Newsgroups: comp.ai.games
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!uunet!munnari.oz.au!cs.mu.OZ.AU!mundoe!arc
From: arc@mundoe.maths.mu.OZ.AU (Andrew Conway)
Subject: Re: Risk-Type game logic
Message-ID: <9510613.20366@mulga.cs.mu.OZ.AU>
Sender: news@cs.mu.OZ.AU (CS-Usenet)
Organization: Computer Science, University of Melbourne, Australia
References: <3mfbo1$2eb@news.vcd.hp.com> <1995Apr14.072438.11199@jarvis.cs.toronto.edu> <3mls7p$611@ixnews2.ix.netcom.com>
Date: Sun, 16 Apr 1995 03:05:23 GMT
Lines: 39

dionkk@ix.netcom.com (Dion Kurczek) writes:

> [I'm working on a global-level strategic wargame similar to Risk]
>In thinking about the AI for the computer player, I decided to 
>investigate using neural networks.  I was successful in training a 
>neural network to arrive at a "confidence level" for victory on any 
>attack, given the number of attacking and defending units.

Isn't this a rather trivial mathematical algorithm? Why approximate it with
a complex neural network, when an exact answer is probably much simpler.
Never use nural networks to do something that can be done much more simply
some other way.

>This network can give me a pretty good "weight" value when evaluating 
>possible moves.  I'm now thinking about what kinds of techniques to 
>employ for the overall player strategy.  I'm not sure a neural net would 
>be a good choice for this phase of the AI, and am leaning more toward an 
>expert system/rule-based solution.

Play the game several times with real people.
See what they do.
Make up some rules. 
Implement them.
do {
	play game agianst computer players.
	Notice obvious mistakes by the computer
	Fix them
} while (not acceptable computer players)

>I've heard some talk of existing Risk-type strategies being available on 
>FTP, but I never saw any locations.  Anyone have any ideas?

I made some pretty effective RISK opponents using the above mentioned 
algorithm.

I hope this is useful.

Andrew.

