Hi, package optim-1.0.11 has been submitted to the release forum.
Only leasqr.m has changed since the previous version. Please see second item for a possibly important bugfix. Changes: - leasqr.m (additional files gjp.m and cpiv.m): Linear inequality constraints implemented, adapting a quadratic programming algorithm of Y. Bard (see cpiv.m for the reference) to the Levenberg/Marquardt algorithm with svd decomposition of leasqr.m. Bounds are now a special case. Much more efficient (in terms of number of iterations, eg. 43->6 or so in a test case) than the previous bounds implementation of simply cutting off parameters. Also, the previous bounds implementation could fail to converge since the algorithm of leasqr is equivalent to multiplying the gradient with a positive-definite, but not diagonal matrix for "epsL"->Inf, so there is not in general a fallback to a descent in each gradient component (this could be easily changed, but without obvious advantages with the current implementation of constraints). BTW when I last looked at Matlabs Levenberg/Marquardt method of lsqcurvefit, it had _no_ implementation of constraints ... - also leasqr.m: Bugfix in computation of additional statistics for the result --- not the final parameters, but those from the step before were used. Sorry, that was my fault. Olaf ------------------------------------------------------------------------------ Download Intel® Parallel Studio Eval Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta. http://p.sf.net/sfu/intel-sw-dev _______________________________________________ Octave-dev mailing list Octave-dev@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/octave-dev
