chess 960 engines vs humans

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
lkaufman
Posts: 4324
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: chess 960 engines vs humans

Post by lkaufman » Thu Aug 18, 2016 2:31 am

George wrote:
lkaufman wrote:
George wrote:
lkaufman wrote:
George wrote:
lkaufman wrote:
George wrote:
lkaufman wrote:
Modern Times wrote:Logically I would have thought so, because the human can't rely on memorised opening theory, but if that is true in practice I don't know.

Perhaps the next Komodo vs human match can be chess960.... not that that would answer the question, but it would be good nevertheless :)
Actually I have been thinking about a Komodo vs. GM (or IM) match in chess960. The problem is that it is totally obvious to me (as a former U.S. Open chess960 champion and a GM in normal chess) that the human will need a larger handicap in chess960. Human GMs are very familiar with all the typical ideas and pawn structures that arise from normal openings in standard chess, but are quite uncomfortable in typical chess960 positions. It's not just that they can't rely on memorized openings (although that's already a big problem); it's more the sheer weirdness of the starting positions. For a computer this makes no difference.
It seems the only easy handicap for chess960 is knight odds, where the computer chooses which knight to remove after the position is chosen. But even at chess960 knight odds is probably too much for a GM; maybe it's okay for an IM, although even that seems questionable to me since the break-even point for knight odds in normal chess appears to be in the 1900s FIDE. But odds of a pawn or two would be too position-dependent, and the Exchange appears not to be enough even for strong GMs in normal chess.
I'm open to suggestions. If I hear something that sounds reasonable and close to fair we might try to work it into our upcoming match on chess.com with IM Daniel Rensch. I feel he's a bit too strong for knight odds even at chess960 (at 45' plus inc).
I spoke with GM Nakamura and he said that with any two pawns Stockfish with the same time control is beating him but with a knight odd even with his super fast PC he is sharing wins with stockfish using the same time control when he beat GM Aronian back in 2009 to become the last Chess960 Champion. ==> https://en.chessbase.com/post/che-claic ... ampionship
What was the time control in these Nakamura vs Stockfish handicap games? Are you saying that results at knight odds were about even? If this was at a non-blitz tc I am surprised. Chess960 must be even harder than I thought it was for humans.
I suppose that Chess960 is harder for humans and the more time you allow the Monster the deeper it calculates. The time control was 20 minutes + 5s game, and his latest PC is an Intel Core i7 Extreme Edition i7-4960X Hexa-core (6 Core) 3.60 GHz Processor.

PS: I don't know if he is better off playing at faster time control where the monster can hardly calculate as deep, like in these games ===>
https://www.youtube.com/watch?v=2k4V_LJTvqM

GM Wesley So lost terribly with Knight Odd Vs Stockfish in Bullet Chess, but he is rated over 100 points less than Nakamura in Bullet Chess, and this version of Stockfish is rated over 200 points above the Rybka that GM Nakamura played back in 2008 :-)
https://www.youtube.com/watch?v=xu-pSaN62lc

Now compared to S L OW E R Time Control
https://www.youtube.com/watch?v=nQDQNBEsJ80

https://www.youtube.com/watch?v=_ue_woDfr9A
Bullet chess for humans has almost nothing to do with any normal time control chess. Two things are very clear:
1. In human vs. computer chess, longer time controls always favor the human (within reason; of course if it's too slow without a break fatigue is an issue, but no one uses time controls that slow). Computers were beating GMs in blitz long before they could do so in rapid, and were winning in rapid long before they could do so in standard chess.
2. In handicap play, longer time controls always favor the handicap receiver. This is obvious, since the handicap giver has to hope for blunders. In normal (not 960) chess, at 3' + 2" Komodo is about even with 2600 level players at knight odds, but below 2000 level at 45' + 15".

Chess 960 is much less familiar for humans, so it should take a much stronger player than 2000 to win a knight odds match at 45' + 15". But if Nakamura was only around equal at 20' + 5", that suggests that a 2700 level human might be a fair match at knight odds 45' + 15". Frankly that is very hard for me to believe. Chess 960 is different from normal chess, but I can't believe it's so different that the break-even rating for a human would rise from less than 2000 to 2700. I would have guessed 2300.

I guess we will have to try it out in a match in the near future. Maybe I'll try a game or two myself just to get a feel for how difficult it is.
Those were only the first two games, later he played 4 more games always playing different positions and won 3 and drew one.
Okay, so the computer performed around 2600 level, which at 45' + 15" would perhaps drop to 2500 level. So knight odds in 960 is reasonable for a below-average GM or an IM. Good to know.
Mr: Kaufman Here is the initial chess960 position where Booot has the e7 and f7 pawns odds from Stockfish + plus the initial f4 move for White. Just by comparison I believe that either GM Nakamura or Carlsen can be given the odds of 2 pawns in Chess960 at longer time control like game in 30 minutes per side.
[D]nrnkrbbq/pppp2pp/8/8/5P2/8/PPPPP1PP/NRNKRBBQ b - - 0 1[D]

[pgn][Event "Computer chess game"]
[Site "HP"]
[Date "2016.08.17"]
[Round "3"]
[White "Booot6_32"]
[Black "Stockfish 7 x64"]
[Result "1/2-1/2"]
[BlackElo "2828"]
[WhiteElo "3100"]
[TimeControl "5400"]
[SetUp "1"]
[FEN "nrnkrbbq/pppp2pp/8/8/5P2/8/PPPPP1PP/NRNKRBBQ b - - 0 1"]
[WhiteType "program"]
[BlackType "human"]

