next up previous

4.3.7 Algorithm Performance

Figure 24 shows the progress of an ES on the two-dimensional Rosenbrock function, . Each member of the 1st, 10th, 20th and 30th generations is shown (by a symbol), although, in fact, most of the members of the 1st generation lie outside the bounds of the figure. The convergence of the population to the neighborhood of the optimum at is readily apparent. Notice how the 20th generation appears to be converging on , but in the next 10 generations the search successfully progresses along the shallow valley to the true optimum.

For this run a (20, 100)-ES was used. The two control variables were subject to discrete recombination (of pairs of parents), while the strategy parameters were subject to global intermediate recombination.

Figure 24: Minimization of the Two-Dimensional Rosenbrock Function by an Evolution Strategy: Population Distributions.