Okay, when the program "displaying" the genetic algorithm terminates, describe the "solution". When does the genetic algorithm program terminate? Why does it terminate?
When you have a solution that matches your parameters for a successful solution. In #4, PGS said
The fact that there is a pre-determined “stop executing” point is not a function of the evolutionary nature of a GA; it is simply a recognition that there’s not need to continue looking for answers after you have one...As I understand it, that means you know what an answer will look like, but you don't know what the answer is. I'm sure he can explain better.And the fact there is a “stop executing” point does not mean your answer is pre-ordained. I’ve run GAs on the same problem set, with the same initial seeding, and come up with different - but still very viable and useful - results.
I meant to add to my roulette comparison: in real roulette, you don’t know what the answer will be, just that there will be an answer. But if you wanted to model betting strategies, it’d be perfectly reasonable to pick a single answer and use a computer model to see how often you’d win with a particular strategy. It still wouldn’t require manipulating the results to get that answer.
Define for me how much salt to add to 2 pounds of mashed potatoes. How much is the right amount? How do you know to stop adding salt? Is that amount of salt correct for GGG? For me? For everyone? Or only for you? Is that after or before addition of milk and butter, and what type of potatoes?
A GA is terminated (it doesn't end on its own) when the programmer/researcher decides he has an answer he can work with.
Imagine I give you a 30 element network to create a 20th order bandpass filter. You have unlimited choices of resistors, capacitors, and inductors to use. I want that filter to have a Chebychev response with a defined passband gain, passband ripple, and attenuation outside the passband.
Now, I also want you to make the filter as low cost as possible! However, since this is going on a research vessel, weight matters, as does size - we can't have a perfect fit filter that weighs more than 5 pounds, and occupies more than 200 cubic inches! And budgets being what they are, lower cost is always wanted.
In fact, if you can make it significantly smaller, we'd be willing to trade off some cost, or maybe some performance. If you can make it lighter, that's even better than small but not too much better. Light AND small together is really good - better than either one by themselves, or even together.
So how do you find the filter that best trades off functionality (the electrical passband/stopband performance) for cost, size, and weight? Each component you select has electrical, fiscal, and physical dimensions, and you have to solve for a solution that works.
There isn't just ONE solution here - given the complexity of the problem, there can be literally INFINITE valid solutions! But your problem is to figure out how to make one that YOU believe is pretty darn good.
THAT is how you use a GA. For solving problems where you have an idea of "how much salt is enough", but to exactly quantify it is essentially impossible and in fact irrelevant. You need a "good enough" answer that you - and your customer/boss - will be happy with.