1. ... g5 2. g3 Bh6 3. Nab3 gxf4 4. Bd4 Bg7 5. Bxg7 Qxg7 6. gxf4 Nd6 7. Nd3
Qh6 8. e4 b6 9. Qf3 c5 10. Ne5 Nc7 11. Qg3 Bf7 12. Kc1 a5 13. Bh3 a4 14.
Bxd7 Rxe5 15. fxe5 Kxd7 16. Rf1 Rg8 17. Rxf7+ Nxf7 18. Qxg8 Nxe5 19. Qg3
Qf6 20. Nxc5+ bxc5 21. b3 axb3 22. axb3 Kc6 23. Qg2 Ne6 24. Qh3 Nf4 25. Qf5
Qxf5 26. exf5 Kd6 27. Ra1 Nd7 28. c3 Ke5 29. Ra7 Kd6 30. Ra5 Nd3+ 31. Kd1
Nf4 32. Ra4 Ke5 33. Ra7 Kd6 34. Kc2 h5 35. b4 cxb4 36. cxb4 Nd5 37. Kb3
N5b6 38. Ra5 h4 39. h3 Nd5 40. d4 Nf4 41. Ra6+ Kd5 42. Kc3 Nxh3 43. Rh6 Ng5
44. Rxh4 Ne4+ 45. Kd3 Nd6 46. Rh7 Nb8 47. Rh5 Nd7 48. Rg5 Nf6 49. Rg6 Nh5
50. Rg4 Nf6 51. Rg2 Nxf5 52. b5 Ne4 53. Rh2 Nf6 54. Rf2 Nxd4 55. Rxf6 Nxb5
56. Rf5+ Kc6 57. Rg5 Nd6 58. Kd4 Nb5+ 59. Ke5 Kc5 60. Rg1 Kc4 61. Rc1+ Nc3
62. Rc2 Kb3 63. Rd2 Kc4 64. Rd4+ Kc5 65. Rd7 Kc4 66. Rc7+ Kd3 67. Re7 Kc4
68. Rd7 Nb5 69. Rd1 Nc3 70. Rd4+ Kc5 71. Rf4 Nb5 72. Rf2 Kc4 73. Rh2 Nc3
74. Rh1 Kc5 75. Rh8 Kc4 76. Rc8+ Kd3 77. Rc5 Ne2 78. Rc7 Nc3 79. Ra7 Kc4
80. Re7 Nb5 81. Rg7 Kb4 82. Rg2 Kc4 83. Rb2 Kc5 84. Rc2+ Kb6 85. Kd5 Nc7+
86. Kc4 Kc6 87. Re2 Kd6 88. Rh2 Ke7 89. Kc5 Ne6+ 90. Kd5 Nf4+ 91. Ke5 Ng6+
92. Kf5 Nf8 93. Rh1 Kf7 94. Re1 Nd7 {White forfeits on time} 1/2-1/2 Thank you Graham, I followed your direction and here is the result between Booot vs Stockfish giving two pawns odds.[/pgn]
You started this game removing the two pawns most favorable to the odds-giver, with both bishops and a rook alreadyi active. I don't doubt that Komodo or Stockfish could give this handicap to Carlsen or Nakamura, but if they chose the pawns to be removed it would be a different story. I think two pawn handicap is not very suitable for chess960, because the size of the handicap is too dependent on which pawns are removed. With knight odds, that's only a minor issue. Without knowing the start position, any handicap involving pawns is too dependent on how the choice is made and what the start position is.
Komodo rules!

User avatar
George
Posts: 682
Joined: Sun Oct 13, 2013 11:44 pm

Re: chess 960 engines vs humans

Post by George » Thu Aug 18, 2016 10:50 am

lkaufman wrote:
George wrote:
lkaufman wrote:
George wrote:
lkaufman wrote:
George wrote:
lkaufman wrote:
George wrote:
lkaufman wrote:
Modern Times wrote:Logically I would have thought so, because the human can't rely on memorised opening theory, but if that is true in practice I don't know.

Perhaps the next Komodo vs human match can be chess960.... not that that would answer the question, but it would be good nevertheless :)
Actually I have been thinking about a Komodo vs. GM (or IM) match in chess960. The problem is that it is totally obvious to me (as a former U.S. Open chess960 champion and a GM in normal chess) that the human will need a larger handicap in chess960. Human GMs are very familiar with all the typical ideas and pawn structures that arise from normal openings in standard chess, but are quite uncomfortable in typical chess960 positions. It's not just that they can't rely on memorized openings (although that's already a big problem); it's more the sheer weirdness of the starting positions. For a computer this makes no difference.
It seems the only easy handicap for chess960 is knight odds, where the computer chooses which knight to remove after the position is chosen. But even at chess960 knight odds is probably too much for a GM; maybe it's okay for an IM, although even that seems questionable to me since the break-even point for knight odds in normal chess appears to be in the 1900s FIDE. But odds of a pawn or two would be too position-dependent, and the Exchange appears not to be enough even for strong GMs in normal chess.
I'm open to suggestions. If I hear something that sounds reasonable and close to fair we might try to work it into our upcoming match on chess.com with IM Daniel Rensch. I feel he's a bit too strong for knight odds even at chess960 (at 45' plus inc).
I spoke with GM Nakamura and he said that with any two pawns Stockfish with the same time control is beating him but with a knight odd even with his super fast PC he is sharing wins with stockfish using the same time control when he beat GM Aronian back in 2009 to become the last Chess960 Champion. ==> https://en.chessbase.com/post/che-claic ... ampionship
What was the time control in these Nakamura vs Stockfish handicap games? Are you saying that results at knight odds were about even? If this was at a non-blitz tc I am surprised. Chess960 must be even harder than I thought it was for humans.
I suppose that Chess960 is harder for humans and the more time you allow the Monster the deeper it calculates. The time control was 20 minutes + 5s game, and his latest PC is an Intel Core i7 Extreme Edition i7-4960X Hexa-core (6 Core) 3.60 GHz Processor.

PS: I don't know if he is better off playing at faster time control where the monster can hardly calculate as deep, like in these games ===>
https://www.youtube.com/watch?v=2k4V_LJTvqM

GM Wesley So lost terribly with Knight Odd Vs Stockfish in Bullet Chess, but he is rated over 100 points less than Nakamura in Bullet Chess, and this version of Stockfish is rated over 200 points above the Rybka that GM Nakamura played back in 2008 :-)
https://www.youtube.com/watch?v=xu-pSaN62lc

Now compared to S L OW E R Time Control
https://www.youtube.com/watch?v=nQDQNBEsJ80

https://www.youtube.com/watch?v=_ue_woDfr9A
Bullet chess for humans has almost nothing to do with any normal time control chess. Two things are very clear:
1. In human vs. computer chess, longer time controls always favor the human (within reason; of course if it's too slow without a break fatigue is an issue, but no one uses time controls that slow). Computers were beating GMs in blitz long before they could do so in rapid, and were winning in rapid long before they could do so in standard chess.
2. In handicap play, longer time controls always favor the handicap receiver. This is obvious, since the handicap giver has to hope for blunders. In normal (not 960) chess, at 3' + 2" Komodo is about even with 2600 level players at knight odds, but below 2000 level at 45' + 15".

Chess 960 is much less familiar for humans, so it should take a much stronger player than 2000 to win a knight odds match at 45' + 15". But if Nakamura was only around equal at 20' + 5", that suggests that a 2700 level human might be a fair match at knight odds 45' + 15". Frankly that is very hard for me to believe. Chess 960 is different from normal chess, but I can't believe it's so different that the break-even rating for a human would rise from less than 2000 to 2700. I would have guessed 2300.

