Author: Uri Blass
Date: 01:53:33 04/13/01
Go up one level in this thread
On April 13, 2001 at 00:54:03, Dann Corbit wrote: >On April 13, 2001 at 00:40:47, Uri Blass wrote: > >>It is possible to play gmaes at fixed depth under Deep Fritz. >> >>Did someone try to play tournament between engines at fixed depth in order to >>find out which engine earns more from getting more plies and if there is a >>diminishing return from bigger depth? >> >>I think that it will be intresting to have a rating list of the following >>programs: >> >>Deep Fritz(1 ply) >>Deep Fritz(2 plies) >>Deep Fritz(3 plies) >>... >>Deep Fritz(10 plies) >>Deep Fritz(11 plies) >>Deep Fritz(12 plies) >>Hiarcs7.32(1 ply) >>Hiarcs7.32(2 plies) >>Hiarcs7.32(3 plies) >>... >>Hiarcs7.32(10 plies) >>Gambittiger(x plies) >>Tiger14(x plies) >> >>You can add more programs like Nimzo8 or the latest chessbase version of Crafty. >>The rating of the programs should be based only on games between different >>programs so there should not be games between Deep Fritz(6 plies) and Deep >>Fritz(7 plies) because I remember that when Fritz6 played against itself people >>criticized the experiment and said that it does not prove diminishing returns >>from bigger depth. > >Two problems: >1. Everyone counts depth differently. I know it but it still interesting to know which programs earn more from more plies. >2. How do you set up a tournament to play fixed depths? It is easy under Deep Fritz gui Click on File,New engine match and you have options to set fixed depth that is not the same for the engines. Here is an example for a game Deep Fritz(1 ply) against Crafty18.03(2 plies) from the nunn match The game took only few seconds. I used 1 mbytes hash for both sides. Crafty18.03(2 plies) won the match 11-9 repetition in a good position is one of the weaknesses of crafty at this depth all the 4 draws in the match were result of repetition when Crafty had the advantage and in3 of them it was a big advantage. Deep Fritz - Crafty 18.03 [B33] pIII800, 2ply/1ply tel-aviv (2), 13.04.2001 W=1.0 ply; 13kN/s B=2.1 ply; 10kN/s 1.e4 c5 2.Nf3 e6 3.d4 cxd4 4.Nxd4 Nc6 5.Nb5 Nf6 6.N1c3 d6 7.Bf4 e5 8.Bg5 a6 9.Na3 b5 10.Bxf6 gxf6 11.Nd5 f5 12.c3 Bg7 13.exf5 Bxf5 Both last book move 14.Bd3 0.28/1 0 e4 -0.05/2 0 15.Qe2 (Be2) 0.25/1 0 15...0-0 0.89/2 0 16.Bxe4 1.28/1 0 Re8 0.46/2 0 17.f3 0.72/1 0 Qh4+ 0.30/2 0 18.g3 0.78/1 0 Qh5 0.55/2 0 19.Qd3 (Qe3) 0.94/1 0 19...Qg6 0.36/2 0 20.0-0-0 0.66/1 0 Bxe4 0.36/2 0 21.fxe4 0.66/1 0 Qxe4 (Rxe4) 0.53/2 0 22.Rhf1 0.78/1 0 Rac8 0.45/2 0 23.Rf6 (Kb1) 0.72/1 0 23...Qxd3 -1.72/2 0 24.Rxd3 0.78/1 0 Ne5 -1.63/2 0 25.Rd4 (Re3) 0.59/1 0 25...Ng6 -1.49/2 0 26.Kd2 0.97/1 0 Rc6 (Kf8) -1.41/2 0 27.c4 (Nc2) 0.53/1 0 27...bxc4 -0.98/2 0 28.Nxc4 (Rxc4) 0.53/1 0 28...Rb8 -0.40/2 0 29.Na5 (Nxd6) -0.69/1 0 29...Rxb2+ -2.95/2 0 30.Ke3 -0.69/1 0 Rcc2 -2.78/2 0 31.Nc4 (Rc4) -1.44/1 0 31...Re2+ (Rxa2) -3.53/2 0 32.Kd3 (Kf3) -1.44/1 0 32...Rxa2 -3.18/2 0 33.Nxd6 -2.12/1 0 Ne5+ -3.70/2 0 34.Kc3 -2.31/1 0 Rec2+ (Rac2+) -3.86/2 0 35.Kb3 -2.28/1 0 Rab2+ -3.86/2 0 36.Ka4 -2.28/1 0 Bxf6 -4.01/2 0 37.Nxf6+ -2.28/1 0 Kg7 -4.04/2 0 38.Nfe8+ (Nd5) -2.59/1 0 38...Kg6 -3.34/2 0 39.h3 -2.56/1 0 Ra2+ -3.34/2 0 40.Kb3 -2.56/1 0 Rab2+ -3.50/2 0 41.Ka4 0.00/1 0 Ra2+ -3.34/2 0 42.Kb3 0.00/1 0 Rab2+ -3.50/2 0 43.Ka4 Draw accepted 0.00/1 0 ½-½ > >The Crafty goes deep and DarkThought goes deep experiments are the only data I >have seen published on increasing ply verses ply+1 self-play. > >What happened with the Fritz tournament? Where is the data? I remember that it was proved with 95% confidence that there is a diminishing return from depth from the fritz experiment but it was only Fritz6 against itself at different depthes. I remember that the number of wins for the weaker side did not increase but the number of draws increased. 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.