Author: Bas Hamstra
Date: 07:23:44 11/06/03
Go up one level in this thread
On November 06, 2003 at 06:08:09, José Carlos wrote: >On November 06, 2003 at 05:52:49, martin fierz wrote: > >>On November 06, 2003 at 04:51:38, José Carlos wrote: >> >>> This position is easy, but interesting: >>> >>>[D]r2qk2r/pp3ppp/2pbpn2/4Nb2/PnBP2P1/2N1P3/1P3P1P/R1BQ1RK1 b kq g3 0 1 >>> >>> Black plays Ng4 with big advantage. >>> Anubis (complex qsearch) finds it in iteration 8, while Averno (simple >>>qsearch) needs 12. Studying what the program was searching in qsearch in this >>>position has been very interesting for me. >>> How does your program here? >>> >>> José C. >> >>I pick it up on ply 7 (but probably later than you because my program is >>dog-slow!). but my program has lots of extensions, specially when it's something >>king-related. i also generate checks in qsearch if that's what you mean with a >>complex qsearch.... >> >>cheers >> martin >> >>Muse 0.841 on P4 2.4GHz >> >>FEN: r2qk2r/pp3ppp/2pbpn2/4Nb2/PnBP2P1/2N1P3/1P3P1P/R1BQ1RK1 b kq - 0 0 >>depth eval time nodes kN/s QS BR NR pv >> 1/14 33 0.20s 123 2.1 72 96 0 d6e5 . >> 2/16 39 0.44s 954 8.6 75 96 70 b4c2 a1a2 . >> 3/16 39 0.64s 2020 8.6 64 96 75 b4c2 a1a2 d6e5 . >> 4/19 41 0.95s 11071 27.3 58 95 88 f5c2 d1f3 e8g8 f3h3 . >> 5/21 44 1.34s 29075 42.8 49 94 91 f5c2 d1e2 e8g8 g4g5 d6e5 . >> 6/29 46 2.39s 90849 84.5 55 95 92 f5c2 d1e2 e8g8 g4g5 f6d7 e5f3 . >> 7/29 57 6.02s 367188 124.8 51 94 91 f6g4 e5g4 d8h4 c3b5 c6b5 c4b5 >>e8e7 g4e5 . >> 8/30 62 14.05s 1034613 137.5 46 94 93 f6g4 e5g4 d8h4 h2h3 h4h3 g4e5 >>e8g8 a1a3 . >> 9/30 99 37.45s 3425111 148.3 38 93 93 f6g4 e5g4 d8h4 h2h3 h7h5 g4e5 >>f5h3 e3e4 h3f1 . >>10/30 119 103.98s 10498260 154.4 35 94 93 f6g4 e5g4 d8h4 h2h3 h7h5 c4e2 >>e8c8 g4h2 f5h3 f1e1 . > > Mine's not specially fast either. This is on a PIII 933 > Anubis 0.00.0034 > >6 -0.14 0.76 52534 f5c2 d1f3 e8g8 e3e4 c6c5 c1e3 c5d4 e3d4 >7 -0.14 1.67 136973 f5c2 d1f3 e8g8 e3e4 c6c5 c1e3 c5d4 e3d4 >8 0.03 5.49 497234 f5c2 d1f3 e8g8 f3h3 d6e5 d4e5 f6d7 h3h5 > c2g6 >8 0.34 9.21 877142 f6g4 e5g4 d8h4 h2h3 h7h5 g4e5 d6e5 d4e5 > h4c4 g1h2 >9 0.34 14.73 1423051 f6g4 e5g4 d8h4 h2h3 h7h5 g4e5 d6e5 d4e5 > h4c4 g1h2 >10 0.52 46.98 4530358 f6g4 e3e4 d6e5 d4e5 d8d1 f1d1 g4e5 e4f5 > e5c4 f5e6 f7e6 b2b3 c4a5 a1b1 > > Checks in qsearch is a wide concept, I mean, how many plies do you try checks, >what creteria you follow to generate this or that check, when do you try all >moves to escape the check or only captures, how do you handle mate threats, when >do you accept stand-pat and when not... I have a complex logic in Anubis, with a >lot of decisions for each node, while in Averno I do captures and promotions and >stand-pat with eval >= beta. > > José C. I have done quite some experimenting, IMO you can easily overdo it and make the qsearch too heavy. I do checks only the first 2 ply. When I try to do more (like "extending" for 1 rep checks), it does do worse on suites. Bas.
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.