I guess we will have to try it out in a match in the near future. Maybe I'll try a game or two myself just to get a feel for how difficult it is.
Those were only the first two games, later he played 4 more games always playing different positions and won 3 and drew one.
Okay, so the computer performed around 2600 level, which at 45' + 15" would perhaps drop to 2500 level. So knight odds in 960 is reasonable for a below-average GM or an IM. Good to know.
Mr: Kaufman Here is the initial chess960 position where Booot has the e7 and f7 pawns odds from Stockfish + plus the initial f4 move for White. Just by comparison I believe that either GM Nakamura or Carlsen can be given the odds of 2 pawns in Chess960 at longer time control like game in 30 minutes per side.
[D]nrnkrbbq/pppp2pp/8/8/5P2/8/PPPPP1PP/NRNKRBBQ b - - 0 1[D]

[pgn][Event "Computer chess game"]
[Site "HP"]
[Date "2016.08.17"]
[Round "3"]
[White "Booot6_32"]
[Black "Stockfish 7 x64"]
[Result "1/2-1/2"]
[BlackElo "2828"]
[WhiteElo "3100"]
[TimeControl "5400"]
[SetUp "1"]
[FEN "nrnkrbbq/pppp2pp/8/8/5P2/8/PPPPP1PP/NRNKRBBQ b - - 0 1"]
[WhiteType "program"]
[BlackType "human"]

1. ... g5 2. g3 Bh6 3. Nab3 gxf4 4. Bd4 Bg7 5. Bxg7 Qxg7 6. gxf4 Nd6 7. Nd3
Qh6 8. e4 b6 9. Qf3 c5 10. Ne5 Nc7 11. Qg3 Bf7 12. Kc1 a5 13. Bh3 a4 14.
Bxd7 Rxe5 15. fxe5 Kxd7 16. Rf1 Rg8 17. Rxf7+ Nxf7 18. Qxg8 Nxe5 19. Qg3
Qf6 20. Nxc5+ bxc5 21. b3 axb3 22. axb3 Kc6 23. Qg2 Ne6 24. Qh3 Nf4 25. Qf5
Qxf5 26. exf5 Kd6 27. Ra1 Nd7 28. c3 Ke5 29. Ra7 Kd6 30. Ra5 Nd3+ 31. Kd1
Nf4 32. Ra4 Ke5 33. Ra7 Kd6 34. Kc2 h5 35. b4 cxb4 36. cxb4 Nd5 37. Kb3
N5b6 38. Ra5 h4 39. h3 Nd5 40. d4 Nf4 41. Ra6+ Kd5 42. Kc3 Nxh3 43. Rh6 Ng5
44. Rxh4 Ne4+ 45. Kd3 Nd6 46. Rh7 Nb8 47. Rh5 Nd7 48. Rg5 Nf6 49. Rg6 Nh5
50. Rg4 Nf6 51. Rg2 Nxf5 52. b5 Ne4 53. Rh2 Nf6 54. Rf2 Nxd4 55. Rxf6 Nxb5
56. Rf5+ Kc6 57. Rg5 Nd6 58. Kd4 Nb5+ 59. Ke5 Kc5 60. Rg1 Kc4 61. Rc1+ Nc3
62. Rc2 Kb3 63. Rd2 Kc4 64. Rd4+ Kc5 65. Rd7 Kc4 66. Rc7+ Kd3 67. Re7 Kc4
68. Rd7 Nb5 69. Rd1 Nc3 70. Rd4+ Kc5 71. Rf4 Nb5 72. Rf2 Kc4 73. Rh2 Nc3
74. Rh1 Kc5 75. Rh8 Kc4 76. Rc8+ Kd3 77. Rc5 Ne2 78. Rc7 Nc3 79. Ra7 Kc4
80. Re7 Nb5 81. Rg7 Kb4 82. Rg2 Kc4 83. Rb2 Kc5 84. Rc2+ Kb6 85. Kd5 Nc7+
86. Kc4 Kc6 87. Re2 Kd6 88. Rh2 Ke7 89. Kc5 Ne6+ 90. Kd5 Nf4+ 91. Ke5 Ng6+
92. Kf5 Nf8 93. Rh1 Kf7 94. Re1 Nd7 {White forfeits on time} 1/2-1/2 Thank you Graham, I followed your direction and here is the result between Booot vs Stockfish giving two pawns odds.[/pgn]
You started this game removing the two pawns most favorable to the odds-giver, with both bishops and a rook alreadyi active. I don't doubt that Komodo or Stockfish could give this handicap to Carlsen or Nakamura, but if they chose the pawns to be removed it would be a different story. I think two pawn handicap is not very suitable for chess960, because the size of the handicap is too dependent on which pawns are removed. With knight odds, that's only a minor issue. Without knowing the start position, any handicap involving pawns is too dependent on how the choice is made and what the start position is.
Here is a new Chess960 starting position ===>
[D]nrnkrbbq/1p1ppppp/8/8/8/8/PPPPPPPP/NRNKRBBQ w - - 0 1[D]
[pgn][Event "Computer chess game"]
[Site "HP"]
[Date "2016.08.18"]
[Round "?"]
[White "Booot6_32"]
[Black "Stockfish 7 x64"]
[Result "*"]
[BlackElo "3100"]
[WhiteElo "2828"]
[TimeControl "180"]
[SetUp "1"]
[FEN "nrnkrbbq/1p1ppppp/8/8/8/8/PPPPPPPP/NRNKRBBQ w - - 0 1"]


1. Nab3 e5 2. f4 e4 3. e3 Nc7 4. Nd4 Nd6 5. Nce2 f5 6. Ra1 g6 7. g4 Bg7 8.
gxf5 Nxf5 9. Bh3 Nxd4 10. Nxd4 Bd5 11. Qg2 Kc8 12. Qf1 Qf8 13. Kc1 Qd6 14.
Qf2 Ra8 15. Qh4 Rxa2 16. Rxa2 Bxa2 17. c3 Bxd4 18. cxd4 Bb3 19. Bxd7+ Qxd7
20. Qg4 Re6 21. Kb1 Qa4 22. Qxe6+ Nxe6 23. Rc1+ {White resigns} *

I decided to remove the two pawns in front of the Knights a7 and c7, and hwere is the result between Booot vs Stockfish [/pgn]

Modern Times
Posts: 2598
Joined: Thu Jun 07, 2012 9:02 pm

Re: chess 960 engines vs humans

Post by Modern Times » Thu Aug 18, 2016 11:46 am

Booot plays chess960 ? I didn't know that.
.

Opinions expressed here are my own, and not necessarily those of the CCRL Group.

lkaufman
Posts: 4324
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: chess 960 engines vs humans

Post by lkaufman » Thu Aug 18, 2016 2:45 pm

George wrote:
lkaufman wrote:
George wrote:
lkaufman wrote:
George wrote:
lkaufman wrote:
George wrote:
lkaufman wrote:
George wrote:
lkaufman wrote:
Modern Times wrote:Logically I would have thought so, because the human can't rely on memorised opening theory, but if that is true in practice I don't know.

