I tried to use Galapagos to find a solution to a problem with a lot of input parameters (defined as genomes in galapagos) and a single fitness parameter. The algorithm eventually finds pretty good designs but one problem I observed is, that the initial generation of genomes seems to not be entirely random. The sliders for all the input parameters or genomes allow three different integer values. However the initial generation only contains designs that use the first two values of each genome. That is a huge limitation of the design space and pushes the optimization in a difficult direction right from the start. I am guessing that there is some degree of random mutation of the genomes with each generation. So at some point there are designs using genomes representing the third value of the sliders but in order to keep those designs “alive” I have to keep the population fairly big and the variety high which conflicts with the ideal settings that would perform better and find an optimum faster.
Is there any way I can solve this problem? Is there a better way to set this up so that I do not run into this problem?