Actually this game also exposes the weaknesses of engines quite a bit. Stockfish and other engines thought the position after Nd2 was just losing for white.
And this Qa4 position is another case in point, what if black just plays 42.. h5 after Qa4 trading the g-pawn. I tried with engines and yes you will see a +0.5 which keeps going down, but the engines are just evaluating the position, there is no win of course.. Shipov of course has caught on the engine disease. I might be proven wrong but positions like occurred in the game are confusing for both humans and engines. For the same reason, I do not trust the +0.44 after Qh3 either. Just shows how complex a game chess is.
For several other moves in the game, I turned on the engine, but found unreliable evaluations.
I do not see +0.5 after Qa4 h5 gxh5+ Kxh5 Qa7
Of course score of more than +3 is no proof that white wins and I did not analyze enough to be sure that Qa4 wins but you can be sure that nobody is going to claim that white can win based on +0.5
I can add that I did not see evaluation by engines that suggested that white is losing after Nd2.
I am using Rybka 3 and after Qa4 h5 gh5+ Kh5 Qa7, I see a +0.5 score. I did not let it run overnight. However, for a human such as much, I do not see a clear way to win as white has to bank on the c3 pawn and winning at least one pawn.
Here is an analysis of Stockfish 1.7.1 after 42.Qa4 h5 43.gxh5+ Kxh5 44.Qa7:
[d]
The analysis is not perfect since 62.Qf4+? (two plies before the line ends) is not optimal and the final position of that analysis is draw based on perpetual check, but white can simply play 62.Qxe2 instead and win, so the overall result would be that your variation does also lose for black provided that the remaining part of the analysis is correct. It is difficult, though, I admit, and your suggested move h5 does not look too bad.
JuLieN wrote:Yes, I remember that SF gave 42. Qa4 a mark above +2 for white. White could have won this game. The thing is that such moves are nearly unreachable for humans, as their consequences are not obvious at all.
Could this +2 score merely be the result of a partial search and evaluation? In other words, the engine finds a strong line and shows that score, but as it evaluates more lines at that same depth, it finds that that line is not so strong?
I've seen this happen before when letting the engine search for a long time. Until it's actually searched the entire tree, you don't have an accurate evaluation and so you can only look at the score at the end of each depth search, and not in the middle. (I hope somebody understands what I'm saying here).
JuLieN wrote:Yes, I remember that SF gave 42. Qa4 a mark above +2 for white. White could have won this game. The thing is that such moves are nearly unreachable for humans, as their consequences are not obvious at all.
Could this +2 score merely be the result of a partial search and evaluation? In other words, the engine finds a strong line and shows that score, but as it evaluates more lines at that same depth, it finds that that line is not so strong?
I've seen this happen before when letting the engine search for a long time. Until it's actually searched the entire tree, you don't have an accurate evaluation and so you can only look at the score at the end of each depth search, and not in the middle. (I hope somebody understands what I'm saying here).
Yeah, I do. It failed high that's all. It was a good move but winning I doubt it.
Actually this game also exposes the weaknesses of engines quite a bit. Stockfish and other engines thought the position after Nd2 was just losing for white.
And this Qa4 position is another case in point, what if black just plays 42.. h5 after Qa4 trading the g-pawn. I tried with engines and yes you will see a +0.5 which keeps going down, but the engines are just evaluating the position, there is no win of course.. Shipov of course has caught on the engine disease. I might be proven wrong but positions like occurred in the game are confusing for both humans and engines. For the same reason, I do not trust the +0.44 after Qh3 either. Just shows how complex a game chess is.
For several other moves in the game, I turned on the engine, but found unreliable evaluations.
I do not see +0.5 after Qa4 h5 gxh5+ Kxh5 Qa7
Of course score of more than +3 is no proof that white wins and I did not analyze enough to be sure that Qa4 wins but you can be sure that nobody is going to claim that white can win based on +0.5
I can add that I did not see evaluation by engines that suggested that white is losing after Nd2.
I am using Rybka 3 and after Qa4 h5 gh5+ Kh5 Qa7, I see a +0.5 score. I did not let it run overnight. However, for a human such as much, I do not see a clear way to win as white has to bank on the c3 pawn and winning at least one pawn.
Here is an analysis of Stockfish 1.7.1 after 42.Qa4 h5 43.gxh5+ Kxh5 44.Qa7:
[d]
The analysis is not perfect since 62.Qf4+? (two plies before the line ends) is not optimal and the final position of that analysis is draw based on perpetual check, but white can simply play 62.Qxe2 instead and win, so the overall result would be that your variation does also lose for black provided that the remaining part of the analysis is correct. It is difficult, though, I admit, and your suggested move h5 does not look too bad.
Sven
I advise you analyze with Rybka as well. Rybka 2.2 n2 is free. I normally would try to find holes in the analysis, but I found that turning on Rybka is more than enough to silence Stockfish in most endgame lines. Stockfish is still far behind Rybka in the endgame, though of course I am very happy to see Stockfish improving rapidly. In this particular endgame, I think you will see the benefits soon.
Brief comment is that Black should not have to play Kg6 and Kf5 voluntarily. However, if there is a Rybka certified line to the win, I am very interested to know.
JuLieN wrote:Yes, I remember that SF gave 42. Qa4 a mark above +2 for white. White could have won this game. The thing is that such moves are nearly unreachable for humans, as their consequences are not obvious at all.
