[R] Problem with new(er) R version's matrix package
maechler at stat.math.ethz.ch
Sat Apr 26 22:00:13 CEST 2014
>>>>> Arne Henningsen <arne.henningsen at gmail.com>
>>>>> on Sat, 26 Apr 2014 08:15:37 +0200 writes:
> On 25 April 2014 20:15, David Winsemius
> <dwinsemius at comcast.net> wrote:
>> On Apr 25, 2014, at 9:17 AM, Werner W. wrote:
>>> Dear Rs,
>>> I am re-executing some older code. It does work in the
>>> ancient R 2.12.0 which I still have on my PC but with
>>> the new version R 3.1.0 it does not work any more (but
>>> some other new stuff, which won't work with 2.12).
>>> The problem arises in context with the systemfit package
>>> using the matrix package. In R 3.1.0 the following error
>>> is thrown: Error in as.matrix(solve(W, tol =
>>> solvetol)[1:ncol(xMat), 1:ncol(xMat)]) : error in
>>> evaluating the argument 'x' in selecting a method for
>>> function 'as.matrix': Error in .solve.sparse.dgC(as(a,
>>> "dgCMatrix"), b = b, tol = tol) : LU computationally
>>> singular: ratio of extreme entries in |diag(U)| =
>>> However, I have no clue what I can do about this. Was
>>> there some change in the defaults of the matrix package?
>>> I couldn't find anything apparent in the changelog. As
>>> the same code works in R 2.12.0, I suppose that the
>>> problem is not my data.
>> You have not told us what version of the Matrix package
>> you were using. As such I would suggest that you review
>> the Changelog which is a link for the CRAN page for
>> pkg:Matrix and go back 4 years or so since R major
>> versions change about once a year.
> In addition, please provide a minimal, self-contained,
> reproducible example.
Yes, please do. As maintainer of the Matrix package, I'm
willing to look into the situation of course.
As was mentioned, many things have changed in 4 years.
The error message above looks like you'd want to invert a
(very close to) singular matrix, and there could be quite few
reasons why parts of the older code gave slightly different
Without a reproducible example, we can't get started though.
Martin Maechler, ETH Zurich
More information about the R-help