Author: Vincent Diepeveen
Date: 13:25:15 10/27/02
Go up one level in this thread
On October 26, 2002 at 21:00:55, Dan Andersson wrote: >Misere games are games where the winning condition becomes the losing one. Or Can you give some example here which gets practiced by say more than a million people world wide? >more broadly when the purpose is to lose the possibility to make a legal move. >The last loose definition makes nullmove impossible. I.e. you can not assume you >have already won. > A question for you. How the heck do you use nullmove in Go? Passing is a move This is very easy to do in GO. In fact the nullmove is a legal move, with this exception that you also reduce depth now. Usually nullmove is the worst game in GO, so the algorithm works great there! It reduces branching factor *bigtime*. The go players complain too much about their game being 10^1000 positions. In fact there are less than 3^361 positions. *way* less. It's more like. 10^100. The crucial positions are most likely the openings positions. The first few moves really the whole game gets decided. It's here where the branching factor must get measured IMHO. With an empty board! It's here where i measure a branching factor of 10.0 with my GO program. That's the first 8 ply or so. From experience we know that above 10 ply this number will reduce a lot more even when hashtables start working better then. I do not mirror at all any board positions yet. I do not plan to do that either. That would theoretically however make the size of the tree smaller. >AFAIK. I can imagine algorithms that decide that not making a move is the best >option based on a less than full depth search. But that would he closer to B* >than nullmove. It is pretty easy, but not obvious, to proof in chess that nullmove is a simple way to prune in a correct way, where vaste majority of pruning is not losing search depth, and only a few irrelevant positions get pruned away which take you a lot of plies more, and a small number of positions you need 1 or a very few plies more (an extra depth you win by nullmove anyway). Same proof applies to GO obviously. Everything applying to chess algorithms applies to GO too. Of course the sad thing for the GO world is that the commercial and scientific interest world wide in GO is far less than chess. Where GO gets practiced basically in Japan, Korea and China; there the chess rules are known by billions world wide. Obviously all interest went and still goes to chess and far less to GO therefore. As a chessplayer i am happy with that of course, but objectively it doesn't matter at all whether development is done for GO programs or for chess. The same lemma's are valid for both games. >MvH Dan Andersson
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.