Could this +2 score merely be the result of a partial search and evaluation? In other words, the engine finds a strong line and shows that score, but as it evaluates more lines at that same depth, it finds that that line is not so strong?
I've seen this happen before when letting the engine search for a long time. Until it's actually searched the entire tree, you don't have an accurate evaluation and so you can only look at the score at the end of each depth search, and not in the middle. (I hope somebody understands what I'm saying here).
Yeah, I do. It failed high that's all. It was a good move but winning I doubt it.
I think so as well, as implied by my post. If anyone can find a clear win after Qa4 h5 with Rybka (Stockfish has a lot to learn in the endgames I am afraid), let me know.
Actually this game also exposes the weaknesses of engines quite a bit. Stockfish and other engines thought the position after Nd2 was just losing for white.
And this Qa4 position is another case in point, what if black just plays 42.. h5 after Qa4 trading the g-pawn. I tried with engines and yes you will see a +0.5 which keeps going down, but the engines are just evaluating the position, there is no win of course.. Shipov of course has caught on the engine disease. I might be proven wrong but positions like occurred in the game are confusing for both humans and engines. For the same reason, I do not trust the +0.44 after Qh3 either. Just shows how complex a game chess is.
For several other moves in the game, I turned on the engine, but found unreliable evaluations.
I do not see +0.5 after Qa4 h5 gxh5+ Kxh5 Qa7
Of course score of more than +3 is no proof that white wins and I did not analyze enough to be sure that Qa4 wins but you can be sure that nobody is going to claim that white can win based on +0.5
I can add that I did not see evaluation by engines that suggested that white is losing after Nd2.
I am using Rybka 3 and after Qa4 h5 gh5+ Kh5 Qa7, I see a +0.5 score. I did not let it run overnight. However, for a human such as much, I do not see a clear way to win as white has to bank on the c3 pawn and winning at least one pawn.
Here is an analysis of Stockfish 1.7.1 after 42.Qa4 h5 43.gxh5+ Kxh5 44.Qa7:
[d]
The analysis is not perfect since 62.Qf4+? (two plies before the line ends) is not optimal and the final position of that analysis is draw based on perpetual check, but white can simply play 62.Qxe2 instead and win, so the overall result would be that your variation does also lose for black provided that the remaining part of the analysis is correct. It is difficult, though, I admit, and your suggested move h5 does not look too bad.
Sven
I advise you analyze with Rybka as well. Rybka 2.2 n2 is free. I normally would try to find holes in the analysis, but I found that turning on Rybka is more than enough to silence Stockfish in most endgame lines. Stockfish is still far behind Rybka in the endgame, though of course I am very happy to see Stockfish improving rapidly. In this particular endgame, I think you will see the benefits soon.
Brief comment is that Black should not have to play Kg6 and Kf5 voluntarily. However, if there is a Rybka certified line to the win, I am very interested to know.
Based on my knowledge stockfish is better than rybka in the endgame
because of stockfish deep search.
The superiority of rybka is in the middlegame and not in the endgame.
Not sure about the deep search stuff in stockfish, but I have played endgames against Stockfish and Rybka (without any handicap) and have been able to win equal endgames against Stockfish far more frequently than against Rybka. I feel that Stockfish does not have sufficient knowledge in many positions whereas Rybka does. I can post examples if you want.
That being said, I must state that I do not win most games against computers of course, I only have a chance if I play openings that lead into endgames and even then its a mine field.
shiv wrote:I advise you analyze with Rybka as well. Rybka 2.2 n2 is free. I normally would try to find holes in the analysis, but I found that turning on Rybka is more than enough to silence Stockfish in most endgame lines. Stockfish is still far behind Rybka in the endgame, though of course I am very happy to see Stockfish improving rapidly. In this particular endgame, I think you will see the benefits soon.
Brief comment is that Black should not have to play Kg6 and Kf5 voluntarily. However, if there is a Rybka certified line to the win, I am very interested to know.
Based on my knowledge stockfish is better than rybka in the endgame
because of stockfish deep search.
The superiority of rybka is in the middlegame and not in the endgame.
Uri
Sounds like a Rybka-Stockfish match from this endgame is called for. I'm curious which would prevail: the raw depth of Stockfish, or the additional knowledge from Rybka's eval and endgame databases.
shiv wrote:I advise you analyze with Rybka as well. Rybka 2.2 n2 is free. I normally would try to find holes in the analysis, but I found that turning on Rybka is more than enough to silence Stockfish in most endgame lines. Stockfish is still far behind Rybka in the endgame, though of course I am very happy to see Stockfish improving rapidly. In this particular endgame, I think you will see the benefits soon.
Brief comment is that Black should not have to play Kg6 and Kf5 voluntarily. However, if there is a Rybka certified line to the win, I am very interested to know.
Based on my knowledge stockfish is better than rybka in the endgame
because of stockfish deep search.
The superiority of rybka is in the middlegame and not in the endgame.
Uri
Sounds like a Rybka-Stockfish match from this endgame is called for. I'm curious which would prevail: the raw depth of Stockfish, or the additional knowledge from Rybka's eval and endgame databases.
Stockfish would win, Rybka needs it's tablebases in order to equalize the match.
"Good decisions come from experience, and experience comes from bad decisions."
__________________________________________________________________
Ted Summers
