Amazing results of LeelaKnightOdds-Dev

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

Moderators: hgm, chrisw, Rebel

Father
Posts: 1660
Joined: Sun Mar 19, 2006 4:39 am
Location: Colombia
Full name: Pablo Ignacio Restrepo

Re: Amazing results of LeelaKnightOdds-Dev

Post by Father »

[pgn][/pgn]
Father wrote: Tue Nov 26, 2024 2:01 am
Uri Blass wrote: Mon Nov 25, 2024 2:26 am It seems that it is not as easy as I thought.

I played 5 more games when I decided to play differnet first move 1.c4 1.c3 and 1.f4 and lost 4-1 against the bot at 15+10.
I went back to 1.e4 and won 2-1 later

part of my losses were loss on time in a winning position when I tried to calculate the win that was not simple did not like the move that I wanted to play and finally when I decided to play a move it was too late and I lost on time.

one loss was a checkmate when I saw the mating line before the losing mistake but thought maybe I have some defence and decided that I had not time to calculate and did not like to play at time trouble a line that I did not calculate that was better.
[pgn] [Event "Casual bullet game"]
[Site "https://lichess.org/MzuI1HRw"]
[Date "2024.11.26"]
[White "LeelaQueenOdds"]
[Black "Catecan"]
[Result "1/2-1/2"]
[UTCDate "2024.11.26"]
[UTCTime "00:51:24"]
[WhiteElo "2000"]
[BlackElo "2030"]
[WhiteTitle "BOT"]
[Variant "From Position"]
[TimeControl "60+0"]
[ECO "?"]
[Opening "?"]
[Termination "Normal"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[SetUp "1"]
[Annotator "lichess.org"]

1. b3 d5 2. Bb2 e6 3. g4 c6 4. f4 f5 5. g5 Ne7 6. h4 Rg8 7. h5 g6 8. e3 Bg7 9. Nc3 Nd7 10. Nf3 Nf8 11. d4 b6 12. a4 a6 13. a5 b5 14. b4 Bb7 15. Ne2 Rc8 16. Nc1 Qc7 17. Nd3 Qb8 18. Be2 Qc7 19. Nc5 Qb8 20. O-O-O Qc7 21. Ne5 Qb8 22. Ned3 Qc7 23. Kb1 Qb8 24. Ne5 Qc7 25. Ned3 Qb8 26. Nb3 Qc7 27. Ndc5 Qb8 28. Nd3 Qc7 29. Ndc5 Qb8 30. Bd3 Qc7 31. Rh2 Qb8 32. Be2 Qc7 33. Rdh1 Qb8 34. Nd3 Qc7 35. Nbc5 Qb8 36. Ne5 Qc7 37. Ned3 Qb8 38. Bc3 Qc7 39. Be1 Qb8 40. Bc3 Qc7 41. Be1 Qb8 42. Bf3 Qc7 43. Bd1 Qb8 44. Be2 Qc7 45. Bg3 Qb8 46. Be1 Qc7 47. Bf2 Qb8 48. Bf3 Qc7 49. Be1 Qb8 50. Bd1 Qc7 51. Kb2 Qb8 52. Ka2 Qc7 53. Kb1 Qb8 54. Kb2 Qc7 55. Ka2 Qb8 56. Kb3 Qc7 57. Kb2 Qb8 58. Kc1 Qc7 59. Bc3 Qb8 60. Be1 Qc7 61. Bc3 Qb8 62. Ne5 Qc7 63. hxg6 hxg6 64. Ned3 Qb8 65. Ne5 Qc7 66. Ned3 Qb8 67. Bb2 Qc7 68. Ne5 Qb8 69. Be2 Qc7 70. Kb1 Qb8 71. Ka2 Qc7 72. Kb1 Qb8 73. Bc3 Qc7 74. Kb2 Qb8 75. Kb3 Qc7 76. Kb2 Qb8 77. Kc1 Qc7 78. Kb1 Qb8 79. Kb2 Qc7 80. Kc1 Qb8 81. Kb2 Qc7 82. Ned3 Qb8 83. Be1 Qc7 84. Bc3 Qb8 85. Kb3 Qc7 86. Ne5 Qb8 87. Ned3 Qc7 88. Be1 Qb8 89. Ka2 Qc7 90. Kb1 Qb8 91. Ka2 Qc7 92. Kb1 Qb8 93. Kb2 Qc7 94. Ka2 Qb8 95. Kb1 Qc7 96. Kb2 Qb8 97. Kb1 Qc7 98. Kb2 Qb8 99. Kb3 Qc7 100. Bc3 Qb8 101. Bb2 Qc7 102. Bc3 Qb8 103. Be1 Qc7 104. Bf3 Qb8 105. Ka2 Qc7 106. Kb1 Qb8 107. Kc1 Qc7 108. Kd1 Qb8 109. Kc1 Qc7 110. Kb1 Qb8 111. Kb2 Qc7 112. c3 Qb8 113. Ne5 Qc7 114. Ned3 Qb8 115. Kb1 Qc7 116. Bg3 Qb8 117. Ne5 Qc7 118. Ned3 Qb8 119. Ne5 Qc7 120. Kb2 Qb8 121. Kc2 Qc7 122. Ned3 Qb8 123. Ne5 Qc7 124. Ned3 Qb8 125. Bd1 Qc7 126. Kd2 Qb8 127. Kc2 Qc7 128. Kb2 Qb8 129. Kb3 Qc7 130. Kc2 Qb8 131. Kb2 Qc7 132. Kc2 Qb8 133. Kb3 Qc7 134. Ka2 Qb8 135. Ka3 Qc7 136. Kb3 Qb8 137. Kb2 Qc7 138. Kb3 Qb8 139. Bf3 Qc7 140. Kb2 Qb8 141. Kc2 Qc7 142. Kd1 Qb8 143. Kc2 Qc7 144. Kc1 Qb8 145. Kd1 Qc7 146. Kc1 Qb8 147. Bd1 Qc7 148. Rh3 Qb8 149. Bf3 Qc7 150. Bd1 Qb8 151. Be1 Qc7 152. Kb1 Qb8 153. Kb2 Qc7 154. Kc1 Qb8 155. Kb1 Qc7 156. Kb2 Qb8 157. Kc1 Qc7 158. Kb1 Qb8 159. Kb2 Qc7 160. Kc2 Qb8 161. Ne5 Qc7 162. Ned3 { The game is a draw. } 1/2-1/2[/pgn]
[pgn] [Event "Casual bullet game"]
[Site "https://lichess.org/ICAzgGYH"]
[Date "2024.11.26"]
[White "LeelaQueenOdds"]
[Black "Catecan"]
[Result "1/2-1/2"]
[UTCDate "2024.11.26"]
[UTCTime "04:38:43"]
[WhiteElo "2000"]
[BlackElo "2030"]
[WhiteTitle "BOT"]
[Variant "From Position"]
[TimeControl "60+0"]
[ECO "?"]
[Opening "?"]
[Termination "Normal"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[SetUp "1"]
[Annotator "lichess.org"]

