Komodo Dragon vs. GM Perelshteyn, knight odds match.

[Chessqueen]

Revisiting my thoughts on this match, my opinion has now changed given that the match does not start until Nov 6th and that the GM does have a version of Komodo to practice with, I now think he should have good chances to win the match. I don't believe I gave enough thought to this little detail.

Since he is practicing with Komodo Dragon 2.5.1, he only need ChrisW opening book and Mr. Kaufman 32 core Threadripper, but I strongly recommend Mr. Kaufman to set his Komodo Dragon xxx.x to play at (5'+2") instead of (3' + 2")

I've run enough simulations to say that using more than 16 threads or more than 3' +2" (with 16 threads) is totally useless for knight odds. Imagine if we had a 32 man tablebase (from a million years in the future!). How would that help us in a knight odds match? Sure, if we got down to less than a pawn deficit in a game it might see a weird draw that normal search can't find, but against a knight-odds human opponent there is no point in assuming that he will play perfectly, it's better not to see every little detail in the far future that he will never even think of. If we get within drawing range we will surely not lose even if we play imperfectly, and if we reach a position where a win is possible we will probably win even if we miss the first win. Knight odds chess requires a different mental model than normal chess. Also, the ChrisW opening book is of no use for a human preparing, it is only for getting variety for computer vs computer matches. The positions (mostly) won't actually occur in real knight odds games.

If that is the case how come you will use a 32 core Threadripper instead of a 16 threads >

If we go with 16 threads then indeed I could use my fast 16 core machine instead. But everything is set up on my threadripper now, so there is no advantage to switching, and I might decide to use maybe 24 threads, depending on testing.

By now you probably have done several testing with 16 threads and with 24 threads; which seems to be more effective at Knight Odds against an engine of similar strength Gaviota-win32-0.84 = GM Perelshteyn ?

So far it seems 16 threads is best for this, but there are many ways to limit the search, by time, by depth, by Skill level, or by threads. So it's not a simple question, I'm trying to find the best combination. I'm hoping that we can draw the match, but I'm not predicting that.

So in the best case scenario all that you are hoping for is to get a draw with Knight Odds against a 2500+ rated player https://www.chess.com/players/eugene-perelshteyn

You mean a drawn match, not a draw. Of course I hope to win the match, just as I hoped to win the World Senior Championship in 2008, but didn't expect to win it. But I did win it, and it's possible we might win this match; it just wouldn't be a good bet in my opinion.

Why did you decided to let him use the latest release of Komodo Dragon, unless you have a much better version that plays better against human with Knight Odds ? I still do NOT understand why Komodo Dragon does not play better with more than 16 threads with Knight Odds, Is there also a limit as far as performing better with more than 7 pieces endgame for future Knight Odds matches

We just gave him Dragon 2.5.1, which he could buy anyway. We do have a better version, but not much better. With normal settings, in standard mode, Dragon won't do any better with big tablebases or more cores or more time. With MCTS mode, it does better with more time or more threads, but I'm not sure whether the time limit we are giving ourselves for this match is long enough for MCTS to do better than standard mode. However, I'm still trying to find ways to utilize the full power of the machine for this match, and I'm reasonably hopeful that I'll find a better solution than just letting it run in normal mode with 16 threads.

I can not wait to see if Komodo Dragon xxx.x will be able to get at least a drawn match with Knight odds in 2 days. I suppose that if a GM like GM Eugene Perelshteyn play 100 games at Knight Odds vs Komodo Dragon 2.5.1 he will benefit more than if Gaviota-win32-0.84 play 1000 games against Komodo Dragon 2.5.1 at Knight Odds which clearly tell me that human GMs benefit or learn more than similar engines

[lkaufman]

Revisiting my thoughts on this match, my opinion has now changed given that the match does not start until Nov 6th and that the GM does have a version of Komodo to practice with, I now think he should have good chances to win the match. I don't believe I gave enough thought to this little detail.

