[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: eigs arpack arpack++ and Debian
From: |
David Bateman |
Subject: |
Re: eigs arpack arpack++ and Debian |
Date: |
Tue, 21 Mar 2006 14:15:44 +0100 |
User-agent: |
Mozilla Thunderbird 1.0.6-7.5.20060mdk (X11/20050322) |
Christophe Prud'homme wrote:
>David,
>
>| I spent a large time about a year ago trying to get eigs running in
>| octave. I have a full implementation but it is unstable. There is a
>| thread about it on the maintainers list in April and May last year.
>I saw a thread on this and your post. I was wondering if Debian disabled
>eigs/arpack or if it was actually implemented for good in octave.
>
>
It was never included in octave as I don't consider it sufficiently stable.
>|
>| I believe there is something I'm doing incorrectly with arpack that
>| makes the results unstable, and I'm not sure what. For real PD matrices,
>| this means that the algorithm seems to be slower to converge than the
>| matlab equivalent, whereas as for other types of matrices, the incorrect
>| eigenvalues are returned. Someone who knows more about the Arnoldi
>| algorithm or even better arpack itself than me really needs to look at
>| this code to fix.
>|
>| I attach my latest version of eigs.cc to this message. To build it run
>|
>| % mkoctfile eigs.cc -larpack
>I will have a look
>I never wanted to use arpack directly but rather arpack++ which simplifies
>things a lot. Moreover you have different strategies depending on the nature
>of the problem which is easier for a human being to handle with arpack++
>
>
I tried to avoid arpack++ to try and avoid the need for an additional
dependency. I suppose that arpack++ would be ok, if it made things
simpler...
I handled all of the cases in eigs.cc that are handled by the same
function in matlab, and that is what I meant by a "full
implementation"... I'm missing cholB flags treatment, as I didn't have a
sparse cholesky at the time I wrote eigs.cc. I do now and so that should
be easy to add. There is also another small case that wasn't implemented
for reasons that escape me at the moment.
>|
>| There are integrated tests in the code that can test its full
>| functionality. that can be run from the octave prompt as
>|
>| % test eigs.cc
>|
>| though you'll either need the octave CVS (or 2.9.5 that is coming out
>| very shortly), or octave-forge to have the test function available.
>I will get octave/cvs and see what I can do
>
>
2.9.5 should be in the next day or so.
>| If you have experience with arpack, then perhaps you might like to take
>| a look at this code and see if you can see what I did wrong...
>I have some experience with arpack/arpack++ which has been positive so far
>As I said in my previous email I have a nice testsuite in arpack++ for a wide
>range of problems that could be helpful to debug eigs() in octave.
>
>
This is great news... The test I have are octave specific, and should
all run (but don't). If your tests allow comparison with similar results
in octave, that would be a great help.
Regards
David
>Best regards
>C.
>
>
--
David Bateman address@hidden
Motorola Labs - Paris +33 1 69 35 48 04 (Ph)
Parc Les Algorithmes, Commune de St Aubin +33 6 72 01 06 33 (Mob)
91193 Gif-Sur-Yvette FRANCE +33 1 69 35 77 01 (Fax)
The information contained in this communication has been classified as:
[x] General Business Information
[ ] Motorola Internal Use Only
[ ] Motorola Confidential Proprietary
-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.
Octave's home on the web: http://www.octave.org
How to fund new projects: http://www.octave.org/funding.html
Subscription information: http://www.octave.org/archive.html
-------------------------------------------------------------