Le 23/11/2011 12:22, Mikkel Meyer Andersen a écrit : > 2011/11/23 annaykay <[email protected]>: >> Hello everyone,
Hi Annakay, >> I am working on the optimization of some model parameters in my simulation, >> which simulates the impact of communication on the attitude towards a >> specific topic. >> I want to optimize 17 parameters in a non-linear function to get a minimal >> error-value. Therefor I implemented two different Optimization-Algorithms in >> my Simulation. >> First I tried the Nelder-Mead Algorithm. But in this case I have the >> problem, that first my error-value increases, then decreases again and >> starts to stagnate on a non-satisfying value. Is this even possible for the >> Nelder-Mead method, that the error increases however I want to minimize it? >> Then I also tried a different Optimization-Algorithm, the >> Levenberg-Marquardt-Algorithm. Here the problem is that the changes in the >> parameters are too small, so that the optimization already stops after one >> iteration. >> Do you maybe have an idea about approaching this problem or do you know a >> different Optimization-Algorithm that could suit my problem? >> >> Thanks in advance! >> >> >> -- >> View this message in context: >> http://apache-commons.680414.n4.nabble.com/math-Optimization-Nelder-Mead-and-Levenberg-Marquardt-tp4099186p4099186.html >> Sent from the Commons - User mailing list archive at Nabble.com. >> >> --------------------------------------------------------------------- >> To unsubscribe, e-mail: [email protected] >> For additional commands, e-mail: [email protected] >> >> > > Hi, > > If the function you want to optimize has several local maxima, then it > is (almost) always problematic. Especially with 17 parameters, that is > a lot. Are you sure that you cannot obtain an analytical solution? > Have you tried different starting values for the Nelder-Mead > Algorithm? One way to try different starting values as suggested by Mikkel is to use the multi-start wrappers around the raw optimizers. Luc > > Cheers, Mikkel. > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
