Dear Jan,
This routine (and also mdsvd) returns the full SVD decomposition by
default. If you just want singular vectors issue
($U, $S, $VT) = msvd($A,2,2)
The perldl shell helps a lot when you want to see the documentation of a
function. A simple 'help msvd' should do the job. Remember that each
time you add a new PDL package you have to run 'scantree.pl' and
'mkhtmldoc.pl'. These scripts should be where PDL is installed, if not
you can retrieve them from the pdl source tarball.
For example 'help msvd' should return :
Module PDL::LinearAlgebra
msvd
Computes SVD. Can compute singular values, either U or V or neither.
Return singulars values in scalar context else left (U), singular
values, right (V' (hermitian for complex) singulars vector and info.
Supports threading. If only singular values are requested, info is
returned in array context. Uses gesvd or cgesvd from Lapack.
( (PDL(U)), PDL(s), (PDL(V), PDL(info)) = msvd(PDL, SCALAR(jobu),
SCALAR(jobv))
jobu : 0 = Doesn't compute U
1 = computes full SVD (square U)
2 = computes right singular vectors
default = 1
jobv : 0 = Doesn't compute V
1 = computes full SVD (square V)
2 = computes left singular vectors
default = 1
my $a = random(10,10);
my ($u, $s, $v) = msvd($a);
Docs from /usr/lib/perl5/PDL/LinearAlgebra.pm
Glad to see your interest,
Greg
PS: And if you wonder why each routine name begins with a 'm' this is
intended to be used with an automatic completion component which helps
when you want to know what routines are available for matrices.
Le jeudi 04 juin 2009 à 23:16 +0200, Jan Hoogenraad a écrit :
> I didn't know that.
>
> This saves me a lot of work, as I see that most of the routines I
> actually need are there.
>
> This should be added to the feature request:
> https://sourceforge.net/tracker/index.php?func=detail&aid=2774886&group_id=612&atid=350612
>
> I've run a short test, and was surprised to get a 4x4 matrix back from
> the msvd routine. I was expecting to have a 3x4 returned.
>
> Do you know what is going on there ?
>
> #short test
> my $m=4;
> my $n=3;
> my $A = PDL->zeroes(double ,$n,$m);
> $A .= ($A->yvals*$n)+($A->xvals+1);
> $qmAt= pdl [
> [ 1, 2 ,3 ],
> [ 4, 5 ,6 ],
> [ 7, 8 ,9 ],
> [ 10, 11, 12 ]];
> print "$A <> $qmAt\n" if ! all approx $qmAt, $A;
> ($U, $S, $VT) = msvd($A);
>
> $qmSt= pdl [ 25.4624, 1.29066 ,2.21303e-15] ;
> $qmVTt= pdl [
> [ -0.140877, 0.824714, 0.540521 ],
> [ -0.343946, 0.426264, -0.654711 ],
> [ -0.547016, 0.0278135, -0.312142 ],
> [ -0.750086, -0.370637, 0.426331 ] ] ;
> $qmUt= pdl [
> [ -0.504533, -0.574516 ,-0.644498] ,
> [ -0.760776, -0.0571405, 0.646495] ,
> [ -0.408248, 0.816497 ,-0.4082481] ];
>
> -------------
>
> I've now more or less figured out how to do parse all .f files
> automatically. I was now starting the hard phase: figuring out how
> swapping arguments for arrays should be done.
> PDL and FORTRAN use opposite ways to store the data: in PDL, the FIRST
> dimension is contiguous in memory, while in FORTRAN, the LAST dimension
> steps by 1 through memory.
>
> I see that I can now stop that effort
>
> Is there somebody I can make happy with the parsing code ?
> I'll feed some improvements I found for the lapack .f files back to the
> lapack development.
