Author: Dave Gomboc
Date: 00:35:34 10/10/01
Go up one level in this thread
On October 09, 2001 at 21:51:01, Dann Corbit wrote: >On October 09, 2001 at 21:26:54, Slater Wold wrote: >[snip] >>You would think. But it looks like TB's *might* be hindering the discovery of a >>quicker mate. > >For sure. When there are a lot of pieces on the board, chess programs will >simplify by trying direct exchanges. They will even toss a queen for *no >reason* except that it knows how to solve without the queen. > >Imagine that you have kqqqkr, and on the board we have kqqqqqKR. >The chess engine will very likely toss out two queens, if it sees a sure mate >using the kqqqkr tablebase. > >And why is it so hard to shake out of it? > >Because of alpha beta. One thing about Alpha-beta, it knows a bird in the hand >is worth two in the bush. If I have a mate in 35, and another path might put me >up 3 queens, it won't even consider the other path -- it will exit right away >because it sees: >If I chose door #1, I get a checkmate in 35 (value 32767-5) >If I choose door #2, I get a 3 queen advantage (value +30) >Therefore, choose door #1. Do not continue to search until the next iteration. >Also, if I have an advantage of a sure checkmate, any other choice will look >like a massive loser if it is not a shorter checkmate. > >Now, there will be 3 queen advantages that lose (pretty rare of course). > >Tablebase wins can be clear beyond hideous all the way to commical. But they >will be absolutely certain to win, and so the goal is achieved. As your own example shows, this is not really a problem with alpha-beta, but rather the fault of an evaluation function that gives a higher value to a non-special (e.g. not stalemate) KQQKR position than to a non-special KQQQQKR position. Dave
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.