Perhaps the next Komodo vs human match can be chess960.... not that that would answer the question, but it would be good nevertheless :)
Actually I have been thinking about a Komodo vs. GM (or IM) match in chess960. The problem is that it is totally obvious to me (as a former U.S. Open chess960 champion and a GM in normal chess) that the human will need a larger handicap in chess960. Human GMs are very familiar with all the typical ideas and pawn structures that arise from normal openings in standard chess, but are quite uncomfortable in typical chess960 positions. It's not just that they can't rely on memorized openings (although that's already a big problem); it's more the sheer weirdness of the starting positions. For a computer this makes no difference.
It seems the only easy handicap for chess960 is knight odds, where the computer chooses which knight to remove after the position is chosen. But even at chess960 knight odds is probably too much for a GM; maybe it's okay for an IM, although even that seems questionable to me since the break-even point for knight odds in normal chess appears to be in the 1900s FIDE. But odds of a pawn or two would be too position-dependent, and the Exchange appears not to be enough even for strong GMs in normal chess.
I'm open to suggestions. If I hear something that sounds reasonable and close to fair we might try to work it into our upcoming match on chess.com with IM Daniel Rensch. I feel he's a bit too strong for knight odds even at chess960 (at 45' plus inc).
I spoke with GM Nakamura and he said that with any two pawns Stockfish with the same time control is beating him but with a knight odd even with his super fast PC he is sharing wins with stockfish using the same time control when he beat GM Aronian back in 2009 to become the last Chess960 Champion. ==> https://en.chessbase.com/post/che-claic ... ampionship
What was the time control in these Nakamura vs Stockfish handicap games? Are you saying that results at knight odds were about even? If this was at a non-blitz tc I am surprised. Chess960 must be even harder than I thought it was for humans.
I suppose that Chess960 is harder for humans and the more time you allow the Monster the deeper it calculates. The time control was 20 minutes + 5s game, and his latest PC is an Intel Core i7 Extreme Edition i7-4960X Hexa-core (6 Core) 3.60 GHz Processor.

PS: I don't know if he is better off playing at faster time control where the monster can hardly calculate as deep, like in these games ===>
https://www.youtube.com/watch?v=2k4V_LJTvqM

GM Wesley So lost terribly with Knight Odd Vs Stockfish in Bullet Chess, but he is rated over 100 points less than Nakamura in Bullet Chess, and this version of Stockfish is rated over 200 points above the Rybka that GM Nakamura played back in 2008 :-)
https://www.youtube.com/watch?v=xu-pSaN62lc

Now compared to S L OW E R Time Control
https://www.youtube.com/watch?v=nQDQNBEsJ80

https://www.youtube.com/watch?v=_ue_woDfr9A
Bullet chess for humans has almost nothing to do with any normal time control chess. Two things are very clear:
1. In human vs. computer chess, longer time controls always favor the human (within reason; of course if it's too slow without a break fatigue is an issue, but no one uses time controls that slow). Computers were beating GMs in blitz long before they could do so in rapid, and were winning in rapid long before they could do so in standard chess.
2. In handicap play, longer time controls always favor the handicap receiver. This is obvious, since the handicap giver has to hope for blunders. In normal (not 960) chess, at 3' + 2" Komodo is about even with 2600 level players at knight odds, but below 2000 level at 45' + 15".

Chess 960 is much less familiar for humans, so it should take a much stronger player than 2000 to win a knight odds match at 45' + 15". But if Nakamura was only around equal at 20' + 5", that suggests that a 2700 level human might be a fair match at knight odds 45' + 15". Frankly that is very hard for me to believe. Chess 960 is different from normal chess, but I can't believe it's so different that the break-even rating for a human would rise from less than 2000 to 2700. I would have guessed 2300.

I guess we will have to try it out in a match in the near future. Maybe I'll try a game or two myself just to get a feel for how difficult it is.
Those were only the first two games, later he played 4 more games always playing different positions and won 3 and drew one.
Okay, so the computer performed around 2600 level, which at 45' + 15" would perhaps drop to 2500 level. So knight odds in 960 is reasonable for a below-average GM or an IM. Good to know.
Mr: Kaufman Here is the initial chess960 position where Booot has the e7 and f7 pawns odds from Stockfish + plus the initial f4 move for White. Just by comparison I believe that either GM Nakamura or Carlsen can be given the odds of 2 pawns in Chess960 at longer time control like game in 30 minutes per side.
[D]nrnkrbbq/pppp2pp/8/8/5P2/8/PPPPP1PP/NRNKRBBQ b - - 0 1[D]

[pgn][Event "Computer chess game"]
[Site "HP"]
[Date "2016.08.17"]
[Round "3"]
[White "Booot6_32"]
[Black "Stockfish 7 x64"]
[Result "1/2-1/2"]
[BlackElo "2828"]
[WhiteElo "3100"]
[TimeControl "5400"]
[SetUp "1"]
[FEN "nrnkrbbq/pppp2pp/8/8/5P2/8/PPPPP1PP/NRNKRBBQ b - - 0 1"]
[WhiteType "program"]
[BlackType "human"]

1. ... g5 2. g3 Bh6 3. Nab3 gxf4 4. Bd4 Bg7 5. Bxg7 Qxg7 6. gxf4 Nd6 7. Nd3
Qh6 8. e4 b6 9. Qf3 c5 10. Ne5 Nc7 11. Qg3 Bf7 12. Kc1 a5 13. Bh3 a4 14.
Bxd7 Rxe5 15. fxe5 Kxd7 16. Rf1 Rg8 17. Rxf7+ Nxf7 18. Qxg8 Nxe5 19. Qg3
Qf6 20. Nxc5+ bxc5 21. b3 axb3 22. axb3 Kc6 23. Qg2 Ne6 24. Qh3 Nf4 25. Qf5
Qxf5 26. exf5 Kd6 27. Ra1 Nd7 28. c3 Ke5 29. Ra7 Kd6 30. Ra5 Nd3+ 31. Kd1
Nf4 32. Ra4 Ke5 33. Ra7 Kd6 34. Kc2 h5 35. b4 cxb4 36. cxb4 Nd5 37. Kb3
N5b6 38. Ra5 h4 39. h3 Nd5 40. d4 Nf4 41. Ra6+ Kd5 42. Kc3 Nxh3 43. Rh6 Ng5
44. Rxh4 Ne4+ 45. Kd3 Nd6 46. Rh7 Nb8 47. Rh5 Nd7 48. Rg5 Nf6 49. Rg6 Nh5
50. Rg4 Nf6 51. Rg2 Nxf5 52. b5 Ne4 53. Rh2 Nf6 54. Rf2 Nxd4 55. Rxf6 Nxb5
56. Rf5+ Kc6 57. Rg5 Nd6 58. Kd4 Nb5+ 59. Ke5 Kc5 60. Rg1 Kc4 61. Rc1+ Nc3
62. Rc2 Kb3 63. Rd2 Kc4 64. Rd4+ Kc5 65. Rd7 Kc4 66. Rc7+ Kd3 67. Re7 Kc4
68. Rd7 Nb5 69. Rd1 Nc3 70. Rd4+ Kc5 71. Rf4 Nb5 72. Rf2 Kc4 73. Rh2 Nc3
74. Rh1 Kc5 75. Rh8 Kc4 76. Rc8+ Kd3 77. Rc5 Ne2 78. Rc7 Nc3 79. Ra7 Kc4
80. Re7 Nb5 81. Rg7 Kb4 82. Rg2 Kc4 83. Rb2 Kc5 84. Rc2+ Kb6 85. Kd5 Nc7+
86. Kc4 Kc6 87. Re2 Kd6 88. Rh2 Ke7 89. Kc5 Ne6+ 90. Kd5 Nf4+ 91. Ke5 Ng6+
92. Kf5 Nf8 93. Rh1 Kf7 94. Re1 Nd7 {White forfeits on time} 1/2-1/2 Thank you Graham, I followed your direction and here is the result between Booot vs Stockfish giving two pawns odds.[/pgn]
You started this game removing the two pawns most favorable to the odds-giver, with both bishops and a rook alreadyi active. I don't doubt that Komodo or Stockfish could give this handicap to Carlsen or Nakamura, but if they chose the pawns to be removed it would be a different story. I think two pawn handicap is not very suitable for chess960, because the size of the handicap is too dependent on which pawns are removed. With knight odds, that's only a minor issue. Without knowing the start position, any handicap involving pawns is too dependent on how the choice is made and what the start position is.
Here is a new Chess960 starting position ===>
[D]nrnkrbbq/1p1ppppp/8/8/8/8/PPPPPPPP/NRNKRBBQ w - - 0 1[D]
[pgn][Event "Computer chess game"]
[Site "HP"]
[Date "2016.08.18"]
[Round "?"]
[White "Booot6_32"]
[Black "Stockfish 7 x64"]
[Result "*"]
[BlackElo "3100"]
[WhiteElo "2828"]
[TimeControl "180"]
[SetUp "1"]
[FEN "nrnkrbbq/1p1ppppp/8/8/8/8/PPPPPPPP/NRNKRBBQ w - - 0 1"]


1. Nab3 e5 2. f4 e4 3. e3 Nc7 4. Nd4 Nd6 5. Nce2 f5 6. Ra1 g6 7. g4 Bg7 8.
gxf5 Nxf5 9. Bh3 Nxd4 10. Nxd4 Bd5 11. Qg2 Kc8 12. Qf1 Qf8 13. Kc1 Qd6 14.
Qf2 Ra8 15. Qh4 Rxa2 16. Rxa2 Bxa2 17. c3 Bxd4 18. cxd4 Bb3 19. Bxd7+ Qxd7
20. Qg4 Re6 21. Kb1 Qa4 22. Qxe6+ Nxe6 23. Rc1+ {White resigns} *

I decided to remove the two pawns in front of the Knights a7 and c7, and hwere is the result between Booot vs Stockfish [/pgn]
OK, this is a better test, it's a full two pawn handicap. Was this game/30' or some otheer TC, and were both engines single-cpu? In any case, it shows something I reported on here a while back, that Komodo can give material handicaps to surprisingly strong unrelated engines, but has much more trouble giving older or time-handicapped versions of itself such handicaps. Probably same is true for Stockfish. I concluded that time-handicapped self-play was a better predictor of human results vs Komodo than Komodo vs. weaker engines. Probably the reason is that unrelated engines have very different evals, which leads them to play much more complicated games than same eval, and complications favor the handicap giver.
Anyway, I suppose for chess 960 we can have two two-pawn handicaps; namely "human choice of pawns" and "engine choice of pawns". Obviously the first is a much larger handicap. But some positions will make the handicaps more severe than others.
I played a couple quick games of chess960 myself with Komodo at knight odds, and lost. It is now obvious to me that it is MUCH harder to win with knight odds at chess960 than at normal chess, and I am willing to try knight odds in 960 vs. IM or GM (not too strong GM). I agree with you now that even with choice of pawns Carlsen or Nakamura would be underdog at chess960 two pawn handicap 45' + 15". Not only is chess960 much more unfamiliar for humans, but also in normal chess the knights can develop towards the center without seriously inpeding the other pieces from developing, which is not generally true in chess960.
By the way, the convention is that when we say "knight odds" or "two pawn odds" or "pawn odds" it means the odds-giver has White; if he has Black you call it "knight and move" or "two pawns and move" etc.
Komodo rules!

User avatar
George
Posts: 682
Joined: Sun Oct 13, 2013 11:44 pm

Re: chess 960 engines vs humans

Post by George » Thu Aug 18, 2016 4:11 pm

lkaufman wrote:
George wrote:
lkaufman wrote:
George wrote:
lkaufman wrote:
George wrote:
lkaufman wrote:
George wrote:
lkaufman wrote:
George wrote:
lkaufman wrote:
Modern Times wrote:Logically I would have thought so, because the human can't rely on memorised opening theory, but if that is true in practice I don't know.

Perhaps the next Komodo vs human match can be chess960.... not that that would answer the question, but it would be good nevertheless :)
Actually I have been thinking about a Komodo vs. GM (or IM) match in chess960. The problem is that it is totally obvious to me (as a former U.S. Open chess960 champion and a GM in normal chess) that the human will need a larger handicap in chess960. Human GMs are very familiar with all the typical ideas and pawn structures that arise from normal openings in standard chess, but are quite uncomfortable in typical chess960 positions. It's not just that they can't rely on memorized openings (although that's already a big problem); it's more the sheer weirdness of the starting positions. For a computer this makes no difference.
It seems the only easy handicap for chess960 is knight odds, where the computer chooses which knight to remove after the position is chosen. But even at chess960 knight odds is probably too much for a GM; maybe it's okay for an IM, although even that seems questionable to me since the break-even point for knight odds in normal chess appears to be in the 1900s FIDE. But odds of a pawn or two would be too position-dependent, and the Exchange appears not to be enough even for strong GMs in normal chess.
I'm open to suggestions. If I hear something that sounds reasonable and close to fair we might try to work it into our upcoming match on chess.com with IM Daniel Rensch. I feel he's a bit too strong for knight odds even at chess960 (at 45' plus inc).
I spoke with GM Nakamura and he said that with any two pawns Stockfish with the same time control is beating him but with a knight odd even with his super fast PC he is sharing wins with stockfish using the same time control when he beat GM Aronian back in 2009 to become the last Chess960 Champion. ==> https://en.chessbase.com/post/che-claic ... ampionship
What was the time control in these Nakamura vs Stockfish handicap games? Are you saying that results at knight odds were about even? If this was at a non-blitz tc I am surprised. Chess960 must be even harder than I thought it was for humans.
I suppose that Chess960 is harder for humans and the more time you allow the Monster the deeper it calculates. The time control was 20 minutes + 5s game, and his latest PC is an Intel Core i7 Extreme Edition i7-4960X Hexa-core (6 Core) 3.60 GHz Processor.

