
>  > > > data$W<- 5-log(1+data$CV)
>  > > > full1.w.mm <- lmer(ESW ~ Wingspan * Dataset * VCS + (1|Site) +
>  > (1|Species),
>  > > > data=data, weights=W)
>  > > >
>  > > >
>  > > >
>  > > > Produces a crash with the following message:
>  > > >
>  > > >  *** caught bus error ***
>  > > > address 0x0, cause 'non-existent physical address'
>  > > >
>  > > > Traceback:
>  > > >  1: .Call(mer_optimize, ans, verbose)
>  > > >  2: mer_finalize(ans, verbose)
>  > > >  3: lmer(ESW ~ Wingspan * Dataset * VCS + (1 | Site) + (1 | Species),
>  > > > data = data, weights = W)
>  > > >
>  > > >
>  > > >
>  > > > I repeated the exercise on another machine, running  the version on
>  > CRAN:
>  > > >
>  > > >
>  > > > > sessionInfo()
>  > > > R version 2.6.2 (2008-02-08)
>  > > > i386-apple-darwin8.10.1
>  > > >
>  > > > locale:
>  > > > en_GB.UTF-8/en_GB.UTF-8/en_GB.UTF-8/C/en_GB.UTF-8/en_GB.UTF-8
>  > > >
>  > > > attached base packages:
>  > > > [1] stats     graphics  grDevices utils     datasets  methods   base
>  > > >
>  > > > other attached packages:
>  > > > [1] lme4_0.99875-9    Matrix_0.999375-4 lattice_0.17-4
>  > > >
>  > > >
>  > > >
>  > > >
>  > > > The lmer fit the model with a warning:
>  > > >
>  > > > Warning message:
>  > > > In .local(x, ..., value) :
>  > > >   Estimated variance for factor 'Site' is effectively zero
>  > > >
>  > > >
>  > > >
>  > > > The warning reflects the fact that this model is rather over-specified.
>  > > > However, the behaviour in the development version is rather unexpected.
>  > > > Best wishes, Nick Isaac
>  > > >
>  > >
>  > > > _______________________________________________
>  > > > R-sig-mixed-models at r-project.org mailing list
>  > > > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>  > > >
>  > > >
>  > >
>  >
>  >
>