Hi, I have a deterministic function (computer code) of a couple of hundred variables (the maximum size of the problem under normal conditions). Each evaluation takes about 50ms. It is not very expensive.
Of the 200 variables, 40 are real numbers between 0 and 1 (actually 39 and the 40th is just 1 - sum of 39 others), another 20 also reals between 0 and 0.5.. The rest are strictly positive integers between 1 and 500 (for the largest upper bound) I currently have a brute force solution which generates enumerates all the space based on user input. I estimate a reasonable run though to take 1 month with an accurate enough sample of the real variables space. For some of the variables, the behaviour of the function is smooth. For others (especially the integer ones), it is non smooth. This is computer code and it doesn't have gradients. The function is quite multimodal. I'm unsure which way to go. A downhill simplex doesn't seem to bring much. I wouldn't be able to use gradient methods. Kriging methods seem interesting as well as a DACE paper by Jones, D., Schonlau, M., Welch, W., (1998) ``Efficient Global Optimization of Expensive Black-Box Functions''. Journal of Global Optimization} Vol. 13, 455-492 Any advice of an area I should investigate given the nature of my problem, its dimensionality and the nature of variables would be appreciated It would be great to obtain an optimum within 1hour. Regards, MM
_______________________________________________ NLopt-discuss mailing list [email protected] http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/nlopt-discuss
