Author: Stuart Cracraft
Date: 20:51:36 03/06/06
Go up one level in this thread
On March 06, 2006 at 23:20:21, Dann Corbit wrote: >On March 06, 2006 at 23:04:24, Dann Corbit wrote: > >>*** Problem Solution(s): Nxd5 (bm) >>[D] r1q2rk1/p3bppb/3p1n1p/2nPp3/1p2P1P1/6NP/PP2QPB1/R1BNK2R b KQ - bm Nxd5 >>*** Problem Solution(s): Nxd5 (bm) >>BR ** BQ ** -- BR BK ** >>BP -- ** -- BB BP BP BB >>-- ** -- BP -- BN -- BP >>** -- BN WP BP -- ** -- >>-- BP -- ** WP ** WP ** >>** -- ** -- ** -- WN WP >>WP WP -- ** WQ WP WB ** >>WR -- WB WN WK -- ** WR >>mv 1 stage 0, black to move, computer plays black >>hash=3a81fdded6269e62 >>pawnhash=2466670e8a77b5de >>Alpha=-111 Beta=689 Maxdepth=9999999 MaxTime=1000 >>Ply/Max Mv Score Time Nodes PV >> 1/ 6>f6e4 689 0.00 45 f6e4 >> 1/ 6 c8d8 1117 0.01 124 c8d8 >> 2/ 6 c8d7 981 0.01 256 c8d7 >> 3/ 9<c8d7 580 0.02 1136 c8d7 c1d2 >> 3/ 9 c8d7 537 0.03 2049 c8d7 b2b3 >> 4/10 c8d7 636 0.07 4563 c8d7 g4g5 h6g5 c1g5 >> 5/13 c8d8 762 0.17 13572 c8d8 g4g5 f6d7 >> 6/19 c8c7 681 0.71 59228 c8c7 g4g5 h6g5 c1d2 a8b8 d2g5 >> 7/19 c8a6 602 2.45 206288 c8a6 e2a6 c5a6 c1d2 a8b8 mtmt >> 8/23 c8d8 688 4.37 375249 c8d8 f2f3 a8c8 c1d2 >> 9/25 Tc8d8 508 10.01 855040 c8d8 e1g1 a8c8 g1h2 g8h8 f2f3 >>nps=85410 h/p/q=27.00%/97.00%/0.00% q=80.0% bc=47% >>br=3.59mp=1921/3040/-2129/-1311 >>pawnx=9340 recapx=5270 qcheckx=0 checkx=10159 qfutilx=224976 onereplyx=272 >>mthrx=246 futilx=0 bmx=0 >>c8d8 e1g1 a8c8 g1h2 g8h8 f2f3 - 10.00/24.60 0% 0/5 bf=2.58 >>h/p/q=27.80%/98.00%/0.00% 50.07 5432343 1086469/10/108499 >>34809/18912/738/120154/1064146/6169/0/2714 >> >>There is something very intersting here. You have a large positive score for >>the queen move. The best move (according to most programs) is the knight sac, >>which leads, after a convoluted exchange, to an advantage of less than one pawn. >> Perhaps you have discovered something interesting here. > >Is your score reported in millipawns? If not, then something does not make >sense with your scores up above. Or perhaps I am reading the output wrong. > >Interesting that Aristarch like c8d7 for a while: > >Analysis from E:\bust.epd >3/6/2006 8:05:17 PM Level: Infinite >Analyzing engine: Aristarch 4.50 > >1) Nxd5; > Searching move: Nf6xd5 > Best move (Aristarch 4.50): Nf6xd5 > identical moves! Found in: 01:35 > 2 00:00 2 2 -0.42 f6g4 ? e2g4 > 2 00:00 38 38 -3.52 f6g4 e2g4 > 2 00:00 76 76 -3.51 f6e4 !! g3e4 > 2 00:00 113 113 -3.32 f6e4 g2e4 > 2 00:00 150 150 -3.31 f6d5 !! e4d5 > 2 00:00 186 186 -3.11 f6d5 !! e4d5 > 2 00:00 222 222 -2.85 f6d5 e4d5 > 2 00:00 268 268 -2.84 a7a5 !! c1h6 > 2 00:00 303 303 -2.64 a7a5 !! c1h6 > 2 00:00 341 341 +0.11 a7a5 g3f5 > 2 00:00 387 387 +0.12 b4b3 !! a2b3 > 2 00:00 424 424 +0.18 b4b3 g3f5 > 2 00:00 530 530 +0.19 h7g6 !! g3f5 > 2 00:00 566 566 +0.24 h7g6 g3f5 > 3 00:00 705 705 +0.32 h7g6 g3f5 c8d7 > 3 00:00 903 903 +0.33 c8b7 !! g3f5 a8c8 > 3 00:00 1.090 1.090 +0.33 c8b7 g4g5 h6g5 > 3 00:00 1.153 1.153 +0.34 c8d7 !! g3f5 a8c8 > 3 00:00 1.309 1.309 +0.41 c8d7 g4g5 h6g5 > 4 00:00 3.046 152.300 +0.08 c8d7 c1d2 a7a5 a1c1 > 4 00:00 3.233 161.650 +0.09 c8b7 !! g4g5 h6g5 c1g5 > 4 00:00 3.717 185.850 +0.26 c8b7 e1g1 a8c8 g3f5 > 5 00:00 10.930 546.500 +0.14 c8b7 f2f4 e5f4 c1f4 a8c8 g3f5 > 5 00:00 11.895 594.750 +0.15 c8d7 !! c1d2 a7a5 a1c1 a8c8 > 5 00:00 13.702 456.733 +0.35 c8d7 !! c1d2 a7a5 a1c1 a8c8 > 5 00:00 18.092 603.066 +0.44 c8d7 c1d2 a7a5 a2a3 d7a4 > 6 00:00 40.018 666.966 +0.28 c8d7 c1d2 b4b3 e2c4 b3a2 b2b4 d7a4 > 6 00:00 51.968 649.600 +0.29 c8b7 !! f2f4 e5f4 c1f4 g7g5 f4c1 a8c8 > 6 00:00 62.322 692.466 +0.40 c8b7 f2f4 e5f4 c1f4 g7g5 f4d2 a8c8 > 7 00:00 176.309 705.236 -0.02 c8b7 ? h3h4 b7d7 g4g5 f6g4 f2f3 h6g5 > 7 00:00 188.594 698.496 -0.02 c8b7 h3h4 b4b3 g4g5 f6d7 g5h6 a8c8 > 7 00:00 195.383 697.796 -0.01 c8d7 !! c1d2 b4b3 e2c4 b3a2 b2b4 d7a4 >b4c5 > 7 00:00 207.649 692.163 +0.19 c8d7 !! c1d2 b4b3 e2c4 b3a2 b2b4 d7a4 >b4c5 > 7 00:00 261.244 725.677 +0.43 c8d7 c1d2 a7a5 g3f5 a8c8 a1c1 h7g6 > 7 00:00 306.316 729.323 +0.44 f6d5 !! e4d5 c5d3 e1d2 d3f4 e2f3 c8c2 >d2e1 f4d3 e1f1 > 7 00:00 320.513 728.438 +0.64 f6d5 !! e4d5 c5d3 e1d2 d3f4 e2f3 c8c2 >d2e1 f4d3 e1f1 > 7 00:01 414.511 753.656 +0.97 f6d5 g3f5 h7f5 g4f5 d5f6 c1d2 c8b7 f2f3 > 8 00:01 516.011 747.842 +0.68 f6d5 g3f5 h7f5 e4f5 c8b7 e2c4 e5e4 g2e4 >c5e4 > 9 00:01 808.025 784.490 +0.89 f6d5 g3f5 h7f5 e4f5 c8b7 e2c4 e5e4 e1g1 >d5f6 d1e3 > 10 00:02 1.346.429 826.030 +0.85 f6d5 g3f5 h7f5 e4f5 c8b7 e1g1 a8d8 e2c4 >d5b6 c4b4 d6d5 > 11 00:03 2.503.074 848.499 +0.67 f6d5 g3f5 h7f5 e4f5 c8b7 e1g1 a8c8 e2c4 >e5e4 g2e4 > 12 00:05 4.658.176 846.941 +0.25 f6d5 ? g3f5 h7f5 e4f5 c8b7 e2c4 e5e4 >g2e4 c5e4 c4e4 b7b5 d1e3 d5e3 f2e3 > 12 00:06 5.760.049 853.340 +0.22 f6d5 g3f5 h7f5 e4f5 c8b7 e2c4 e5e4 g2e4 >c5e4 c4e4 a8c8 d1e3 c8c5 e1g1 > 12 00:08 7.184.824 845.273 +0.23 c8d7 !! g3f5 b4b3 > 12 00:18 14.912.697 795.343 +0.43 c8d7 !! g3f5 a8c8 > 12 00:29 22.597.382 779.220 +0.38 c8d7 e1g1 f8c8 b2b3 a7a5 d1b2 e7d8 b2c4 >h6h5 g3f5 h7f5 e4f5 a5a4 > 13 01:12 56.881.398 790.019 +0.47 c8d7 g4g5 h6g5 c1d2 > 13 01:35 76.500.639 805.269 +0.48 f6d5 !! e4d5 c5d3 e1d2 d3f4 e2f3 c8c2 >d2e1 f4d3 e1f1 > 13 01:39 80.309.833 811.210 +0.52 f6d5 g3f5 h7f5 e4f5 c8b7 e1g1 a8d8 f1e1 >b4b3 d1e3 d5f4 