PS: I don't know if he is better off playing at faster time control where the monster can hardly calculate as deep, like in these games ===>
https://www.youtube.com/watch?v=2k4V_LJTvqM

GM Wesley So lost terribly with Knight Odd Vs Stockfish in Bullet Chess, but he is rated over 100 points less than Nakamura in Bullet Chess, and this version of Stockfish is rated over 200 points above the Rybka that GM Nakamura played back in 2008 :-)
https://www.youtube.com/watch?v=xu-pSaN62lc

Now compared to S L OW E R Time Control
https://www.youtube.com/watch?v=nQDQNBEsJ80

https://www.youtube.com/watch?v=_ue_woDfr9A
Bullet chess for humans has almost nothing to do with any normal time control chess. Two things are very clear:
1. In human vs. computer chess, longer time controls always favor the human (within reason; of course if it's too slow without a break fatigue is an issue, but no one uses time controls that slow). Computers were beating GMs in blitz long before they could do so in rapid, and were winning in rapid long before they could do so in standard chess.
2. In handicap play, longer time controls always favor the handicap receiver. This is obvious, since the handicap giver has to hope for blunders. In normal (not 960) chess, at 3' + 2" Komodo is about even with 2600 level players at knight odds, but below 2000 level at 45' + 15".

Chess 960 is much less familiar for humans, so it should take a much stronger player than 2000 to win a knight odds match at 45' + 15". But if Nakamura was only around equal at 20' + 5", that suggests that a 2700 level human might be a fair match at knight odds 45' + 15". Frankly that is very hard for me to believe. Chess 960 is different from normal chess, but I can't believe it's so different that the break-even rating for a human would rise from less than 2000 to 2700. I would have guessed 2300.

I guess we will have to try it out in a match in the near future. Maybe I'll try a game or two myself just to get a feel for how difficult it is.
Those were only the first two games, later he played 4 more games always playing different positions and won 3 and drew one.
Okay, so the computer performed around 2600 level, which at 45' + 15" would perhaps drop to 2500 level. So knight odds in 960 is reasonable for a below-average GM or an IM. Good to know.
Mr: Kaufman Here is the initial chess960 position where Booot has the e7 and f7 pawns odds from Stockfish + plus the initial f4 move for White. Just by comparison I believe that either GM Nakamura or Carlsen can be given the odds of 2 pawns in Chess960 at longer time control like game in 30 minutes per side.
[D]nrnkrbbq/pppp2pp/8/8/5P2/8/PPPPP1PP/NRNKRBBQ b - - 0 1[D]

[pgn][Event "Computer chess game"]
[Site "HP"]
[Date "2016.08.17"]
[Round "3"]
[White "Booot6_32"]
[Black "Stockfish 7 x64"]
[Result "1/2-1/2"]
[BlackElo "2828"]
[WhiteElo "3100"]
[TimeControl "5400"]
[SetUp "1"]
[FEN "nrnkrbbq/pppp2pp/8/8/5P2/8/PPPPP1PP/NRNKRBBQ b - - 0 1"]
[WhiteType "program"]
[BlackType "human"]

1. ... g5 2. g3 Bh6 3. Nab3 gxf4 4. Bd4 Bg7 5. Bxg7 Qxg7 6. gxf4 Nd6 7. Nd3
Qh6 8. e4 b6 9. Qf3 c5 10. Ne5 Nc7 11. Qg3 Bf7 12. Kc1 a5 13. Bh3 a4 14.
Bxd7 Rxe5 15. fxe5 Kxd7 16. Rf1 Rg8 17. Rxf7+ Nxf7 18. Qxg8 Nxe5 19. Qg3
Qf6 20. Nxc5+ bxc5 21. b3 axb3 22. axb3 Kc6 23. Qg2 Ne6 24. Qh3 Nf4 25. Qf5
Qxf5 26. exf5 Kd6 27. Ra1 Nd7 28. c3 Ke5 29. Ra7 Kd6 30. Ra5 Nd3+ 31. Kd1
Nf4 32. Ra4 Ke5 33. Ra7 Kd6 34. Kc2 h5 35. b4 cxb4 36. cxb4 Nd5 37. Kb3
N5b6 38. Ra5 h4 39. h3 Nd5 40. d4 Nf4 41. Ra6+ Kd5 42. Kc3 Nxh3 43. Rh6 Ng5
44. Rxh4 Ne4+ 45. Kd3 Nd6 46. Rh7 Nb8 47. Rh5 Nd7 48. Rg5 Nf6 49. Rg6 Nh5
50. Rg4 Nf6 51. Rg2 Nxf5 52. b5 Ne4 53. Rh2 Nf6 54. Rf2 Nxd4 55. Rxf6 Nxb5
56. Rf5+ Kc6 57. Rg5 Nd6 58. Kd4 Nb5+ 59. Ke5 Kc5 60. Rg1 Kc4 61. Rc1+ Nc3
62. Rc2 Kb3 63. Rd2 Kc4 64. Rd4+ Kc5 65. Rd7 Kc4 66. Rc7+ Kd3 67. Re7 Kc4
68. Rd7 Nb5 69. Rd1 Nc3 70. Rd4+ Kc5 71. Rf4 Nb5 72. Rf2 Kc4 73. Rh2 Nc3
74. Rh1 Kc5 75. Rh8 Kc4 76. Rc8+ Kd3 77. Rc5 Ne2 78. Rc7 Nc3 79. Ra7 Kc4
80. Re7 Nb5 81. Rg7 Kb4 82. Rg2 Kc4 83. Rb2 Kc5 84. Rc2+ Kb6 85. Kd5 Nc7+
86. Kc4 Kc6 87. Re2 Kd6 88. Rh2 Ke7 89. Kc5 Ne6+ 90. Kd5 Nf4+ 91. Ke5 Ng6+
92. Kf5 Nf8 93. Rh1 Kf7 94. Re1 Nd7 {White forfeits on time} 1/2-1/2 Thank you Graham, I followed your direction and here is the result between Booot vs Stockfish giving two pawns odds.[/pgn]
You started this game removing the two pawns most favorable to the odds-giver, with both bishops and a rook alreadyi active. I don't doubt that Komodo or Stockfish could give this handicap to Carlsen or Nakamura, but if they chose the pawns to be removed it would be a different story. I think two pawn handicap is not very suitable for chess960, because the size of the handicap is too dependent on which pawns are removed. With knight odds, that's only a minor issue. Without knowing the start position, any handicap involving pawns is too dependent on how the choice is made and what the start position is.
Here is a new Chess960 starting position ===>
[D]nrnkrbbq/1p1ppppp/8/8/8/8/PPPPPPPP/NRNKRBBQ w - - 0 1[D]
[pgn][Event "Computer chess game"]
[Site "HP"]
[Date "2016.08.18"]
[Round "?"]
[White "Booot6_32"]
[Black "Stockfish 7 x64"]
[Result "*"]
[BlackElo "3100"]
[WhiteElo "2828"]
[TimeControl "180"]
[SetUp "1"]
[FEN "nrnkrbbq/1p1ppppp/8/8/8/8/PPPPPPPP/NRNKRBBQ w - - 0 1"]


