Author: Robert Hyatt
Date: 15:15:29 12/01/99
Go up one level in this thread
On December 01, 1999 at 14:57:54, Robert Hyatt wrote: >On December 01, 1999 at 13:09:37, Ed Schröder wrote: > >>I ran Howard's test, Rebel Century had no problems with all 9 positions. I >>will comment the interesting ones and paste the crucial lines of the LOG. >> >>r1bb2k1/p1q2pp1/4p1np/2ppP2Q/8/2P2N2/2P2PPP/R1B1R1K1 w - - am Bxh6; >>00:00:14 9.00 0.33 1.c4 d4 2.Bd2 a5 3.Ra3 a4 4.Rea1 Bd7 5.Rb1 >> >>r3q1k1/ppp1rpp1/2n1b2p/8/2P2B2/3B4/PPPQ1RPP/5RK1 w - - bm f4h6; >>00:00:00 1.00 1.02 1.Bxh6 gxh6 2.Qxh6 >>00:00:00 4.04 0.88 1.Bxh6 gxh6 2.Qxh6 f5 3.Bxf5 Bxc4 4.Bg6 >> >>The position which started the whole issue. >>---------------- >> >>r3q1k1/ppp1rpp1/4b1np/8/2P2B2/3B4/PPPQ1RPP/5RK1 w - - am BXH6; >>00:00:16 8.00 0.26 1.Bg3 Rd8 2.Qb4 b6 3.Bxg6 fxg6 4.Bh4 g5 5.Bg3 Bf7 >> >>3rrnk1/ppp1qpp1/4b2p/8/2P2B2/3B4/PPPQ1RPP/5RK1 w - - id BXH6spe; >>00:00:14 8.00 0.48 1.Qa5 c6 2.Qxa7 Ra8 3.Qd4 c5 4.Qc3 Rxa2 5.Bg3 >> >>3r1rk1/p2q1ppp/2pb1n2/5b2/1nB5/2N1BN1P/PP2QPP1/R4RK1 b - - bm BXH3; >>00:00:52 9.00 0.31 1..Rfe8 2.Nd4 Bxh3 3.gxh3 Qxh3 4.Nf3 Qg4+ >>00:02:19 9.19 0.41 1..Bxh3 2.gxh3 Qxh3 3.Rac1 Ng4 4.Ne4 Nxe3 5.Qxe3 Qg4+ >>6.Ng3 Bf4 >> >>Rebel Century needs SEARCH + EVAL to find the find the sac. >>----------- >> >>3r1rk1/p2q1ppp/2pb4/p4b2/1nB5/4BN1P/PP2QPP1/R4RK1 b - - am BXH3; >>00:00:54 9.00 0.82 1..Rfe8 2.Rfd1 Nc2 3.Ng5 Re7 4.Rac1 Nxe3 >> >>1qr1b1k1/4bpp1/pn2p2p/1p1nN3/3P4/P2BBN1Q/1P3PPP/4R1K1 w - - bm Bxh6; >>00:00:00 3.01 0.20 1.Bxh6 gxh6 2.Qxh6 Nf6 3.Ng5 a5 >>00:00:00 4.00 0.23 1.Bxh6 gxh6 2.Qxh6 Nf6 3.Ng5 Qd6 4.Bh7+ Nxh7 5.Qxh7+ Kf8 >>6.Nexf7 >>00:00:00 5.00 0.30 1.Bxh6 gxh6 2.Qxh6 Nf6 >>00:00:05 6.00 1.30 1.Bxh6 Nf6 2.Bg5 Qd6 3.Bxf6 Bxf6 4.Qh7+ Kf8 5.Qh8+ Ke7 >>00:00:08 7.00 1.39 1.Bxh6 Nf6 2.Bg5 Nbd5 3.Nd2 Bd6 4.b4 >>00:00:21 8.00 1.39 1.Bxh6 Nf6 2.Bf4 Qb7 3.Ng4 Nbd5 4.Nxf6+ Nxf6 5.Ne5+ >>00:00:49 9.00 1.42 1.Bxh6 Nf6 2.Bg5 Nbd5 3.Bxf6 Bxf6 4.Qh7+ Kf8 5.Qh8 Ke7 >>6.Ng4 >> >>Another working case of an early positional sacrifice. SEARCH proves Rebel >>is right. >>--------------- >> >>1qr1b1k1/4bpp1/p3p2p/1p1nN3/3P4/P3BN1Q/1P3PPP/4R1K1 w - - am Bxh6; >>00:00:00 1.01 0.13 1.Bxh6 gxh6 2.Qxh6 >>00:00:00 2.00 -0.34 1.Bxh6 gxh6 2.Qxh6 f6 3.Ng6 >>00:00:00 3.00 -1.02 1.Bxh6 gxh6 2.Qxh6 Bf8 3.Qh5 Bg7 >>00:00:00 3.01 -0.39 1.Qg3 Nxe3 2.fxe3 Rc2 >>00:00:00 4.00 -0.40 1.Qg3 Nxe3 2.fxe3 Rc2 3.b4 >>00:00:00 4.02 -0.34 1.Bd2 Rc2 2.Qg4 Nf6 3.Qg3 Bd6 >>00:00:00 5.00 -0.28 1.Bd2 Rc2 2.Qg3 g5 3.b4 >>00:00:01 6.00 -0.34 1.Bd2 Bd6 2.Qg4 Nf6 3.Qg3 Rc2 >>00:00:03 