[SciPy-User] scipy.sparse.linalg.lobpcg broken?

Nico Schlömer nico.schloemer@gmail....
Wed Oct 13 15:57:49 CDT 2010


Bisection action yields:

Line 131ff:

    print 'gramVBV', gramVBV
    sla.inv( gramVBV, overwrite_a = True )
    print 'gramVBV', gramVBV

return

gramVBV [[ 2.23606798]]
gramVBV [[ 2.23606798]]

in the first iteration where it should be

   gramVBV [[ 2.23606798]]
   gramVBV [[ 0.4472136]]

Right? It looks like "overwrite_a = True" is of no effect on my
machine. A quick check independent of lobpcg confirms this.

--Nico


On Wed, Oct 13, 2010 at 10:48 PM, Bruce Southey <bsouthey@gmail.com> wrote:
>  On 10/13/2010 03:30 PM, Nico Schlömer wrote:
>>> Did you use a random matrix as input? If yes, did you set the seed first,
>>> so that it is testing the same matrix?
>> I'm using the identity matrix, as in
>>
>> =========================== *snip* ===========================
>> from scipy.sparse.linalg import lobpcg
>> from scipy.sparse import identity
>> import numpy as np
>>
>> n = 5
>> X = np.ones( (n, 1) )
>> A = identity( n )
>> lobpcg( A, X, verbosityLevel = 10 )
>> =========================== *snap* ===========================
>>
>> Bug report at<http://projects.scipy.org/scipy/ticket/1304>.
>>
>> Cheers,
>> Nico
>>
>>
>>
>> On Wed, Oct 13, 2010 at 9:46 PM, Pauli Virtanen<pav@iki.fi>  wrote:
>>> Wed, 13 Oct 2010 21:23:19 +0200, Nico Schlömer wrote:
>>>>> It actually works for me:
>>>> Okay, nice. I tried to run the script on another machine too, where it
>>>> appears to work alright.
>>> Did you use a random matrix as input? If yes, did you set the seed first,
>>> so that it is testing the same matrix?
>>>
>>>> This exact same code fails on my own machine locally, meaning I'm
>>>> probably missing some crucial dependencies (of Scipy?). Using Scipy
>>>> 0.8.0 on Gentoo.
>>>> Any idea what that could possibly be?
>>> The LOBPCG code in Scipy is pretty self-contained. So I can't guess much
>>> else than issues with BLAS or LAPACK?
>>>
>>> I'd maybe make a copy of lobpcg.py and put some checks in the symeig
>>> routine. But this is just guessing, I don't know in detail how LOBPCG
>>> works.
>>>
>>> --
>>> Pauli Virtanen
>>>
>>> _______________________________________________
>>> SciPy-User mailing list
>>> SciPy-User@scipy.org
>>> http://mail.scipy.org/mailman/listinfo/scipy-user
>>>
>> _______________________________________________
>> SciPy-User mailing list
>> SciPy-User@scipy.org
>> http://mail.scipy.org/mailman/listinfo/scipy-user
>
> Due to past history, I see eigenvalue related problems I suspect
> lapack/blas/altas so please check your installation these packages.
>
> Can you please provide details on your system?
> Like OS (gentoo is insufficient), 32 vs 64 bit, python version, numpy
> version, lapack/blas/altas versions etc.,
> such the output from np.show_config()
>
> Also, have you tried the dev versions of scipy?
>
> Really you need to find someone else to confirm this.
>
> Bruce
>
>
>
> _______________________________________________
> SciPy-User mailing list
> SciPy-User@scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-user
>


More information about the SciPy-User mailing list