[Scipy-tickets] [SciPy] #409: LinAlgError: SVD did not converge

SciPy scipy-tickets@scipy....
Wed May 2 11:03:26 CDT 2007


#409: LinAlgError: SVD did not converge
--------------------------+-------------------------------------------------
 Reporter:  wnbell        |       Owner:  somebody                         
     Type:  defect        |      Status:  new                              
 Priority:  normal        |   Milestone:                                   
Component:  scipy.linalg  |     Version:  devel                            
 Severity:  normal        |    Keywords:  svd converge linalg factorization
--------------------------+-------------------------------------------------
 I make use of linalg.pinv2() which in turn uses linalg.svd().
 Occasionally I get the following error:


 {{{
       2
       3 A = scipy.io.mmread("A.mtx")
 ----> 4 x = scipy.linalg.svd(A)
       5
       6

 /usr/lib/python2.4/site-packages/scipy/linalg/decomp.py in svd(a,
 full_matrices, compute_uv, overwrite_a)
     503     else: # 'clapack'
     504         raise NotImplementedError,'calling gesdd from %s' %
 (gesdd.module_name)
 --> 505     if info>0: raise LinAlgError, "SVD did not converge"
     506     if info<0: raise ValueError,\
     507        'illegal value in %-th argument of internal gesdd'%(-info)

 LinAlgError: SVD did not converge
 WARNING: Failure executing file: <test_svd.py>
 }}}


 I've attached a matrix and script that generates the problem.  The matrix
 is fairly pedestrian - (264x264) and symmetric positive semidefinite, so a
 robust factorization like SVD should not be problematic.  Other methods
 like pinv() (which uses QR), lu(), and eig() seem to work fine.

-- 
Ticket URL: <http://projects.scipy.org/scipy/scipy/ticket/409>
SciPy <http://www.scipy.org/>
SciPy is open-source software for mathematics, science, and engineering.


More information about the Scipy-tickets mailing list