bad exit() from LinearAlgebra.singular_value_decomposition

 This appears to be the unresolved bug posted here:

http://sourceforge.net/tracker/index.php?func=detail&aid=923048&group_id=1369&atid=450446

please could you check if python-numpy is an alternative for you?

On Wed, 2008-02-27 at 12:00 +0000, Matthias Klose wrote:
> please could you check if python-numpy is an alternative for you?
>
>

It wouldn't be an ideal solution as it's not in the dapper repositories,
and for various strategic reasons I want to avoid (or at least minimise)
the number of tarball installs connected with this project.

I'll have a quick look and see if I can replicate the problem with numpy
though.

More soon and thanks for your attention,

Mark

> ** Also affects: python-numeric via
> http://sourceforge.net/support/tracker.php?aid=923048
> Importance: Unknown
> Status: Unknown
>
> ** Changed in: python-numeric (Ubuntu)
> Importance: Undecided => Medium
> Status: New => Confirmed
>
--
Dr. Mark Poolman
Senior Research Fellow
Cell Systems Modelling Group
School of Life Science
Oxford Brookes University

On Wed, 2008-02-27 at 12:00 +0000, Matthias Klose wrote:
> please could you check if python-numpy is an alternative for you?
>

OK, fresh install of NumPy-1.0.4, nothing else to install, same machine
as previously, the bug is absent, BUT now affected by
https://bugs.dogfood.launchpad.net/scipy/+bug/134277 (which is why I'd
fallen back to numarray in the first place ;-) ).

Cheers,

Mark

--
Dr. Mark Poolman
Senior Research Fellow
Cell Systems Modelling Group
School of Life Science
Oxford Brookes University

does this mean you don't see the error when the atlas packages are not installed?
would it be possible to recheck on hardy (with or without atlas)? all fortran libs are now built using gfortran instead of g77.

On Wed, 2008-02-27 at 15:08 +0000, Matthias Klose wrote:
> does this mean you don't see the error when the atlas packages are not installed?

Assume you mean [Bug 134277] ?
The test script for that analysed two matrices, one of which was OK with
atlas but not without, the other failed with atlas but was OK without.
This is the same for the numpy version (attached).

[Bug 195396] is not affected by presence/absence of atlas.

> would it be possible to recheck on hardy (with or without atlas)? all
> fortran libs are now built using gfortran instead of g77.
>

I can have a look for diagnostic purposes (probably not 'till next week
though), but simply upgrading the hardy would not be at all a preferred
solution (I do want the LTS).

My solution to date has been to simply do the error check and fall back
to Gauss-Jordan (+/- Gramm-Schmidt) if it fails. Although this isn't the
most elegant solution, it does yield a tolerable performance, so maybe I
should just stick with that.

/best/*

Mark

--
Dr. Mark Poolman
Senior Research Fellow
Cell Systems Modelling Group
School of Life Science
Oxford Brookes University

I tend to close this report as won't fix. there is no Numeric development anymore, the next LTS is released.