25. ....g3!!! Sacrifice a pawn to open "g"file.
78.....Bd7 (sacrifice center pawn to open "e" file.
107....Re4!!! Sacrifice a rook! Leela see a knight a center is better than a rook, so sacrificed a rook and keep her knight.
From Stockfish POV, Stockfish also sacrificed a knight in the centre, in exchange for two centre pawn, but Leela did not take the knight
Second pawn sacrifice wasnt a sacrifice though, as a4 drops. Proper term for the Re4 move is an exchange sacrifice (rook for bishop or knight). Rook sacrifice generally implies only pawns and favorable board transformation as compensation.
Probably SFs unusual Nd2-f3-e5 maneuver was my favorite move sequence in the game. I don't know if the knight can be taken in any way at any point, but there is something fascinating about how long (almost 30 full moves!) that knight was en-prise, with neither side seeming to care.
25. ....g3!!! Sacrifice a pawn to open "g"file.
78.....Bd7 (sacrifice center pawn to open "e" file.
107....Re4!!! Sacrifice a rook! Leela see a knight a center is better than a rook, so sacrificed a rook and keep her knight.
From Stockfish POV, Stockfish also sacrificed a knight in the centre, in exchange for two centre pawn, but Leela did not take the knight
Thanks for this game. Strange how SF didn't claim (at moves 140, 142 and 143) a three-fold repetition as pointed out (see game comment) above. There were three opportunities to claim for a draw and yet SF didn't do it. Maybe SF felt it is winning due to material advantage. Conversely, Lc0 should have avoided those repetitions if it (Lc0) knew it was winning. Would be nice to see the logs if indeed both of them were thinking they were winning. But if SF did indeed consider it was losing at that point (move 140) then the correct move is to just claim a draw via a three-fold-repetition.
Both conditions should be met to prove that both Stockfish 18122721 and LCZero v20-rc2-32329 don't have a 3-fold repetition blindness (bug).
-SF thought it was winning (SO NO DRAW CLAIM) and also Lc0 v20 32329 thought it was just drawish game (hence it allowed 3-fold repetition many times) so it reached up to move 192.
I told my wife that a husband is like a fine wine; he gets better with age. The next day, she locked me in the cellar.
25. ....g3!!! Sacrifice a pawn to open "g"file.
78.....Bd7 (sacrifice center pawn to open "e" file.
107....Re4!!! Sacrifice a rook! Leela see a knight a center is better than a rook, so sacrificed a rook and keep her knight.
From Stockfish POV, Stockfish also sacrificed a knight in the centre, in exchange for two centre pawn, but Leela did not take the knight
Thanks for this game. Strange how SF didn't claim (at moves 140, 142 and 143) a three-fold repetition as pointed out (see game comment) above. There were three opportunities to claim for a draw and yet SF didn't do it. Maybe SF felt it is winning due to material advantage. Conversely, Lc0 should have avoided those repetitions if it (Lc0) knew it was winning. Would be nice to see the logs if indeed both of them were thinking they were winning. But if SF did indeed consider it was losing at that point (move 140) then the correct move is to just claim a draw via a three-fold-repetition.
Both conditions should be met to prove that both Stockfish 18122721 and LCZero v20-rc2-32329 don't have a 3-fold repetition blindness (bug).
-SF thought it was winning (SO NO DRAW CLAIM) and also Lc0 v20 32329 thought it was just drawish game (hence it allowed 3-fold repetition many times) so it reached up to move 192.
Sorry, all your speculations about bugs and 3-fold are obsolete.
The OP just used a GUI, which doesn't know the complete chess rules (e.g. Arena - it does not consider the right to move! for 3-fold and always adds faulty 'repetitions' to the pgns).
This happens already a few times per month now (posts with wrong 3-fold claims), since LC0 loves shuffling
and users love Arena. (just search for wrong repetition...)
SF claimed white is better since the opening and about half of the whole game, meanwhile LCo think black is better since the end of book move!
It was a really tense and confusing game! See the evaluation graph in the middle.
The question become "Why Lco think black is better" since the beginning whereas SF disagreed!
People were mad whose evaluation to be trusted!
phpJGWomv.jpeg
Last edited by Nay Lin Tun on Wed Jan 02, 2019 3:46 pm, edited 1 time in total.
25. ....g3!!! Sacrifice a pawn to open "g"file.
78.....Bd7 (sacrifice center pawn to open "e" file.
107....Re4!!! Sacrifice a rook! Leela see a knight a center is better than a rook, so sacrificed a rook and keep her knight.
From Stockfish POV, Stockfish also sacrificed a knight in the centre, in exchange for two centre pawn, but Leela did not take the knight
Thanks for this game. Strange how SF didn't claim (at moves 140, 142 and 143) a three-fold repetition as pointed out (see game comment) above. There were three opportunities to claim for a draw and yet SF didn't do it. Maybe SF felt it is winning due to material advantage. Conversely, Lc0 should have avoided those repetitions if it (Lc0) knew it was winning. Would be nice to see the logs if indeed both of them were thinking they were winning. But if SF did indeed consider it was losing at that point (move 140) then the correct move is to just claim a draw via a three-fold-repetition.
