Author: Dieter Buerssner
Date: 06:02:35 09/27/00
Go up one level in this thread
On September 27, 2000 at 05:53:59, Uri Blass wrote: >On September 27, 2000 at 04:21:52, Dieter Buerssner wrote: > >>On September 26, 2000 at 16:51:33, Uri Blass wrote: >> >>>[D]3r3k/p7/1p6/2p2Q1p/2P1R3/2P5/P2KP3/6q1 w - - 0 1 >>> >>>Hiarcs played Kc2 and had to resign some moves later because the queen cannot be >>>saved[the game continued Kc2 Qd1+ Kb2 Rd2+ Ka3 Qc1+ Ka4 b5+ cxb5 Qc2+ Ka5 Qxc3+ >>>and white resigned because of the line Ka6 Rd6+ Kb7 Qg7+ Kb8(forced because Kc8 >>>is leading to mate) Rd8+]. >>> >>>How much time do programs need to avoid this tactical error?(some program like >>>hiarcs can play Rd4 after a short time but later change their mind to Kc2 so the >>>right test is to give them at least some minutes) >> >>Interesting position. I think, it is not unlikely for many engines to switch >>moves here. It needs some relatively deep search to see that the queen is lost. >>Before this, Yace sees tactical strong lines after Kc2, but in between, at just >>one depth, they seem to be refuted. >> >>Yace needs 64 seconds on AMD K6-2 475 MHz, 30MB hash. >>[...] >> 8756 0.131 -1.43 4t : Re4d4 Rd8xd4+ c3xd4 Qg1xd4+ Kd2e1 Qd4c3+ Ke1f2 >> Qc3xc4 Qf5xh5+ Kh8g8 [-100] >>[stays with Rd4 until] >> 3464157 16.90 -0.66 10t : Re4d4 Rd8xd4+ c3xd4 Qg1xd4+ Kd2e1 Qd4h4+ Ke1f1 >> Qh4xc4 Qf5xh5+ Kh8g8 Qh5e8+ Kg8g7 Qe8e5+ Kg7g6 >> Qe5d6+ Kg6f5 Qd6d7+ Kf5e5 Qd7xa7 Qc4c1+ Kf1f2 >> b6b5 [0] >>[...] >> 9973064 53.77 -0.36 10t : Kd2c2 Qg1d1+ Kc2b2 Rd8d2+ Kb2a3 Qd1c1+ Ka3a4 >> b6b5+ c4xb5 Qc1c2+ Ka4a5 Qc2xc3+ Ka5a6 Rd2d6+ >> Ka6b7 Qc3g7+ Kb7c8 Qg7g8+ Kc8b7 Rd6b6+ Kb7c7H >> Qg8b8+H Kc7d7H Rb6d6+H Kd7e7H <HT> [0] > >I do not understand the score of -0.36 >When I look in the final position of the pv even without searching deeper I >evaluate it as a very big advantage for black. I agree, that this is no good evaluation. But it may not be as bad as it looks. >Reasons: >1)The white king is in a very bad square and I am afraid from mate even before >seeing it by search(it is a problem when the sides have queen and rook that is >often enough to mate). But almost the same is true for the black king. With white to move, this is a mate in 3. But of course ... >2)black is the attacker(it is black to move and black is not in check) >I define the attacker as the side to move unless the side to move is in check. when the eval takes this into account, things are different. I hope, that the search or search extensions would solve this problem. This may not be optimal and I will think about the concept of side to move for this sort of eval terms. Nevertheless, if I calculate extensions correctly, one ply deeper, the mate in this line would be seen. >It seems to me that a better piece square table with big positional scores can >help to have a better evaluation. Don't see this here. You want to punish the advanced king very much? >I do not unsderstand the reason that a lot of programs evaluate the final >position of the pv as almost equal. > >Here is the final position of the pv > >[D]1q5k/p3K3/3r4/1Pp2Q1p/4R3/8/P3P3/8 b - - 0 1 > >evaluation of programs at depth 1: >CometB20 1.61: pawns for black This really is interesting. Perhaps Ulrich Türke can tell, if he looks at the side to move for king safty. >This kind of evaluations(except Comet) make me wonder if programmers think >before writing the piece square tables because I expect more than one pawn bonus >for king safety problems in this case and if you do not consider king safety >then I also see positional advantage for black because black has 2 passed pawns >when white has only one. How much is one passed pawn (not connected) worth in an open position with queens on board, where there are allways threats for perpetual checks? >Are programmers afraid to have big bonuses of more than 1 pawn in the piece >square tables? Where would you put bonuses of more than one pawn in piece square tables (besides perhaps pawns on rank 7)?
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.