Skip to content

current r-forge version fails R CMD check ... ?

5 messages · Fabian Scheipl, Martin Maechler, Reinhold Kliegl +2 more

Original
Martin Maechler
# 
FS> Comparison of m2 and m3 works for me.  I use R-2.9.0
    FS> with lme4_0.999375-31, Matrix_0.999375-30 on Windoze XP
    FS> Professional SP2.

    FS> Code is below.

    FS> Best, Fabian

Thank you, Fabian (and Reinhold and Ben).

I think the version of Matrix and lme4 do not matter here,
really, but rather the compilers and libraries used to build
*and* run R.

Notably I have small suspicion that the case which show the
buggy behavior are linked to an optimized version of BLAS and
LAPACK, and these might be the culprit.
My (self-compiled) versions of R always use the R-builtin
non-optimized BLAS+LAPACK, so that would fit the picture that I
never see the problem.
Also, AFAIK, the Mac versions *do* use optimized libraries, and
Ubuntu often does too (because Debian does).

Martin
      
    >> sessionInfo()
    FS> R version 2.9.0 (2009-04-17) i386-pc-mingw32

    FS> locale:
    FS> LC_COLLATE=German_Germany.1252;LC_CTYPE=German_Germany.1252;LC_MONETARY=German_Germany.1252;LC_NUMERIC=C;LC_TIME=German_Germany.1252

    FS> attached base packages: [1] stats graphics grDevices
    FS> utils datasets methods base

    FS> other attached packages: [1] lme4_0.999375-31
    FS> Matrix_0.999375-30 lattice_0.17-25

    FS> loaded via a namespace (and not attached): [1]
    FS> grid_2.9.0 tools_2.4.1

    >> all(replicate(50, {
    FS> + set.seed(1) + D <- data.frame(y= rnorm(20,10), ff =
    FS> gl(4,5), + x1=rnorm(20,3), x2=rnorm(20,7), +
    FS> x3=rnorm(20,1)) + m2 <- lmer(y ~ x1 + (x2|ff), data = D)
    FS> + m3 <- lmer(y ~ (x2|ff) + x1, data = D) + +
    FS> identical(ranef(m2), ranef(m3)) + })) [1] TRUE

    >> set.seed(1) all(replicate(50, {
    FS> + D <- data.frame(y= rnorm(20,10), ff = gl(4,5), +
    FS> x1=rnorm(20,3), x2=rnorm(20,7), + x3=rnorm(20,1)) + m2
    FS> <- lmer(y ~ x1 + (x2|ff), data = D) + m3 <- lmer(y ~
    FS> (x2|ff) + x1, data = D) + + identical(ranef(m2),
    FS> ranef(m3)) + })) [1] TRUE

    FS> 	[[alternative HTML version deleted]]

    FS> _______________________________________________
    FS> R-sig-mixed-models at r-project.org mailing list
    FS> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
Reinhold Kliegl
# 
Not sure this helps, but I was told once that the following should hold:

model at ST  # Check on proper convergence: no zeros on diagonal

The dataframe D clearly violates this expectation. I seem to recall
that I have seen different solutions for degenerated data before;
context was a comparison between SAS and lmer.

I do NOT get  different results across runs for various models that
are well behaved with respect to the above criterion.

Reinhold

On Mon, Aug 3, 2009 at 4:36 PM, Fabian
Scheipl<Fabian.Scheipl at stat.uni-muenchen.de> wrote:
Ben Bolker
# 
Is there a simple way to determine the BLAS/LAPACK versions in use?
  I believe at the moment I have the Debian binary installed, I can build
a different version locally (although I'm not sure which BLAS/LAPACK versions
it will pick up from my system).

  Is this going to be fixable/workable-around?

  cheers
    Ben
Douglas Bates
# 
First, I apologize for my prolonged absence from the mailing list.  I
managed to get a bad cold at the end of my stay in Europe and still
had it when, shortly after returning home, we moved to a new home.  At
present I don't have internet access unless I go to the public library
or a coffee shop downstairs from where I now live.  Between unpacking
at the new home and fixing up our old home to get it ready to put on
the market, I have gone on hiatus from reading email.  I hope to have
an Internet connection by the end of this week, after which I can
start responding again.
On Mon, Aug 3, 2009 at 1:41 PM, Bolker,Benjamin Michael<bolker at ufl.edu> wrote:

            

      
      
      
    

        
I'll leave that to others to answer.  My netbook running Ubuntu is
upstairs and if I were to write out the instructions without checking
on the system I would likely get them wrong.