The Stochastic Radial Basis Function Algorithm aims at solving computationally expensive continuous black-box global optimization problems with box constraints. The algorithm uses radial basis functions to approximate the true objective function and to decide at which points in the variable domain the costly objective function should be evaluated. The algorithm uses a scoring criterion to select sample points, hence no auxiliary problem needs to be solved. The algorithm can do more than one function evaluation in parallel in each iteration if desired.
Julie (2020). Stochastic Radial Basis Function Algorithm for Global Optimization (https://www.mathworks.com/matlabcentral/fileexchange/42090-stochastic-radial-basis-function-algorithm-for-global-optimization), MATLAB Central File Exchange. Retrieved .
Great work, thank you very much for sharing!
Michal, if you scale your variables to [0,1] for the optimization and use lower bounds as 0 and upper bounds as 1, this works pretty well. You can scale your variables back to the original scale when calling your objective function evaluation, i.e. x in [0,1], scale it to original interval by using z= xlow + x*(xup-xlow), evaluate f(z).
Variable "sigma_stdev_default" should be estimated for each variable range in every dimension.
Current value corresponding to
minxrange = min(xrange)
as smallest variable range is not appropriate!!!
What about constrained version???
well done !!!