Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: adjusting bounds

Author: Dieter Buerssner

Date: 12:43:26 09/03/01

Go up one level in this thread


On September 02, 2001 at 18:22:26, Georg v. Zimmermann wrote:

>The NullMove thing might be not too much of a problem. Shouldn't it happen very
>very seldom ?

I agree, that it should happen only rarely. Otherwise, we couldn't use null
moves efficiently. But it can (an will) change results. And adjusting the window
bounds from the hash complicates this again a bit. That was actually all, I
wanted to point out.

Some numbers from 22 postions of WAC searched for 2 million nodes. These were
run with R=2, disabling any double null move tricks, and also disabling null
move cutoffs. Just "probe" the null move and remember, if you would have gotten
a cutoff. Then compare to the result of a normal search. Remember the cases,
when the null move cutoff would have been wrong. Because I normally use fail
soft, some more stringent comparision would ideally be needed. But I was too
lazy for that now. I just looked, if both scores were >= beta. Also other
pruning techniques should be taken into account, and how they interfere with
null move. Still, the numbers may get a small idea.

You will see a comparable low nomber of wrong null move fail highs, but still
they can have a significant result. For examply WAC 141 (test.ci 7 down there)
is usually found at ply 8 with null move, without null move it is found at ply 6
with my program.

The 22 positions:

8/7p/5k2/5p2/p1p2P2/Pr1pPK2/1P1R3P/8 b - - bm Rxb2; id "test.ci 1";
2kr3r/pp1q1ppp/5n2/1Nb5/2Pp1B2/7Q/P4PPP/1R3RK1 w - - bm Nxa7+; id "test.ci 2";
8/p3k1p1/4r3/2ppNpp1/PP1P4/2P3KP/5P2/8 b - - bm Rxe5; id "test.ci 3";
r4rk1/1p2ppbp/p2pbnp1/q7/3BPPP1/2N2B2/PPP4P/R2Q1RK1 b - - bm Bxg4; id "test.ci
4";
8/k1b5/P4p2/1Pp2p1p/K1P2P1P/8/3B4/8 w - - bm b6+ Be3; id "test.ci 5";
r2r2k1/pb3ppp/1p1bp3/7q/3n2nP/PP1B2P1/1B1N1P2/RQ2NRK1 b - - bm Qxh4; id "test.ci
6";
4r1k1/p1qr1p2/2pb1Bp1/1p5p/3P1n1R/1B3P2/PP3PK1/2Q4R w - - bm Qxf4; id "test.ci
7";
r1bq4/1p4kp/3p1n2/p4pB1/2pQ4/8/1P4PP/4RRK1 w - - bm Re8; id "test.ci 8";
5rk1/p4ppp/2p1b3/3Nq3/4P1n1/1p1B2QP/1PPr2P1/1K2R2R w - - bm Ne7+; id "test.ci
9";
5rk1/2p4p/2p4r/3P4/4p1b1/1Q2NqPp/PP3P1K/R4R2 b - - bm Qg2+; id "test.ci 10";
rr4k1/p1pq2pp/Q1n1pn2/2bpp3/4P3/2PP1NN1/PP3PPP/R1B1K2R b KQ - bm Nb4; id
"test.ci 11";
r1br2k1/pp2nppp/2n5/1B1q4/Q7/4BN2/PP3PPP/2R2RK1 w - - bm Bxc6 Rfd1; id "test.ci
12";
2rqrn1k/pb4pp/1p2pp2/n2P4/2P3N1/P2B2Q1/1B3PPP/2R1R1K1 w - - bm Bxf6; id "test.ci
13";
2b5/1r6/2kBp1p1/p2pP1P1/2pP4/1pP3K1/1R3P2/8 b - - bm Rb4; id "test.ci 14";
5r2/1p1RRrk1/4Qq1p/1PP3p1/8/4B3/1b3P1P/6K1 w - - bm Qxf7+ Rxf7+; id "test.ci
15";
r5k1/pQp2qpp/8/4pbN1/3P4/6P1/PPr4P/1K1R3R b - - bm Rc1+; id "test.ci 16";
2rq1rk1/pp3ppp/2n2b2/4NR2/3P4/PB5Q/1P4PP/3R2K1 w - - bm Qxh7+; id "test.ci 17";
2r1k2r/2pn1pp1/1p3n1p/p3PP2/4q2B/P1P5/2Q1N1PP/R4RK1 w k - bm exf6; id "test.ci
18";
2kr2nr/pp1n1ppp/2p1p3/q7/1b1P1B2/P1N2Q1P/1PP1BPP1/R3K2R w KQ - bm axb4; id
"test.ci 19";
2r1r1k1/pp1q1ppp/3p1b2/3P4/3Q4/5N2/PP2RPPP/4R1K1 w - - bm Qg4; id "test.ci 20";
3q1rk1/4bp1p/1n2P2Q/3p1p2/6r1/Pp2R2N/1B4PP/7K w - - bm Ng5; id "test.ci 21";
1nbq1r1k/3rbp1p/p1p1pp1Q/1p6/P1pPN3/5NP1/1P2PPBP/R4RK1 w - - bm Nfg5; id
"test.ci 22";

And the corresponding statistics for null moves:

null move: tried   199461, fail highs   130646, wrong      102
null move: tried   137714, fail highs    79226, wrong        4
null move: tried    82124, fail highs    69131, wrong        1
null move: tried    95967, fail highs    54452, wrong        0
null move: tried   273495, fail highs   189388, wrong      245
null move: tried   105239, fail highs   103096, wrong        4
null move: tried   155019, fail highs   101862, wrong       35
null move: tried   140807, fail highs    94361, wrong        4
null move: tried   163565, fail highs   103156, wrong        0
null move: tried   177757, fail highs   116430, wrong       42
null move: tried   112171, fail highs    61919, wrong        5
null move: tried    89585, fail highs    50509, wrong        0
null move: tried   103539, fail highs    63705, wrong        8
null move: tried   296831, fail highs   191852, wrong      326
null move: tried   196593, fail highs   136806, wrong       21
null move: tried   135310, fail highs   117776, wrong       11
null move: tried   129382, fail highs   120070, wrong        8
null move: tried   168999, fail highs    91865, wrong       10
null move: tried   134935, fail highs    86521, wrong        2
null move: tried   228132, fail highs   154049, wrong       18
null move: tried    74631, fail highs    65437, wrong        0
null move: tried    94449, fail highs    84651, wrong        0

Regards,
Dieter



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.