[SciPy-user] generalized eig algorithm did not converge

manouchk manouchk at gmail.com
Tue Mar 21 05:57:08 CST 2006


Well I don't know if there is a link but I try to build scipy-0.4.6 on my 
machine ,during testing it could finish this :
check_simple (scipy.linalg.tests.test_decomp.test_eig)

(It just would pass this test, there is maybe a link?)

See herer my old report:
   
Hi,

I'm again trying to build scipy. Now I get much further. It does last a long 
time first to compile full lapack and blas stuff and then build scipy 0.4.6. 
After it runs the tests it fails "only" once :
W.II.A.0. Print ROUND with only one digit. ... FAIL

but few tests later it does stop at that line :
check_simple (scipy.linalg.tests.test_decomp.test_eig)

and "never" passes this test or it is a "very long" test (more than 20 
minutes with 1.5GHz dothan proc).

To build scipy with complete lapack, blas I use the way, the old mandrake RPM 
package does, it builds lapack and blas and link them inside the tree of 
scipy.

I guess I'm still missing something?


Le Mardi 21 Mars 2006 04:52, Nils Wagner a écrit :
> Nils Wagner wrote:
> > Hi all,
> >
> > What is the reason for this message ?
> >
> > Traceback (most recent call last):
> >   File "gh.py", line 4, in ?
> >     w = linalg.eigvals(G,H)
> >   File "/usr/lib64/python2.4/site-packages/scipy/linalg/decomp.py", line
> > 244, in eigvals
> >     return eig(a,b=b,left=0,right=0,overwrite_a=overwrite_a)
> >   File "/usr/lib64/python2.4/site-packages/scipy/linalg/decomp.py", line
> > 121, in eig
> >     return _geneig(a1,b,left,right,overwrite_a,overwrite_b)
> >   File "/usr/lib64/python2.4/site-packages/scipy/linalg/decomp.py", line
> > 72, in _geneig
> >     if info>0: raise LinAlgError,"generalized eig algorithm did not
> > converge"
> > scipy.linalg.basic.LinAlgError: generalized eig algorithm did not
> > converge
> >
> > Matlab is able to solve this generalized eigenvalue problem without any
> > warning.
> >
> > Afaik, Matlab and scipy's eig is based on LAPACK. So what is the reason
> > for the difference ?
> >
> > Any idea ?
> >
> > Nils
> >
> > ------------------------------------------------------------------------
> >
> > from scipy import *
> > G = io.mmread('G.mtx')
> > H = io.mmread('H.mtx')
> > w = linalg.eigvals(G,H)
> > ------------------------------------------------------------------------
> >
> > _______________________________________________
> > SciPy-user mailing list
> > SciPy-user at scipy.net
> > http://www.scipy.net/mailman/listinfo/scipy-user
>
> I forgot to ask if someone can reproduce this problem.
>
> Travis, Pearu, Robert K., do you have a clue why Matlab can solve this
> eigenproblem while scipy cannot cope with that task.
>
> Nils
>
>
>
> _______________________________________________
> SciPy-user mailing list
> SciPy-user at scipy.net
> http://www.scipy.net/mailman/listinfo/scipy-user



More information about the SciPy-user mailing list