[SciPy-dev] Please help!

Fernando Perez fperez.net@gmail....
Sun Aug 23 00:32:52 CDT 2009


On Fri, Aug 21, 2009 at 9:50 AM, Aric Hagberg<aric.hagberg@gmail.com> wrote:

> It should work for complex double.  I just ran your example code and
> it executed without failure for me.  It seems like it is giving the
> correct answer (d and dreal, v and vreal are the same).
>
> It may be that you have a miscompiled ARPACK in scipy or some part of
> the complex double LAPACK/BLAS that ARPACK calls?
>
> What platform are you using? I ran this test with stock Ubuntu
> versions of LAPACK and BLAS and the dev versions of scipy and numpy.

Aric, I bet you're on Ubuntu 8.10 or prior, and  the OP is on 9.04.
It does fail on 9.04:

maqroll[Desktop]> python test_complex.py
[ -1.33333333e+00 +0.00000000e+00j  -1.33333333e+00 +0.00000000e+00j
   0.00000000e+00 +0.00000000e+00j  -1.23857134e-32 +0.00000000e+00j
  -6.49706555e-33 +2.10704681e-33j  -6.49706555e-33 -2.10704681e-33j
  -5.24286184e-34 +2.27743942e-33j  -5.24286184e-34 -2.27743942e-33j
   2.01220438e-33 +0.00000000e+00j   5.24878615e-32 +0.00000000e+00j]
Traceback (most recent call last):
  File "test_complex.py", line 98, in <module>
    (d,v) = eigen(A,k=10,which='SR')
  File "/home/fperez/usr/opt/lib/python2.6/site-packages/scipy/sparse/linalg/eigen/arpack/arpack.py",
line 220, in eigen
    raise RuntimeError("Error info=%d in arpack"%info)
RuntimeError: Error info=-9999 in arpack


And the problem is not your fault , it's a Debian/Ubuntu bug:

https://bugs.launchpad.net/ubuntu/+source/atlas/+bug/363510

The bad news is nothing much seems to be happening upstream.  The good
news is that  Matthew  Brett has  been pushing for ways to help with
this, and both David Cournapeau and Ondrej stepped up to fight it,
and  I think they've finally made progress (David told me on Friday
afternoon at the conference that they'd been able to crack some of the
problems related to the build).  I'm sure they'll post soon with more
details.

Cheers,

f


More information about the Scipy-dev mailing list