Hello all, This is my first email to a mailing list, so I apologise in advance if I am using it wrongly - of course, any constructive criticism is welcome.
1. Mail Archives The NLopt-discuss Archives website ( http://ab-initio.mit.edu/pipermail/nlopt-discuss/) seems to be down. Is there any way to access it, or is there any forum-like alternative? 2. Integration of the CMA-ES algorithm The covariance matrix adaptation evolution strategy (CMA-ES) algorithm is a very robust stochastic global optimizer, showing impressive results for especially difficult problems (non-convex, non-smooth), as seen in the classic article "Derivative-free optimization: a review of algorithms and comparison of software implementations" (link <https://link.springer.com/article/10.1007/s10898-012-9951-y>). As the algorithm itself is rather famous <https://scholar.google.co.uk/scholar?hl=en&as_sdt=0%2C5&q=cma-es&btnG=> (at least in terms of stochastic optimizers) and there is an official, maintained ANSI C implementation <https://github.com/cma-es/c-cmaes>, is there any particular reason it was not yet integrated into the NLopt library? Otherwise, how would one go with doing this? I would be interested in doing the coding work if someone would be willing to offer a bit of guidance integration-wise. 3. Implementation of the MCS algorithm This is mainly a personal curiosity, though it might end up being a nice project: the multilevel coordinate search <https://www.mat.univie.ac.at/~neum/ms/mcs.pdf> (MCS) algorithm is a deterministic global optimiser that consistently shows up in comparisons and applications (the above review article named it the "best" non-commercial optimiser in 2012; 1 <https://link.springer.com/article/10.1007/s10898-012-9951-y>, 2 <https://www.sciencedirect.com/science/article/pii/S0098135416302848?casa_token=91rcDLOcg3MAAAAA:tKn5_ivySy2hJ-G75lyK7UQ7lMiH0dC0G-7BUf0s4RJWuijymQG_DYHHKbfQWvJYW4IOFiAM>) with impressively small numbers of function evaluations. However, it only has a single MATLAB implementation by the authors - freely available here <https://www.mat.univie.ac.at/~neum/software/mcs/>. While MCS does rely on two other algorithms - bound-constrained quadratic program solver and global line search - it seems strange that such a good algorithm did not have any other implementation since 2008. Is there any reason for this? Personal opinions are welcome too. On a related note, writing a C implementation would most certainly be a fun project. Best wishes, Leonard --- Andrei Leonard Nicusan College of Engineering and Physical Sciences University of Birmingham
_______________________________________________ NLopt-discuss mailing list [email protected] http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/nlopt-discuss