1. d4 d5 2. e4 dxe4 3. Bg5 Nf6 4. Nc3 e6 5. O-O-O Be7 6. h4 c6 7. g4 Nxg4 8. Nh3 Bxg5+ 9. hxg5 Nxf2 10. Nxf2 Qxg5+ 11. Kb1 f5 12. Bc4 Qf6 13. Nh3 Nd7 14. Bb3 Nb6 15. Rdf1 Nd5 16. Nxe4 fxe4 17. Rxf6 Nxf6 18. Rg1 O-O 19. Re1 Kh8 20. Nf4 Bd7 21. Rh1 Rae8 22. Ng6+ Kg8 23. Ne5 Kh8 24. a4 a6 25. a5 Kg8 26. Ka2 Kh8 27. Ka3 Kg8 28. Kb4 Rb8 29. Kc5 Rfd8 30. Rf1 Rf8 31. Re1 Rfd8 32. c3 Rf8 33. Kb6 Rfd8 34. Kc5 Rf8 35. Kb6 Rfd8 36. Kc7 Rdc8+ 37. Kd6 Rd8 38. Ke7 Re8+ 39. Kd6 Red8 40. Ke7 Re8+ 41. Kd6 Red8 { The game is a draw. } 1/2-1/2[/pgn]
I am thinking chess is in a coin.Human beings for ever playing in one face.Now I am playing in the other face:"Antichess". Computers are as a fortres where owner forgot to close a little door behind. You must enter across this door.Forget the front.
Father
Posts: 1660
Joined: Sun Mar 19, 2006 4:39 am
Location: Colombia
Full name: Pablo Ignacio Restrepo

Re: Amazing results of LeelaKnightOdds-Dev

Post by Father »

Father wrote: Tue Nov 26, 2024 5:42 am [pgn][/pgn]
Father wrote: Tue Nov 26, 2024 2:01 am
Uri Blass wrote: Mon Nov 25, 2024 2:26 am It seems that it is not as easy as I thought.

