Re: Freely available optimization code
>>>>> "PA" == Pierre Aubert <email@example.com> writes:
PA> Douglas Bates <firstname.lastname@example.org> writes:
>> Does anyone know of high quality freely-available code for
>> numerical optimization? I am part of the core group of
>> developers for a statistical programming environment called R
>> (see the r-base Debian package or http://cran.stat.wisc.edu).
>> For optimization we have been using the Fortran code that
>> supplements Dennis and Schnabel's book "Numerical Methods for
>> Unconstrained Optimization and Nonlinear Equations". When I
>> compare it against some code from the Port library, it doesn't
>> perform as well. I would like to replace it.
>> I would also much prefer to use C, C++ or Java but I do have
>> two constraints on code. Since R is GPL'd we would need code
>> with a license that allowed us to use it in a GPL'd product.
>> Also, we tend to store matrices as unidimensional arrays with
>> the Fortran-style column-major ordering.
PA> You may try CFSQP a good SQP code: Ask Prof. Andre L. Tits:
PA> (Rq: I don't think the code is GPL)
CFSQP is good and free, but redistribution is highly restricted (which
killed off 2 software routines I'd developed, sigh... anyone know of a
decent free (in the Debian sense) nonlinear programming library?
Actually, this reminds me that there is libwn in Debian,
which does conjugate gradient (and uses boundary methods for NLP).
You might want to check that out, Doug. I'm not certain about overall
quality; it works, probably could be professionalized a bit.
A.J. Rossini Research Assistant Professor of Biostatistics
Center for AIDS Research UW Biostatistics
206-720-4282 (4209=fax) 206-543-1044 (xxxx=fax)