Author: Rich Van Gaasbeck
Date: 11:14:55 06/08/01
Go up one level in this thread
On June 07, 2001 at 20:27:36, Dann Corbit wrote: >The two most popular strategies to thump a computer are: > >1. Make a closed formation. Now, you have lots of time to plan, and the >computer won't be able to see any further anyway. If the "big stuff" happens a >full nine moves down the road, the computer will probably not see what you are >doing. It would take a whale of a hardware getup to solve this problem. Closed >formations are the best way to beat computers. Current programs reward (or avoid, if it is the opponents pieces) positional elements even if they can't see an immediate material gain (e.g. doubled rooks, rooks on the 7th, rooks on open files). Aren't there positional elements that could be put in the evaluation function that would work for closed positions? How about something like ignoring all the opponents pawns and counting how many squares are attacked near the program's king. >However, some programs will now >try like the dickens to keep you from getting a closed formation. Try it >against crafty, for instance, and you will see that crafty is watching for it! Also the default start.pgn is essentially an anti-stonewall book. Maybe Crafty tries too hard to avoid the Stonewall and closed positions. If so, it seems you could study what it does and build an ant-anti-stonewall book. > >2. Make a big buildup of pieces on the kingside. You are willing to toss every >bit of material you have if it ends in a mating attack. The computer will >calculate: >A. He'll lose his queen, a knight, a rook and a bishop if he starts out that >way -- there's no way that he will do that. >So it won't keep calculating down that road. > >Computers battle this by: >A. Delay castling. Just common sense (really) that you don't castle until you >are ready and it is a good idea. >B. Watch for material buildup on the king side. This one seems obvious, but >computers seem to have a problem with it. >C. Don't castle into trouble. Seems obvious, but apparently it's not. > >Currently, computers are only so-so at detecting mass attacks on the king. I >expect algorithms can eventually solve this one completely.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.