I played 5 more games when I decided to play differnet first move 1.c4 1.c3 and 1.f4 and lost 4-1 against the bot at 15+10.
I went back to 1.e4 and won 2-1 later

part of my losses were loss on time in a winning position when I tried to calculate the win that was not simple did not like the move that I wanted to play and finally when I decided to play a move it was too late and I lost on time.

one loss was a checkmate when I saw the mating line before the losing mistake but thought maybe I have some defence and decided that I had not time to calculate and did not like to play at time trouble a line that I did not calculate that was better.
[pgn] [Event "Casual bullet game"]
[Site "https://lichess.org/MzuI1HRw"]
[Date "2024.11.26"]
[White "LeelaQueenOdds"]
[Black "Catecan"]
[Result "1/2-1/2"]
[UTCDate "2024.11.26"]
[UTCTime "00:51:24"]
[WhiteElo "2000"]
[BlackElo "2030"]
[WhiteTitle "BOT"]
[Variant "From Position"]
[TimeControl "60+0"]
[ECO "?"]
[Opening "?"]
[Termination "Normal"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[SetUp "1"]
[Annotator "lichess.org"]

1. b3 d5 2. Bb2 e6 3. g4 c6 4. f4 f5 5. g5 Ne7 6. h4 Rg8 7. h5 g6 8. e3 Bg7 9. Nc3 Nd7 10. Nf3 Nf8 11. d4 b6 12. a4 a6 13. a5 b5 14. b4 Bb7 15. Ne2 Rc8 16. Nc1 Qc7 17. Nd3 Qb8 18. Be2 Qc7 19. Nc5 Qb8 20. O-O-O Qc7 21. Ne5 Qb8 22. Ned3 Qc7 23. Kb1 Qb8 24. Ne5 Qc7 25. Ned3 Qb8 26. Nb3 Qc7 27. Ndc5 Qb8 28. Nd3 Qc7 29. Ndc5 Qb8 30. Bd3 Qc7 31. Rh2 Qb8 32. Be2 Qc7 33. Rdh1 Qb8 34. Nd3 Qc7 35. Nbc5 Qb8 36. Ne5 Qc7 37. Ned3 Qb8 38. Bc3 Qc7 39. Be1 Qb8 40. Bc3 Qc7 41. Be1 Qb8 42. Bf3 Qc7 43. Bd1 Qb8 44. Be2 Qc7 45. Bg3 Qb8 46. Be1 Qc7 47. Bf2 Qb8 48. Bf3 Qc7 49. Be1 Qb8 50. Bd1 Qc7 51. Kb2 Qb8 52. Ka2 Qc7 53. Kb1 Qb8 54. Kb2 Qc7 55. Ka2 Qb8 56. Kb3 Qc7 57. Kb2 Qb8 58. Kc1 Qc7 59. Bc3 Qb8 60. Be1 Qc7 61. Bc3 Qb8 62. Ne5 Qc7 63. hxg6 hxg6 64. Ned3 Qb8 65. Ne5 Qc7 66. Ned3 Qb8 67. Bb2 Qc7 68. Ne5 Qb8 69. Be2 Qc7 70. Kb1 Qb8 71. Ka2 Qc7 72. Kb1 Qb8 73. Bc3 Qc7 74. Kb2 Qb8 75. Kb3 Qc7 76. Kb2 Qb8 77. Kc1 Qc7 78. Kb1 Qb8 79. Kb2 Qc7 80. Kc1 Qb8 81. Kb2 Qc7 82. Ned3 Qb8 83. Be1 Qc7 84. Bc3 Qb8 85. Kb3 Qc7 86. Ne5 Qb8 87. Ned3 Qc7 88. Be1 Qb8 89. Ka2 Qc7 90. Kb1 Qb8 91. Ka2 Qc7 92. Kb1 Qb8 93. Kb2 Qc7 94. Ka2 Qb8 95. Kb1 Qc7 96. Kb2 Qb8 97. Kb1 Qc7 98. Kb2 Qb8 99. Kb3 Qc7 100. Bc3 Qb8 101. Bb2 Qc7 102. Bc3 Qb8 103. Be1 Qc7 104. Bf3 Qb8 105. Ka2 Qc7 106. Kb1 Qb8 107. Kc1 Qc7 108. Kd1 Qb8 109. Kc1 Qc7 110. Kb1 Qb8 111. Kb2 Qc7 112. c3 Qb8 113. Ne5 Qc7 114. Ned3 Qb8 115. Kb1 Qc7 116. Bg3 Qb8 117. Ne5 Qc7 118. Ned3 Qb8 119. Ne5 Qc7 120. Kb2 Qb8 121. Kc2 Qc7 122. Ned3 Qb8 123. Ne5 Qc7 124. Ned3 Qb8 125. Bd1 Qc7 126. Kd2 Qb8 127. Kc2 Qc7 128. Kb2 Qb8 129. Kb3 Qc7 130. Kc2 Qb8 131. Kb2 Qc7 132. Kc2 Qb8 133. Kb3 Qc7 134. Ka2 Qb8 135. Ka3 Qc7 136. Kb3 Qb8 137. Kb2 Qc7 138. Kb3 Qb8 139. Bf3 Qc7 140. Kb2 Qb8 141. Kc2 Qc7 142. Kd1 Qb8 143. Kc2 Qc7 144. Kc1 Qb8 145. Kd1 Qc7 146. Kc1 Qb8 147. Bd1 Qc7 148. Rh3 Qb8 149. Bf3 Qc7 150. Bd1 Qb8 151. Be1 Qc7 152. Kb1 Qb8 153. Kb2 Qc7 154. Kc1 Qb8 155. Kb1 Qc7 156. Kb2 Qb8 157. Kc1 Qc7 158. Kb1 Qb8 159. Kb2 Qc7 160. Kc2 Qb8 161. Ne5 Qc7 162. Ned3 { The game is a draw. } 1/2-1/2[/pgn]
[pgn] [Event "Casual bullet game"]
[Site "https://lichess.org/ICAzgGYH"]
[Date "2024.11.26"]
[White "LeelaQueenOdds"]
[Black "Catecan"]
[Result "1/2-1/2"]
[UTCDate "2024.11.26"]
[UTCTime "04:38:43"]
[WhiteElo "2000"]
[BlackElo "2030"]
[WhiteTitle "BOT"]
[Variant "From Position"]
[TimeControl "60+0"]
[ECO "?"]
[Opening "?"]
[Termination "Normal"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[SetUp "1"]
[Annotator "lichess.org"]

