Author: Uri Blass
Date: 03:11:32 08/07/04
Go up one level in this thread
On August 07, 2004 at 03:52:23, Stuart Cracraft wrote: >On August 07, 2004 at 03:47:37, Stuart Cracraft wrote: > >>So here is an 8 ply search of the opening >>position without SEE: >> >>Alpha=-250 Beta=250 Maxdepth=8 MaxTime=9999999 >> 1/ 2 e2e4 0.00 35 41 e2e4 >> 2/ 4 e2e4 0.00 0 149 e2e4 e7e5 >> 3/ 9 e2e4 0.01 17 450 e2e4 e7e5 f1c4 >> 4/14 e2e4 0.02 0 2283 e2e4 e7e5 f1c4 f8c5 >> 5/17 e2e4 0.06 17 7260 e2e4 e7e5 g1f3 b8c6 f1c4 >> 6/23 e2e4 0.40 0 46344 e2e4 e7e5 g1f3 b8c6 f1c4 f8c5 >> 7/26 e2e4 1.08 17 119943 e2e4 e7e5 g1f3 b8c6 f1c4 g8f6 b1c3 >> 8/31 e2e4 6.11 0 699160 e2e4 e7e5 g1f3 b8c6 f1c4 g8f6 b1c3 f8c5 >>e2e4 6.11 0 699160 e2e4 e7e5 g1f3 b8c6 f1c4 g8f6 b1c3 f8c5 >>nps=114484 ha=17.00% q=57.0% bc=71% br=NaN% mp=108<>99 >>pawnx=0 recapx=0 qcheckx=27853 checkx=0 futilx=0 qfutilx=0 >> >>And with SEE: >> >>Alpha=-250 Beta=250 Maxdepth=8 MaxTime=9999999 >> 1/ 2 e2e4 0.00 35 41 e2e4 >> 2/ 4 e2e4 0.00 0 149 e2e4 e7e5 >> 3/ 5 e2e4 0.00 17 439 e2e4 e7e5 f1c4 >> 4/10 e2e4 0.02 0 2093 e2e4 e7e5 f1c4 f8c5 >> 5/13 e2e4 0.05 17 6550 e2e4 e7e5 g1f3 b8c6 f1c4 >> 6/20 e2e4 0.31 0 37096 e2e4 e7e5 g1f3 b8c6 f1c4 f8c5 >> 7/24 e2e4 0.81 17 90371 e2e4 e7e5 g1f3 b8c6 f1c4 g8f6 b1c3 >> 8/34 e2e4 6.48 4 740555 e2e4 g8f6 b1c3 e7e5 g1f3 b8c6 f1c4 f8c5 >>e2e4 6.48 4 740555 e2e4 g8f6 b1c3 e7e5 g1f3 b8c6 f1c4 f8c5 >>nps=114244 ha=35.00% q=50.0% bc=68% br=NaN% mp=105<>106 >>pawnx=0 recapx=0 qcheckx=9862 checkx=0 futilx=0 qfutilx=0 >> >>The result is worse (6.11 seconds vs. 6.48 seconds). This is >>true of a larger test of problem positions. >> >>The use of SEE was in the quiescence search only and >>involved rejecting any search of a capture whose SEE >>value to the to square of the capture was <0, otherwise >>the move would be made, legality tested, and if legal, >>then the recursive capture search performed (with the >>same see test.) >> >>Is this an abnormal result? What could explain it? I thought >>SEE was a shoe-in? >> >>Stuart > >The only thing I can think is that the SEE routine is not >implemented correctly. For example, after e4, d5 I do a SEE >on d5 and it returns -8000 instead of 0. > >So the SEE itself is clearly broken. Your SEE is obviously broken. The right way to implement SEE is first to test it not in the search but on chess positions(like the position after 1.e4 d5) and only if you get correct results to start to test if it helps in the search. You still may have bugs but at least you reduce the chances of bugs(it is possible to continue to have bugs in SEE without doing it for a long time because the advantage in correct cases may be more than the disadvantage in cases that you have the bug) 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.