1. Nab3 e5 2. f4 e4 3. e3 Nc7 4. Nd4 Nd6 5. Nce2 f5 6. Ra1 g6 7. g4 Bg7 8.
gxf5 Nxf5 9. Bh3 Nxd4 10. Nxd4 Bd5 11. Qg2 Kc8 12. Qf1 Qf8 13. Kc1 Qd6 14.
Qf2 Ra8 15. Qh4 Rxa2 16. Rxa2 Bxa2 17. c3 Bxd4 18. cxd4 Bb3 19. Bxd7+ Qxd7
20. Qg4 Re6 21. Kb1 Qa4 22. Qxe6+ Nxe6 23. Rc1+ {White resigns} *

I decided to remove the two pawns in front of the Knights a7 and c7, and hwere is the result between Booot vs Stockfish [/pgn]
OK, this is a better test, it's a full two pawn handicap. Was this game/30' or some otheer TC, and were both engines single-cpu? In any case, it shows something I reported on here a while back, that Komodo can give material handicaps to surprisingly strong unrelated engines, but has much more trouble giving older or time-handicapped versions of itself such handicaps. Probably same is true for Stockfish. I concluded that time-handicapped self-play was a better predictor of human results vs Komodo than Komodo vs. weaker engines. Probably the reason is that unrelated engines have very different evals, which leads them to play much more complicated games than same eval, and complications favor the handicap giver.
Anyway, I suppose for chess 960 we can have two two-pawn handicaps; namely "human choice of pawns" and "engine choice of pawns". Obviously the first is a much larger handicap. But some positions will make the handicaps more severe than others.
I played a couple quick games of chess960 myself with Komodo at knight odds, and lost. It is now obvious to me that it is MUCH harder to win with knight odds at chess960 than at normal chess, and I am willing to try knight odds in 960 vs. IM or GM (not too strong GM). I agree with you now that even with choice of pawns Carlsen or Nakamura would be underdog at chess960 two pawn handicap 45' + 15". Not only is chess960 much more unfamiliar for humans, but also in normal chess the knights can develop towards the center without seriously inpeding the other pieces from developing, which is not generally true in chess960.
By the way, the convention is that when we say "knight odds" or "two pawn odds" or "pawn odds" it means the odds-giver has White; if he has Black you call it "knight and move" or "two pawns and move" etc.
This was on a single computer time control was game in 45 minutes per side. Now here is with a pawn odds in standard chess
[D]rnbqkbnr/ppppp1pp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1[D]
[pgn][Event "Computer chess game"]
[Site "HP"]
[Date "2016.08.18"]
[Round "?"]
[White "Booot6_32"]
[Black "Stockfish 7 x64"]
[Result "*"]
[BlackElo "3100"]
[WhiteElo "2828"]
[TimeControl "900"]
[SetUp "1"]
[FEN "rnbqkbnr/ppppp1pp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1"]
[WhiteType "human"]
[BlackType "human"]

1. e4 Nc6 2. d4 d5 3. Nc3 dxe4 4. d5 Ne5 5. Bb5+ Bd7 6. Bxd7+ Qxd7 7. Nge2
Nf6 8. O-O c6 9. dxc6 Qxc6 10. Nd4 Qc4 11. Be3 e6 12. a4 Bb4 13. Nce2 O-O
14. b3 Qc8 15. Bg5 Nh5 16. c3 Bd6 17. Qd2 Qe8 18. Rad1 h6 19. Be3 Qg6 20.
Kh1 Rad8 21. f4 exf3 22. gxf3 Bc5 23. Bf4 Bxd4 24. cxd4 Nxf3 25. Rxf3 e5
26. Qe3 exf4 27. Nxf4 Nxf4 28. Rxf4 Rxf4 29. Qxf4 Qc6+ 30. Kg1 Rf8 31. d5
Qc2 32. Qd2 Qxb3 33. Qd3 Qxa4 34. d6 Qd7 35. h3 b5 36. Ra1 b4 37. Rf1 Rd8
38. Re1 a5 39. Re7 Qc6 40. Qd4 Qc1+ 41. Kh2 Qg5 42. Re2 Qf5 43. Rg2 Rd7 44.
Re2 Kh7 45. Re7 Qc2+ 46. Kg1 Qg6+ 47. Kf2 Rxe7 48. dxe7 Qf7+ 49. Ke1 Qxe7+
50. Kd1 a4 51. Qd3+ Kh8 52. Qc2 a3 53. Qa4 Qe4 54. h4 Qb1+ 55. Kd2 a2 56.
Qe8+ {White resigns} [/pgn]

lkaufman
Posts: 4324
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: chess 960 engines vs humans

Post by lkaufman » Thu Aug 18, 2016 4:27 pm

Just one criticism: you should never run tests like this without some meaningful increment. Otherwise one engine may lose a very long game simply due to running out of time, either forfeiting or having to make instant moves and blunder. Perhaps a strong engine may have hardly ever been tested at sudden death, and it simply is much weaker at that level than with an increment. I think the 45' + 15" we use is close to optimum in terms of getting the best overall level of play for the likely amount of time it will take. Of couse 30' + 10" would also be fine. This 180 to 1 ratio for base time vs. increment is quite good, except for very fast games where the ratio should be smaller, for example one minute plus 0.5 or 0.6" is much better than 1 min plus 1/3 of a second. Anyway I doubt this affected the games you posted much.
Komodo rules!

