 "Dimitri Pourbaix" <pourbaix@astro.ulb.ac.be> a écrit :
> Luc,
>
> > I think there are two parts in the isse.
> > The first one is related to sparse matrix and we don't have an
> answer
> > yet. The second part is related to compute a partial set of
> singular
> > values. This is used for example in image compression or to find a
> > matrix with reduce rank that is the closest possible to an input
> matrix.
> > for this part, we may have an answer.
>
> What do you mean by partial set of singular values? If you mean
> setting
> all the singular values which are either below some threshold, of
> index
> above some value (rank), ... to zero and to compute the resulting
> product
> as an approximation of the original matrix, this is no longer the
> business
> of SVD but rather the user business as (s)he decides what (s)he does
> with
> the decomposition. However, one could add a method to SVD which
> would
> return such a 'product'.
This is the business of SVD. See for example <http://public.lanl.gov/mewall/kluwer2002.html>,
or <http://www2.imm.dtu.dk/~pch/Projekter/tsvd.html>, or even <http://en.wikipedia.org/wiki/Singular_value_decomposition#Matrix_approximation>
and <http://en.wikipedia.org/wiki/Singular_value_decomposition#Reduced_SVDs>.
The point in these methods is not to have an exact representation of A = USVt but to truncate
it in order to have a matrix A' with some interesting properties. If the initial matrix has
a very large dimension, rather than computing the full decomposition and later withdrawing
smaller singular values and associated vectors, we prefer to not compute them.
Luc
>
> Regards,
> Dim.
> 
> Dimitri Pourbaix *
> Institut d'Astronomie et d'Astrophysique * Don't worry, be happy
> CP 226, office 2.N4.211, building NO * and CARPE DIEM.
> Universite Libre de Bruxelles *
> Boulevard du Triomphe * Tel : +322650.35.71
> B1050 Bruxelles * Fax :
> +322650.42.26
> http://sb9.astro.ulb.ac.be/~pourbaix *
> mailto:pourbaix@astro.ulb.ac.be
>
> 
> To unsubscribe, email: devunsubscribe@commons.apache.org
> For additional commands, email: devhelp@commons.apache.org

To unsubscribe, email: devunsubscribe@commons.apache.org
For additional commands, email: devhelp@commons.apache.org
