Re: optimizers

[Heber Farnsworth]

I have been using some that I got from the Matlab ftp site.  They were
contributed by users.  They can be found in 

ftp://ftp.mathworks.com/pub/contrib/optim/dsmax

there are three functions: adsmax.m, mdsmax.m, nmsmax.m.  The third is
just like matlab's fmins except that it maximizes instead of minimizes.
All three routines are multivariate direct search maximization routines
which means that they do not use derivatives to get to an optimum.
Although this can make them slower I like them because they are less
likely to get "stuck" on wierd functions then Gauss-Newton type
optimizers.  On the other hand if you have a "nice" function (smooth, and
convex for minimization or concave for maximization) then the fastest
thing I've found is to use octave's fsolve on the gradient of
your function.

If you have contraints of course then you need something else.  None of
the above will handle constraints although there may be other things on
that same ftp site that will.

  Heber Farnsworth                               | Department of Finance
  Univerity of Washington                        | Box 353200
  tele:  (206) 528-0793 home                     | Seattle, WA 98195-3200
  tele:  (206) 543-4773 finance     web: http://weber.u.washington.edu/~heberf
  fax:   (206) 685-9392             email: address@hidden

On Fri, 11 Apr 1997, Colin Telmer wrote:

> Is there any freely available nonlinear optimizers for octave available
> that are already coded to fit into oct (ie. I know there are some free f77
> optimizers, but I do not know how to code them into octave)? I am a
> student and cannot afford NPSOL. Cheers, Colin.
> 
> --
>            Colin R. Telmer, Department of Economics, Dunning Hall
>            Queen's University, Kingston, Ontario, Canada, K7L-3N6
>           Phone: (613)545-2273 Email: address@hidden
>                     <URL:http://terrapin.econ.queensu.ca>
>      PGP Fingerprint = 09 E9 DA 66 9C EE 33 DC  B8 3B 97 0E 01 BC EC 0B
>              PGP Public Key: finger address@hidden
> 
>