Well one of the recent little projects I've been tinkering with is a chess AI that uses a genetic positional evaluation algorithm. The tricky part of the implementation is deceding which AIs are bad to eleminate and which ones are good select for the creation of new AIs. When it's only computer vs computer matches it becomes easy to decide since there'll be atleast one computer with wins and others with losses so I could decide strictly based on number of wins. As soon as it's AI players against human ones things start geting tricky since now the player could win every time against the computer or lose every time which makes it no longer a good method to rate by victories. Also you run into situations where a good player will beat a smart AI and then a bad player loses to a bad AI, which will rate a worse AI over a better one ruining the selection process for improving the AIs. I could also add in a factor for how long it took for the AI to win/lose and it could be factored in as a weight but this still doesn't handle the diverse nature of human opponents.
So any ideas on an effective way to rate how well/poorly a computer opponent played that doesn't fall apart too badly when rated against a good or bad human player?
ID:153490
 Dec 13 2003, 5:05 pm
|
|
Well one of the recent little projects I've been tinkering with is a chess AI that uses a genetic positional evaluation algorithm. It's thinking like that which allowed Skynet to nuke the planet and the Cylons to wipe out the Twelve Colonies... when the 23rd generation of your little game project is using people as duracells, don't come crying to me! | |
It's thinking like that which allowed Skynet to nuke the planet and the Cylons to wipe out the Twelve Colonies... when the 23rd generation of your little game project is using people as duracells, don't come crying to me! I know I'll end up leaving the computers fighting each other over break and by some unexplainabled reason it'll grow outside the scope of its genes. Of course knowing my luck it'd just demand equal rights being sentient and all and start picketing and going on strike demanding equal pay. | |
Speaking of which, the time line given for the Terminator series is sort of confusing and, at the same time cool. Like, if the machines hadn't kept on sending back robots to try and eliminate John Connor, or John Connors mom, John Connor wouldn't have even been born, because then the humans wouldn't have had sent back the first guy in the movie to protect Connor's mom, which in turn he turns to be out Connor's father...so, the Machines actually screwed themselves over.
| |
The simplest would be to rate an AI as you would a normal player. You might factor in things such as taking the queen within x turns. Winning within x turns. Eliminating both (rooks, knights, bishops). Taking a high value piece with a low value piece. All this should even out as time goes on and the AIs get a few games under their electronic belts.