1. d4 d5 2. e4 dxe4 3. Bg5 Nf6 4. Nc3 e6 5. O-O-O Be7 6. h4 c6 7. g4 Nxg4 8. Nh3 Bxg5+ 9. hxg5 Nxf2 10. Nxf2 Qxg5+ 11. Kb1 f5 12. Bc4 Qf6 13. Nh3 Nd7 14. Bb3 Nb6 15. Rdf1 Nd5 16. Nxe4 fxe4 17. Rxf6 Nxf6 18. Rg1 O-O 19. Re1 Kh8 20. Nf4 Bd7 21. Rh1 Rae8 22. Ng6+ Kg8 23. Ne5 Kh8 24. a4 a6 25. a5 Kg8 26. Ka2 Kh8 27. Ka3 Kg8 28. Kb4 Rb8 29. Kc5 Rfd8 30. Rf1 Rf8 31. Re1 Rfd8 32. c3 Rf8 33. Kb6 Rfd8 34. Kc5 Rf8 35. Kb6 Rfd8 36. Kc7 Rdc8+ 37. Kd6 Rd8 38. Ke7 Re8+ 39. Kd6 Red8 40. Ke7 Re8+ 41. Kd6 Red8 { The game is a draw. } 1/2-1/2[/pgn]
The thinking of our human brain is mega flexible, it mutates like the butterfly, it flies with the wind, draws can simply become the foundation on which victories will be built.
I am thinking chess is in a coin.Human beings for ever playing in one face.Now I am playing in the other face:"Antichess". Computers are as a fortres where owner forgot to close a little door behind. You must enter across this door.Forget the front.
Father
Posts: 1660
Joined: Sun Mar 19, 2006 4:39 am
Location: Colombia
Full name: Pablo Ignacio Restrepo

Re: Amazing results of LeelaKnightOdds-Dev

Post by Father »

Marcus91 wrote: Thu Nov 21, 2024 5:12 pm
Uri Blass wrote: Thu Nov 21, 2024 4:21 pm
Father wrote: Thu Nov 21, 2024 4:01 pm
Marcus91 wrote: Thu Nov 21, 2024 3:00 pm
Brunetti wrote: Thu Nov 21, 2024 11:05 am
Marcus91 wrote: Thu Nov 21, 2024 10:40 am maybe I should add an arbitrary rule, like halve 'K' for all draws?
Doing this is like saying that a win is worth 4 draws, which means you're changing the rules of chess and, consequently, the consistency of the Elo formula. I'm not saying this because I'm currently drawing games, but because that's just how it is :)