User avatar
George
Posts: 682
Joined: Sun Oct 13, 2013 11:44 pm

Re: chess 960 engines vs humans

Post by George » Thu Aug 18, 2016 4:45 pm

lkaufman wrote:Just one criticism: you should never run tests like this without some meaningful increment. Otherwise one engine may lose a very long game simply due to running out of time, either forfeiting or having to make instant moves and blunder. Perhaps a strong engine may have hardly ever been tested at sudden death, and it simply is much weaker at that level than with an increment. I think the 45' + 15" we use is close to optimum in terms of getting the best overall level of play for the likely amount of time it will take. Of couse 30' + 10" would also be fine. This 180 to 1 ratio for base time vs. increment is quite good, except for very fast games where the ratio should be smaller, for example one minute plus 0.5 or 0.6" is much better than 1 min plus 1/3 of a second. Anyway I doubt this affected the games you posted much.
Now based on the last standard position with the f7 pawn removed and a program close to the strength of GM Nakamura Booot 2828, do you believe that the f7 pawn odds giving by your team with Komodo vs GM Nakamura was fair?
https://www.youtube.com/watch?v=PmOaSLJU7aQ

lkaufman
Posts: 4324
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: chess 960 engines vs humans

Post by lkaufman » Thu Aug 18, 2016 5:09 pm

George wrote:
lkaufman wrote:Just one criticism: you should never run tests like this without some meaningful increment. Otherwise one engine may lose a very long game simply due to running out of time, either forfeiting or having to make instant moves and blunder. Perhaps a strong engine may have hardly ever been tested at sudden death, and it simply is much weaker at that level than with an increment. I think the 45' + 15" we use is close to optimum in terms of getting the best overall level of play for the likely amount of time it will take. Of couse 30' + 10" would also be fine. This 180 to 1 ratio for base time vs. increment is quite good, except for very fast games where the ratio should be smaller, for example one minute plus 0.5 or 0.6" is much better than 1 min plus 1/3 of a second. Anyway I doubt this affected the games you posted much.
Now based on the last standard position with the f7 pawn removed and a program close to the strength of GM Nakamura Booot 2828, do you believe that the f7 pawn odds giving by your team with Komodo vs GM Nakamura was fair?

Well I already knew and reported well before that match that Komodo could give f7 odds successfully to other engines rated well above it, but had much more trouble doing the same with older or time/crippled Komodo versions. When engines play each other with a handicap they don't know that they are playing a stronger opponent and should avoid unnecessary complications, but humans do know this and avoid complications as much as possible. This makes it very hard to make predictions of engine vs human games based on engine vs engine games. The only way to tell what is "fair" for engine vs. human games is trial and error. I think that based on the games we have run so far vs. GMs that 2800 (close to Nakamura) is roughly the rating that should break even with Komodo givning f7 odds at 45' + 15". If I have time I might do an exact calculation later.
https://www.youtube.com/watch?v=PmOaSLJU7aQ
Komodo rules!

User avatar
George
Posts: 682
Joined: Sun Oct 13, 2013 11:44 pm

Re: chess 960 engines vs humans

Post by George » Fri Aug 19, 2016 3:14 pm

lkaufman wrote:
George wrote:
lkaufman wrote:Just one criticism: you should never run tests like this without some meaningful increment. Otherwise one engine may lose a very long game simply due to running out of time, either forfeiting or having to make instant moves and blunder. Perhaps a strong engine may have hardly ever been tested at sudden death, and it simply is much weaker at that level than with an increment. I think the 45' + 15" we use is close to optimum in terms of getting the best overall level of play for the likely amount of time it will take. Of couse 30' + 10" would also be fine. This 180 to 1 ratio for base time vs. increment is quite good, except for very fast games where the ratio should be smaller, for example one minute plus 0.5 or 0.6" is much better than 1 min plus 1/3 of a second. Anyway I doubt this affected the games you posted much.
Now based on the last standard position with the f7 pawn removed and a program close to the strength of GM Nakamura Booot 2828, do you believe that the f7 pawn odds giving by your team with Komodo vs GM Nakamura was fair?

Well I already knew and reported well before that match that Komodo could give f7 odds successfully to other engines rated well above it, but had much more trouble doing the same with older or time/crippled Komodo versions. When engines play each other with a handicap they don't know that they are playing a stronger opponent and should avoid unnecessary complications, but humans do know this and avoid complications as much as possible. This makes it very hard to make predictions of engine vs human games based on engine vs engine games. The only way to tell what is "fair" for engine vs. human games is trial and error. I think that based on the games we have run so far vs. GMs that 2800 (close to Nakamura) is roughly the rating that should break even with Komodo givning f7 odds at 45' + 15". If I have time I might do an exact calculation later.
https://www.youtube.com/watch?v=PmOaSLJU7aQ
Where Can I get one previous version older than the current and latest version of Komodo? I would like to test different Engines with different odds in Standard chess as well as Chess960

lkaufman
Posts: 4324
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: chess 960 engines vs humans

Post by lkaufman » Fri Aug 19, 2016 3:22 pm

George wrote:
lkaufman wrote:
George wrote:
lkaufman wrote:Just one criticism: you should never run tests like this without some meaningful increment. Otherwise one engine may lose a very long game simply due to running out of time, either forfeiting or having to make instant moves and blunder. Perhaps a strong engine may have hardly ever been tested at sudden death, and it simply is much weaker at that level than with an increment. I think the 45' + 15" we use is close to optimum in terms of getting the best overall level of play for the likely amount of time it will take. Of couse 30' + 10" would also be fine. This 180 to 1 ratio for base time vs. increment is quite good, except for very fast games where the ratio should be smaller, for example one minute plus 0.5 or 0.6" is much better than 1 min plus 1/3 of a second. Anyway I doubt this affected the games you posted much.
Now based on the last standard position with the f7 pawn removed and a program close to the strength of GM Nakamura Booot 2828, do you believe that the f7 pawn odds giving by your team with Komodo vs GM Nakamura was fair?

Well I already knew and reported well before that match that Komodo could give f7 odds successfully to other engines rated well above it, but had much more trouble doing the same with older or time/crippled Komodo versions. When engines play each other with a handicap they don't know that they are playing a stronger opponent and should avoid unnecessary complications, but humans do know this and avoid complications as much as possible. This makes it very hard to make predictions of engine vs human games based on engine vs engine games. The only way to tell what is "fair" for engine vs. human games is trial and error. I think that based on the games we have run so far vs. GMs that 2800 (close to Nakamura) is roughly the rating that should break even with Komodo givning f7 odds at 45' + 15". If I have time I might do an exact calculation later.
https://www.youtube.com/watch?v=PmOaSLJU7aQ
Where Can I get one previous version older than the current and latest version of Komodo? I would like to test different Engines with different odds in Standard chess as well as Chess960
If you bought a subscription to Komodo, all older versions are free. Komodo 8 is free for everyone at komodochess.com. If you bought Komooo 10.1 from us you are not supposed to get Komodo 10 or 9 for free, but if you send me an email along with a copy of your order I'll send whichever one you want to you free.
Komodo rules!

Post Reply