Since he is practicing with Komodo Dragon 2.5.1, he only need ChrisW opening book and Mr. Kaufman 32 core Threadripper, but I strongly recommend Mr. Kaufman to set his Komodo Dragon xxx.x to play at (5'+2") instead of (3' + 2")

I've run enough simulations to say that using more than 16 threads or more than 3' +2" (with 16 threads) is totally useless for knight odds. Imagine if we had a 32 man tablebase (from a million years in the future!). How would that help us in a knight odds match? Sure, if we got down to less than a pawn deficit in a game it might see a weird draw that normal search can't find, but against a knight-odds human opponent there is no point in assuming that he will play perfectly, it's better not to see every little detail in the far future that he will never even think of. If we get within drawing range we will surely not lose even if we play imperfectly, and if we reach a position where a win is possible we will probably win even if we miss the first win. Knight odds chess requires a different mental model than normal chess. Also, the ChrisW opening book is of no use for a human preparing, it is only for getting variety for computer vs computer matches. The positions (mostly) won't actually occur in real knight odds games.

If that is the case how come you will use a 32 core Threadripper instead of a 16 threads >

If we go with 16 threads then indeed I could use my fast 16 core machine instead. But everything is set up on my threadripper now, so there is no advantage to switching, and I might decide to use maybe 24 threads, depending on testing.

By now you probably have done several testing with 16 threads and with 24 threads; which seems to be more effective at Knight Odds against an engine of similar strength Gaviota-win32-0.84 = GM Perelshteyn ?

So far it seems 16 threads is best for this, but there are many ways to limit the search, by time, by depth, by Skill level, or by threads. So it's not a simple question, I'm trying to find the best combination. I'm hoping that we can draw the match, but I'm not predicting that.

So in the best case scenario all that you are hoping for is to get a draw with Knight Odds against a 2500+ rated player https://www.chess.com/players/eugene-perelshteyn

You mean a drawn match, not a draw. Of course I hope to win the match, just as I hoped to win the World Senior Championship in 2008, but didn't expect to win it. But I did win it, and it's possible we might win this match; it just wouldn't be a good bet in my opinion.

Why did you decided to let him use the latest release of Komodo Dragon, unless you have a much better version that plays better against human with Knight Odds ? I still do NOT understand why Komodo Dragon does not play better with more than 16 threads with Knight Odds, Is there also a limit as far as performing better with more than 7 pieces endgame for future Knight Odds matches

We just gave him Dragon 2.5.1, which he could buy anyway. We do have a better version, but not much better. With normal settings, in standard mode, Dragon won't do any better with big tablebases or more cores or more time. With MCTS mode, it does better with more time or more threads, but I'm not sure whether the time limit we are giving ourselves for this match is long enough for MCTS to do better than standard mode. However, I'm still trying to find ways to utilize the full power of the machine for this match, and I'm reasonably hopeful that I'll find a better solution than just letting it run in normal mode with 16 threads.

I can not wait to see if Komodo Dragon xxx.x will be able to get at least a drawn match with Knight odds in 2 days. I suppose that if a GM like GM Eugene Perelshteyn play 100 games at Knight Odds vs Komodo Dragon 2.5.1 he will benefit more than if Gaviota-win32-0.84 play 1000 games against Komodo Dragon 2.5.1 at Knight Odds which clearly tell me that human GMs benefit or learn more than similar engines

Yes, but if we could update dragon each day based on what we learned dragon would also get better at giving knight odds, so we don’t know which side would do better after 50 games.

[AdminX]

Our next human opponent for a handicap match with Komodo Dragon will be GM Eugene Perelshteyn, born 1980, current FIDE 2494, peak FIDE 2555 in 2008. He is a chess pro, developing a course for chess.com and coaching. The handicap will be full knight odds (alternating b1/g1), standard scoring. Time control as usual Rapid 15' + 10"; however in a new twist Dragon will be set to play as if it were blitz (3' + 2"), so five to one time odds on top of the knight odds (clocks won't show it, but it will be obvious that Dragon is playing blitz)! Games to be played on chess.com on Nov. 6 starting 11 a.m. Eastern Time. Six game match, total playing time shouldn't exceed 3 hours, total match time probably not more than four hours with time for brief comments after each game. Perelshteyn streaming, but due to the time odds he'll be on move most of the time so may not comment much during actual play.