Alex
The Elo formula remains consistent if you consider draws as half the weight. Four draws are not equal to one win, but rather one win and one loss. I'm not making any difference between players. If you draw against a bot rated higher than you, you will continue to rise, just slowly. By giving half the weight to draws the model assumes a better fit, this means that those who drew a lot were overestimated and those who drew little were under estimated, given that lichess blitz rating is the reference here
Thank you Marcus91 for your time and your message. You know that I am not a mathematician, so I would like you to explain it well and clear up my doubts. Let's take an example context: I find myself in a man versus machine challenge in which myself and the hundred best ranked human and machine players in the world participate, all fighting against the machines in the Harvard Cup style, of a group of coffee and coffee players. not federated, some of us are self-taught and others with non-school education, obtained outside of schools, others excellent representatives of schools. We all know in the competition that it is a race for knowledge, to obtain the
higher score. If in such a hypothetical 20-game competition I obtain 20 draws and Carlsen obtains 9 victories and one draw, does it mean that I would be the champion of the hypothetical tournament and the one with the highest rating given my best performance? If your answer is yes, then my duty is to continue playing, otherwise I don't see any point in doing so. Thank you in advance Marcus91. I remind everyone: "Any protectorate that closes the path to runners for the sole reason of competing without shoes in the marathon is an absurdity."

I understand that basically 50% will give you equal rating to the machine if you play enough games.
20 games are not enough to get a stable rating but I think that 200 games are enough.

The main problem is when you get less than 50%
If you get a draw and a loss in every 2 games that is 25% your rating is going to be smaller than a player who get 20% by a win and 4 losses every 5 games assuming a lot of games by both sides.

I think that it is not fair even if I get a better place in the table relative to humans by the new rule.
It's an arbitrary rule, like in soccer where a win earns three points and a draw earns one. The rationale behind this system is as follows:

1. Discourages excessively drawish playstyles

2. Promotes balanced time controls – This system pushes players toward time controls where they perform closer to their true skill level, discouraging the trend of favoring ultra-fast formats simply because they make it easier to grind out draws and rating points.

3. Improves rating accuracy – By adjusting the rating system to reflect more dynamic play, it aligns better with the Lichess blitz rating, ensuring a more accurate representation of a player's actual strength.

It's a thoughtful adjustment designed to balance the competitive landscape and encourage more engaging play, even if it might feel arbitrary
Good morning Marcus91. I hope you woke up well and continue well.
I would like you to help me with two questions, I will be very grateful in advance:
- The first question is for me to know the reason why my Elo score dropped approximately between yesterday and today by 7 points despite having reached 30 draws yesterday, most of these with my black.
- The second is if we were applying the standard chess rules on Elo scoring, at this point what would my Elo be?
- The third is that LeelaQuernOdds definitely seems to refuse to play with the black chips.
In advance I thank you and the entire LeelaOdds team. Happy day I wish you.
I am thinking chess is in a coin.Human beings for ever playing in one face.Now I am playing in the other face:"Antichess". Computers are as a fortres where owner forgot to close a little door behind. You must enter across this door.Forget the front.
lkaufman
Posts: 6108
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA

Re: Amazing results of LeelaKnightOdds-Dev

Post by lkaufman »

Father wrote: Tue Nov 26, 2024 2:47 pm
Marcus91 wrote: Thu Nov 21, 2024 5:12 pm
Uri Blass wrote: Thu Nov 21, 2024 4:21 pm
Father wrote: Thu Nov 21, 2024 4:01 pm
Marcus91 wrote: Thu Nov 21, 2024 3:00 pm
Brunetti wrote: Thu Nov 21, 2024 11:05 am
Marcus91 wrote: Thu Nov 21, 2024 10:40 am maybe I should add an arbitrary rule, like halve 'K' for all draws?
Doing this is like saying that a win is worth 4 draws, which means you're changing the rules of chess and, consequently, the consistency of the Elo formula. I'm not saying this because I'm currently drawing games, but because that's just how it is :)

Alex
The Elo formula remains consistent if you consider draws as half the weight. Four draws are not equal to one win, but rather one win and one loss. I'm not making any difference between players. If you draw against a bot rated higher than you, you will continue to rise, just slowly. By giving half the weight to draws the model assumes a better fit, this means that those who drew a lot were overestimated and those who drew little were under estimated, given that lichess blitz rating is the reference here
Thank you Marcus91 for your time and your message. You know that I am not a mathematician, so I would like you to explain it well and clear up my doubts. Let's take an example context: I find myself in a man versus machine challenge in which myself and the hundred best ranked human and machine players in the world participate, all fighting against the machines in the Harvard Cup style, of a group of coffee and coffee players. not federated, some of us are self-taught and others with non-school education, obtained outside of schools, others excellent representatives of schools. We all know in the competition that it is a race for knowledge, to obtain the
higher score. If in such a hypothetical 20-game competition I obtain 20 draws and Carlsen obtains 9 victories and one draw, does it mean that I would be the champion of the hypothetical tournament and the one with the highest rating given my best performance? If your answer is yes, then my duty is to continue playing, otherwise I don't see any point in doing so. Thank you in advance Marcus91. I remind everyone: "Any protectorate that closes the path to runners for the sole reason of competing without shoes in the marathon is an absurdity."

