Hi Monks
I've already posted once on this topic today. I have gotten a little further but alas have got somewhat stuck. Basically, I have a dataset in the form of a matrix and I want to perform SVD on this dataset and then reconstruct the matrix with the most signficant data. I already have the code in the form of a Matlab M script so have been trying to convert it into perl using the PDL module. Here is the perl script so far (please bear in mind that I'm using a test dataset here, the real one will be much bigger).

#!/usr/bin/perl 

use PDL;


my $a = pdl [
           [ 0.6490,    0.0330,    0.1670,    0.3700],
           [ 0.1150,    0.0290,    0.0480,    0.0650],
           [ 0.0070,    0.3570,    0.3870,    0.2010],
           [ 0.0630,    0.0760,    0.0540,    0.0560],
               [ 0,         0.0400,    0.0160,    0.0160],
               [ 0.0120,    0.0620,    0.0430,    0.0380],
               [ 0.1530,    0.4020,    0.2850,    0.2550],

           ];


($U, $S, $V) = svd($a);

# create square matrix for 'S'

@valuesS = list $S;

$SqS = pdl [

        [ $valuesS[0], 0, 0, 0],
            [ 0, $valuesS[1], 0, 0],
        [ 0, 0, $valuesS[2], 0],
        [ 0, 0, 0, $valuesS[3]], 

        ];

# transpose V

$Vt = $V->transpose();

$Us = $U->slice('0:1,0:6');

$Ss = $SqS->slice('0:1,0:1');

$Vs = $Vt->slice('0:3,0:1');

print "$Us\n";

print "$Ss\n";

print "$Vs\n";

$recA = $Us * $Ss * $Vs;

print "$recA\n";
[download]

Most of this works just fine, but not this line

$recA = $Us * $Ss * $Vs;
[download]

Where I get the following error
PDL: PDL::Ops::mult(a,b,c): Parameter 'b' PDL barfed: Mismatched implicit thread dimension 1: should be 7, is 2
I *think* this is due to the fact that the three matrixes I am trying to multiply are of differeing dimensions. I am trying to emulate the following matlab command

recA = U(:,1:ns)*S(1:ns,1:ns)*Vt(1:ns,:)
[download]

Any suggestions as to a work around appreciated folks :)