When I’ve experienced an issue like that, it has usually been because my own code was trying to access a portion of memory it shouldn’t. Sometimes the accessed memory would be zeroed out and the code would work, other times it wouldn’t. I am not familiar with Simulink, but Matlab prevents that sort of behavior as far as I know.
Race conditions are another issue where you get that behavior. If you are doing parallel code, there could be an issue with that. Of course, it could also be an issue in NLOPT. Not very helpful, sorry. > On Jan 6, 2015, at 9:47 AM, Thomas Mielke <[email protected]> wrote: > > Hi, > > I'm using the NLopt-C-Interface inside Matlab Simulink for optimizing > trajectories. The algorithm I'm using is COBYLA. The objective function is > a polynom with 7 optimization parameters. From Simulink, I pass 4 > parameters to Nlopt, which affect the constraints. > So far, the optimization is working fine. Now inside Simulink, Nlopt is > called every 10ms with different input parameters. Now I noticed, that > sometimes, there are wrong results, which don't fit in and violate my > constraints. I found out the corresponding input parameters and repeatedly > called NLopt with them. As the input parameters and every other > configuration settings inside nlopt were the same, the optimization result > from nlopt should have been the same aswell everytime. What happened was > that the output was now correct most of the time, but after a few thousand > NLopt calls, suddenly there was a wrong result. So after about 8000 > correct identical results there was one wrong different result. This looks > like somekind of bug to me. Has anybody ever experienced something similar > or knows why this could happen? > > Thanks! > > > > _______________________________________________ > NLopt-discuss mailing list > [email protected] > http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/nlopt-discuss _______________________________________________ NLopt-discuss mailing list [email protected] http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/nlopt-discuss