I understand that basically 50% will give you equal rating to the machine if you play enough games.
20 games are not enough to get a stable rating but I think that 200 games are enough.

The main problem is when you get less than 50%
If you get a draw and a loss in every 2 games that is 25% your rating is going to be smaller than a player who get 20% by a win and 4 losses every 5 games assuming a lot of games by both sides.

I think that it is not fair even if I get a better place in the table relative to humans by the new rule.
It's an arbitrary rule, like in soccer where a win earns three points and a draw earns one. The rationale behind this system is as follows:

1. Discourages excessively drawish playstyles

2. Promotes balanced time controls – This system pushes players toward time controls where they perform closer to their true skill level, discouraging the trend of favoring ultra-fast formats simply because they make it easier to grind out draws and rating points.

3. Improves rating accuracy – By adjusting the rating system to reflect more dynamic play, it aligns better with the Lichess blitz rating, ensuring a more accurate representation of a player's actual strength.

It's a thoughtful adjustment designed to balance the competitive landscape and encourage more engaging play, even if it might feel arbitrary
Good morning Marcus91. I hope you woke up well and continue well.
I would like you to help me with two questions, I will be very grateful in advance:
- The first question is for me to know the reason why my Elo score dropped approximately between yesterday and today by 7 points despite having reached 30 draws yesterday, most of these with my black.
- The second is if we were applying the standard chess rules on Elo scoring, at this point what would my Elo be?
- The third is that LeelaQuernOdds definitely seems to refuse to play with the black chips.
In advance I thank you and the entire LeelaOdds team. Happy day I wish you.
I'll try to answer, Marcus91 may have better answers. 1.You have played enough games so that your rating will only depend on your draw percentage (and color choice), unless you change time controls or win a game. The number of draws is not relevant, only the percentage of draws. If you improve that, your rating will go up. 2. If the draws were fully counted instead of half counted, I believe your rating would be about 120 elo higher. But if they were fully counted, we probably would have modified the settings to have Leela try harder to avoid draws. 3. Leela has played various players today as Black. Probably it just happened that when you tried to play as White, four people were already playing the bot. I would like to add though that your games suggest that the difference between playing White and Black is even more than the 100 elo we assumed. My calculations indicated that the gap was at least 150 elo in general, but we didn't yet have enough data to justify using such a high value. Maybe we'll redo the math in the near future and adjust this. It is remarkable to me how much easier it is for you to make draws when you have White than with Black. Of course in normal chess it is easier to draw with White than Black, but I wouldn't guess that it would be such a dramatic difference at queen odds! I'll have to check whether this applies to most players or just to you. Naturally the rules are based on the overall experience, not individual players.
Komodo rules!
Father
Posts: 1660
Joined: Sun Mar 19, 2006 4:39 am
Location: Colombia
Full name: Pablo Ignacio Restrepo

Re: Amazing results of LeelaKnightOdds-Dev

Post by Father »

lkaufman wrote: Tue Nov 26, 2024 10:00 pm
Father wrote: Tue Nov 26, 2024 2:47 pm
Marcus91 wrote: Thu Nov 21, 2024 5:12 pm
Uri Blass wrote: Thu Nov 21, 2024 4:21 pm
Father wrote: Thu Nov 21, 2024 4:01 pm
Marcus91 wrote: Thu Nov 21, 2024 3:00 pm
Brunetti wrote: Thu Nov 21, 2024 11:05 am
Marcus91 wrote: Thu Nov 21, 2024 10:40 am maybe I should add an arbitrary rule, like halve 'K' for all draws?
Doing this is like saying that a win is worth 4 draws, which means you're changing the rules of chess and, consequently, the consistency of the Elo formula. I'm not saying this because I'm currently drawing games, but because that's just how it is :)

Alex
The Elo formula remains consistent if you consider draws as half the weight. Four draws are not equal to one win, but rather one win and one loss. I'm not making any difference between players. If you draw against a bot rated higher than you, you will continue to rise, just slowly. By giving half the weight to draws the model assumes a better fit, this means that those who drew a lot were overestimated and those who drew little were under estimated, given that lichess blitz rating is the reference here
Thank you Marcus91 for your time and your message. You know that I am not a mathematician, so I would like you to explain it well and clear up my doubts. Let's take an example context: I find myself in a man versus machine challenge in which myself and the hundred best ranked human and machine players in the world participate, all fighting against the machines in the Harvard Cup style, of a group of coffee and coffee players. not federated, some of us are self-taught and others with non-school education, obtained outside of schools, others excellent representatives of schools. We all know in the competition that it is a race for knowledge, to obtain the
higher score. If in such a hypothetical 20-game competition I obtain 20 draws and Carlsen obtains 9 victories and one draw, does it mean that I would be the champion of the hypothetical tournament and the one with the highest rating given my best performance? If your answer is yes, then my duty is to continue playing, otherwise I don't see any point in doing so. Thank you in advance Marcus91. I remind everyone: "Any protectorate that closes the path to runners for the sole reason of competing without shoes in the marathon is an absurdity."

