Some comments for RGA - tests
NB! All of functions are considered as black boxes: no use of their analytical forms, functions just return their values (when corrupted with noise, the first and second moments are assumed known ) in the points of calls.
The OF, chosen for the first and second examples, belongs to a family of 2D* analytical functions with 7 random parameters and is defined in a region [ 0. : 1. ] x [ 0. : 1. ].
- The first example: 1-st movie (Fig.1, the bottom) shows the evolution of regularized approximation of the OF with RGA: starting from 11 random points it comes to 200 points with a fairly good estimation of a global minimum.
- The second example differs from the first one with two subitems:
- a) to illustrate that the result is fairly robust with respect to initial set, it is chosen another one, i.e. another realization of 11 random points;
- b) RGA acts in a fully automatic mode: with automatic zooming and autotermination. The latter costs 621 calls of OF instead of 200, but provides us with more refine reconstruction of the OF in the area of global minimum.
- There are 3 movies ( Fig.2 ): the evolution of OF-approximation in the entire region, and two sequential auto-'zooming in' .
