Author: Vincent Diepeveen
Date: 13:23:15 07/05/01
Go up one level in this thread
On July 05, 2001 at 13:22:18, Gunnar Andersson wrote: >On July 05, 2001 at 07:54:24, Vincent Diepeveen wrote: > >>Regrettably its only usage was in a game where the one eyed program >>with probcut was king in the land of the blinds, as we use to say >>in dutch. > >That one eyed program beat the human world champion 6-0 in a match without >breaking a sweat. > >Try playing that program, Logistello, or indeed any strong program (Kitty, >WZebra, etc) searching 4 plies without extensions. Unless you're a very strong >player, the program will outplay you positionally. > >/ Gunnar I'm not trying to take down the efforts put in programs. What i mean is that if you compare heuristics in a game where a few heuristics can completely search the game space, then what a 'search enhancement' gives can be wrongly understood as a good enhancement in games where you cannot completely search the game space. In chess if my program says that king safety is -13.000 pawns for one side, then i still cannot reliable prune on alpha there, because 1 move for this side might be mating my opponent. So even if my own king is 1 move away from the mate, i might have *missed* something. In draughts the board is much bigger as checkers. Checkers is a pretty simple game compared to draughts, let's put it that way. There are no professional checker players as far as i know. Buggy's successes and victories in draughts, even though i don't play draughts myself anyhow, i appreciate way more, as i know the effort put into trying to beat a professional draughts player is way tougher as beating an amateur in a game which you already can search nearly completely all positions to EGTB. Chinook would have been a lot less dominating if there was money to earn writing checker programs! Tinsley died, so did checkers die. In checkers a queen up is a way bigger advantage as in draughts, BECAUSE YOU CANNOT CAPTURE backwards! in draughts you CAN capture backwards, so the chance you can get back that queen within a few moves is far bigger! In other words it is when you consider you have 20 stones at the start of the game, a queen is usually not worth more as 3 stones. that's 3 / 20 = less as 1/6 of what you start with! In chess a pawn is worth even less, in go a move is worth even less as that. If an algorithm works for a simple game, doesn't mean it's a correct algorithm in another game. It might even only mean that the program where it was used for was so bad that it didn't work! Nullmove assumption works in any game where doing nothing is usually the worst move. Draughts & checkers both are a game where doing nothing is the goal they all try to achieve. If i can potentially make more moves as you, so if my stones are more backwards as yours are, then with some exceptions, that means that this is in general better for me! If i have more tempi as you in short you can resign in draughts& checkers, so nullmove doesn't work obviously. Nullmove is a correct way of searching however in chess & go, for the obvious reason that in chess you can detect zugzwang by another nullmove, and in go not making a move is a legal move even. The reason why it works correct is because the dubious thing it does, reducing search depth, is done such that the opponent has the move. Whereas pruning on alpha is highly dubious (though a lot of people use it out of a desperate measure). Suppose i can get a search depth beyond which tactics no longer are relevant. In draughts that is like 12 ply or so, just like in chess. That means simply that if i go use algorithm which work incorrect & dubious and might give back wrong values to the root, that i'm doing a quite stupid thing. However when a game is nearly solved, or when heuristics in a game are dead simple, then things change. Then the game is simply too easy to take serious IMHO to experiment with that algorithm if the search algorithm is getting applied to the entire search space! Best regards, Vincent
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.