I understand that basically 50% will give you equal rating to the machine if you play enough games.
20 games are not enough to get a stable rating but I think that 200 games are enough.

The main problem is when you get less than 50%
If you get a draw and a loss in every 2 games that is 25% your rating is going to be smaller than a player who get 20% by a win and 4 losses every 5 games assuming a lot of games by both sides.

I think that it is not fair even if I get a better place in the table relative to humans by the new rule.
It's an arbitrary rule, like in soccer where a win earns three points and a draw earns one. The rationale behind this system is as follows:

1. Discourages excessively drawish playstyles

2. Promotes balanced time controls – This system pushes players toward time controls where they perform closer to their true skill level, discouraging the trend of favoring ultra-fast formats simply because they make it easier to grind out draws and rating points.

3. Improves rating accuracy – By adjusting the rating system to reflect more dynamic play, it aligns better with the Lichess blitz rating, ensuring a more accurate representation of a player's actual strength.

It's a thoughtful adjustment designed to balance the competitive landscape and encourage more engaging play, even if it might feel arbitrary
Good morning Marcus91. I hope you woke up well and continue well.
I would like you to help me with two questions, I will be very grateful in advance:
- The first question is for me to know the reason why my Elo score dropped approximately between yesterday and today by 7 points despite having reached 30 draws yesterday, most of these with my black.
- The second is if we were applying the standard chess rules on Elo scoring, at this point what would my Elo be?
- The third is that LeelaQuernOdds definitely seems to refuse to play with the black chips.
In advance I thank you and the entire LeelaOdds team. Happy day I wish you.
I'll try to answer, Marcus91 may have better answers. 1.You have played enough games so that your rating will only depend on your draw percentage (and color choice), unless you change time controls or win a game. The number of draws is not relevant, only the percentage of draws. If you improve that, your rating will go up. 2. If the draws were fully counted instead of half counted, I believe your rating would be about 120 elo higher. But if they were fully counted, we probably would have modified the settings to have Leela try harder to avoid draws. 3. Leela has played various players today as Black. Probably it just happened that when you tried to play as White, four people were already playing the bot. I would like to add though that your games suggest that the difference between playing White and Black is even more than the 100 elo we assumed. My calculations indicated that the gap was at least 150 elo in general, but we didn't yet have enough data to justify using such a high value. Maybe we'll redo the math in the near future and adjust this. It is remarkable to me how much easier it is for you to make draws when you have White than with Black. Of course in normal chess it is easier to draw with White than Black, but I wouldn't guess that it would be such a dramatic difference at queen odds! I'll have to check whether this applies to most players or just to you. Naturally the rules are based on the overall experience, not individual players.

Thank you very much Mr. Larry Kaufman for your time and important information. What I am observing is that I had undertaken to follow an exhausting and apparently wrong route on the climb route towards the summit of Mount Enverest, that is, towards the first place in the table of the top 100. I would like to ask you What would I consider my steps to follow should be in order to climb the table towards a better position than my current one of #5. As a "street fighter" I don't care that in the group of the Top 100 whether or not they are representatives of the best Fide humans. Here in the law of anti-machine challenges, the law is different, here there is nothing that is already written, kinetic entities lack feelings and emotions, and in a millisecond they can unleash against us humans, a barrage of stronger and harder blows. That the blows that the world human chess champions could give us... I would appreciate them guiding me in my climb... for my part, I am forever grateful.
I am thinking chess is in a coin.Human beings for ever playing in one face.Now I am playing in the other face:"Antichess". Computers are as a fortres where owner forgot to close a little door behind. You must enter across this door.Forget the front.
Uri Blass
Posts: 10632
Joined: Thu Mar 09, 2006 12:37 am
Location: Tel-Aviv Israel

Re: Amazing results of LeelaKnightOdds-Dev

Post by Uri Blass »

