Author: Uri Blass
Date: 07:24:17 04/10/02
Go up one level in this thread
On April 10, 2002 at 09:54:16, Andrew Williams wrote: >On April 10, 2002 at 09:14:48, Steve Maughan wrote: > >>Andrew, >> >>>I don't understand what you mean by the "fail-soft move"? >> >>This is the move that gets the highest score when all moves fail low i.e it will >>have a score below alpha but may possibly be the best move. Ed Shroeder did >>store this move in the hash table but from Rebel 10? onward found it better to >>store 'nil'. I wonder if in this situation, the lack of moves stored in the >>hash for positions resulting from - say Qb8+, that freezes the search of many >>engines. >> > >Aha! If I'd stopped to think, I'd have worked out what you meant. I don't store >a move in such situations. > >>>It doesn't take PostModernist (on my Athlon 1200) very long at all to see the >>>problem with Qxh6: >>> >>>10> 198 0 14906 1.Qxh6 b4 2.Qh3 h5 3.Kg3 Nf5 4.Kf4 Kf7 >>>10= 215 0 41916 1.Qxh6 Ne6 2.Kg3 b4 3.Kf2 f5 4.Ke2 a4 5.Qh2 >>>11> 208 0 75603 1.Qxh6 Ne6 2.Kg3 b4 3.Kf2 a4 4.Qh2 f5 5.Qd6 >>>11= 214 0 79964 1.Qxh6 Ne6 2.Kg3 b4 3.Kf2 a4 4.Qh2 f5 5.Qd6 >>>12> 214 0 99721 1.Qxh6 Ne6 2.Kg3 b4 3.Kf2 a4 4.Qh2 f5 5.Qd6 >>>12= 219 0 139014 1.Qxh6 Ne6 2.Kg3 b4 3.Kf2 a4 4.Qh2 f5 5.Qd6 >>>13> -522 2 513766 1.Qb8 Ne8 2.Qxe8 Qf8 3.Qxb5 f5 4.Kf4 Qd6 >>> 5.Kg5 a4 6.Qxa4 Qg3 7.Kf6 Qxe3 >>>13= -419 4 746241 1.Qb8 Ne8 2.Qxe8 Qf8 3.Qxb5 h5 4.Kg3 Qd6 >>> 5.Kh3 Qd8 6.Qb7 g5 >>>14> -439 6 1198282 1.Qb8 Ne8 2.Qxe8 Qf8 3.Qxb5 h5 4.Kg3 Qd8 >>> 5.Qb7 f5 6.Kh3 g5 7.Kg3 h4 8.Kh3 >>>14= -429 6 1264249 1.Qb8 Ne8 2.Qxe8 Qf8 3.Qxb5 h5 4.Kg3 Qd8 >>> 5.Qb7 f5 6.a3 g5 7.a4 h4 8.Kh3 >> >>This is impressive!! I notice that the PV from ply 12 does not include Nh5, >>which is essential to meaningfully reject the position. Does PostModernist >>revert to another move (Qb8) instantly due to the heavy fail low? Does >>PostModernist recognise that the Queen is trapped after Nh5? >> > >My program recreates its PV from the hash table, and it only extracts it at >certain points (first fail-high, new best move and completed iteration). I've >quickly changed my MT driver to show a bit more information about fail-lows: > >11= 214 0 79964 1.Qxh6 Ne6 2.Kg3 b4 3.Kf2 a4 4.Qh2 f5 5.Qd6 >12> 214 0 99721 1.Qxh6 Ne6 2.Kg3 b4 3.Kf2 a4 4.Qh2 f5 5.Qd6 >12< 221 0 138235 1.Qxh6 Nh5 >12< 220 0 139014 1.Qxh6 Nh5 >12= 219 0 139014 1.Qxh6 Nh5 >13< 45 1 177912 1.Qxh6 Nh5 >13> -522 3 513766 1.Qb8 Ne8 2.Qxe8 Qf8 3.Qxb5 f5 4.Kf4 Qd6 > 5.Kg5 a4 6.Qxa4 Qg3 7.Kf6 Qxe3 >13< -417 4 730785 1.Qb8 Ne8 2.Qxe8 Qf8 3.Qxb5 h5 4.Kg3 Qd6 > 5.Kh3 Qd8 6.Qb7 g5 >13< -418 4 746241 1.Qb8 Ne8 2.Qxe8 Qf8 3.Qxb5 h5 4.Kg3 Qd6 > 5.Kh3 Qd8 6.Qb7 g5 >13= -419 4 746241 1.Qb8 Ne8 2.Qxe8 Qf8 3.Qxb5 h5 4.Kg3 Qd6 > 5.Kh3 Qd8 6.Qb7 g5 > >You can see that the last thing it sees about Qxh6 is Nh5. I have some queen >trapping code in PM, but it's been commented out for some time now. So I'm not >sure why PM is fast in this position. I guess the results of those other >programs beg the question, how long does it take to see Nh5 if you show them the >position after Qxh6? PM sees it in well under one second. Here's the position >after Qxh6: > >[D] 6k1/6np/5ppQ/pp1p4/3Pp1K1/4P3/P7/8 b - - 0 1 > > >Cheers > >Andrew The position after Qxh6 is clearly easier. The problem of the stupid chess programs(and it includes also the latest version of my program) is that they waste almost all of their time on moves that are different than Qxh6 so they do not discover that Qxh6 is bad. Another problem after finding that Qxh6 is bad is to find a move that is good enough to have a better score. a special weak version of my program for this position(extending 4 plies for the first move that it searches at the first ply) could find that Qxh6 is bad at depth 8 but finished depth 8 without finding a better move and only at depth 9 could find a better move. It is also a positional problem because the difference between the score of Qb8+ and the score after Qxh6 may be small. Foir my program the score at depth 11 for the position after Qxh6 is +2.83 when the score after Qb8+ at the same depth is slightly better only at depth 12 after Qxh6 is 3.30 and only at depth 13 it can see that Qxh6 is losing tactically(score +6.13) Uri
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.