This is the mail archive of the xconq7@sources.redhat.com mailing list for the Xconq project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: ...and the AI's decision-process around resignation


On Sat, 2002-06-08 at 17:01, Jim Kingdon wrote:
> Well, it doesn't resign in the first 10 turns no matter what.

That would partially explain why no matter what the situation, the AI
always resigns around Turns 10-15.

> 
> Tweaking the point values might be the best "quick fix".

Well, I tried implementing point values similar to those in the standard
game, and it now plays even worse than before.

My current point values are:
city: 25 points
arcology: 50 points
floating-city: 35 points
radar: 2 points
spaceport: 10 points
settler: 1 point
engineer: 1 point
infantry: 1 point
armor: 2 points
rover: 0 points
transport ship: 0 points
mothership: 1 point
saucer: 0 points
fighter: 0 points

If the AI's resigning because it thinks it's too far behind in the game,
it would be correct (but it could easily get up to speed if it would
build a few cities!).
Do you want a copy of the revised game module?

> 
> Coming up with a really good resignation algorithm might be tricky,
> but I suppose it could tie into more sophisticated goal-setting in
> general, perhaps.  The algorithm we have now sometimes errs on the
> side of resigning too slowly as well as too quickly.

If the planning strategy I suggested in my last e-mail is ever
implemented, it could be useful to tie that into the resignation
algorithm.  With that planning strategy, it should be easy to decide
whether or not to resign based on how difficult it is to accomplish its
goals.

In the example scenario I gave, one factor that might cause the AI to
consider resignation would be if it neither had a city to build a
fighter, nor a settler to build a city.  Of course, it shouldn't
actually resign before considering other ways (i.e. capturing) that it
could acquire a city or a fighter.  And even then, it should consider
how important it really is to have a fighter in the grand scheme of
things.

This kind of an AI would undoubtedly be difficult to create, but it
would in the end be the smartest AI ever created!


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]