I just set a reminder for tomorrows game.

[lkaufman]

Our next human opponent for a handicap match with Komodo Dragon will be GM Eugene Perelshteyn, born 1980, current FIDE 2494, peak FIDE 2555 in 2008. He is a chess pro, developing a course for chess.com and coaching. The handicap will be full knight odds (alternating b1/g1), standard scoring. Time control as usual Rapid 15' + 10"; however in a new twist Dragon will be set to play as if it were blitz (3' + 2"), so five to one time odds on top of the knight odds (clocks won't show it, but it will be obvious that Dragon is playing blitz)! Games to be played on chess.com on Nov. 6 starting 11 a.m. Eastern Time. Six game match, total playing time shouldn't exceed 3 hours, total match time probably not more than four hours with time for brief comments after each game. Perelshteyn streaming, but due to the time odds he'll be on move most of the time so may not comment much during actual play.

I just set a reminder for tomorrows game.

Based on unofficial test games, I believe that the break-even human FIDE rating for knight odds with dragon in somewhere in the 2400s for Rapid and in the 2600s for normal blitz (like 3' + 2" or so). Since Eugene is at the top of that Rapid range, he is the favorite but hopefully it will be competitive.

[Chessqueen]

Our next human opponent for a handicap match with Komodo Dragon will be GM Eugene Perelshteyn, born 1980, current FIDE 2494, peak FIDE 2555 in 2008. He is a chess pro, developing a course for chess.com and coaching. The handicap will be full knight odds (alternating b1/g1), standard scoring. Time control as usual Rapid 15' + 10"; however in a new twist Dragon will be set to play as if it were blitz (3' + 2"), so five to one time odds on top of the knight odds (clocks won't show it, but it will be obvious that Dragon is playing blitz)! Games to be played on chess.com on Nov. 6 starting 11 a.m. Eastern Time. Six game match, total playing time shouldn't exceed 3 hours, total match time probably not more than four hours with time for brief comments after each game. Perelshteyn streaming, but due to the time odds he'll be on move most of the time so may not comment much during actual play.

I just set a reminder for tomorrows game.

Based on unofficial test games, I believe that the break-even human FIDE rating for knight odds with dragon in somewhere in the 2400s for Rapid and in the 2600s for normal blitz (like 3' + 2" or so). Since Eugene is at the top of that Rapid range, he is the favorite but hopefully it will be competitive.

This is if the GM has plenty of practice with the Latest Komodo Dragon xxx.x , but what about at 5'+2" you would say around 2550 ? You can ask him to play three extra games at 5'+2 and see if he agrees? He is Not rated at Rapid or Blitz only in standard TC https://ratings.fide.com/profile/2012529

Note: This are some of his chess games at chess.com https://www.chess.com/players/eugene-perelshteyn
Not too convincing against the English opening ==> https://www.chess.com/games/view/14359009
This is an interesting opening if he prepared some other Openings against Komodo Dragon xxx.x ==>

[lkaufman]

Our next human opponent for a handicap match with Komodo Dragon will be GM Eugene Perelshteyn, born 1980, current FIDE 2494, peak FIDE 2555 in 2008. He is a chess pro, developing a course for chess.com and coaching. The handicap will be full knight odds (alternating b1/g1), standard scoring. Time control as usual Rapid 15' + 10"; however in a new twist Dragon will be set to play as if it were blitz (3' + 2"), so five to one time odds on top of the knight odds (clocks won't show it, but it will be obvious that Dragon is playing blitz)! Games to be played on chess.com on Nov. 6 starting 11 a.m. Eastern Time. Six game match, total playing time shouldn't exceed 3 hours, total match time probably not more than four hours with time for brief comments after each game. Perelshteyn streaming, but due to the time odds he'll be on move most of the time so may not comment much during actual play.

I just set a reminder for tomorrows game.

