Author: Vincent Diepeveen
Date: 05:00:45 10/13/00
Go up one level in this thread
On October 13, 2000 at 05:37:48, Georg v. Zimmermann wrote: >On October 12, 2000 at 19:32:22, Vincent Diepeveen wrote: > >>On October 11, 2000 at 22:15:23, Daniel Chancey wrote: >> >>>NM DragonSlayr from chess.net claims that white has a forced win in the >>>beginning of a wild 5 game. Wild 5 is the original setup of pieces, but the >>>pawns are 1 square away from queening and the pieces are in front of the pawns! >>> >>>I'm not sure if Computers can be accurate to disprove DragonSlayr's claim. Is >>>Wild 5 a win for white? >>> >>>Castle2000 >> >>the openingsadvantage is real huge. if i let diep search parallel >>about an hour at the position then the score rises and rises and rises >>from +1.x initially to +5.x after an hour. i can't imagine this game >>is NOT won for white easily. You can directly get a huge openings >>advantage. obviously 2 queens against 1 queen is a simple win and >>exactly that is what you can achieve there. Now the problem is that >>the branching factor gets quite huge after a few moves as queen moves >>are there plenty, so you can't give 'scientific' proof that you win >>easily. But a queen versus a rook already in evaluation up after a >>few ply of search in the start position? >> >>that's clearly a game which is won by white yeah. > >Hello Vincent, > >I talked with some wild5 expert a while ago about why computers can't play it >very well. It is very similar to the problems we face in crazyhouse (you know >the game where you can place pieces you captured from opponent - bit like >Shogi). One thing is the obvious high branching factor, another big thing is >that games are much more decided by mating attacks, and sacks of whole pieces >for a tempo or two are common. Now bots completely freak out when you have >poseval() function which can reach +-500 pawn units (at least thats what happens >to me) so they do not see that some of the good wild5 opening lines for black >include a knight sack very early when it is not forced already. > >Wild5 might well be a win for white theoretically like normal chess might be a >draw. But prove it ?! Bah. i think on fics diep is regurarly logged in (after the weekend again changing from room now) if fics allows this type of game, try it against DIEP dual. DIEP knows a few things about passed pawns and king safety and checks in qsearch which makes it extremely hard to beat in this kind of games. Nevertheless the openingsadvantage is so huge directly that anyone who plays it for the first time is mated in 3 moves from the start. Now that's a way to start with white. Directly putting black mate in 1 first so many games people play it for the first time. I've never lost in 3 moves in normal chess. Secondly the plan is so dead simple: if i can promote more queens as you can and if i can exchange thereby more light pieces to a queen of yours, then you're history. Just 'surviving' it tactically to the endgame is a dead sure win then as i keep left with queen versus bishop or so. The 6 men EGTB from Bob/Nalimov come in handy then. Now the problem is of course: there are no wild5 programs and sure no program has been made with the insight that's needed to play with 8 pawns on the 7th rank, but with a very simple evaluation and a fast search, which does some checks in qsearch you can already create the unbeatable program here with white. The branching factor is only true in 'stupid' lines or when your goal is already reached. If you are willing to pay me some money i sure wanna bet that i can create a program that's kicking your ass in a 6 game match (6 games with computer having white). if it doesn't kick you i pay you back. if it does kick you, then i keep the money. 500$ ok? Because as a chessplayer+programmer is realize *very* well exactly what i need to modify to win this game. It's so simple. Note that some features of diep's eval *already* are handy here. this is just coincidence that it works for wild5 a bit. >Regards, >Georg (Tecumseh @ Fics , TRT @ ICC)
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.