Author: Uri Blass
Date: 11:09:48 12/07/02
Go up one level in this thread
On December 07, 2002 at 11:36:03, Ingo Lindam wrote: >On December 07, 2002 at 11:25:03, Uri Blass wrote: > >>On December 07, 2002 at 05:04:33, Ingo Lindam wrote: >> >>>Hello all, >>> >>>concerning the current discussion about testing positional skills >>>I like to add the following: >>> >>>I think to evaluate/test positional skills of an chess engine you don't need to >>>have positions without tactics (what I think is impossible, because for >>>everything true in a chess position there is a tactical proof if you are just >>>able to look deep enough into the tree). >>> >>>We agree that also in a nearly 'tactical solution free' (within a certain scope) >>>position with a one and only best move it could happen that the engine plays the >>>right move for the wrong reason. >>> >>>So why not test the other direction? Assume that you have a true evaluation for >>>a position. This true evaluation doesn't need to be exact it can also be: "White >>>is atleast equal (or better)." So as long as the engine claims a clear advantage >>>for black, it fails the test. If the score corresponds with your own evaluation >>>you are ofcourse not able to judge about the engine as long as you not checked >>>the reason, why it gives this evaluation. Ofcourse your test positions prefers >>>the engine to give the wrong evaluation instantly because of e.g. being a pawn >>>up. The deeper the engine goes into the search tree keeps to give the wrong >>>evaluation the worse is the positional evaluation of the maschine (at least at >>>this special kind of positions). And because you know at least one reason, why >>>the position shouldn't be judge that worse as the engine does, you should also >>>be able to name some/a positional feature/s the engine doesn't evaluate right. >>> >>>Giving an example: >>> >>>[D] r3kb1r/1p1q1ppp/p1n1p3/3pN1B1/3Pb3/1Q2P3/PP2BPPP/2R1K2R b Kkq - 0 1 >>> >>>position after 12.Se5 >>> >>>Analysis by Fritz 7: >>>(sorry for the german letters: S=N,L=B,D=Q,T=R and K=K) >>> >>>12...Sxe5 13.f3 Sg6 14.fxe4 h6 15.Lf4 Sxf4 16.exf4 dxe4 >>>³ (-0.69) Tiefe: 8/21 00:00:00 151kN >>>12...Sxe5 13.dxe5 Lxg2 14.Tg1 Le4 15.f3 h6 16.Lh4 Lf5 17.Kf1 >>>³ (-0.69) Tiefe: 9/24 00:00:00 400kN >>>12...Sxe5 13.f3 Sg6 14.fxe4 h6 15.Lf4 Sxf4 16.exf4 dxe4 17.De3 Lb4+ >>>µ (-0.78) Tiefe: 10/27 00:00:01 935kN >>>12...Sxe5 13.dxe5 Lxg2 14.Tg1 Le4 15.f3 h6 16.Lf4 Lf5 17.Kf1 Tc8 18.Kg2 >>>µ (-0.75) Tiefe: 11/29 00:00:04 2493kN >>>12...Sxe5 13.dxe5 Lxg2 14.Tg1 Le4 15.f3 h6 16.Lf4 Lg6 17.Kf1 Le7 18.Ld3 Lxd3+ >>>19.Dxd3 >>>µ (-0.78) Tiefe: 12/31 00:00:10 5134kN >>>12...Sxe5 13.dxe5 Lxg2 14.Tg1 Le4 15.f3 h6 16.Lf4 Lf5 17.Kf1 Tc8 18.Txc8+ Dxc8 >>>µ (-0.84) Tiefe: 13/36 00:00:28 15940kN >>>12...Sxe5 13.dxe5 Lxg2 14.Tg1 Le4 15.f3 h6 16.Lf4 Lf5 17.Kf1 Tc8 18.Txc8+ Dxc8 >>>µ (-0.78) Tiefe: 14/38 00:01:06 39095kN >>>12...Sxe5 13.dxe5 Lxg2 14.Tg1 Le4 15.f3 Lf5 16.Kf2 b5 17.Lf4 Tc8 18.Txc8+ Dxc8 >>>µ (-0.84) Tiefe: 15/42 00:02:47 102460kN >>>12...Sxe5 13.dxe5 Lxg2 14.Tg1 Le4 15.Db6 Tc8 16.Txc8+ Dxc8 17.Kd2 d4 18.Dxd4 >>>Dc2+ >>>µ (-0.72) Tiefe: 16/46 00:08:08 293734kN >>> >>>Still at ply 16 the machine evaluates (in my opinion wrongly) an advantage for >>>black, although the PV contains th ebest moves for white in the beginning. >>>(going one move back in the position you will see that the engine will not play >>>Ne5 within 17 plys). >>> >>>Giving the engine some more (white) moves towards 16. Kd2 the evaluation will >>>change into the right direction. >>> >>>in the game it follows: >>> >>>12... Nxe5 13. dxe5 Bxg2 14. Rg1 >>>Be4 15. Qb6 Rc8 >> >> >>The question is if 12.Ne5 was the only best move. >>I doubt it >> >>Movei prefers 15...h6 >> >>Is there a win for white after 15...h6? >> >>A possible line is 16.Rc7 Qa4 17.a3 hxg5 18.Qxb7 Bb4+ 19.Kf1 0-0 20.Qxb4 Qxb4 >>21.axb4 and the endgame seems to be equal. >> >>Uri > >My question was not whether Ne5 is doubtless the one and only best move. I am >sure...that black is not better here, after white plays Ne5 and nevertheless the >engine (Fritz 8) gives a clear advantage to black. And (if I am right) that >should be a sign that the positional evaluation of the engine is not optimal is >this position here, shouldn't it? I am not so sure about it. The moves of Fritz in the pv may not be the best moves for black(I suspect that 13...Bxg2 is a mistake). I guess that the fact that 15.Qb6 can force a draw is a tactical observation that Fritz is not able to see. I checked with movei and here is analysis 2 plies later Movei is unable to see the problem at depth 12 and wants to take g2. The problem is lack of important extensions Bxg2 should trigger an extension because Rg1 is forced but movei does not know it. Movei can see at depth 13 that Bxg2 Rg1 Be4 is not good and the pv is changed to Bxg2 Rg1 h6 You can see by the score that it is tactics(the score dropped from 0.89 to 0.13). Movei fails high on h6 later depth=12 +0.89 e4g2 h1g1 g2e4 f2f3 e4f5 b3b6 a8c8 e1d2 c8c1 g1c1 h7h6 c1c7 f8b4 d2c1 Nodes: 17303621 NPS: 161068 Time: 00:01:47.43 depth=13 +0.13 e4g2 h1g1 h7h6 g5h6 h8h6 g1g2 a8c8 c1d1 d7c7 b3a4 e8d8 a4f4 h6h3 f4g5 d8e8 e1f1 Nodes: 47304313 NPS: 159569 Time: 00:04:56.45 depth=13 +0.14 h7h6 Nodes: 69727812 NPS: 160422 Time: 00:07:14.65 depth=13 +0.43 h7h6 Nodes: 77614594 NPS: 159585 Time: 00:08:06.35 depth=13 +0.58 h7h6 f2f3 e4f3 e2f3 h6g5 e3e4 d5d4 f3e2 f8e7 b3b6 d4d3 c1d1 e7d8 b6b4 Nodes: 95817407 NPS: 158874 Time: 00:10:03.10 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.