Based on unofficial test games, I believe that the break-even human FIDE rating for knight odds with dragon in somewhere in the 2400s for Rapid and in the 2600s for normal blitz (like 3' + 2" or so). Since Eugene is at the top of that Rapid range, he is the favorite but hopefully it will be competitive.

This is if the GM has plenty of practice with the Latest Komodo Dragon xxx.x , but what about at 5'+2" you would say around 2550 ? You can ask him to play three extra games at 5'+2 and see if he agrees? He is Not rated at Rapid or Blitz only in standard TC https://ratings.fide.com/profile/2012529

Note: This are some of his chess games at chess.com https://www.chess.com/players/eugene-perelshteyn
Not too convincing against the English opening ==> https://www.chess.com/games/view/14359009
This is an interesting opening if he prepared some other Openings against Komodo Dragon xxx.x ==>

At 5' + 2" I would say FIDE blitz 2600 is needed to break even with Dragon at knight odds, assuming some experience with doing this. I have some confidence in this estimate.

[Chessqueen]

This is an interesting opening if he prepared some other Openings against Komodo Dragon xxx.x [TimeControl "180+2"] ==>

[pgn]Event "Computer chess game"]
[Site "MININT-UB2PIMJ"]
[Date "2021.11.05"]
[Round "?"]
[White "Dragon-2.5-64bit-avx2"]
[Black "Gaviota-win32-0.84"]
[Result "1-0"]
[BlackElo "2600"]
[Time "15:46:44"]
[WhiteElo "3600"]
[TimeControl "180+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "81"]
[WhiteType "program"]
[BlackType "program"]


1. c4 e5 2. g3 Nc6 3. Bg2 Nf6 4. a3 Bc5 5. e3 a5 6. Qa4 Ra6 7. Ne2 O-O 8.
Qc2 d6 9. d3 Bg4 10. b3 Bf5 11. h3 Nd7 12. O-O Qf6 13. g4 Bg6 14. Qd1 h5
15. f4 Ba7 16. f5 Bh7 17. Ng3 hxg4 18. hxg4 e4 19. d4 Nxd4 20. Ra2 Nf3+ 21.
Bxf3 exf3 22. Qxf3 Qc3 23. g5 Ne5 24. Qh5 Bxe3+ 25. Kh1 Bg1 26. Rg2 Qxb3
27. Ne4 Rd8 28. g6 fxg6 29. fxg6 Bxg6 30. Rxg6 Nxg6 31. Ng5 Bf2 32. Qh7+
Kf8 33. Qxg6 Qh3+ 34. Nxh3 d5 35. Rxf2+ Rf6 36. Rxf6+ gxf6 37. Qxf6+ Kg8
38. Bh6 Rd7 39. Ng5 Rf7 40. Qxf7+ Kh8 41. Qe8# 1-0[/pgn]

[Chessqueen]

This is an interesting opening if he prepared some other Openings against Komodo Dragon xxx.x [TimeControl "180+2"] ==>

[pgn][Event "Computer chess game"]
[Site "MININT-UB2PIMJ"]
[Date "2021.11.05"]
[Round "?"]
[White "Dragon-2.5-64bit-avx2"]
[Black "Gaviota-win32-0.84"]
[Result "1/2-1/2"]
[BlackElo "2550"]
[Time "16:11:13"]
[WhiteElo "3600"]
[TimeControl "180+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKB1R w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "290"]
[WhiteType "program"]
[BlackType "program"]

