🚀 go-pugleaf

In article <Bsx9rB.F6H@news.cso.uiuc.edu>, tmkk@uiuc.edu (Khan) writes:
|> 
|> >Has anyone tried using genetic algorithms for CCP's?
|> >
|> >I'll bet you have to play thousands of games to train a good one, unless you
|> >prime the pump with abover average players first.
|> >
|> >Any comments?
|> 
|> Ooh, a neural net computer opponent, which would learn from the
|> programmer and beta testers while they play test the game, so by the
|> time the actual release occurred, it would be TOUGH. As a player, I
|> wouldn't want ALL the computer opponents to be like this, but perhaps
|> one or two, which appear later in the game (towards the end of the
|> puzzle chain) or which "guard" certain key pieces of info/puzzle, would
|> add quite a bit to the game.

With genetic algorithms, you might run the for GA 10,000 generations (???),
and you could pick out easy opponents at 3,000, and then medium level ones
at 7,000 and then the toughies at 10,000.  

Now... has anyone done this?  I'd really like to do something like this
but I couldn't come up with parameters (genes), and I haven't ever written
a GA program before.  Perhaps I'll use a really simple game (like the 
prisoner's dilemma) and try a GA on it, and then if it works well, I
can try a more complex game.  

Ideas?  Comments?  Please respond!  We can drown out the VGA questions!  ;)

	Amit

-- 
/\\ Amit J Patel, amitp@owlnet.rice.edu
\\/ I think I'm at Rice University