Thread Subject:

regularized SVD on a sparse matrix.

Hi All,

I want to factorize a large sparse matrix using regularized svd. Is
there any package that I could use, I am currently using svds() to
calculate the svd over a sparse matrix.

I did try to Google but it did not give any clear results that do it
hence the question.

Regards,
Atul.

What is "regularized svd"?

Bruno

"Bruno Luong" <b.luong@fogale.findmycountry> wrote in message <gpq61q$bap$1@fred.mathworks.com>...
> What is "regularized svd"?
> Bruno

  There's a discussion about it in this site:

 http://alias-i.com/lingpipe/demos/tutorial/svd/read-me.html

I don't know if Mathworks has any function that uses this method.

Roger Stafford

"Roger Stafford" <ellieandrogerxyzzy@mindspring.com.invalid> wrote in message <gpq6u0$aj4$1@fred.mathworks.com>...
> "Bruno Luong" <b.luong@fogale.findmycountry> wrote in message <gpq61q$bap$1@fred.mathworks.com>...
> > What is "regularized svd"?
> > Bruno
>
> There's a discussion about it in this site:
>
> http://alias-i.com/lingpipe/demos/tutorial/svd/read-me.html
>
> I don't know if Mathworks has any function that uses this method.
>
> Roger Stafford

I read quickly this site, but I'm still unsure where as it is just Tikhonov regularization applied on the least square or not. If it is the case, then regularized SVD is nothing more nothing less than adding a constant in the diagonal of matrix S.

Am I completely off?

Bruno

"Bruno Luong" <b.luong@fogale.findmycountry> wrote in message <gpql1p$99b$1@fred.mathworks.com>...
> "Roger Stafford" <ellieandrogerxyzzy@mindspring.com.invalid> wrote in message <gpq6u0$aj4$1@fred.mathworks.com>...
> > "Bruno Luong" <b.luong@fogale.findmycountry> wrote in message <gpq61q$bap$1@fred.mathworks.com>...
> > > What is "regularized svd"?
> > > Bruno
> >
> > There's a discussion about it in this site:
> >
> > http://alias-i.com/lingpipe/demos/tutorial/svd/read-me.html
> >
> > I don't know if Mathworks has any function that uses this method.
> >
> > Roger Stafford
>
> I read quickly this site, but I'm still unsure where as it is just Tikhonov regularization applied on the least square or not. If it is the case, then regularized SVD is nothing more nothing less than adding a constant in the diagonal of matrix S.
>
> Am I completely off?
>
That's what I got out of it too. But with S you can of course play around a little bit more freely than just doing straight Tikhonov regularization. I can for example imagin the use of only apply damping to eigenvalues smaller than 1.

Bjoern