Komodo Dragon Vs. GM Perelshteyn Where is the Link ?

[lkaufman]

I already finished testing 1000 games with each contempt = 80 and 85 against Gaviota-0.86-win64, and contempt 85 is only slightly better by 2%

I decided to run another pair of 1000 games with each contempt = 80 and 85 against Gaviota-0.86-win64, and this time contempt 80 performed slightly better by 10%

I ran 1000 games with Contempt = 85 (using latest dev. version and other optimized settings, one thread, 2' + 1") against Gaviota-0.86-win64, and got the best result yet (-4 elo, compared with -10 elo for Contempt = 80 with same other settings). I'm using the ChrisW knight odds book to get the needed variety. Are you also doing that, or if not what are you doing to avoid getting the same openings game after game? Also, when you say 10% better, do you mean 10 percentage points better? That would be an enormous swing, 70 elo, totally beyond anything plausible for a tiny change to Contempt.

Oh Boy I did NOT checked how many repeated games I got out of the 1000, but Now we are in the right track having the best Contempt, and with your next upgraded version of Komodo and to only play 2 games with b1 Knight Missing and 4 games with the g1, I believe that a 2510 rated GMs will have a hard time against Komodo Dragon xxx.x. Another thing that can be done is to take all the losing opening out of ChristW knight odds book and only leave the opening leading to a win. You already have played enough games to be able to filter all the losing Knight odds openings of ChrisW knight odds book against Gaviota, and to keep only the openings that lead to a win

The ChrisW book is for engine testing and has no relation to my book used for playing humans.

I thought that you used the ChrisW book when Komodo played Versus Chris W and he was making a move per day

I used no book for that game. My book doesn’t even cover 1...Nf6 at knight odds.

I suppose that by the next GM Knight odds challenge against Komodo Dragon, it should be ready to challenge at least a 2525 rated players after you have chosen the best Opening that lead to a win

All openings end in dead lost positions for White at knight odds with proper play by Black. All we can do with the book is to aim for positions that are harder to convert than others.

So the best that we can hope for is that Komodo Dragon can at least win 2 games out of 6 and draw at least 2 games versus a 2550 rated players in 5 months from now.

Well, even that would be quite optimistic if we just improve the Elo of Dragon and tweak the book and settings. But I'm hopeful that we will find a way to make Dragon set much greater practical problems for the GM. There are many ideas for doing this, it's just a question of finding one that has major benefits, not just a few elo. No one knows what the upper limit might be, just that it will require clever ideas, not just parameter tuning.

[Uri Blass]

Everybody talk about improvement in Komodo Dragon to get a better score.
What about improvement of human players?

I do not think that human players are static target and it is possible to write a software to teach them to play better in preventing practical chances of the opponent for a draw or for a win when the opponent is without a knight.

[AdminX]

Everybody talk about improvement in Komodo Dragon to get a better score.
What about improvement of human players?

I do not think that human players are static target and it is possible to write a software to teach them to play better in preventing practical chances of the opponent for a draw or for a win when the opponent is without a knight.

I am interested in using Dragon in the new Fritz 18 interface, it could be a match made in heaven for human players who wish to improve their game.

[Chessqueen]

Everybody talk about improvement in Komodo Dragon to get a better score.
What about improvement of human players?

I do not think that human players are static target and it is possible to write a software to teach them to play better in preventing practical chances of the opponent for a draw or for a win when the opponent is without a knight.

I do believe that in the case of GM Perelshteyn that he gained all the knowledge that he was going to get on how to beat Komodo Dragon, version 2.5.1, but if Mr. Kaufman get another match against him a week before he releases the Next version lets say version 3.0 all that he will have gained knowledge is based on version 2.5.1, Plus for the previous match Mr. Kaufman used contempt 125 instead of contempt = 85 and with different Openings tweaked more in the direction that are harder to convert for a human, and there are many ideas of finding one that has major benefits with much clever ideas.

[lkaufman]

Everybody talk about improvement in Komodo Dragon to get a better score.
What about improvement of human players?

I do not think that human players are static target and it is possible to write a software to teach them to play better in preventing practical chances of the opponent for a draw or for a win when the opponent is without a knight.

Surely with enough analysis of knight odds openings using top engines and memorizing the best lines, grandmasters will score better than otherwise. My own results playing dragon at knight odds are better than they should be (roughly equal draws and losses with an occasional win) based on my current ratings due to having played and analyzed knight odds so much. But it is incredible that we are even talking about GMs needing to memorize computer analysis to beat Dragon at knight odds; the initial position is resignable by normal criteria! It is enough for me if we can defeat an active (non-elderly) GM at knight odds in a Rapid match, I don't feel the need to prove that we can do so against one who devotes hundreds of hours studying the handicap.

[Chessqueen]

Everybody talk about improvement in Komodo Dragon to get a better score.
What about improvement of human players?

I do not think that human players are static target and it is possible to write a software to teach them to play better in preventing practical chances of the opponent for a draw or for a win when the opponent is without a knight.

Surely with enough analysis of knight odds openings using top engines and memorizing the best lines, grandmasters will score better than otherwise. My own results playing dragon at knight odds are better than they should be (roughly equal draws and losses with an occasional win) based on my current ratings due to having played and analyzed knight odds so much. But it is incredible that we are even talking about GMs needing to memorize computer analysis to beat Dragon at knight odds; the initial position is résignable by normal criteria! It is enough for me if we can defeat an active (non-elderly) GM at knight odds in a Rapid match, I don't feel the need to prove that we can do so against one who devotes hundreds of hours studying the handicap.

I was wondering if Komodo Dragon can give a Knight plus any pawn odds, for example g1 Knight plus b2, or c2 pawn but not the the f2 pawn to a 2300 rated player and draw a 6 games match at 15'+10"? Does Komodo Dragon has any chance of winning at least 1 game out of 6


[fen]rnbqkbnr/pppppppp/8/8/8/8/PP1PPPPP/RNBQKB1R w KQkq - 0 1

[lkaufman]

Everybody talk about improvement in Komodo Dragon to get a better score.
What about improvement of human players?

I do not think that human players are static target and it is possible to write a software to teach them to play better in preventing practical chances of the opponent for a draw or for a win when the opponent is without a knight.

Surely with enough analysis of knight odds openings using top engines and memorizing the best lines, grandmasters will score better than otherwise. My own results playing dragon at knight odds are better than they should be (roughly equal draws and losses with an occasional win) based on my current ratings due to having played and analyzed knight odds so much. But it is incredible that we are even talking about GMs needing to memorize computer analysis to beat Dragon at knight odds; the initial position is résignable by normal criteria! It is enough for me if we can defeat an active (non-elderly) GM at knight odds in a Rapid match, I don't feel the need to prove that we can do so against one who devotes hundreds of hours studying the handicap.

I was wondering if Komodo Dragon can give a Knight plus any pawn odds, for example g1 Knight plus b2, or c2 pawn but not the the f2 pawn to a 2300 rated player and draw a 6 games match at 15'+10"? Does Komodo Dragon has any chance of winning at least 1 game out of 6


[fen]rnbqkbnr/pppppppp/8/8/8/8/PP1PPPPP/RNBQKB1R w KQkq - 0 1

The b or c pawn adds roughly 250 elo to the handicap at the level we are talking about, so since we seem to be currently at about 2450 FIDE when giving knight odds to humans in Rapid, a 2200 FIDE rated human would be a fair opponent at knight plus b2 or c2 pawn, which is not a lot less than rook odds. Probably knight plus f2 is about the same as rook odds. Of course we might win 1 out of 6 from a 2300, just as a 2200 human might win 1 out of six (plus some draws) against a 2300 with no handicap. Actually according to engine vs. engine tests I've run, the difference between knight odds and rook odds was only 235 elo, so maybe knight plus b2 or c2 is roughly the same as rook odds, which of course is a much more normal game.

[Chessqueen]

Kingfisher is about 2300 by CCRL at Contempt = 85 Komodo drew

[pgn][Event "Computer chess game"]
[Site "MININT-UB2PIMJ"]
[Date "2021.11.18"]
[Round "1"]
[White "Dragon-2.5-64bit-avx2"]
[Black "Kingfisher.v1.0"]
[Result "1/2-1/2"]
[BlackElo "2300"]
[Time "19:23:07"]
[WhiteElo "2300"]
[TimeControl "180+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PP1PPPPP/RNBQKB1R w KQkq - 1 1"]
[Termination "normal"]
[PlyCount "106"]
[WhiteType "program"]
[BlackType "program"]

1. d4 d5 2. a3 Nc6 3. f4 Bf5 4. e3 e6 5. Qb3 Rb8 6. Be2 Qh4+ 7. g3 Qh3 8.
Bf1 Qg4 9. Bb5 Ne7 10. Nd2 a6 11. Be2 Qh3 12. Qc3 g6 13. b4 Bg7 14. Bb2 f6
15. Bf1 Qh5 16. Bg2 O-O 17. a4 b5 18. a5 g5 19. O-O Rbd8 20. Rae1 gxf4 21.
Rxf4 e5 22. Rh4 Qg5 23. Bf3 Rd6 24. Bc1 Bg6 25. Bd1 h5 26. Nb3 Bf7 27. Rf1
f5 28. Rh3 Qh6 29. Kh1 Kh7 30. Nc5 e4 31. Nxa6 Nxa5 32. bxa5 Rxa6 33. Ba3
Rc6 34. Qe1 Bf6 35. g4 Ra6 36. Qc3 fxg4 37. Bxg4 Bg6 38. Be2 c6 39. Rg1 h4
40. Bd6 Rfa8 41. Bf4 Qf8 42. Bg4 Nf5 43. Qe1 Bd8 44. Bc7 Bxc7 45. Bxf5 Qxf5
46. Qxh4+ Kg7 47. Qh7+ Kf6 48. Qxc7 Ke6 49. Rh6 Qf3+ 50. Rg2 Qf1+ 51. Rg1
Qf3+ 52. Rg2 Qf1+ 53. Rg1 Qf3+ {3-fold repetition} 1/2-1/2[/pgn]

[Chessqueen]

I did not wanted to continue the match but it is equal as you stated

[pgn][Event "Computer chess game"]
[Site "MININT-UB2PIMJ"]
[Date "2021.11.18"]
[Round "?"]
[White "Dragon-2.5-64bit-avx2"]
[Black "Dragon-2.5-64bit-avx2"]
[Result "*"]
[BlackElo "2100"]
[Time "22:30:25"]
[WhiteElo "2100"]
[TimeControl "180+2"]
[SetUp "1"]
[FEN "1nbqkbnr/pppppppp/8/8/8/8/PPPPP1PP/RNBQKB1R w KQk - 1 1"]
[Termination "unterminated"]
[PlyCount "73"]
[WhiteType "program"]
[BlackType "program"]

1. d4 e5 2. dxe5 Qh4+ 3. g3 Qe4 4. Rg1 Qxe5 5. Bg2 Nc6 6. Rf1 Bc5 7. Nc3
Bb4 8. Rf3 Nge7 9. Bf4 Qc5 10. Qd3 Nd4 11. Rf2 Bxc3+ 12. bxc3 Ne6 13. Be3
Qa5 14. Qc4 d6 15. Qb4 Qe5 16. Bf4 Nxf4 17. gxf4 Qe3 18. Qd4 Nf5 19. Qd3 b6
20. Be4 Qc5 21. Bxf5 Qxf5 22. O-O-O O-O 23. e4 Qc5 24. Rg2 Re8 25. Kb2 g6
26. Re1 Bd7 27. Rd2 Qc6 28. Rdd1 Bg4 29. Ra1 Qa4 30. a3 Be6 31. Rad1 Bg4
32. Rd2 Bd7 33. h4 Kf8 34. Re3 Qb5+ 35. Qxb5 Bxb5 36. Rd4 Re7 37. c4 *[/pgn]

[Chessqueen]

I am almost positive that Komodo dragon can give Carlsen, MVL or Caruana the f2 pawn Odds at T/C 15'+10"

[pgn][Event "Computer chess game"]
[Site "MININT-UB2PIMJ"]
[Date "2021.11.19"]
[Round "?"]
[White "Dragon-2.5-64bit-avx2"]
[Black "Nemorino_6.00_win64_avx2"]
[Result "1/2-1/2"]
[BlackElo "2000"]
[Time "12:35:15"]
[WhiteElo "3600"]
[TimeControl "180+3"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PP1PPPPP/RNBQKBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "329"]
[WhiteType "program"]
[BlackType "program"]

1. d4 Nf6 2. Nc3 d5 3. Bf4 c5 4. e3 cxd4 5. exd4 a6 6. Nf3 Nc6 7. Ne5 e6 8.
Be2 Bd7 9. O-O Rc8 10. Re1 Be7 11. Qd3 Qb6 12. Qg3 g6 13. Rad1 Qxb2 14. Rb1
Qa3 15. Nxd7 Nxd7 16. Nxd5 Qxg3 17. hxg3 exd5 18. Rxb7 Nxd4 19. Bxa6 Rc5
20. a4 Ne6 21. Bb5 Rxb5 22. axb5 Bc5 23. Rc7 Kd8 24. Rc6 Re8 25. Bd6 d4 26.
Ra6 Bxd6 27. Rxd6 h5 28. Rd5 Kc7 29. Rxd4 Nxd4 30. Rxe8 Nxb5 31. Re7 Nd6
32. Kh2 Kd8 33. Re1 Nf6 34. f3 Kd7 35. Rd1 Ke7 36. Rd4 Nf5 37. Ra4 Kf8 38.
Ra8+ Kg7 39. Ra4 Ne3 40. Kg1 Kf8 41. Ra6 Ke7 42. Ra4 Nfd5 43. Kf2 Nd1+ 44.
Kg1 N1c3 45. Ra7+ Ke6 46. Ra6+ Ke5 47. Ra7 Kf6 48. Ra6+ Kg7 49. Ra3 Nb5 50.
Ra5 Nbc7 51. Rc5 Kf8 52. Kh2 Ke7 53. Kg1 Kd6 54. Rc4 Ne6 55. Rh4 Nf6 56.
Kh2 Nc5 57. Rc4 Kd5 58. Rb4 Ke6 59. Kh3 Nd5 60. Rb8 Nd7 61. Re8+ Ne7 62.
Ra8 Ne5 63. Ra6+ Kf5 64. Ra5 Kf6 65. Ra6+ Kg7 66. Ra5 f6 67. Ra7 Kf7 68.
Kh4 Kf8 69. Ra8+ Kg7 70. Ra7 Kf7 71. Rb7 Ke8 72. Rb8+ Kd7 73. Rb7+ Kd6 74.
Rb6+ N5c6 75. g4 hxg4 76. Kxg4 Nd5 77. Ra6 Nb4 78. Ra3 Ke5 79. Ra8 Nd5 80.
Re8+ Nce7 81. Kg3 Kd4 82. Ra8 Nf5+ 83. Kf2 g5 84. Ra4+ Kd3 85. Ra6 Nfe3 86.
Ra3+ Kd2 87. Ra2+ Nc2 88. Ra6 Kd3 89. Rd6 Kd4 90. Rd8 Nb4 91. Kg3 Nc6 92.
Ra8 Nce7 93. Kg4 Ne3+ 94. Kg3 N7d5 95. Kf2 Nd1+ 96. Ke2 N5c3+ 97. Ke1 Ne3
98. Kf2 Kd3 99. Ra6 Ned5 100. Ra8 Nd1+ 101. Ke1 N1e3 102. Kf2 Nc2 103. Rd8
Kd4 104. Re8 Ncb4 105. Kg3 Nd3 106. Re4+ Kc5 107. Kg4 Nc1 108. Kf5 Nb3 109.
Kg6 Nd4 110. Re8 Nc6 111. Kf5 Kc4 112. Re4+ Kb5 113. Kg6 Kc5 114. Kf5 Nb8
115. Re6 Nd7 116. Ra6 N7b6 117. g3 Kc6 118. Ra1 Kd6 119. Re1 Kd7 120. Re6
Nc4 121. Rxf6 Nxf6 122. Kxf6 Kc6 123. Kxg5 Nd2 124. Kg4 Kc7 125. Kf4 Nc4
126. Ke4 Kd6 127. Kd4 Nb6 128. g4 Ke7 129. f4 Kf6 130. Ke4 Na4 131. Kf3 Nc5
132. Kg2 Na6 133. f5 Kf7 134. g5 Nc5 135. g6+ Kf6 136. Kh3 Kg7 137. Kh4 Kf6
138. Kh5 Nd7 139. Kh6 Kxf5 140. Kg7 Ke4 141. Kg8 Kd4 142. g7 Ke5 143. Kh7
Kf4 144. Kg6 Ne5+ 145. Kf6 Ng4+ 146. Ke7 Nh6 147. Kd7 Ke3 148. Kc6 Kd2 149.
Kb5 Kc1 150. Kb4 Kb1 151. Kb3 Ka1 152. Kc2 Ka2 153. Kd3 Ka1 154. Ke4 Kb1
155. Kf4 Ka1 156. Kg5 Ng8 157. Kh4 Kb1 158. Kg3 Ka1 159. Kf2 Kb1 160. Kg1
Ka1 161. Kf1 Kb1 162. Ke1 Ka1 163. Kf1 Kb1 164. Ke1 Ka1 165. Kf1 {3-fold
repetition} 1/2-1/2[/pgn]