Komodo Dragon 2.6 released

[Leto]

Does UCI Elo + UCI LimitStrength work with the personalities?

[Chessqueen]

Does UCI Elo + UCI LimitStrength work with the personalities? Yes under Arena 3.5.1 you have to have UCI LimitStrenght Checked and the UCI Elo set to what human=rating strength you want otherwise it will NOT move instantly.

Merry Christmas and have an anticipated New Year everybody, I do NOT have a choice than to be happy while shoveling my Garage for 3 hours, we had a storm

[Chessqueen]

Does UCI Elo + UCI LimitStrength work with the personalities?

Yes under Arena 3.5.1 you have to have UCI LimitStrenght Checked and the UCI Elo set to what human=rating strength you want otherwise it will NOT move instantly. Here is an example. I took this position from Nakamura vs Carlsen game and wanted to know if it was going to end in a draw while they were playing, gave Naka 2836 and Carlsen 2850 Elo rating in Rapid and set Komodo Dragon to play it out under Arena by clicking Demo under game.

[pgn][Event "Computer chess game"]
[Date "2021.12.28"]
[Round "?"]
[White "Dragon-2.6-64bit-avx2"]
[Black "Dragon-2.6-64bit-avx2"]
[Result "1/2-1/2"]
[BlackElo "2850"]
[Time "19:30:46"]
[WhiteElo "2836"]
[TimeControl "900+10"]
[SetUp "1"]
[FEN "8/8/5pk1/R6p/5K1P/p5P1/r7/8 b - - 0 1"]
[Termination "normal"]
[PlyCount "71"]
[WhiteType "program"]
[BlackType "program"]

1. ... Ra1 2. Kf3 a2 3. Kg2 Kf7 4. Kh2 Ke6 5. Ra6+ Kd5 6. Ra5+ Kc4 7. Ra3
Kb4 8. Ra7 Kb3 9. Rb7+ Kc3 10. Ra7 Kd4 11. Ra4+ Kd3 12. Ra5 Ke3 13. Ra4 Kd3
14. Ra5 Ke2 15. Kg2 Ke3 16. Ra4 Kd3 17. Kh2 Ke3 18. Kg2 Kd2 19. Kh2 Kc2 20.
Ra7 Kc3 21. g4 hxg4 22. h5 Kb2 23. Rb7+ Kc3 24. Ra7 f5 25. h6 Rd1 26. h7
Rd2+ 27. Kg3 Rd3+ 28. Kf4 Rh3 29. Rxa2 Rxh7 30. Kxf5 g3 31. Ra3+ Kd2 32.
Rxg3 Re7 33. Ra3 Re8 34. Ra4 Ke2 35. Ra3 Re3 36. Rxe3+ Kxe3 {Insufficient
material} 1/2-1/2[/pgn]

[Chessqueen]

My opinion might not be correct, but probably other can agree with me, even if you set Komodo Dragon 2.6 to 2800 settings in Rapid it should beat Carlsen. My opinion is based on the only fact that human including Super GM and the best human player like Carlsen, once in a while makes an inferior move Not necessarily a blunder but not the best move and Komodo at a setting of 2800 in rapid will never blunder a minor piece or any piece at all, plus that being human and knowing that you are playing an engine is enough to make the human nervous.

I'm not saying that your conclusion is wrong, but the reason you give is not a valid argument because you could say the same thing if we set the elo to 2000, where it would still not make obvious piece blunders. The point is, if we run a match of (for example) Dragon with Elo=2500 vs 2500 rated GM in Rapid and get a tied score after 8 games, it doesn't matter if the engine wins two games by blunders while losing two games by being outplayed. The rating is supposed to be the rating a human needs to achieve an even score against it in a serious Rapid match. If I have set the values correctly, I would expect that in a match the human would more often get the advantage but would occasionally blunder. The engine will also "blunder" some advantages away, but they will at least be subtle blunders, not ones that a novice could see.

I agree with your last statement, but as far as human either they get over confident that they are winning or too nervous and top GM sometimes blunders terribly, which is why Komodo Dragon 2.6 should easily win a 8 games match at T/C 15 minutes 10 sec bonus . I will post this game where GM Svane Ramus was winning with 2 pawns extra versus GM Anish Giri and there was a mate in 2 moves, and if you give this position to any players rated between 1250 to 1450 they will not move their White King to h5? trapping it since they love to check and as long as they continue checking with their Queen ( Notice that the lower the rating of the players the more they love to Check and continue checking hoping to trade either Queens or check mate their opponents and eventually or by coincidence after several checks to the Black King they will see the 2 moves Mate that GM Ramus missed against GM Anish Giri )

If you are right about a top gm (or any gm) losing badly to dragon 2.6 with elo set equal to GM rating, in a Rapid match, it would just mean that I set param values too high. By definition, if I set them properly such a match should be a tossup.

I do NOT know about matching the human elo setting in UCI_ELO equal a GM of lets say 2580 thru 2600 and how well the param values are set either too low or too high, but I decided to match more than 30 games where I set the UCI_ELO = 2325 against Kingsfisher version 1.1.1 which is rated around 2315, but it seems that the param values are too low and Komodo Dragon 2.6 at that Elo setting lost every single game at T/C 15 minutes and 10 seconds bonus, I even tried 10 games at 5 minutes Blitz since Komodo Dragon 2.6 moves instantly and it also lost every game to Kingsfisher 1.1.1 http://ccrl.chessdom.com/ccrl/4040/cgi/ ... 1_1_64-bit

[Chessqueen]

Here is a game where Dumb rated 2296 versus Komodo Dragon 2.6 UCI_ELO = 2325

[pgn][Event "Computer chess game"]
[Site "MININT-UB2PIMJ"]
[Date "2021.12.29"]
[Round "?"]
[White "Dumb-win64"]
[Black "Dragon-2.6-64bit-avx2"]
[Result "1-0"]
[BlackElo "2315"]
[ECO "A06"]
[Opening "Reti Opening"]
[Time "09:31:36"]
[Variation "1...d5 2.e3"]
[WhiteElo "2296"]
[TimeControl "900+10"]
[Termination "normal"]
[PlyCount "65"]
[WhiteType "program"]
[BlackType "program"]

1. Nf3 d5 2. e3 Nc6 3. d4 Qd6 4. Nc3 a6 5. e4 dxe4 6. Nxe4 Qg6 7. Bd3 Qxg2
8. Rg1 Qh3 9. Rg3 Qd7 10. Nc5 Qd6 11. c3 Nf6 12. Qa4 Rb8 13. Be3 b5 14. Qc2
g6 15. a4 e5 16. axb5 axb5 17. Qe2 Nh5 18. Rg5 Qd5 19. Be4 Qd6 20. dxe5
Nxe5 21. Rxe5+ Be7 22. Rd1 O-O 23. Rxd6 Bxd6 24. Rxh5 gxh5 25. Bc2 Re8 26.
Ng5 h6 27. Qxh5 Rxe3+ 28. fxe3 hxg5 29. Qh6 b4 30. Bh7+ Kh8 31. Bg6+ Kg8
32. Qh7+ Kf8 33. Qxf7# 1-0[/pgn]

[lkaufman]

My opinion might not be correct, but probably other can agree with me, even if you set Komodo Dragon 2.6 to 2800 settings in Rapid it should beat Carlsen. My opinion is based on the only fact that human including Super GM and the best human player like Carlsen, once in a while makes an inferior move Not necessarily a blunder but not the best move and Komodo at a setting of 2800 in rapid will never blunder a minor piece or any piece at all, plus that being human and knowing that you are playing an engine is enough to make the human nervous.

I'm not saying that your conclusion is wrong, but the reason you give is not a valid argument because you could say the same thing if we set the elo to 2000, where it would still not make obvious piece blunders. The point is, if we run a match of (for example) Dragon with Elo=2500 vs 2500 rated GM in Rapid and get a tied score after 8 games, it doesn't matter if the engine wins two games by blunders while losing two games by being outplayed. The rating is supposed to be the rating a human needs to achieve an even score against it in a serious Rapid match. If I have set the values correctly, I would expect that in a match the human would more often get the advantage but would occasionally blunder. The engine will also "blunder" some advantages away, but they will at least be subtle blunders, not ones that a novice could see.

I agree with your last statement, but as far as human either they get over confident that they are winning or too nervous and top GM sometimes blunders terribly, which is why Komodo Dragon 2.6 should easily win a 8 games match at T/C 15 minutes 10 sec bonus . I will post this game where GM Svane Ramus was winning with 2 pawns extra versus GM Anish Giri and there was a mate in 2 moves, and if you give this position to any players rated between 1250 to 1450 they will not move their White King to h5? trapping it since they love to check and as long as they continue checking with their Queen ( Notice that the lower the rating of the players the more they love to Check and continue checking hoping to trade either Queens or check mate their opponents and eventually or by coincidence after several checks to the Black King they will see the 2 moves Mate that GM Ramus missed against GM Anish Giri )

If you are right about a top gm (or any gm) losing badly to dragon 2.6 with elo set equal to GM rating, in a Rapid match, it would just mean that I set param values too high. By definition, if I set them properly such a match should be a tossup.

I do NOT know about matching the human elo setting in UCI_ELO equal a GM of lets say 2580 thru 2600 and how well the param values are set either too low or too high, but I decided to match more than 30 games where I set the UCI_ELO = 2325 against Kingsfisher version 1.1.1 which is rated around 2315, but it seems that the param values are too low and Komodo Dragon 2.6 at that Elo setting lost every single game at T/C 15 minutes and 10 seconds bonus, I even tried 10 games at 5 minutes Blitz since Komodo Dragon 2.6 moves instantly and it also lost every game to Kingsfisher 1.1.1 http://ccrl.chessdom.com/ccrl/4040/cgi/ ... 1_1_64-bit

CCRL Rapid ratings may be pretty close to FIDE human ratings at Magnus Carlsen (2850) level, but at lower levels the CCRL ratings are much stronger than the FIDE ratings. That's why you saw that Dragon was a good match for humans with ratings up to 2300, but it is not a match for CCRL engines when the ratings are set to match. In other words, if you matched a 2300 human vs a CCRL engine with a 2300 Rapid rating in 15' + 10" chess, the human would get demolished. The human scale is more compressed than the engine scale. How much more compressed is the big question.

[Chessqueen]

This time I set Komodo Dragon UCI_ELO to 2325 versus Dumb rated 2296 and it still lost

[pgn][Event "Computer chess game"]
[Site "MININT-UB2PIMJ"]
[Date "2021.12.29"]
[Round "?"]
[White "Dragon-2.6-64bit-avx2"]
[Black "Dumb-win64"]
[Result "0-1"]
[BlackElo "2296"]
[ECO "C10"]
[Opening "French"]
[Time "09:48:11"]
[Variation "Rubinstein, 1.e4 e6 2.d4 d5 3.Nc3 dxe4 4.Nxe4"]
[WhiteElo "2325"]
[TimeControl "900+10"]
[Termination "normal"]
[PlyCount "97"]
[WhiteType "program"]
[BlackType "program"]

1. d4 d5 2. Nc3 e6 3. e4 dxe4 4. Nxe4 Nc6 5. c3 Qd5 6. Qe2 Nf6 7. Nxf6+
gxf6 8. Nh3 Bd7 9. Nf4 Qd6 10. Nh5 e5 11. dxe5 fxe5 12. Bg5 Be7 13. Rd1 Qg6
14. Bxe7 Kxe7 15. Nf4 Qg7 16. Qe3 Rae8 17. Be2 Kd8 18. Qh3 Re7 19. Nh5 Qg6
20. Bg4 Nd4 21. Nf6 Qxf6 22. O-O Ne6 23. Qe3 c5 24. Bxe6 Qxe6 25. Qxc5 Qb6
26. Qa3 Re6 27. Qa4 Qc7 28. Rfe1 Kc8 29. Qg4 Rg6 30. Qe2 Re8 31. f3 Qb6+
32. Kh1 Bc6 33. Rd2 Kb8 34. Qd3 a5 35. b4 axb4 36. cxb4 Qxb4 37. Qe3 f5 38.
a3 Qh4 39. Rc1 f4 40. Qb6 Rh6 41. Qg1 e4 42. fxe4 Rxe4 43. Rb2 Re3 44. a4
Rh3 45. Qa7+ Kxa7 46. Kg1 Rc3 47. Rxb7+ Bxb7 48. Rb1 Qxh2+ 49. Kf2 {White
resigns} *[/pgn]

[lkaufman]

This time I set Komodo Dragon UCI_ELO to 2325 versus Dumb rated 2296 and it still lost

[pgn][Event "Computer chess game"]
[Site "MININT-UB2PIMJ"]
[Date "2021.12.29"]
[Round "?"]
[White "Dragon-2.6-64bit-avx2"]
[Black "Dumb-win64"]
[Result "0-1"]
[BlackElo "2296"]
[ECO "C10"]
[Opening "French"]
[Time "09:48:11"]
[Variation "Rubinstein, 1.e4 e6 2.d4 d5 3.Nc3 dxe4 4.Nxe4"]
[WhiteElo "2325"]
[TimeControl "900+10"]
[Termination "normal"]
[PlyCount "97"]
[WhiteType "program"]
[BlackType "program"]

1. d4 d5 2. Nc3 e6 3. e4 dxe4 4. Nxe4 Nc6 5. c3 Qd5 6. Qe2 Nf6 7. Nxf6+
gxf6 8. Nh3 Bd7 9. Nf4 Qd6 10. Nh5 e5 11. dxe5 fxe5 12. Bg5 Be7 13. Rd1 Qg6
14. Bxe7 Kxe7 15. Nf4 Qg7 16. Qe3 Rae8 17. Be2 Kd8 18. Qh3 Re7 19. Nh5 Qg6
20. Bg4 Nd4 21. Nf6 Qxf6 22. O-O Ne6 23. Qe3 c5 24. Bxe6 Qxe6 25. Qxc5 Qb6
26. Qa3 Re6 27. Qa4 Qc7 28. Rfe1 Kc8 29. Qg4 Rg6 30. Qe2 Re8 31. f3 Qb6+
32. Kh1 Bc6 33. Rd2 Kb8 34. Qd3 a5 35. b4 axb4 36. cxb4 Qxb4 37. Qe3 f5 38.
a3 Qh4 39. Rc1 f4 40. Qb6 Rh6 41. Qg1 e4 42. fxe4 Rxe4 43. Rb2 Re3 44. a4
Rh3 45. Qa7+ Kxa7 46. Kg1 Rc3 47. Rxb7+ Bxb7 48. Rb1 Qxh2+ 49. Kf2 {White
resigns} *[/pgn]

In my Rapid testing Dragon set to 2300 is only about an even match with Snowy (rated about 2000 Rapid by ccrl), but around the 2850 level Dragon matches CCRL. This is based on the notion that 100 elo gap on ccrl is only about 60 elo gap in human ratings. Whether that is accurate is the question.

[Chessqueen]

This time I set Komodo Dragon UCI_ELO to 2325 versus Dumb rated 2296 and it still lost

[pgn][Event "Computer chess game"]
[Site "MININT-UB2PIMJ"]
[Date "2021.12.29"]
[Round "?"]
[White "Dragon-2.6-64bit-avx2"]
[Black "Dumb-win64"]
[Result "0-1"]
[BlackElo "2296"]
[ECO "C10"]
[Opening "French"]
[Time "09:48:11"]
[Variation "Rubinstein, 1.e4 e6 2.d4 d5 3.Nc3 dxe4 4.Nxe4"]
[WhiteElo "2325"]
[TimeControl "900+10"]
[Termination "normal"]
[PlyCount "97"]
[WhiteType "program"]
[BlackType "program"]

1. d4 d5 2. Nc3 e6 3. e4 dxe4 4. Nxe4 Nc6 5. c3 Qd5 6. Qe2 Nf6 7. Nxf6+
gxf6 8. Nh3 Bd7 9. Nf4 Qd6 10. Nh5 e5 11. dxe5 fxe5 12. Bg5 Be7 13. Rd1 Qg6
14. Bxe7 Kxe7 15. Nf4 Qg7 16. Qe3 Rae8 17. Be2 Kd8 18. Qh3 Re7 19. Nh5 Qg6
20. Bg4 Nd4 21. Nf6 Qxf6 22. O-O Ne6 23. Qe3 c5 24. Bxe6 Qxe6 25. Qxc5 Qb6
26. Qa3 Re6 27. Qa4 Qc7 28. Rfe1 Kc8 29. Qg4 Rg6 30. Qe2 Re8 31. f3 Qb6+
32. Kh1 Bc6 33. Rd2 Kb8 34. Qd3 a5 35. b4 axb4 36. cxb4 Qxb4 37. Qe3 f5 38.
a3 Qh4 39. Rc1 f4 40. Qb6 Rh6 41. Qg1 e4 42. fxe4 Rxe4 43. Rb2 Re3 44. a4
Rh3 45. Qa7+ Kxa7 46. Kg1 Rc3 47. Rxb7+ Bxb7 48. Rb1 Qxh2+ 49. Kf2 {White
resigns} *[/pgn]

In my Rapid testing Dragon set to 2300 is only about an even match with Snowy (rated about 2000 Rapid by ccrl), but around the 2850 level Dragon matches CCRL. This is based on the notion that 100 elo gap on ccrl is only about 60 elo gap in human ratings. Whether that is accurate is the question.

What about in the 2600 just in case you match it against a human 4 games g1 knight odds and 4 games matching the UCI ELO

[lkaufman]

This time I set Komodo Dragon UCI_ELO to 2325 versus Dumb rated 2296 and it still lost

[pgn][Event "Computer chess game"]
[Site "MININT-UB2PIMJ"]
[Date "2021.12.29"]
[Round "?"]
[White "Dragon-2.6-64bit-avx2"]
[Black "Dumb-win64"]
[Result "0-1"]
[BlackElo "2296"]
[ECO "C10"]
[Opening "French"]
[Time "09:48:11"]
[Variation "Rubinstein, 1.e4 e6 2.d4 d5 3.Nc3 dxe4 4.Nxe4"]
[WhiteElo "2325"]
[TimeControl "900+10"]
[Termination "normal"]
[PlyCount "97"]
[WhiteType "program"]
[BlackType "program"]

1. d4 d5 2. Nc3 e6 3. e4 dxe4 4. Nxe4 Nc6 5. c3 Qd5 6. Qe2 Nf6 7. Nxf6+
gxf6 8. Nh3 Bd7 9. Nf4 Qd6 10. Nh5 e5 11. dxe5 fxe5 12. Bg5 Be7 13. Rd1 Qg6
14. Bxe7 Kxe7 15. Nf4 Qg7 16. Qe3 Rae8 17. Be2 Kd8 18. Qh3 Re7 19. Nh5 Qg6
20. Bg4 Nd4 21. Nf6 Qxf6 22. O-O Ne6 23. Qe3 c5 24. Bxe6 Qxe6 25. Qxc5 Qb6
26. Qa3 Re6 27. Qa4 Qc7 28. Rfe1 Kc8 29. Qg4 Rg6 30. Qe2 Re8 31. f3 Qb6+
32. Kh1 Bc6 33. Rd2 Kb8 34. Qd3 a5 35. b4 axb4 36. cxb4 Qxb4 37. Qe3 f5 38.
a3 Qh4 39. Rc1 f4 40. Qb6 Rh6 41. Qg1 e4 42. fxe4 Rxe4 43. Rb2 Re3 44. a4
Rh3 45. Qa7+ Kxa7 46. Kg1 Rc3 47. Rxb7+ Bxb7 48. Rb1 Qxh2+ 49. Kf2 {White
resigns} *[/pgn]

In my Rapid testing Dragon set to 2300 is only about an even match with Snowy (rated about 2000 Rapid by ccrl), but around the 2850 level Dragon matches CCRL. This is based on the notion that 100 elo gap on ccrl is only about 60 elo gap in human ratings. Whether that is accurate is the question.

What about in the 2600 just in case you match it against a human 4 games g1 knight odds and 4 games matching the UCI ELO

I expect we'll have matches vs. GMs soon at knight odds and also with the UCI Elo option, though probably they will be separate events against different opponents. Knight odds is an ongoing challenge to see if and when we can draw or win a match with a GM in Rapid; the Elo match would only be for the purpose of seeing whether I got the parameters set properly.