So after some time the people at work chose the name Skynet for my “artificially intelligent” doctor schedule planner. Ha ha. While it is not as intelligent as “genetic” or “natural evolution” style of program its goof dor laugh if something actually went wrong in the future.
Anyway. Way back i gave serious thought about fitness function and its limits which i rated and it turns out, that it might actually be for the worse. Aparently if i worked with different ratings for different soft constraints or hard constraints like “50 150 300 1000” the algorithm had troubles advancing and fine tuning the specimens since introducing once hard constraint rating of 1000 (-1000) basically destroyed a specimen that was otherwise solid and could be improved upon.
So after trying i assigned mostly low integer values like “1 2 3” and the results improved dramatically. This led to a feeling of needing to introduce many more constraints for richness and broader spectrum of choices and outcomes. I believe this will improve the overall results of the Skynet significantly.
I allready am poking around the thought in which my genetic algorithm actually randomly assigns rating coeficients and find the best one for my use on its own. SCARY.