Re: unnormalized covariances

[Jason Stover]

You are right. For the double pass, we should just call
cm_to_gsl (cov). 

On Sat, Mar 06, 2010 at 08:34:39PM +0000, John Darrington wrote:
> On Thu, Mar 04, 2010 at 05:37:47PM -0500, Jason Stover wrote:
>      I need to use the "un-normalized" covariances for the
>      regression, meaning just dot products, not divided by
>      sample sizes. Does anyone mind if I add the following functions
>      to covariance.c?
>      
>      static const gsl_matrix *
>      covariance_calculate_double_pass_unnormalized (struct covariance *cov)
>      {
>        size_t i, j;
>        for (i = 0 ; i < cov->dim; ++i)
>          {
>            for (j = 0 ; j < cov->dim; ++j)
>            {
>              int idx;
>         double *x = gsl_matrix_ptr (cov->moments[MOMENT_VARIANCE], i, j);
>      
>           idx = cm_idx (cov, i, j);
>             if ( idx >= 0)
>                 {
>                       x = &cov->cm [idx];
>                 }
>           }
>          }
>      
> So far as I can see, this code has no effect.
> 
> 
> We need to be very clear about what exactly the function
> const gsl_matrix *covariance_moments (const struct covariance *cov, int m);
> 
> is returning.  Otherwise a lot of other things will break.
> 
> Perhaps we should have a different function to return non-normalised moments.
> 
> -- 
> PGP Public key ID: 1024D/2DE827B3 
> fingerprint = 8797 A26D 0854 2EAB 0285  A290 8A67 719C 2DE8 27B3
> See http://pgp.mit.edu or any PGP keyserver for public key.
> 
>