Both conditions should be met to prove that both Stockfish 18122721 and LCZero v20-rc2-32329 don't have a 3-fold repetition blindness (bug).
-SF thought it was winning (SO NO DRAW CLAIM) and also Lc0 v20 32329 thought it was just drawish game (hence it allowed 3-fold repetition many times) so it reached up to move 192.
Sorry, all your speculations about bugs and 3-fold are obsolete.
The OP just used a GUI, which doesn't know the complete chess rules (e.g. Arena - it does not consider the right to move! for 3-fold and always adds faulty 'repetitions' to the pgns).
This happens already a few times per month now (posts with wrong 3-fold claims), since LC0 loves shuffling
and users love Arena. (just search for wrong repetition...)
Thanks a lot for the reply. I rechecked the positions and you are indeed right, the side to move was different though the position was the same. So the Arena bug you mention is correct. Sorry, missed the memo. Strange how this old bug of Arena was not fixed.
Happy new year and thanks again for the clarification.
I told my wife that a husband is like a fine wine; he gets better with age. The next day, she locked me in the cellar.
25. ....g3!!! Sacrifice a pawn to open "g"file.
78.....Bd7 (sacrifice center pawn to open "e" file.
107....Re4!!! Sacrifice a rook! Leela see a knight a center is better than a rook, so sacrificed a rook and keep her knight.
From Stockfish POV, Stockfish also sacrificed a knight in the centre, in exchange for two centre pawn, but Leela did not take the knight
Thanks for this game. Strange how SF didn't claim (at moves 140, 142 and 143) a three-fold repetition as pointed out (see game comment) above. There were three opportunities to claim for a draw and yet SF didn't do it. Maybe SF felt it is winning due to material advantage. Conversely, Lc0 should have avoided those repetitions if it (Lc0) knew it was winning. Would be nice to see the logs if indeed both of them were thinking they were winning. But if SF did indeed consider it was losing at that point (move 140) then the correct move is to just claim a draw via a three-fold-repetition.
Both conditions should be met to prove that both Stockfish 18122721 and LCZero v20-rc2-32329 don't have a 3-fold repetition blindness (bug).
-SF thought it was winning (SO NO DRAW CLAIM) and also Lc0 v20 32329 thought it was just drawish game (hence it allowed 3-fold repetition many times) so it reached up to move 192.
There is no code in SF that sends a draw claim to the GUI. It assumes the GUI will handle those cases appropriately. The UCI protocol calls for the GUI to act as arbiter and declare a game drawn after a threefold repetition has occurred. The GUI used in the games above is either not fully UCI compliant or perhaps some setting was not set properly. I use xBoard and one may set various adjudications settings, including 3 fold.
25. ....g3!!! Sacrifice a pawn to open "g"file.
78.....Bd7 (sacrifice center pawn to open "e" file.
107....Re4!!! Sacrifice a rook! Leela see a knight a center is better than a rook, so sacrificed a rook and keep her knight.
From Stockfish POV, Stockfish also sacrificed a knight in the centre, in exchange for two centre pawn, but Leela did not take the knight
Thanks for this game. Strange how SF didn't claim (at moves 140, 142 and 143) a three-fold repetition as pointed out (see game comment) above. There were three opportunities to claim for a draw and yet SF didn't do it. Maybe SF felt it is winning due to material advantage. Conversely, Lc0 should have avoided those repetitions if it (Lc0) knew it was winning. Would be nice to see the logs if indeed both of them were thinking they were winning. But if SF did indeed consider it was losing at that point (move 140) then the correct move is to just claim a draw via a three-fold-repetition.
Both conditions should be met to prove that both Stockfish 18122721 and LCZero v20-rc2-32329 don't have a 3-fold repetition blindness (bug).
-SF thought it was winning (SO NO DRAW CLAIM) and also Lc0 v20 32329 thought it was just drawish game (hence it allowed 3-fold repetition many times) so it reached up to move 192.
There is no code in SF that sends a draw claim to the GUI. It assumes the GUI will handle those cases appropriately. The UCI protocol calls for the GUI to act as arbiter and declare a game drawn after a threefold repetition has occurred. The GUI used in the games above is either not fully UCI compliant or perhaps some setting was not set properly. I use xBoard and one may set various adjudications settings, including 3 fold.
Thanks for that info. Major difference in human-human chess. Where it is up to players to detect a 3-fold and mutually agree for a draw, or if the opposing player won't accept, then you call the arbiter to prove that 3-fold did indeed occur, else if nobody notices the 3-fold then the game continues and won't be stopped by arbiter nor by DGT board.
I told my wife that a husband is like a fine wine; he gets better with age. The next day, she locked me in the cellar.