[Hunble]

You Sir, also have an invitation to help me improve my software. Obviously, you have practical knowledge that I am lacking, and I will be more than happy to implement any suggestions you have. I have absolutely no problem implementing the "God subroutine" if it helps to speed up this software. All you have to do, is let me know is what that algorithm is. PLEASE HELP! Medved always has outstanding new theories, so if I can apply them, he is more then welcome to help out.

[AndrewC]

PLEASE HELP!

Sounds like a missle homing problem. Try those algorithms.

[gore3000]

I have absolutely no problem implementing the "God subroutine" if it helps to speed up this software. All you have to do, is let me know is what that algorithm is.

Sorry I cannot be of help. I know a lot about AlGoreisms, but not about algorithms.

[NukeMan]

Hello...

I think the problem is you and others on this rant are talking apples and oranges. An evolutionary algorithm is indeed a powerful search tool, but it only bears a glancing resemblance to the literally real-life theory of neo-Darwinism (NDT). For example:

in real life mutation frequency cannot be so high (1/200)
in real life selection pressure cannot be so high (50%)
in real life selection is not immediate and certain
in real life DNA sequences are very much longer than 200
in real life there is no guarantee the intermediates will actually function
in real life there is no distinct goal & simple scoring mechanism such as comparing shared pixels

It is tempting to look at the success of an EA and extend it to the natural world, but the EA you describe is simple enough that it cannot fail, and the extension to real-life is perilous. After all, it was greatly simplified from NDT in order to make it tractable.

And all this is assuming that NDT is correct, which it may well not be.

If I read your posts correctly, what you want is an even *more* efficient EA. There are EAs that use both intelligent design and stochastic search, and they outperform stochastic search alone in some problems. It is also not clear that EAs are optimal for every problem. Sometimes annealing or hill-climbing is better (The Traveling Salesman Problem is one in which EAs underperform significantly compared to others). Sometimes it is better to combine strategies such as annealing *and* EA. I would first try that combo to improve your pixel problem...