Author: Richard Pijl
Date: 01:36:28 01/10/03
Go up one level in this thread
On January 09, 2003 at 16:51:32, Will Singleton wrote: >>If I force Qxa2 in the Baron, it takes (on a PII-500) about 8 minutes to show a >>good score for white. Before that the score was drawish: >> >> ply time nodes score pv >> >> 13(21). 146 14006096 0 b1a1 a2b3 e1b1 b3c2 b1c1 c2b3 c1b1 <rep> >> 14(23)+ 239 22183609 35 b1a1 a2b3 e1b1 b3c2 b1c1 c2d2 e2d2 f6g4 >> 14(24)& 484 42965262 120 b1a1 a2b3 e1b1 b3c2 b1c1 c2c1 d2c1 f6g4 >> 14(24). 638 55906104 120 b1a1 a2b3 e1b1 b3c2 b1c1 c2c1 d2c1 f6g4 > >Richard, > >You must have a problem here. 8 min is way too long. Perhaps a problem with >repetition and hash. > >[D]r3r1k1/pbpp1pp1/1p3n1p/8/2PPp3/2P1P1P1/q2BQP1P/1R2RBK1 w - - > >Amateur shows +1.7 in 8 secs: > >8 -201 163 584342 c5 d6 cxd6 cxd6 Bg2 Ba6 Qd1 Bd3 >9 -182 425 1508721 c5 Bc6 Ra1 Qe6 cxb6 cxb6 Bg2 d5 Ra6 >9 0 433 1561526 Ra1 Qc2 Rc1 Qb3 Rb1 Qc2 Rc1 >10 0 443 1660070 Ra1 Qc2 Rc1 Qb3 Rb1 Qc2 Rc1 >11 0 479 1853613 Ra1 Qc2 Rc1 Qb3 Rb1 Qc2 Rc1 >12 81 654 2681634 Ra1 Qc2 Rc1 Qb3 Be1 d5 c5 Kf8 >13 177 819 3718292 Ra1 Qc2 Rc1 Qb3 Be1 Nd5 Rb1 Nxc3 Rxb3 Nxe2 Bxe2 Bc6 d5 Bb7 >14 224 1161 5927341 Ra1 Qc2 Rc1 Qb3 Be1 Nd5 Rb1 Nxc3 Rxb3 Nxe2 Bxe2 d5 Ra3 dxc4 >Rxa7 > I'm not sure it's a bug. Repetition detection has been a problem some time ago but I think I corrected that. Also the fact that it finds the draw in 2 seconds supports that. To be sure I will check it again. Hash might be another cause. I'll try and check that as well. What worries me a little is that Amateur seems to need a lot less nodes to go to iteration 13 than the Baron. Amateur needs 3.7M nodes, Baron 14M nodes. As Baron doesn't do futility pruning, and has liberal limits on its extensions that explains only part of it. 0.99.3 (with futility pruning and limited extensions) still needs about 11M nodes to get this deep. A problem could be that Baron is extending the wrong lines in this position. Another one is that it might try to push capture of the closed-in queen over the horizon somehow. To be sure it isn't something I changed lately I also tested the position in Baron 0.99.3 and Baron 0.36. But it's the same story here. It is only found in the 14th iteration (which takes a long time on my slow PII-500 here, just over 5 minutes for 0.99.3). Well, I just have to look at it. I'm focussing on search at the moment so I might find some improvements. At least Baron finds the correct move (Ra1) in only two seconds, although it only expect a draw with it. :-) Richard.
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.