Author: Richard Pijl
Date: 12:19:54 11/26/02
Go up one level in this thread
On November 26, 2002 at 12:26:08, Richard Pijl wrote: >On November 26, 2002 at 05:22:42, Dann Corbit wrote: > >>On November 26, 2002 at 04:43:16, David Rasmussen wrote: >> >>>[D]r4rk1/1p1n1pp1/1bq1bn1p/p1pp4/2P2B2/1NNP2P1/PPQ2PBP/R4RK1 w - - 0 19 >>> > >When looking closer at the position, the possible exchange at d5 and the other >available captures in this position are probably the cause of the fat qsearch. >Just try to build a tree of all possible sequences. As there are also other >captures in the position, almost all of them will be searched, possibly ending >up in the same end position. > >As Baron doesn't limit qsearch either, I guess the use of transposition tables >are helping Baron in limiting the size of the qsearch. I'll try the position >this evening without the hashtables in qsearch ... I'll see (and post :-) ) what >happens. > >Richard. Ok, I did the test again, with and without hashtables as the version I have here (and the computer (!)) is a bit different. Results WITH hashtables (AMD Athlon MP 2000+) : ply time nodes score pv 5( 6)& 0 38340 31 f1e1 a5a4 c4d5 f6d5 b3d2 5( 6). 0 68609 31 f1e1 a5a4 c4d5 f6d5 b3d2 6( 8)& 0 90763 27 f1e1 a5a4 b3d2 f8d8 a1c1 d8e8 6(10). 0 150114 27 f1e1 a5a4 b3d2 f8d8 a1c1 d8e8 7(11)& 0 224203 28 f1e1 a5a4 b3d2 a4a3 c4d5 f6d5 b2b3 7(11). 1 424721 28 f1e1 a5a4 b3d2 a4a3 c4d5 f6d5 b2b3 8(12)& 1 508745 24 f1e1 a5a4 b3d2 a4a3 c4d5 f6d5 c3d5 e6d5 8(12)& 4 1305968 33 a1e1 a5a4 c4d5 f6d5 b3d2 8(12)& 8 2411396 38 a2a4 a8e8 c2d2 e8a8 a1e1 8(12). 8 2449600 38 a2a4 a8e8 c2d2 e8a8 a1e1 9(14)& 15 4419417 55 a2a4 f8d8 f1e1 d7f8 c4d5 f6d5 g2d5 e6d5 e1e7 9(15). 16 4995786 55 a2a4 f8d8 f1e1 d7f8 c4d5 f6d5 g2d5 e6d5 e1e7 10(21)& 41 11949429 40 a2a4 f8d8 f1e1 d7f8 f4e3 f8g6 e1e2 d8d6 a1e1 10(21). 53 15598179 40 a2a4 f8d8 f1e1 d7f8 f4e3 f8g6 e1e2 d8d6 a1e1 Computer's move: a2a4 Nodes: 15598179 Qnodes: 9088473 Ratio qnodes/totalnodes = 58.3% Same version but now without using hashtables in qsearch: ply time nodes score pv 5( 6)& 0 65982 31 f1e1 a5a4 c4d5 f6d5 b3d2 5( 6). 0 114420 31 f1e1 a5a4 c4d5 f6d5 b3d2 6( 8)& 0 159461 27 f1e1 a5a4 b3d2 f8d8 a1c1 d8e8 6(10). 1 273852 27 f1e1 a5a4 b3d2 f8d8 a1c1 d8e8 7(11)& 1 389923 28 f1e1 a5a4 b3d2 a4a3 c4d5 f6d5 b2b3 7(11). 2 702407 28 f1e1 a5a4 b3d2 a4a3 c4d5 f6d5 b2b3 8(12)& 3 908294 24 f1e1 a5a4 b3d2 a4a3 c4d5 f6d5 c3d5 e6d5 8(12)& 9 2525471 33 a1e1 a5a4 c4d5 f6d5 b3d2 8(12)& 13 3598144 38 a2a4 a8e8 c2d2 e8a8 a1e1 8(12). 13 3673547 38 a2a4 a8e8 c2d2 e8a8 a1e1 9(14)& 21 5730691 45 a2a4 a8c8 f1e1 c8e8 c4d5 f6d5 c2d2 e8a8 a1c1 9(15). 24 6733229 45 a2a4 a8c8 f1e1 c8e8 c4d5 f6d5 c2d2 e8a8 a1c1 10(21)& 66 17532252 39 a2a4 f8d8 a1e1 d7f8 f4e3 f8d7 e1e2 c6d6 e3f4 d6c6 10(21). 82 22030146 39 a2a4 f8d8 a1e1 d7f8 f4e3 f8d7 e1e2 c6d6 e3f4 d6c6 Computer's move: a2a4 Nodes: 22030146 Qnodes: 15678588 Ratio qnodes/totalnodes = 71.2% So the hash tables had significant impact on the ratio qnodes/total nodes, but not nearly the figures in Chezzz On second thought, I might have guessed that... Even doubling the amount of Q-nodes would not have gotten it closer than say 75% ... Richard.
This page took 0.01 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.