[R-sig-ME] R-sig-mixed-models Digest, Vol 82, Issue 31

[Ben Bolker]

Iasonas Lamprianou <lamprianou at ...> writes:

 
> Dear all, I have recently faced a problem where lme4 seems to be
> broken on R version 3.0.2 for 64 bits. I run a model which was
> always OK, and I get this message:
 
> m1.4.1 <- glmer(value ~ 1 + Specialist + Qualification + Experience
> + (1 | Writer) + (1 + Gender | Rule), writers.long , binomial,
> verbose = TRUE) Warning in rankMatrix(X) : rankMatrix(<large sparse
> Matrix>, method = 'tolNorm2') coerces to dense matrix. Probably
> should rather use method = 'qrLINPACK' !?

    This is a known false positive warning that actually
emanates from the Matrix package.  Fixing this will require a coordinated
release of a new Matrix, new RcppEigen, and new lme4 package.  You
can feel free to ignore it, though.

> Has anyone seen this message before? Has anyone noticed lme4
> behaving strangely these days after an update. I also get messages
> regarding Cholemsky function and not enough memory, on models which
> used to run very happily. Any help will be appreciated.

  Can't say much about these latter cases without specific examples,
although we would certainly be interested to find out more about
such cases.

  Ben Bolker