[SciPy-user] Not a number in linalg.svdvals

[Nils Wagner]

Pietro Berkes schrieb:
> 
> The whole story is very strange: if the tests pass, then svdvals should be
> allright. Try inserting in svd, line 274 in decomp.py the following line:
> 
> print u,s,v,info
> 
> and maybe also
> 
> print lwork
> 
> Then call linalg.svdvals again, and look if 's' looks fine.

I have modified decomp.py according to your suggestion, that is in
decomp.py line 274

    overwrite_a = overwrite_a or a1 is not a
    gesdd, = get_lapack_funcs(('gesdd',),(a1,))
    if gesdd.module_name[:7] == 'flapack':
        lwork = calc_lwork.gesdd(gesdd.prefix,m,n,compute_uv)[1]
        u,s,v,info = gesdd(a1,compute_uv = compute_uv, lwork = lwork,
                      overwrite_a = overwrite_a)


        print u,s,v,info
        print lwork


Please find attached my small test program. This is 
the output


Compute singular values using svd

[[-0.01064368  0.93221664 -0.35956157  0.03967781]
 [ 0.02760821  0.35939368  0.93265212  0.01529682]
 [ 0.35956157 -0.03967781 -0.01064368  0.93221664]
 [-0.93265212 -0.01529682  0.02760821  0.35939368]] [ 1.29446282 
1.00042554  0.99970394  0.72808433] [[-0.40511624  0.91346254
-0.01377785  0.0357378 ]
 [ 0.02830277  0.01255214  0.93261333  0.35954662]
 [ 0.01552803 -0.03501285 -0.35945512  0.932376  ]
 [-0.91369506 -0.40521936  0.02888879  0.01113738]] 0
140

Singular values using svd

[ 1.29446282  1.00042554  0.99970394  0.72808433]

Compute singular values using svdvals

[ [ 0.]] [              nan               nan              
nan               nan] [ [ 0.]] 0
268

Singular values using svdvals

[              nan               nan               nan              
nan]

Nils

> 
> Pietro.
> 
> On Tuesday 22 October 2002 10:17, you wrote:
> > Pearu,
> >
> > I have removed the whole scipy installation. Then I have reinstalled
> > scipy
> > using latest CVS, but the problem with linalg.svdvals(A)
> >
> > [              nan               nan               nan
> > nan]
> >
> > is still there.
> >
> > svdvals is defined in decomp.py and afaik it uses svd, but svd works
> > well here.
> >
> > It is exasperating !
> >
> > Any idea ?
> >
> > Nils
> >
> > > _______________________________________________
> > > SciPy-user mailing list
> > > SciPy-user at scipy.net
> > > http://www.scipy.net/mailman/listinfo/scipy-user
> >
> > _______________________________________________
> > SciPy-user mailing list
> > SciPy-user at scipy.net
> > http://www.scipy.net/mailman/listinfo/scipy-user
> _______________________________________________
> SciPy-user mailing list
> SciPy-user at scipy.net
> http://www.scipy.net/mailman/listinfo/scipy-user
-------------- next part --------------
from scipy import *
from scipy.linalg import *
A = array(([ 0.  ,  0.  ,  1.  ,  0.  ],
[ 0.  ,  0.  ,  0.  ,  1.  ],
[-0.81,  0.15, -0.02,  0.  ],
[ 0.25, -1.21,  0.  , -0.02]))
print
print 'Compute singular values using svd'
print
u,s,v = svd(A)
print
print 'Singular values using svd'
print
print s
print
print 'Compute singular values using svdvals'
print
s1 = svdvals(A)
print
print 'Singular values using svdvals'
print
print s1