Author: Enrique Irazoqui
Date: 06:31:37 10/18/00
Go up one level in this thread
On October 18, 2000 at 08:46:53, Uri Blass wrote: >On October 18, 2000 at 08:09:28, Enrique Irazoqui wrote: > >>On October 18, 2000 at 03:54:08, Uri Blass wrote: >> >>>On October 17, 2000 at 20:53:45, Robert Hyatt wrote: >>> >>>>On October 17, 2000 at 14:25:32, Chessfun wrote: >>>> >>>>>On October 17, 2000 at 14:23:38, Chessfun wrote: >>>>> >>>>>>On October 17, 2000 at 14:22:19, Chessfun wrote: >>>>>> >>>>>>>On October 17, 2000 at 14:19:56, Enrique Irazoqui wrote: >>>>>>> >>>>>>>>On October 17, 2000 at 14:17:46, Chessfun wrote: >>>>>>>> >>>>>>>>> >>>>>>>>>Rd7+ Re7 Qb3+ Kf8 Rd6 Nxe5 Qc3 b4 >>>>>>>>>depth 11 +3.18 >>>>>>>> >>>>>>>>WHAT??? No way (I hope). >>>>>>>> >>>>>>>>Kramnik,V - Kasparov,G >>>>>>>>8/5k1p/p1nRrp1P/PpP2qp1/4p3/4B3/1P3PP1/3Q2K1 w - - 0 1 >>>>>>>> >>>>>>>>Analysis by Deep Fritz T28: >>>>>>>> >>>>>>>>43.Rd7+ Kg6 44.Rg7+ Kxh6 45.Qd7 Re5 46.Rxh7+ Qxh7 47.Qxc6 Kg6 >>>>>>>> ² (0.44) Depth: 11/28 00:00:50 6786kN >>>>>>>> ² (0.50) Depth: 12/32 00:01:55 15933kN >>>>>>>> >>>>>>>>(Irazoqui, Cadaqués 17.10.2000) >>>>>>> >>>>>>>Still showing >>>>>>>Rd7+ Re7 Qb3+ kf8 Rd6 Nxa5 Qc3 b4 Qxf6+ etc >>>>>>> >>>>>>>Depth 13. +2.84 >>>>>>> >>>>>>>I'll check after Rd7+ Kg6 >>>>>>> >>>>>>>Sarah. >>>>>> >>>>>>Added score. >>>>> >>>>>if Kg6 >>>>>Rg7+ Kxh6 Rc7 Qe5 QD7 Ne7 g4 Kg7 Bd4 Qxe6 Qxc7 Bxf6+ Kf8 >>>>>+3.88 Depth 12 >>>>> >>>>>Sarah. >>>> >>>> >>>>That eval is too big. IE this is another example. I don't recall this >>>>exact position but white is either up 1 or 2 pawns (2 I think) but with >>>>lots of holes and a queen and knight to deal with. >>>> >>>>Most programs had this as +1 to +1.3 in this stage of the game. I became >>>>less and less optimistic for white as I watched, as the queen is simply a real >>>>pain, and the knight is the optimal piece to have working with the queen in a >>>>board with pawns moved everywhere. >>>> >>>>I wasn't surprised by the outcome, particularly. If programs say +2 or less, >>>>draws don't cause any speculation or bug-hunting... >>>> >>>>I think the krnp vs kr might have been easier to win than this game was. Of >>>>course, taking time to eat the a-pawn in this game might have cost white more >>>>than he thought... >>> >>>I suspect that the score of Krnp vs Kr of chess tiger is bigger so the fact that >>>it is more easy to win does not say that the evaluation is wrong. >>> >>>Chesstiger did not like kramnik's moves so the fact that the game was drawn does >>>not prove that the evaluation was wrong. >>> >>>The question is if chesstiger can win other programs. >>> >>>Uri >> >>In this particular case it seems that Gambit's evaluation was a mirage. I made >>it play after Gambit's choice of 47.Rc7, giving Gambit, white, and Deep Fritz >>beta, black, about 5 to 10 minutes/move each on 2 P600E, DF with 184MB hash, >>Gambit with 192MB hash: >> >>Gambit 47.Rc7, +3.62 >>DF 47... Ne7, +0.94 >>Gambit 48.Qd8, +4.38 >>DF 48... Nd5, +0.66 >>Gambit 49.Rb7 (why?), +4.28 >>DF 49... Nxe3, 0.00 >>Gambit 50.Qf8+, +1.83 (failing low) >>DF 50... Kh5, +0.19 >>Gambit 51.fxe3, +1.10 (lower) >>DF 51... Re5, +0.19 >> >>As seen by Gambit itself, the initial evaluations of this line were wrong. >> >>But I don't think that this is the right way to look at Gambit's evals, which in >>these cases are speculative. It is not so much a matter of being accurate, of >>seeing the whole line, but of making Gambit go for the attack, and more often >>than not it succeeds. > >I am not against gambittiger but I believe that it is the right way to look at >gambit's evaluations. To find out about how Gambit works, yes. But I don't think it makes sense to pretend that speculative evals are exact, or else they wouldn't be speculative. That's what I meant. >Christophe explained that he did not finish to tune gambit's evaluation and I >believe that he is going to have better evaluation in the future. >I guess that in this case slower time control can also help tiger to avoid >49.Rb7 because the score dropped one move after it. I left Gambit computing this move for almost an hour, and Rb7 was still the choice. >I see from the fact that tiger used 5-10 minutes per move that tournament time >control is not slow ennough. As I told you the other day, so far there is no evidence showing that Gmabit does better at slow time controls. I tried this playing matches Gambit-Junior 6a at game/10, 40/20 and 40/60 with similar results. Enrique >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.