Author: Uri Blass
Date: 11:03:12 11/26/02
Go up one level in this thread
On November 26, 2002 at 12:51:38, Richard Pijl wrote: >[snip] > >>I can explain reasons that the best move can give a smaller tree. >> >>I have a pruning based on evaluation and other factors. >>The best move may lead to search to reduced depth because of this pruning. >> >>Even programs that only use null move pruning may have search to reduce depth >>based on this reason because after the best move it is more common to have a >>cutoff because of no threat for the opponent. > >This size of the subtree below the first move matters more than the height of >the eval. >Probably made clear the best by an extreme example: > >Suppose the best move will result in a quiet position. The moves leading to that >position are all quiet moves. Alternatives to the PV moves are probably just as >good. > >But now an alternative. This move leads to a PV with many checks and capture >moves. Although the endposition is slightly inferior to the PV described above, >the resulting score is still above beta. The tree for this move is much smaller >than the tree for the best move. > >Well, I agree this is an extreme example, but I think you may see many of those >cases where the smallest tree needed for a cutoff does not start with the best >move. A good capture move may even be good enough I agree that cases when the smallest tree for a cutoff does not begin with the best move are not rare. It does not contradict my theory that in most of the cases the best move produce a smaller tree relative to the move that programs is choosing that is good enough(I talk only about cases when the moves are different). It may be dpendent on the program but I believe that it is the case for programs with good pruning rules. An extreme example is when the best move is a checkmate. Even when the best move wins material programs that use pruning based on evaluation(movei is doing pruning based on evaluation and other factors) may practically need smallest depth so they may need smaller trees. In cases that the move is only slightly positionally better I expect the move to create the same average size of tree but in cases that the move is clearly better I expect smaller tree so the average result that I expect is smaller tree. Uri
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.