Author: Uri Blass
Date: 03:43:30 05/11/03
Go up one level in this thread
On May 11, 2003 at 06:08:23, Geoff wrote: >Hi > >Seems several of us are currently struggling with adding Null move logic to our >code. I have added it to what started off as TSCP. > >My first attempt was pretty hopeless. > >Added new features were ;-) > >1) Failure to return from search in some positions >2) Timeout code was broken as the taking back of moves after a timeout now went >wrong. >3) Depth extension if in check now causes something to go wrong too. > >Anyways I have been tinkering and trying to fix it. Null move is a small amount >of extra code but very tricky to debug and understand. Least it is for me ! > >I am not sure I have got it totally fixed now, but it seems to be working OK at >the moment (only a couple hours of testing) > >The results are amazing, so good in fact I am wondering if I am doing too many >Null moves and potentially adding a bug? A time of 324.6 Secs drops to 9.6 Secs. >Is this plausible ? Yes Of course it is not going always to be the case and there may be cases when you find the solution one ply later or worse than it. > >See results below on a particular test position I tried. The one thing that I >havent figured out yet is does adding Null move screw up the reps. detect. >I suspect it does but dont know what to do to fix it ? Any thoughts ? > > Regards Geoff > > >Test Results > >/* Queen Sacrifice Mate in 07 Apr 2003 Alexandra Kosteniuk (White) v John D. >Bick (Black) > > 1.f5xf7+ f8xf7 2.h2xh7+ f7-g8 3.h7-h8+ g8-f7 4.g4xf4+ e7-f6 5.f4xf6+ f7-e7 > 6.h8-h7+ e7-d8 7.f6-f8# */ > >[D]1r3k2/2q1bb1p/2Bp2p1/3P1QP1/3B1pR1/2P2P2/P1PK3R/1r6 w - - 0 1 > >Before NullMove Mods >Ply Eval Time Nodes Principle Variation > 1: 332 0 66 f5f4 > 2: 293 0 856 f5f4 f8g8 h2h7 g8h7 f4f7 > 3: 325 0 3315 f5f4 f8g8 a2a4 > 4: 323 3 16149 f5f4 f8g8 g4h4 h7h5 g5h6ep e7h4 h2h4 > 5: 323 17 79165 f5f4 f8g8 g4h4 h7h5 g5h6ep e7h4 h2h4 > 6: 494 137 545323 f5f4 h7h5 g5h6ep g6g5 g4g5 e7g5 f4g5 > 7: 533 475 1971525 f5f4 h7h5 g5h6ep g6g5 f4e3 f8g8 a2a4 > 8: 805 2659 10248418 f5f7 f8f7 h2h7 f7g8 h7g7 g8f8 g4f4 e7f6 g7c7 > 9: 805 8301 33627573 f5f7 f8f7 h2h7 f7g8 h7g7 g8f8 g4f4 e7f6 g7c7 >10: 1164 32432 124888939 f5f7 f8f7 h2h7 f7g8 h7g7 g8f8 g4f4 e7f6 g7c7 >b8b2 g5f6 b2a2 > > >After Null move Mods but 1 Null Move per search line only >Ply Eval Time Nodes Principle Variation > 1: 332 0 66 f5f4 > 2: 293 0 856 f5f4 f8g8 h2h7 g8h7 f4f7 > 3: 325 0 3316 f5f4 f8g8 a2a4 > 4: 323 3 14578 f5f4 f8g8 g4h4 h7h5 g5h6ep e7h4 h2h4 > 5: 323 14 61484 f5f4 f8g8 g4h4 h7h5 g5h6ep e7h4 h2h4 > 6: 494 40 166882 f5f4 h7h5 g5h6ep g6g5 g4g5 e7g5 f4g5 > 7: 533 107 435051 f5f4 h7h5 g5h6ep g6g5 f4g3 f8g8 a2a4 > 8: 805 237 958128 f5f7 f8f7 h2h7 f7g8 h7g7 g8f8 g4f4 e7f6 g7c7 > 9: 805 914 3817929 f5f7 f8f7 h2h7 f7g8 h7g7 g8f8 g4f4 e7f6 g7c7 >10: 1164 4659 18642807 f5f7 f8f7 h2h7 f7g8 h7g7 g8f8 g4f4 e7f6 g7c7 >b8b2 g5f6 b2a2 > > >After Null move but more than 1 allowed per search line >Ply Eval Time Nodes Principle Variation > 1: 332 0 66 f5f4 > 2: 293 0 856 f5f4 f8g8 h2h7 g8h7 f4f7 > 3: 325 1 3316 f5f4 f8g8 a2a4 > 4: 323 4 14578 f5f4 f8g8 g4h4 h7h5 g5h6ep e7h4 h2h4 > 5: 323 15 61484 f5f4 f8g8 g4h4 h7h5 g5h6ep e7h4 h2h4 > 6: 494 45 166882 f5f4 h7h5 g5h6ep g6g5 g4g5 e7g5 f4g5 > 7: 533 115 435052 f5f4 h7h5 g5h6ep g6g5 f4g3 f8g8 a2a4 > 8: 805 178 674742 f5f7 f8f7 h2h7 f7g8 h7g7 g8f8 g4f4 e7f6 g7c7 > 9: 805 462 1790313 f5f7 f8f7 h2h7 f7g8 h7g7 g8f8 g4f4 e7f6 g7c7 >10: 1164 960 3758500 f5f7 f8f7 h2h7 f7g8 h7g7 g8f8 g4f4 e7f6 g7c7 >b8b2 g5f6 b2a2 > >Computer's move: f5f7 I tried the same test position Movei finds Qf7+ at depth 2 but I admit that it does a lot of check extensions when part of them are in the qsearch. It does not show the last mate move because a problem in updating the pv in my qsearch but it is no problem in games because the first move is never in qsearch. 1 385 0 93 f5f4 2 382 0 273 f5f4 f8g8 2 383 0 674 f5f7 2 412 0 696 f5f7 2 1243 1 1028 f5f7 f8f7 h2h7 f7g8 h7g7 g8f8 g4f4 e7f6 g7c7 f8g8 f4f6 2 1243 1 1037 f5f7 f8f7 h2h7 f7g8 h7g7 g8f8 g4f4 e7f6 g7c7 f8g8 f4f6 3 1243 1 2301 f5f7 f8f7 h2h7 f7g8 h7g7 g8f8 g4f4 e7f6 g7c7 f8g8 f4f6 3 1243 2 2998 f5f7 f8f7 h2h7 f7g8 h7g7 g8f8 g4f4 e7f6 g7c7 f8g8 f4f6 4 9987 3 5185 f5f7 f8f7 h2h7 f7g8 h7h8 g8f7 g4f4 e7f6 f4f6 f7e7 h8h7 e7d8 4 9987 4 6407 f5f7 f8f7 h2h7 f7g8 h7h8 g8f7 g4f4 e7f6 f4f6 f7e7 h8h7 e7d8 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.