Author: Will Singleton
Date: 17:26:20 09/21/02
Go up one level in this thread
On September 21, 2002 at 16:26:59, Colin Frayn wrote: >I've been messing around with Beowulf recently and I noticed that doing reduced >depth NULL move searches where the depth is cut to less than one ply, and >therefore the search continues directly into the Qsearch, tends to cause search >problems. That is, that the search suddenly avoids the best move, or can't find >it in the same number of ply, or else gets it and then fails (very) low on it. > >Without NULL moves going directly into the Qsearch, Beo doesn't make any of >these mistakes on the positions that I've tested, and gets the correct move >easily. Example positions are WAC 019, 116, 120. These are all easy positions, >but sometimes when I've fiddled with the eval I find that suddenly they aren't >solved any more. > >I'm wondering if there's any known tactical problems with this technique. I >can't just switch it off because then I get a performance decrease (and a >substantial one, too). Is this something to do with the fact that the Qsearch >only returns an approximate score (because of the stand-pat algorithm)? Is >there any way to make this safer? Does anyone else experience this? > >Cheers, >Col Hi Colin, I also use null to go direct into the qsearch. Like you, if I restrict that, the nodes go way up. However, I haven't found any negative side-effects from doing so (that I know of). I don't know if others restrict null to reduction from the frontier or not. Here's my data: bm c6; id "WAC.019"; 1 -102 0.01 51 3060 Bg5 2 -149 0.01 747 44820 Be3 Nf6 3 93 0.01 2253 135180 c6 Qd5 Pxd7 Bxd7 Qxd5 Pxd5 4 93 0.01 5650 339000 c6 Qd5 Pxd7 Bxd7 Qxd5 Pxd5 5 93 0.03 17279 345540 c6 Qd5 Pxd7 Bxd7 Qxd5 Pxd5 6 107 0.07 40405 346320 c6 Qxb5 Nxb5 Bb6 Pxd7 Bxd7 Nd6 7 94 0.14 87144 373440 c6 Qxb5 Nxb5 Bb6 Pxd7 Bxd7 Nd6 Bc6 8 107 0.28 189274 405540 c6 Qxb5 Nxb5 Bb6 Pxd7 Bxd7 Nd6 Bc6 Ra3 9 99 2.01 804782 399060 c6 Qxb5 Nxb5 Bb6 Pxd7 Bxd7 Nd6 Bc6 Ra3 Bd5 10 133 3.09 1314579 417300 c6 Qxb5 Nxb5 Bb6 Pxd7 Bxd7 Nd6 Bc6 Ra3 Rb8 Bb2 11 121 4.54 2108484 430260 c6 Qxb5 Nxb5 Bb6 Pxd7 Bxd7 Nd6 Bc6 Ra3 Bc7 Bf4 Be4 12 121 5.01 2150001 428520 c6 Qxb5 Nxb5 Bb6 Pxd7 Bxd7 Nd6 Bc6 Ra3 Bc7 Bf4 Be4 bm Rd2; id "WAC.116"; 1 79 0.01 192 11520 Nh3 2 82 0.01 767 46020 Nh3 Rg3 3 91 0.01 2335 140100 Nh3 Rg3 c5 Pxb5 Pxb5 Rxa8 Rxa8 Qxb5 Nxf2 Kg1 4 65 0.07 27857 238740 Qf4 Pxb5 Nxf3 Rg3 5 196 0.16 70141 262980 Rd2 Qe3 Rxc2 Qxg5 Qxg5 Rxg5 Rxb2 Pxb5 Rxb5 6 235 0.23 105422 274980 Rd2 Qxd2 Nxf3 Rxg7 Kxg7 Rg1 Nxg1 Kxg1 Pxa4 7 348 0.53 268210 303600 Rd2 Qe3 Rxf2 Rg2 Rxg2 Kxg2 Qh3 Kf2 8 349 1.28 465866 317580 Rd2 Qe3 Rxf2 Rg2 Rxg2 Kxg2 Qh3 Kh1 9 313 2.39 887394 334860 Rd2 Qe3 Rxf2 Rg2 Rxg2 Kxg2 Qh3 Kh1 Nxf3 Rd1 10 313 5.05 1675001 329460 Rd2 Qe3 Rxf2 Rg2 Rxg2 Kxg2 Qh3 Kh1 Nxf3 Rd1 bm g6 Rhg1; id "WAC.120"; 1 270 0.01 84 5040 Bxg7 Qxg7 Qxd6 2 270 0.01 318 19080 Bxg7 Qxg7 Qxd6 3 235 0.01 3211 192660 Bxg7 Qxg7 Qxd6 Re8 4 96 0.07 25597 219360 Re1 Re8 Bxg7 Rxe1 Rxe1 Qxg7 5 96 0.17 71445 252120 Re1 Re8 Bxg7 Rxe1 Rxe1 Qxg7 6 355 1.17 380153 296220 g6 Qxg6 Bxg7 Re8 Rg1 Qf7 7 298 1.26 429942 299940 g6 Qxg6 Bxg7 b3 Bxf8 Pxa2 Kd2 Rxf8 Qxg6 Pxg6 8 373 1.55 612716 319620 g6 Qxg6 Bxg7 b3 Pxb3 Re8 Rg1 Qf7 9 399 3.03 1010870 331380 g6 Qxg6 Bxg7 Qxh6 Bxh6 Rf7 Rg1 Kh8 Ng5 Re7 10 332 5.01 1781257 355020 g6 Qxg6 Bxg7 Qxh6 Bxh6 Rf7 Rg1 Kh8 Re1 c3 Pxc3 Pxc3 found= 3 missed= 0 seconds= 5 Avg n/s= 370860 Avg depth= 11.0 depth ratio= 3.1
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.