Author: Vincent Diepeveen
Date: 16:00:52 01/29/02
Go up one level in this thread
On January 29, 2002 at 00:05:01, Will Singleton wrote: >On January 28, 2002 at 23:20:45, James Swafford wrote: > >> >>Does anybody do this? I just implemented a passed pawn extension >>that extends the depth by 1 if the move was a pawn move to the >>sixth or seventh rank, and that pawn is not blocked by a pawn of >>the opposite color. >> >>The results are pretty disappointing. It really blew up my search. >>I decided to modify the implementation by saying only looking for >>pawn pushes to the seventh rank, and it still can blow up pretty >>bad. >> >>Obviously most over the board positions would not trigger this >>extension, but there are some that would blow up the search like >>mad. >> >>One thought I had is to use a fractional extension scheme. That >>may help, but I'd like to know if anybody else does this extension, >>and what mechanisms they have in place to keep the search bloat >>under control. >> >>It's interesting to note that Bruce lists five different extensions >>on his (excellent) web page; he hasn't actually written about any of >>them yet, but passed pawn isn't even listed. Hmm... >> >>Any ideas or comments? >> >>-- >>James > >Yeah, I found that extending both 6th and 7th rank was too costly in terms of >nodes. But extending just the seventh was ok, and the promo. Are you doing >passed pawns or all pawns? Right it was too costly for DIEP too, i have put the extensions between: /**/ Instead i tried to improve evaluation of passers, with good results by the way. just positions like these rxb2 like in bs2830 are hard. some programs say simply 2 pawns on 3d rank are stronger than a rook (older crafties used to do it not sure about current). Diep only solves this one if it sees a queen which is like 18 ply normal or something here with R=3. In this rxb2 position diep is lucky with singular extensions though: 00:01 0 0 169309 (21804) 8 -0.067 c4-c3 b2xc3 Rd2xa2 Na4xb6 a7xb6 Rb7xb6 Ra2-a3 Rb6-c6 Ra3xc3 Rc6-c7 00:01 0 0 233352 (26964) 9 -0.067 c4-c3 b2xc3 Rd2xa2 Na4xb6 a7xb6 Rb7xb6 Ra2-a3 Rb6-c6 Ra3xc3 Rc6-c7 00:04 0 0 550688 (42782) 10 -0.058 c4-c3 b2xc3 Rd2-d1 Kg1-f2 Rd1-d2 Kf2-g3 Rd2xa 2 Na4xb6 a7xb6 Rb7xb6 Ra2-a3 Kg3-g4 Ra3xc3 Rb6-b8 00:08 0 0 1336381 (126743) 11 -0.030 c4-c3 b2xc3 Rd2xa2 Na4xb6 Ra2-a1 Kg1-f2 Ra1 -a2 Kf2-e3 a7xb6 g2-g4 Ra2xh2 Rb7xb6 Rh2-c2 Ke3-d3 Rc2-g2 Rb6-c6 00:13 0 0 2417480 (380214) 12 -0.030 c4-c3 b2xc3 Rd2xa2 Na4xb6 Ra2-a1 Kg1-f2 Ra1 -a2 Kf2-e3 a7xb6 g2-g4 Ra2xh2 Rb7xb6 Rh2-c2 Ke3-d3 Rc2-g2 Rb6-c6 00:37 0 0 6819763 (1351459) 13 -0.016 c4-c3 b2xc3 Rd2-d1 Kg1-f2 Rd1-d2 Kf2-f1 Rd 2xa2 Na4xb6 a7xb6 Rb7xb6 Ra2-a3 Kf1-f2 Ra3xc3 Rb6-e6 ++ b6-d8 02:03 0 0 21913298 (5306660) 13 0.001 Bb6-d8 h2-h3 Bd8-f6 Rb7-b5 Bf6-d4 Kg1-h2 R d2-d3 g2-g4 Bd4-e5 Kh2-g2 Rd3-g3 Kg2-f1 Rg3xh3 Rb5xc5 Rh3-h1 Kf1-e2 ++ d2-b2 03:30 0 0 36316364 (10392082) 13 2.881 Rd2xb2 Na4-c3 Rb2-c2 Nc3-e4 c4-c3 Ne4-f6 Kh7-h6 Nf6-d5 c5-c4 Nd5xb6 Rc2xa2 Rb7-c7 a7xb6 Rc7xc4 c3-c2 Solving tricks by overreacting in evaluation is very easy. Solving tricks by turning on a few extensions like passed pawn extensions is also very tempting, yet i do not believe it is good for a programs branching factor and/or normal game play. >Will
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.