🚀 go-pugleaf

On Mon, 28 Nov 2005 20:53:40 -0000, Oliver Wong <owong@castortech.com>  
wrote:
>     What you'll want to do is to programmatically enter in every strategy
> you've tried at hooking up with a girl. You'll need to have some metric  
> for
> measuring success, and enter those in for each strategy as well. Then run
> one step of simulation (i.e. one generation) so that the strategies will
> have mutated slightly, with the strategies that have performed the best
> mutating the least (thus, on average, your strategies have become  
> "better").

That's not quite how it works.  All strategies are equally eligible for  
mutation, regardless of how good they are, and there's every chance the  
next generation will be worse on average than the previous one.   
Eventually however, if your selection strategy and fitness function are  
good, the quality of strategies in your population should improve and will  
tend towards an optimum (there are no guarantees though).

>     The advantages of genetic programming are that it's a viable tool for
> when you have no idea where to start looking for a good strategy.
>
>     The disadvantages are that typically it takes a very long before the
> best strategy "evolves" from your initial pool of random strategies.

Genetic programming is just one application of genetic algorithms (or more  
generally, evolutionary algorithms).

There are two pre-requisites.  You need a way of encoding a potential  
solution and you need a way to rank partial solutions.  The approach is  
only really applicable in situations where a good approximate solution is  
acceptable as it may never find the optimal solution.  It can be very slow  
but is typically orders of magnitude faster than a brute force search of  
the solution space for complex problems.

Dan.

-- 
Daniel Dyer
http://www.dandyer.co.uk