[Rd] Re: [R-SIG-Mac] NaN and linear algebra

[David Firth]

I am sorry that I wasn't clear.  All that I meant was that *this* 
problem can result in different behaviour in "ordinary" statistical 
applications.  For example, if the objective function in a call to 
optim() involves calling one of these linear algebra routines, the 
result may be NaN (on systems other than Mac OS X) --- which optim will 
typically handle sensibly --- or something else (an error, or perhaps 
some consequence of getting 0 for the determinant) under Mac OS X .

Probably this was obvious to you.  Apologies if I misled you into 
thinking that there was some other problem I knew about.

Best regards,
David
On 20 Mar, 2005, at 15:08, stefano iacus wrote:
No, blas/veclib is tested, so aprt this extreme case you should report 
some other more commonly used cases in which something fails on OS X.  
This will help us to work it out.
As said, I'll try some tests without using veclib and let you know.

I've fowarded this mail to r-devel, which seems to be the right place, 
so for future msg on the subject please use r-devel.
stefano
On 19/mar/05, at 17:44, David Firth wrote:

Dear Don, Bill and Stefano
Many thanks for your helpful replies on this.  I do think this is 
pretty serious: the example I gave is an extreme one, but in real 
problems (e.g., calls to optim()) this sort of thing can and does 
result in different behaviour on the Mac than on other systems.  And 
that has to be a Bad Thing.

I'm unsure whether it is better to press Apple to improve vecLib, or 
to test R with an alternative BLAS (and if successful, recommend 
using that BLAS in place of vecLib).  Or both.  Unfortunately I don't 
know enough about these routines and the relevant standards to pursue 
either route myself.

Best regards,
David

At 11:57 AM +0000 3/16/05, David Firth wrote:
I don't know whether this is a bug, or a problem with the way I 
built R 2.0.1 (under Mac OS 10.3 on a G5), or something else.  Can 
anyone else confirm (or otherwise) that this happens in their R 
2.0.1 on Mac OS X?

 >  d<-matrix(NaN,3,3)
  d
    [,1] [,2] [,3]
[1,]  NaN  NaN  NaN
[2,]  NaN  NaN  NaN
[3,]  NaN  NaN  NaN
  solve(d)
Error in solve.default(d) : Lapack routine dgesv: system is exactly 
singular
  chol(d)
Error in chol(d) : the leading minor of order 1 is not positive 
definite
  det(d)
[1] 0
Doing the same thing on a Windows setup gave a different (and more 
useful, I think) result

 d<-matrix(NaN,3,3)
 d
    [,1] [,2] [,3]
[1,]  NaN  NaN  NaN
[2,]  NaN  NaN  NaN
[3,]  NaN  NaN  NaN
 solve(d)
    [,1] [,2] [,3]
[1,]  NaN  NaN  NaN
[2,]  NaN  NaN  NaN
[3,]  NaN  NaN  NaN
 chol(d)
    [,1] [,2] [,3]
[1,]  NaN  NaN  NaN
[2,]    0  NaN  NaN
[3,]    0    0  NaN
 det(d)
[1] NaN
Any thoughts?
David
Professor David Firth
Dept of Statistics
University of Warwick
Coventry CV4 7AL
United Kingdom
Voice: +44 (0)247 657 2581
Fax:   +44 (0)247 652 4532
Web:   http://www.warwick.ac.uk/go/dfirth
_______________________________________________
R-SIG-Mac mailing list
[EMAIL PROTECTED]
https://stat.ethz.ch/mailman/listinfo/r-sig-mac

--
--------------------------------------
Don MacQueen
Environmental Protection Department
Lawrence Livermore National Laboratory
Livermore, CA, USA
--------------------------------------
_______________________________________________
R-SIG-Mac mailing list
[EMAIL PROTECTED]
https://stat.ethz.ch/mailman/listinfo/r-sig-mac

______________________________________________
R-devel@stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel