Author: Mridul Muralidharan
Date: 12:18:36 12/10/04
Go up one level in this thread
Inline ... On December 10, 2004 at 15:00:18, Uri Blass wrote: >On December 10, 2004 at 13:50:43, Mridul Muralidharan wrote: > >>On December 09, 2004 at 04:00:11, Eduard Nemeth wrote: >> >>>This position is from a game, that was postet from programmer of Aristarch in >>>the CSS Forum: >>> >>> >>>[D]8/2p5/2P1R3/3k4/4n2p/6pP/6P1/4r1NK w - - 0 1 >>> >>>Analysis by Shredder 7.04: >>> >>>1.Te5+ Kxc6 2.Tc5+ Kd6 3.Tc6+ Ke5 4.Tc5+ Kf4 5.Tf5+ Ke3 6.Te5 Kd4 7.Td5+ Kc4 >>>8.Td4+ Kb5 9.Td5+ Ka6 10.Ta5+ Kb7 11.Ta7+ Kb6 12.Tb7+ Ka5 13.Ta7+ Kb5 14.Tb7+ >>>Ka4 15.Ta7+ >>> -+ (-2.70) Tiefe: 16/34 00:01:17 12978kN, tb=31 >>> >>>Only +2.70. >>> >>>Analysis by Fritz 8.0026: >>> >>>1.Te5+ Kxc6 2.Tc5+ Kd6 3.Td5+ Ke7 4.Te5+ Kd7 5.Td5+ Kc8 6.Td8+ Kb7 7.Tb8+ Kc6 >>>8.Tb6+ >>> -+ (-4.50) Tiefe: 17/68 00:00:39 10770kN, tb=2 >>> >>>(Nemeth, Privat 09.12.2004) >>> >>>Fritz 8 is better! 4.50. >>> >>>Junior 8 the best. >>> >>>Analysis by Junior 8: >>> >>>1.Txe4 Txe4 2.Sf3 Ta4 3.Sg1 Ke4 4.Sf3 Ke3 5.Sg1 Kf2 6.Sf3 Ta1+ 7.Sg1 Txg1# >>> -+ (-#8) Tiefe: 24 00:00:18 5179kN, tb=630 >>>1.Te5+ Kd6 2.Td5+ Kxc6 3.Td6+ Kb7 4.Tb6+ Ka8 5.Ta6+ Kb8 6.Ta2 Sf2+ 7.Txf2 gxf2 >>>8.g4 fxg1D# >>> -+ (-#9) Tiefe: 24 00:00:36 10202kN, tb=764 >>> >>>Mate after 18 seconds! >>> >>>And: What says your program? >>> >>>Eduard >> >>Hi Eduard,Martin , all , >> >>My current program does not resolve this as a draw (assuming it is ?) >> >>depth : 16 , value : -3687 , time : 31625ms 15047657 (70.0289) e6e5 d5c6 e5c5 >>c6d6 c5d5 d6e6 d5e5 e6f6 e5f5 f6e7 f5f7 e7e6 f7e7 e6f6 e7f7 f6e5 f7e7 e5d6 e7d7 >>d6c5 d7c7 c5b6 c7b7 b6a5 b7a7 a5b5 a7b7 b5c6 b7b6 c6c5 b6b2 e4f2 b2f2 g3f2 >> >>I will try leaving it to analyze for higher amount of time ... >> >>The position Martin Baumung posted >>(b2b3R/1np2pp1/Np6/n2K4/p7/N1k2p2/P1p3p1/2B2rqr w - - 0 1) gets resolved in my >>QSearch itself though ;-) so that gets solved at depth 1 ..... > >It seems that you do not limit the depth of the qsearch. >I do not believe that this strategy can be correct for games and the qsearch may >explode. > >Movei solves it very fast but not at depth 1. Ofcourse I limit the depth silly ! Ok , to more description. My qsearch is a combination of a selective search and a plain old cap only qsearch (with check evasion ofcourse ! most people dont consider this so explict mention). The depth limit is of two kinds : a static upper bound (the max you can ever search) and a dynamic bound. The dynamic bound depends on two things : In cap only qsearch : a) Then obviously this is the max number of captures possible - simple and no need to limit here. In selective search Other than (a) , another factor comes in : b) I very severely limit the possible moves and depth extension's here , but then some moves/positions are just way too interesting ! When it comes to checking moves , things get slightly complicated. Initial idea was to use something similar to what Ed mentions in his page. This has potential to blow up so fast I had to modify it quiet extensively. My current code from 60,000 ft high might resemble what Ed has posted ... basic idea is same with extensive adaption to make it more real world friendly (atleast from my expierence that is YMMV) > >>Lots more work yet to be done by me , esp on draw detection in eval and with >>search techniques ! >> >>It would be great if you can post the solution here - can I assume that the pv >>given by yace is the expected result ? (yet to exhaustively check that out ...) >> >>Thanks >>Mridul > >Movei also has problems to see draw score and see almost 1 pawn advantage for >black but at depth 15 it has a fail high and hopefully it will be able to see >draw score at depth 15. > >depth=14 -0.93 e6e5 d5c4 e5c5 c4d3 c5c3 d3d4 c3d3 d4c5 d3d5 c5b4 d5d4 b4b5 d4d5 >b5a4 d5a5 a4b3 a5b5 b3c2 b5c5 c2d2 c5d5 d2e3 d5e5 e3d4 e5d5 d4c3 d5c5 c3b4 c5b5 >b4c4 >Nodes: 360637114 NPS: 848657 >Time: 00:07:04.95 >depth=15 -0.63 e6e5 >Nodes: 1165795377 NPS: 851281 >Time: 00:22:49.46 Interesting .... looks like this position is something I need to look more into ! Thanks for the pv and eval data. Regards Mridul
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.