Author: Dann Corbit
Date: 12:53:18 01/27/99
Go up one level in this thread
On January 27, 1999 at 15:11:46, Don Dailey wrote: [snip] >Hypothesis 1 does not need to be tested. As worded it is false, but >in principle it is true. > >If the programs are "correct" (which I'll define), and you define best >move as a set of best moves then the hypothesis is correct. > >"Correct" means the program is constructed in such a way that it is >guaranteed to eventually find the "best" move. Pure null move >programs may fail this requirement. Probably even the hash table >implementations of most or all programs will defeat this requirement, >but this could be turned off. We are not considering time factors, >any current program/machine combination may require trillions of years >(or many orders of magnitude more to find the best move in some >positions.) > >If the "best" move is interpreted to be ANY move that preserves a >WIN if a win exists, or preserves a draw if a draw is the best one >can achieve given best play, then the hypothesis is true. You bring up some points that I did not think about, such as null move pruning. Of course, in some programs, you can turn it off (e.g. Arasan 5.0a) but in some you can't so in a few situations, you might *never* get equivalent results. I had not thought of that possibility. I was speaking of, in general terms, the best possible move (which is any move leading to checkmate). Here are some examples where the best move in a known test suite is not the same as a move that leads to checkmate. You could harldy say that the checkmate was wrong: 1K2krb1/1P1RP2p/7P/1n2N3/B3Q3/8/8/8 w - - acd 4; acn 55013; acs 1; bm Nf7; ce 32762; id "BWTC.1000"; pv exf8=Q+ Kxf8 Ng6+ hxg6 Qe7#; 1k6/1P3Q2/8/8/8/3K4/8/8 w - - acd 4; acn 10179; acs 1; bm Kc4; ce 32760; id "IMCP.024"; pv Qb3 Kc7 b8=Q+ Kd7 Qd5+ Ke7 Qbd8#; 1r4k1/p4ppp/8/1B6/3rP2K/8/PPP3q1/R1BQ1R2 b - - acd 4; acn 24300; acs 1; bm Rxb5 Rxe4+; ce 32758; id "WCSAC.0901"; pv Rxe4+ Bf4 g5+ Kh5 Qh3+ Kxg5 h6+ Kf6 Re6#; 3q3r/2pb1Qp1/1rpk1b2/3pN3/3Pp3/2N4P/PPP5/R3K2R w KQ - 0 1 bm Nc4+ Nxe4+; acd 4; acn 6795; acs 1; ce 32764; id "F.D.Bloss' Test No.3"; pv Nxe4+ dxe4 Nc4#; 3r4/3N2kr/1p6/pBpn1p2/Q2PR1p1/P7/1P4P1/2q3K1 w - - acd 10; acn 125819802; acs 743; bm Bf1; c0 "HITECH/DEEP THOUGHT, World CCC 1989, move 42W (Bf1)"; ce -32755; id "Crafty POS097"; pv Kf2 fxe4 Qb3 e3+ Kg3 e2 Qxd5 e1=Q+ Kxg4 Rh4+ Kf5 Qf4#; 4B3/5nb1/R7/3r1k2/6NK/3p1P2/4n1N1/1Q6 w - - acd 4; acn 63429; acs 1; bm Qb5; ce 32762; id "CUSTMATE.008"; pv N2e3+ Kf4 Nxd5+ Kxf3 Qh1#; 4r2k/p2Q3p/2p1r1pR/4NpR1/3P4/2P1PP1b/P4q1P/6RK w - - acd 4; acn 27835; acs 1; bm Nxg6+; ce 32766; id ECM.462; pv Qxh7#; c0 This one's funny, don't you think? Even more interesting, here are some "Best move" epd records where if the opponent plays correctly, you are checkmated anyway: 3r4/3N2kr/1p6/pBpn1p2/Q2PR1p1/P7/1P4P1/2q3K1 w - - acd 10; acn 125819802; acs 743; bm Bf1; c0 "HITECH/DEEP THOUGHT, World CCC 1989, move 42W (Bf1)"; ce -32755; id "Crafty POS097"; pv Kf2 fxe4 Qb3 e3+ Kg3 e2 Qxd5 e1=Q+ Kxg4 Rh4+ Kf5 Qf4#; 6nk/R7/8/7K/8/8/8/8 b - - acd 23; acn 186260587; acs 771; bm Nf6+; ce -32759; id "ECE3.0427"; pv Ne7 Rxe7 Kg8 Kg6 Kf8 Re6 Kg8 Re8#; 8/8/8/8/4pk2/5p2/5Kp1/4R3 w - - acd 16; acn 160897080; acs 510; bm Kg1; ce -32747; id "ECE3.0218"; pv Re3 g1=Q+ Kxg1 Kxe3 Kf1 f2 Kg2 Ke2 Kg3 f1=Q Kg4 Qf6 Kh5 Ke3 Kg4 Qg6+ Kh4 Kf4 Kh3 Qg3#;
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.