g2b7 > 14 02:06 102.082.343 810.177 +0.72 f6d5 g3f5 > 15 04:43 225.280.866 796.045 +1.05 f6d5 e4d5 > 3/6/2006 8:15:34 PM, Time for this analysis: 00:05:00, Rated time: 01:35 > >1 of 1 matching moves >3/6/2006 8:15:35 PM, Total time: 12:10:18 AM >Rated time: 01:35 = 95 Seconds Yes -- all millipawns. Here are my evaluation settings: Passers get the credit by the rank squared times a factor related to the attack/defense/occupation status of the square in front of it. This is from the Chess 4.7 chapter in Frey's book, as is almost all the evaluation function. I couldn't think of a passed connected value so I use half a pawn (500). 1, // PAWN_PASSED_R2 : Passed pawn credit 1, // PAWN_PASSED_R3 : Passed pawn credit 1, // PAWN_PASSED_R4 : Passed pawn credit 1, // PAWN_PASSED_R5 : Passed pawn credit 1, // PAWN_PASSED_R6 : Passed pawn credit 1, // PAWN_PASSED_R7 : Passed pawn credit 500, // PAWN_PASSED_CONNECTED : Passed pawns connected No credit for advancing pawns on files a, b, or g. Slight negative for an h-file advance 0, // PAWN_ADVANCED_A : Advanced pawn A file 0, // PAWN_ADVANCED_B : Advanced pawn B file 39, // PAWN_ADVANCED_C : Advanced pawn C file 54, // PAWN_ADVANCED_D : Advanced pawn D file 70, // PAWN_ADVANCED_E : Advanced pawn E file 23, // PAWN_ADVANCED_F : Advanced pawn F file 0, // PAWN_ADVANCED_G : Advanced pawn G file -2, // PAWN_ADVANCED_H : Advanced pawn H file Doubled, isolated are fairly heavy penalties. -80, // PAWN_DOUBLED : Doubled pawn penalty (anywhere on board) -200, // PAWN_ISOLATED : Isolated pawn penalty (anywhere on board) Weak pawns (lite and very) are based on Bob Hyatt's recent comments. I didn't know what to score them so I just picked 1/100th and 1/50th of a pawn. -10, // PAWN_WEAK_LITE : Not defended and square in front is attacked by 1 gt defended -20, // PAWN_WEAK_VERY : Not defended and square in front is attacked by >1 gt defended Attacking a weak pawn gets a small credit. 7, // PAWN_WEAK_ATTACK : Attacks on a weak pawn The next three I believe are all from Crafty by way of GNU but I don't know if the values of -1/10th of a pawn are too small or too big. -100, // PAWN_EIGHT : Still eight pawns remaining on computer's side -100, // PAWN_STONEWALL : Fight people using stonewall -100, // PAWN_LOCKED : Locked pawns in the center and extended center This is the multiplier for knight tropism. Just like Chess 4.7 16, // KNIGHT_CENTER_TROPISM : Multiplier for center tropism 6-2*(4.5-r+4.5-f) Penalties for undevelopment of minors are the same. -94, // KNIGHT_UNDEV : Penalty per undeveloped knight anywhere on back rank -110, // BISHOP_UNDEV : Penalty per undeveloped bishop anywhere on back rank Likewise bishop square control. This is a multiplier that if 1 simply keeps the same value as Chess 4.7. 