Fitness function breakthrough. No doubt i made one of the greatest discoveries in my genetic algorithm career. When i thought about various versions of the Fitness function i came across a problem which i called “Extinction level event”.
Basically, if you are too strict in rating of the current specimen or population, there might be such a low score, that the results are all within few points or even equal ZERO. This usually leads to removal of all specimens in the population, where due to crossover with Empty specimens the specimens become shorter and shorter untill there are NONE. I came up with the solution, where i introduce a benefit for “fourth place” – that means no medal, but still a reward for participating. This fixed value will keep the specimens in population even if they are complete garbage, so that the population can recover from a bad streak. I cant wait to try it out. With this approach, even a fitness function based on “minimal value” among the doctors could be viable.
I didnt come up with anything else, since i had a night shift at work.