1. c4 e6 2. d4 Nf6 3. Nd2 c5 4. dxc5 Bxc5 5. Qc2 b6 6. a3 Bb7 7. e3 Bd6 8.
b4 Qc7 9. Bd3 Bxh2 10. Kf1 a5 11. f4 Ng4 12. Nf3 Bg3 13. Rh3 Bf2 14. Qb2
Rg8 15. Ne5 h5 16. Be2 f5 17. b5 Be4 18. Nxg4 fxg4 19. Rxh5 g3 20. Qd4 Qb7
21. Bg4 Bxg2+ 22. Ke2 Qe4 23. Qxe4 Bxe4 24. Re5 Bb7 25. Bd2 g5 26. f5 Kd8
27. Bc3 exf5 28. Rxf5 Be4 29. Bf6+ Kc7 30. Re5 Bg2 31. Rd1 Rg6 32. Bh8 Rh6
33. Bg7 Rg6 34. Bh8 Rd6 35. Rxd6 Kxd6 36. Bf6 Kc7 37. Re7 Kb7 38. Rg7 Ra7
39. Be5 Ka8 40. Bd6 Be4 41. Bf3 Bxf3+ 42. Kxf3 Kb7 43. Rg8 g4+ 44. Kxg4 Ra8
45. Kf3 a4 46. Ke2 Na6 47. bxa6+ Ka7 48. Rg7 Re8 49. Bxg3 Bxe3 50. Rxd7+
Kxa6 51. Bd6 Bg5+ 52. Kd3 Re1 53. Rg7 Bf6 54. Rg2 Rb1 55. Ke4 Rh1 56. Rf2
Rh4+ 57. Kd5 Rd4+ 58. Ke6 Bh4 59. Rc2 Be1 60. c5 bxc5 61. Rxc5 Ba5 62. Kd7
Bb6 63. Rc1 Rg4 64. Kc6 Rh4 65. Rc2 Re4 66. Rc3 Rg4 67. Be5 Ba5 68. Rc5 Bb6
69. Rc2 Rh4 70. Rg2 Rc4+ 71. Kd5 Rc5+ 72. Ke4 Rc8 73. Kd3 Bc5 74. Bc3 Be7
75. Re2 Bc5 76. Rg2 Be7 77. Re2 Bf8 78. Re4 Kb5 79. Re5+ Rc5 80. Re8 Rf5
81. Rb8+ Kc6 82. Ra8 Rf3+ 83. Kc4 Rf4+ 84. Bd4 Bd6 85. Ra6+ Kd7 86. Kd3 Rg4
87. Kc3 Bxa3 88. Be5 Bc5 89. Kd3 a3 90. Bh8 Kc7 91. Be5+ Kb7 92. Rh6 a2 93.
Kc2 Rg1 94. Kb3 a1=Q 95. Bxa1 Rxa1 96. Kc4 Bf8 97. Rh8 Bd6 98. Kd4 Kc6 99.
Rc8+ Bc7 100. Ke4 Ra4+ 101. Ke3 Rc4 102. Ra8 Kd5 103. Kd2 Rg4 104. Kc3 Rg3+
105. Kb2 Be5+ 106. Kb1 Kc4 107. Kc2 Rg2+ 108. Kd1 Kd3 109. Ke1 Bd4 110.
Ra3+ Ke4 111. Rh3 Rc2 112. Kd1 Rg2 113. Kc1 Bf6 114. Rb3 Rf2 115. Kd1 Be5
116. Ke1 Ra2 117. Kf1 Bf4 118. Rh3 Rb2 119. Ra3 Be3 120. Ke1 Bc5 121. Rh3
Be3 122. Kf1 Rf2+ 123. Kg1 Rc2+ 124. Kf1 Rf2+ 125. Kg1 Rc2+ 126. Kf1 Rd2
127. Rh4+ Bf4 128. Ke1 Rb2 129. Rh5 Kd3 130. Kf1 Be3 131. Rd5+ Ke4 132. Rh5
Kd4 133. Rh7 Rf2+ 134. Ke1 Rb2 135. Rd7+ Ke4 136. Kf1 Rh2 137. Ke1 Bc5 138.
Kd1 Bd4 139. Rf7 Bc5 140. Rb7 Bd6 141. Ke1 Kf5 142. Kd1 Ke5 143. Rb6 Bc5
144. Rb7 Kf6 145. Rb3 Bd4 {50 moves rule} 1/2-1/2[/pgn]

[Chessqueen]

Where can I watch the Match ? Direct Link ?

[lkaufman]

Where can I watch the Match ? Direct Link ?