1, // BISHOP_SQUARE_CONTROL : Bonus per bishop-attacked non-friendly-pawn square This is from GNU. I made knights and bishops 3250 and 3250 instead of Chess 4.7's 3250 and 3500 because (really 325 and 350 since 4.7 was centipawn) because I think the bishop pair is very valuable together but not so useful if the other side has a bishop. 180, // BISHOP_PAIR : Advantage of the bishop pair (B+B) against no bishops The obvious for rooks 16, // ROOK_SQUARE_CONTROL : Bonus per rook-attacked square -16, // ROOK_KING_TROPISM : Penalty for distance to least of enemy king file/column 80, // ROOK_DOUBLING : Bonus for attacking a friendly rook 220, // ROOK_SEVENTH : Bonus for each rook on seventh 80, // ROOK_OPENFILE : Bonus for rook on open file 30, // ROOK_SEMIOPENFILE : Bonus for rook on semi-open file -400, // QUEEN_FORAGING : Queen moving in opening Still more 4.7 stuff for Queen's -8, // QUEEN_KING_TROPISM : Queen tropism to king as function of minimum of taxicab f/r 8, // QUEEN_SQUARE_CONTROL : Credit for each un-enemy-attacked my-queen-controlled sq -250, // QUEEN_ABSENT : Queen is absent King safety is identical to 4.7 as well, including the simplified mopup algorithm of 4.7. These are for the main king safety. -23, // KING_CORNERED : King cornered in endgame -12, // KING_ATTACKED : Enemy king adjacent square attackers -31, // KING_NOSANCTUARY : Friendly king not in sanctuary -78, // KING_PAWNAFFINITY : Friendly king affinity for center of all pawns in endgame -31, // KING_BARE : Friendly king has no friendly pawns or pieces next to it -16, // KING_NEARLYBARE : Friendly king has only one friendly pawn or piece next to it -41, // KING_NUDE : Friendly king has no pawn on same file -36, // KING_NUDER : Friendly king has no pawn on at least one neighbor file -8, // KING_LUFT : Friendly king has no luft on 2nd rank Hung pieces are a little different. 4.7 sets to -16 for one and if more than the worst positional evaluation so far for the side in the search. I can add this when my minposnscore[2] array is maintained accurately and its extreme sizes bug discovered. minposnscore isn't used right now. -16, // HUNG : Hung pieces require tempi to safety (defend, move, ignore, etc.) A small flat credit for mobility, all pieces. This is not 4.7 either. 4 // MOBILITY : Mobility bonus for pseudo-legal moves at terminal nodes Stuart
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.