Douglas Bates
2004-Jan-29 23:36 UTC
[Rd] Developmental version of Matrix package for R-1.9.0
I recently uploaded a developmental version of the Matrix package, Matrix_0.6-1.tar.gz, to CRAN where it is in the src/contrib/1.9.0/Other directory. It requires some of the packages that will appear in R-1.9.0. This version marks a major redesign of the Matrix package to use S4 classes and methods and to incorporate sparse matrix manipulations using routines from TAUCS (http://www.tau.ac.il/~stoledo/taucs/), Metis (http://www-users.cs.umn.edu/~karypis/metis/) and UMFPACK (http://www.cise.ufl.edu/research/sparse/umfpack). Although this package is very much a work in progress and the documentation is, shall we say, sketchy, there are two vignettes included with the package. One of these illustrates the performance of different ways of performing least squares calculations, using an example from Koenker and Ng's SparseM package. The results are, I think, impressive. Although not documented in this release, there are the beginnings of routines to represent large, sparse, pairwise crosstabulations as sparse matrices. I am using these in calculations for mixed-effects models and I understand that others may be interested in them for other applications. I would be happy to correspond off-list if you would like a preview of the capabilities. Metis is a package for partioning unstructured graphs, partitioning meshes, and computing fill-reducing orderings of sparse matrices. Although the entire source code for Metis is included in the Matrix package I am only using one of the functions for computing fill-reducing orderings. Perhaps those who are working on graphical models may want to look at some of the other capabilities of Metis. Certainly it provides standard ways of representing and manipulating unstructured graphs. I've only compiled and tested the package on Linux_x86 environments. I will be interested in whether Uwe is successful in building it for Windows and whether there are problems on 64-bit machines. -- Douglas Bates bates@stat.wisc.edu Statistics Department 608/262-2598 University of Wisconsin - Madison http://www.stat.wisc.edu/~bates/
Jan de Leeuw
2004-Jan-29 23:51 UTC
[Rd] Developmental version of Matrix package for R-1.9.0
Builds OK on OS X, except for 1. The static archives foo.a are not ranlibbed, so you have to do that manually. 2. Cannot build Matrix.so because it uses flat namespaces and has multiple definitions of some symbols. I solved 2 by the usual hack, which is to add the -Xlinker -m flag to the link by the gcc driver, but the preferred solutions would be to use external in the code. Or, even better, to switch the Darwin build to two-level namespaces. On Jan 29, 2004, at 14:35, Douglas Bates wrote:> I recently uploaded a developmental version of the Matrix package, > Matrix_0.6-1.tar.gz, to CRAN where it is in the > src/contrib/1.9.0/Other directory. It requires some of the packages > that will appear in R-1.9.0. > > This version marks a major redesign of the Matrix package to use S4 > classes and methods and to incorporate sparse matrix manipulations > using routines from TAUCS (http://www.tau.ac.il/~stoledo/taucs/), > Metis (http://www-users.cs.umn.edu/~karypis/metis/) and UMFPACK > (http://www.cise.ufl.edu/research/sparse/umfpack). > > Although this package is very much a work in progress and the > documentation is, shall we say, sketchy, there are two vignettes > included with the package. One of these illustrates the performance > of different ways of performing least squares calculations, using an > example from Koenker and Ng's SparseM package. The results are, I > think, impressive. > > Although not documented in this release, there are the beginnings of > routines to represent large, sparse, pairwise crosstabulations as > sparse matrices. I am using these in calculations for mixed-effects > models and I understand that others may be interested in them for > other applications. I would be happy to correspond off-list if you > would like a preview of the capabilities. > > Metis is a package for partioning unstructured graphs, partitioning > meshes, and computing fill-reducing orderings of sparse matrices. > Although the entire source code for Metis is included in the Matrix > package I am only using one of the functions for computing > fill-reducing orderings. Perhaps those who are working on graphical > models may want to look at some of the other capabilities of Metis. > Certainly it provides standard ways of representing and manipulating > unstructured graphs. > > I've only compiled and tested the package on Linux_x86 environments. > I will be interested in whether Uwe is successful in building it for > Windows and whether there are problems on 64-bit machines. > -- > Douglas Bates bates@stat.wisc.edu > Statistics Department 608/262-2598 > University of Wisconsin - Madison > http://www.stat.wisc.edu/~bates/ > > ______________________________________________ > R-devel@stat.math.ethz.ch mailing list > https://www.stat.math.ethz.ch/mailman/listinfo/r-devel > >==Jan de Leeuw; Professor and Chair, UCLA Department of Statistics; Editor: Journal of Multivariate Analysis, Journal of Statistical Software US mail: 8130 Math Sciences Bldg, Box 951554, Los Angeles, CA 90095-1554 phone (310)-825-9550; fax (310)-206-5658; email: deleeuw@stat.ucla.edu homepage: http://gifi.stat.ucla.edu ------------------------------------------------------------------------ ------------------------- No matter where you go, there you are. --- Buckaroo Banzai http://gifi.stat.ucla.edu/sounds/nomatter.au
Prof Brian Ripley
2004-Jan-30 09:29 UTC
[Rd] Developmental version of Matrix package for R-1.9.0
On 29 Jan 2004, Douglas Bates wrote:> I recently uploaded a developmental version of the Matrix package, > Matrix_0.6-1.tar.gz, to CRAN where it is in the > src/contrib/1.9.0/Other directory. It requires some of the packages > that will appear in R-1.9.0....> I've only compiled and tested the package on Linux_x86 environments. > I will be interested in whether Uwe is successful in building it for > Windows and whether there are problems on 64-bit machines.I don't think Uwe builds on R-devel routinely. It didn't work for me and I will send you a patch when I have resolved this. It also did not build under Linux (RH8.0 + gcc 3.3.2), as Metis/proto.h:462: error: conflicting types for `__log2' /usr/include/bits/mathcalls.h:145: error: previous declaration of `__log2' It did build on 64-bit Solaris, but failed to load Matrix.so, with ld.so.1: /home/markov/ripley/new/R/R64/bin/R.bin: fatal: relocation error: file /export/staff/ripley/new/R/Matrix.Rcheck/Matrix/libs/Matrix.so: symbol isinf: referenced symbol not found There appears to be an attempted define for this in src/taucs/taucs.h, but isinf is only used in an earlier define! Again, patch to follow. My impression was that the packages you are using had not been written very carefully, so there may be a lot more of this sort of thing. Brian -- Brian D. Ripley, ripley@stats.ox.ac.uk Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/ University of Oxford, Tel: +44 1865 272861 (self) 1 South Parks Road, +44 1865 272866 (PA) Oxford OX1 3TG, UK Fax: +44 1865 272595
Roger D. Peng
2004-Jan-30 15:21 UTC
[Rd] Developmental version of Matrix package for R-1.9.0
The packages builds, loads, and passes R CMD check fine on
our (64-bit) AMD Opteron machine running SuSE Linux 8.0 and
gcc 3.2.2. I used
> version
_
platform x86_64-unknown-linux-gnu
arch x86_64
os linux-gnu
system x86_64, linux-gnu
status Under development (unstable)
major 1
minor 9.0
year 2004
month 01
day 29
language R
Hope this helps,
-roger
Douglas Bates wrote:> I recently uploaded a developmental version of the Matrix package,
> Matrix_0.6-1.tar.gz, to CRAN where it is in the
> src/contrib/1.9.0/Other directory. It requires some of the packages
> that will appear in R-1.9.0.
>
> This version marks a major redesign of the Matrix package to use S4
> classes and methods and to incorporate sparse matrix manipulations
> using routines from TAUCS (http://www.tau.ac.il/~stoledo/taucs/),
> Metis (http://www-users.cs.umn.edu/~karypis/metis/) and UMFPACK
> (http://www.cise.ufl.edu/research/sparse/umfpack).
>
> Although this package is very much a work in progress and the
> documentation is, shall we say, sketchy, there are two vignettes
> included with the package. One of these illustrates the performance
> of different ways of performing least squares calculations, using an
> example from Koenker and Ng's SparseM package. The results are, I
> think, impressive.
>
> Although not documented in this release, there are the beginnings of
> routines to represent large, sparse, pairwise crosstabulations as
> sparse matrices. I am using these in calculations for mixed-effects
> models and I understand that others may be interested in them for
> other applications. I would be happy to correspond off-list if you
> would like a preview of the capabilities.
>
> Metis is a package for partioning unstructured graphs, partitioning
> meshes, and computing fill-reducing orderings of sparse matrices.
> Although the entire source code for Metis is included in the Matrix
> package I am only using one of the functions for computing
> fill-reducing orderings. Perhaps those who are working on graphical
> models may want to look at some of the other capabilities of Metis.
> Certainly it provides standard ways of representing and manipulating
> unstructured graphs.
>
> I've only compiled and tested the package on Linux_x86 environments.
> I will be interested in whether Uwe is successful in building it for
> Windows and whether there are problems on 64-bit machines.