Author: Robert Hyatt
Date: 13:23:55 05/07/01
Go up one level in this thread
On May 07, 2001 at 15:10:13, Uri Blass wrote: >On May 07, 2001 at 14:29:05, Robert Hyatt wrote: > >>On May 07, 2001 at 12:40:22, Uri Blass wrote: >> >>> >>>For me difference of more than +2 in the evaluation of Deeper blue and Crafty is >>>convincing. >>> >>>I am not talking about the difference of 1.7 and 1.4. >>>There is even a difference of more than 1 pawn between Deep Fritz's evaluation >>>and Crafty's evaluation but Fritz can find the right pv. >> >>Just listen to what you just said. There is a difference of more than one pawn >>but DF has the right PV. > >Difference of more than 1 pawn is not difference of more than 2 pawns and the >difference between Deeper blue and crafty was of more than 2 pawns. Your point would be? I have seen Crafty play Gambit Tiger 1.0 and have a difference of _three_ pawns between their scores. > >Difference of more than 1 pawn can be a result of different evaluation but >difference of more than 2 pawns is usually result of tactics. Nope. Just try Gambit Tiger 1.0 and Crafty on some similar and wide-open positions. 2 pawns will barely get you started. You will see 3 pawn differences at times... Purely positional, too, I might add... > > > With DB you can _not_ see the PV. The first moves >>are probably reasonable. The moves at the end might or might not be reasonable, >>and the last 6 plies + extensions + captures are not seen _ever_. > >I mean to the second move in the pv that is not at the end of the pv and is >almost the start of the pv when I say that deep blue has not the right pv >Rf5+ Ke2... is the wrong pv of deeper blue. Maybe. Maybe Ke2 and Ke3 are equivalent there. Remember the fail low of Crafty was less than half a pawn, which means that the difference between the two moves is not huge and black is already lost... > >Rf5+ Ke3... is the right pv after Rf5+. >> >>Which simply shows that you can _not_ compare DB to _anything_ without all the >>information... >> >> >> >> >>> >>>I cannot say 1.7 of Crafty is better than 1.4 of Deeper blue but I can say that >>>4.x of Crafty and 3.x of other programs are better than 1.x of deeper blue and >>>the evaluation of deeper blue was 1.x if you look at depth 15(the depth that >>>Crafty found the line with Ke3) >> >>Count the pieces on the board, then look at king safety, and everything else. >>Then you can _guess_ at what DB was evaluating compared to Crafty and the other >>micro programs. But you can only _guess_. You are assuming they are seeing >>_less_. What if they are seeing _more_??? How can you tell they aren't? You >>can't. It is _all_ a guess... >> >> >> >> >>> >>>I am almost sure that Crafty can beat Deeper blue if Crafty searches the same >>>depthes that deeper blue searched and my guess is that Crafty on an alpha is >>>clearly better than deeper blue. >>> >>>Uri >> >> >>Not a chance. It would be competitive, but only competitive, in that it would >>not get run off the board in a match. It would lose decisively however... >> >>I ran a 10 game set vs Cray Blitz about a year ago (right after I first got the >>quad xeon). I did it as a favor for someone wanting to test a T90 for >>comparison to something I can't mention. I played the games at a time control >>of 60 moves in 30 minutes. Cray Blitz won 6 and drew 3. There were only a >>couple of games that were tactical busts... most were just endgame losses where >>the Cray searched incredibly deep with its speed advantage... >> >>I call that a pretty decisive result. And that same program got stomped by the >>predecessor of Deep Blue. > >There was only one game when Cray blitz with no bugs lost and one game proves >nothing. > >Uri We lost to them at least three times. The first loss was not necessarily a bug, just a non-reproducible move. I also had lots of private communications with them as they were busy solving difficult tactical positions as well, and they were _always_ far faster/better than we were, and we weren't slouches...
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.