>
> Gregory Vanuxem wrote:
> > Hello Jan, hello all
> >
> > Have you seen that :
> >
> > http://search.cpan.org/~ellipse/PDL-LinearAlgebra-0.06
> >
> >
> > This modules was built incrementally (implemented when needed and when
> > time was available) but of course a better way is to do it automatically
> > (I did it later in lisp). Beside the lot of disadvantage a manual way
> > is, it's relatively well checked, practically checked to be more
> > precise . Each routine was tested, but some bugs remains for sure. The
> > x.t check for the « educational » LinearAlgebra.pm module is on my hard
> > drive, if you want it drop me a mail.
> >
> > Cheers,
> >
> > Greg
> >
> >
> > Le dimanche 31 mai 2009 à 23:30 +0200, Jan Hoogenraad a écrit :
> >> Please help me with starting an interface to lapack.
> >> Chris Marshall (marshallch) requested that I'll put my progress and
> >> questions onto this list.
> >>
> >> I've come as far as installing PDL and starting to use inlpp_link.pl as a
> >> template.
> >>
> >> I set up attached .pl file. This does not compile, as line 64 does not
> >> parse.
> >> I've attached out.pdlpp, as this contains the error code:
> >> Invalid pdl def int n (regex
> >> byte|short|ushort|int|longlong|float|double) at (eval 16) line 7
> >> PDL::PP::PdlParObj::new('PDL::PP::PdlParObj', 'int n',
> >> 'PDL_UNDEF_NUMBER', 0) called at (eval 13) line 7
> >>
> >> I've tried to match what was done at Slatec, but there a different parser
> >> is included.
> >>
> >> The header for the specific function I am trying to call is:
> >> extern void dgesv_(int *n, int *nrhs, double *a, int *lda, int *ipiv,
> >> double *b, int *ldb, int *info);
> >>
> >>
> >>
> >> pièce jointe document texte brut (out.pdlpp)
> >> Invalid pdl def int n (regex byte|short|ushort|int|longlong|float|double)
> >> at (eval 16) line 7
> >> PDL::PP::PdlParObj::new('PDL::PP::PdlParObj', 'int n',
> >> 'PDL_UNDEF_NUMBER', 0) called at (eval 13) line 7
> >> PDL::PP::Signature::parse('int n;int nrhs;double [o]a(lda,n);int
> >> lda;int [o]ipiv(n); dou...', 0) called at
> >> /home/jhh/linsrc/PDL/blib/lib/PDL/PP/Signature.pm line 31
> >> PDL::PP::Signature::new('PDL::PP::Signature', 'int n;int nrhs;double
> >> [o]a(lda,n);int lda;int [o]ipiv(n); dou...', undef) called at
> >> /home/jhh/linsrc/PDL/blib/lib/PDL/PP.pm line 1355
> >> PDL::PP::Pars_nft('int n;int nrhs;double [o]a(lda,n);int lda;int
> >> [o]ipiv(n); dou...', undef) called at
> >> /home/jhh/linsrc/PDL/blib/lib/PDL/PP.pm line 324
> >> PDL::PP::Rule::apply('PDL::PP::Rule=HASH(0x8d73a50)',
> >> 'HASH(0x8b56b20)') called at /home/jhh/linsrc/PDL/blib/lib/PDL/PP.pm line
> >> 3236
> >> PDL::PP::translate('HASH(0x8b56b20)', 'ARRAY(0x8d83a28)') called at
> >> /home/jhh/linsrc/PDL/blib/lib/PDL/PP.pm line 1038
> >> PDL::PP::pp_def('dgsev', 'Pars', 'int n;int nrhs;double [o]a(lda,n);int
> >> lda;int [o]ipiv(n); dou...', 'GenericTypes', 'ARRAY(0x8ab2880)', 'Code',
> >> 'dgesv_($P(n), $P(nrhs), $P(a), $P(lda), $P(ipiv), $P(b), $P(l...') called
> >> at lapack_link_pl_c24d.pd line 17
> >> Inline running PDL::PP version 2.2...
> >> _______________________________________________
> >> Perldl mailing list
> >> [email protected]
> >> http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
> >
> >
>
_______________________________________________
Perldl mailing list
[email protected]
http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