7.00 -0.34 1.Bd2 Rc2 2.Qg4 Kf8 3.b4 Bd6 >>00:00:09 8.00 -0.40 1.Bd2 Rc2 2.Qg4 Kh8 3.Qh5 Rxb2 4.Nxf7+ >>00:00:37 9.00 -0.34 1.Bd2 Rc2 2.Qg4 Kh8 3.b4 Qc7 >>00:02:17 10.00 -0.38 1.Bd2 Rc2 2.Qg4 Kf8 3.b4 Qc7 >> >>This is a very interesting one. The first 2 plies Rebel Century sacs the bishop >>but then is corrected by SEARCH after seeing 2..Bf8 and 3..Bg7 on the third >>ply. Isn't CC sometimes wonderful? >>------------------- >> >>r1br2k1/p1q2pp1/4p1np/2ppP2Q/2n5/2PB1N2/2P2PPP/R1B1R1K1 w - - bm Bxh6; >>00:00:17 8.08 0.00 1.Bxh6 gxh6 2.Qxh6 Rd7 3.Bxg6 >>00:01:43 9.00 0.62 1.Bxh6 a5 2.Bg5 Re8 3.Bxg6 fxg6 4.Qxg6 Rf8 >> >>Thanks Howard for interesting set. If you have more please post. >> >>Ed > > >This output wasn't exactly what I was looking for. I am much more interested >in your 1-2-3 ply results for these position... > >Why? > >Because here is what your program will do: (a) I like Bxh6, but after I get to >depth=7 I see that it is bad... so (b) I will first play two captures that >eat 4 plies, then a check to eat a couple more, and _then_ I can now play Bxh6 >and I don't see anything bad happening. > >This has happened to me on many occasions on ICC. Sure, at the root you can >dismiss Bxh6 maybe, but if you can't dismiss it at the tips with equal accuracy, >you are going to (on occasion) make a committal move at ply-1 that means you >_must_ follow through with Bxh6 at ply=9... > >What I want is as much tactical accuracy as I can get, with some speculative >evaluation terms that are right most of the time, but even when they are wrong >I maintain a chance to not lose. Because then this sort of 'horizon effect' >doesn't hurt. Of course, I might miss a speculative sac that wins, or I might >miss one by my opponent that loses (less likely as I use assymetric king safety) >But at least I try to avoid being "hoisted on my own petard"... I remembered another neat case where this was an issue. Crafty vs program X. Final position was KQP vs KQP. Crafty quickly discovered that (a) it could win the pawn but (b) the resulting KQP vs KQ ending was a dead draw, because of the tablebase hits. It found a way to win the pawn and (at the same time) fooling itself into believing that the game wasn't drawn. I probe in the regular search, but not in the q-search. So what it did was check-check-check until it exhausted the nominal search depth, and then it could play QxP in the q-search and avoid the draw... that kind of 'pushing things off beyond the horizon' is a pain...
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.