Prev: Integration
Next: can't install MCR on ubuntu8.10 amd64
From: omegayen on 10 May 2010 13:15
"John" <j.demello(a)remove.this.imperial.ac.uk> wrote in message <go170u$md5$1(a)fred.mathworks.com>...
> Thanks Oliver - much appreciated!
>
> John
>
> > OK, but including Tikhonov regularization means you're really trying to solve something like:
> > min_x ||Ax - b||^2 + ||Tx||^2
> > x >= 0
> >
> > You can solve this problem with any of the following functions from the optimization toolbox: fmincon, lsqnonlin, and quadprog (at first look). Fmincon will perform poorly I guess, so ignore that. My guess is that quadprog would work best, though lsqnonlin might be faster.
> >
> > Now it's just a question of reading the documentation and setting up the problem correctly.

I would be interested in doing this as well. I took a look at lsqnonlin but havent used it before. Could anyone perhaps show me how to use tikhonov regularization using lsqnonlin or quadprog?

also is this different than using

x = [A;lambda*eye(p,p)] \ [b;zeros(p,1)]

where p is the size of A (if its square). thanks
From: omegayen on 10 May 2010 13:15
"John" <j.demello(a)remove.this.imperial.ac.uk> wrote in message <go170u$md5$1(a)fred.mathworks.com>...
> Thanks Oliver - much appreciated!
>
> John
>
> > OK, but including Tikhonov regularization means you're really trying to solve something like:
> > min_x ||Ax - b||^2 + ||Tx||^2
> > x >= 0
> >
> > You can solve this problem with any of the following functions from the optimization toolbox: fmincon, lsqnonlin, and quadprog (at first look). Fmincon will perform poorly I guess, so ignore that. My guess is that quadprog would work best, though lsqnonlin might be faster.
> >
> > Now it's just a question of reading the documentation and setting up the problem correctly.

I would be interested in doing this as well. I took a look at lsqnonlin but havent used it before. Could anyone perhaps show me how to use tikhonov regularization using lsqnonlin or quadprog?

also is this different than using

x = [A;lambda*eye(p,p)] \ [b;zeros(p,1)]

where p is the size of A (if its square). thanks
 | 
Pages: 1
Prev: Integration
Next: can't install MCR on ubuntu8.10 amd64