Hi, in a laser physics experiment I am building I need to optimize the positions of 4 actuators on some mirrors; these positions are simply controlled by 4 numbers. The optimization needs to be done to maximize a set of metrics that I can encode as a single fitness value. Usually I perform this optimization by assigning it to some PhD students or post-docs. But I'm hoping I can save their time for other tasks and get Julia to do this for me!
I need to essentially solve a bounded global optimization problem. Previously I have used genetic algorithms for such tasks with some success, but I get the impression that these are now out of date and that better solutions exist.
However, when I look at the packages in Julia I quickly get overwhelmed by the range of choices and the optimization jargon that I do not understand (and don't really have the time to study). For example I do not know really what my "model" is or if my problem is convex, quadratic etc. For example the table of solvers at http://www.juliaopt.org/ is essentially useless to me!
Essentially I'd like a routine that I pass my fitness function, that takes 4 numbers and returns a single number to be maximized, which then finds (or has a good chance of finding) the globally optimal set within bounds. I don't mind tuning parameters to the algorithm. As I said I've used genetic algorithms for this before. I also don't have access to the gradients etc. as it is for hardware control. And there is likely some hysteresis, but I'm hoping to ignore that.
It seems that NLOpt might be my best bet, but I'm happy to hear suggestions from the experts!