TalkChess.com

pichy wrote:[D]8/8/7p/3KNN1k/2p4p/8/3P2p1/8 w - - 1 1

Maybe you can experiment with very deep searches or even use the Monte Carlo tree search first used by Shogi programmers and later used in the Fritz/Rybka interface to evaluate the position. A new type of analysis called Monte Carlo is now making computers able to understand when a position is drawn because no more progress can be made. How? The chess engine will start multiple ultra fast games against itself and analyse the results of those games. If they are all draw (50 moves without piece capture or pawn move is a draw), the computer will understand the position is draw. Also, if for example, one move leads to only drawn positions while other moves lead to lost games, the computer will understand that it has to play this one move, even if it does not understand why (fortress? perpetual check? blocked position? does not matter as long as the result is there)

pichy wrote:A new type of analysis called Monte Carlo is now making computers able to understand when a position is drawn because no more progress can be made. How? The chess engine will start multiple ultra fast games against itself and analyse the results of those games. If they are all draw (50 moves without piece capture or pawn move is a draw), the computer will understand the position is draw. Also, if for example, one move leads to only drawn positions while other moves lead to lost games, the computer will understand that it has to play this one move, even if it does not understand why (fortress? perpetual check? blocked position? does not matter as long as the result is there)

ZirconiumX wrote:

pichy wrote:A new type of analysis called Monte Carlo is now making computers able to understand when a position is drawn because no more progress can be made. How? The chess engine will start multiple ultra fast games against itself and analyse the results of those games. If they are all draw (50 moves without piece capture or pawn move is a draw), the computer will understand the position is draw. Also, if for example, one move leads to only drawn positions while other moves lead to lost games, the computer will understand that it has to play this one move, even if it does not understand why (fortress? perpetual check? blocked position? does not matter as long as the result is there)

Not quite. The engine plays purely random moves against itself - a lot of them. A better move will have a higher proportion of winning - this is the move that is chosen.

Matthew:out

Master Om wrote:The solution is not far!!

New game
[d]8/8/7p/3KNN1k/2p4p/8/3P2p1/8 w - - 0 1

Analysis by Houdini 3 Pro x64 L:

1. = (-0.17): 1.Kc6 g1Q 2.Nxh4 Qa1 3.Nhf3 Qa5 4.Kd6 Qb5 5.Ke6 Qc5 6.Kf5 Qd5 7.Kf6 Qb7 8.Ke6 Qb1 9.Kd5 Qa2 10.Ke4 Qb2 11.Kf4 c3 12.dxc3 Qxc3 13.Kf5 Qb4 14.Ke6 Qb3+ 15.Kf6 Qc2 16.Ke6 Qb1 17.Kd5 Qb4 18.Ke6 Qf8 19.Kd5 Qa3 20.Kc4 Qa5 21.Kd4 Qb4+ 22.Ke3 Qc5+ 23.Kf4
2. = (-0.17): 1.Ng7+ Kg5 2.Nf3+ Kg4 3.Ke4 h3 4.Nf5 g1Q 5.Nxh6+ Kh5 6.Nxg1 h2 7.Nf3 h1Q 8.Nf5 Kg4 9.Ne3+ Kg3 10.Nf5+ Kf2 11.N5d4 Qb1+ 12.Kd5 Qa2 13.Kc5 Kg3 14.Kb4 Kf4 15.Kc3 Qa5+ 16.Kxc4 Ke4 17.Kb3 Kd3 18.Nc6 Qb6+ 19.Ka2 Kc4 20.Nfe5+ Kd5 21.Ka3 Qb1 22.Ka4 Qc2+ 23.Kb5 Qb2+ 24.Ka6 Qxd2 25.Kb7

(Prakash, Bhubaneswar 22.02.2013)

Master Om wrote:1. = (-0.17): 1.Kc6 g1Q 2.Nxh4 Qa1 3.Nhf3 Qa5

This is the solution of the problem. It is in PV. If u set fifty move distance to 10 it clearly gives 0.00 after move 4. Hence it is solved.