Author: Robert Hyatt
Date: 12:56:33 06/08/01
Go up one level in this thread
On June 08, 2001 at 14:33:24, Rich Van Gaasbeck wrote: > >>4. While all that goes on, white places his pieces on squares so that even >>though they are not threatening _right now_ they are positioned to quickly >>reach vital squares when needed. If the computer doesn't see this 'build-up' >>and take action soon enough to offset the threat, then it loses, and very >>easily. > >Could an evaluation function reward "well posted" pieces, where well posted >means they are a small number of moves from somewhere "really great". The >smaller the number of moves, the higher the evaluated value. The moves would >have to follow the normal action of the pieces, but could ignore being pinned, >pieces in the way, etc (so that the cost to evaluate isn't equivalent to doing a >tree search). "Really great" could mean somewhere where the SEE gives a good >value, or hand picked locations like 7th rank, near king, etc. They could. But this is _really_ hard to measure. IE you want your knight at f6, and you have a direct 4-move path to get it there. But if your opponent plays the right two moves, he can make it impossible for you to pass through the one critical square you need to get there. And if your eval misses that, it will be thinking "great position" when it really can't do anything about it. That is why, when someone says "Heck, a 20 ply search won't be any better than an 18 ply search." I simply grin and go on. Because I _know_ that extra two plies will help find things like the above. I would suspect that 30 ply searches, with a reasonable evaluation, will make long-term anti-computer kingside attacks a thing of the past. Because they will then not only see the pieces "gathering" they will see the result of this as well. Today you have to just recognize the "gathering" and try to defend against it, where most programs don't even recognize the "gathering" part. With a 30 ply search, you can just look for disrupted kingsides and ignore the gathering stuff, since with 30 plies you will see both the gathering _and_ the result before you get to 30. I do some of this kind of "pathway analysis" in my pawn evaluation code, so that I can figure out which pawns form potential levers and which do not. But the code is somewhat inaccurate since I don't do a tree search in my pawn code to see what squares my opponent can prevent me from advancing over. What I do is far better than "nothing". But it could be even better. a 30 ply search would be a good start. :)
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.