Father wrote: Wed Nov 27, 2024 12:47 am
lkaufman wrote: Tue Nov 26, 2024 10:00 pm
Father wrote: Tue Nov 26, 2024 2:47 pm
Marcus91 wrote: Thu Nov 21, 2024 5:12 pm
Uri Blass wrote: Thu Nov 21, 2024 4:21 pm
Father wrote: Thu Nov 21, 2024 4:01 pm
Marcus91 wrote: Thu Nov 21, 2024 3:00 pm
Brunetti wrote: Thu Nov 21, 2024 11:05 am
Marcus91 wrote: Thu Nov 21, 2024 10:40 am maybe I should add an arbitrary rule, like halve 'K' for all draws?
Doing this is like saying that a win is worth 4 draws, which means you're changing the rules of chess and, consequently, the consistency of the Elo formula. I'm not saying this because I'm currently drawing games, but because that's just how it is :)

Alex
The Elo formula remains consistent if you consider draws as half the weight. Four draws are not equal to one win, but rather one win and one loss. I'm not making any difference between players. If you draw against a bot rated higher than you, you will continue to rise, just slowly. By giving half the weight to draws the model assumes a better fit, this means that those who drew a lot were overestimated and those who drew little were under estimated, given that lichess blitz rating is the reference here
Thank you Marcus91 for your time and your message. You know that I am not a mathematician, so I would like you to explain it well and clear up my doubts. Let's take an example context: I find myself in a man versus machine challenge in which myself and the hundred best ranked human and machine players in the world participate, all fighting against the machines in the Harvard Cup style, of a group of coffee and coffee players. not federated, some of us are self-taught and others with non-school education, obtained outside of schools, others excellent representatives of schools. We all know in the competition that it is a race for knowledge, to obtain the
higher score. If in such a hypothetical 20-game competition I obtain 20 draws and Carlsen obtains 9 victories and one draw, does it mean that I would be the champion of the hypothetical tournament and the one with the highest rating given my best performance? If your answer is yes, then my duty is to continue playing, otherwise I don't see any point in doing so. Thank you in advance Marcus91. I remind everyone: "Any protectorate that closes the path to runners for the sole reason of competing without shoes in the marathon is an absurdity."

I understand that basically 50% will give you equal rating to the machine if you play enough games.
20 games are not enough to get a stable rating but I think that 200 games are enough.

The main problem is when you get less than 50%
If you get a draw and a loss in every 2 games that is 25% your rating is going to be smaller than a player who get 20% by a win and 4 losses every 5 games assuming a lot of games by both sides.

I think that it is not fair even if I get a better place in the table relative to humans by the new rule.
It's an arbitrary rule, like in soccer where a win earns three points and a draw earns one. The rationale behind this system is as follows:

1. Discourages excessively drawish playstyles

2. Promotes balanced time controls – This system pushes players toward time controls where they perform closer to their true skill level, discouraging the trend of favoring ultra-fast formats simply because they make it easier to grind out draws and rating points.

3. Improves rating accuracy – By adjusting the rating system to reflect more dynamic play, it aligns better with the Lichess blitz rating, ensuring a more accurate representation of a player's actual strength.

It's a thoughtful adjustment designed to balance the competitive landscape and encourage more engaging play, even if it might feel arbitrary
Good morning Marcus91. I hope you woke up well and continue well.
I would like you to help me with two questions, I will be very grateful in advance:
- The first question is for me to know the reason why my Elo score dropped approximately between yesterday and today by 7 points despite having reached 30 draws yesterday, most of these with my black.
- The second is if we were applying the standard chess rules on Elo scoring, at this point what would my Elo be?
- The third is that LeelaQuernOdds definitely seems to refuse to play with the black chips.
In advance I thank you and the entire LeelaOdds team. Happy day I wish you.
I'll try to answer, Marcus91 may have better answers. 1.You have played enough games so that your rating will only depend on your draw percentage (and color choice), unless you change time controls or win a game. The number of draws is not relevant, only the percentage of draws. If you improve that, your rating will go up. 2. If the draws were fully counted instead of half counted, I believe your rating would be about 120 elo higher. But if they were fully counted, we probably would have modified the settings to have Leela try harder to avoid draws. 3. Leela has played various players today as Black. Probably it just happened that when you tried to play as White, four people were already playing the bot. I would like to add though that your games suggest that the difference between playing White and Black is even more than the 100 elo we assumed. My calculations indicated that the gap was at least 150 elo in general, but we didn't yet have enough data to justify using such a high value. Maybe we'll redo the math in the near future and adjust this. It is remarkable to me how much easier it is for you to make draws when you have White than with Black. Of course in normal chess it is easier to draw with White than Black, but I wouldn't guess that it would be such a dramatic difference at queen odds! I'll have to check whether this applies to most players or just to you. Naturally the rules are based on the overall experience, not individual players.

Thank you very much Mr. Larry Kaufman for your time and important information. What I am observing is that I had undertaken to follow an exhausting and apparently wrong route on the climb route towards the summit of Mount Enverest, that is, towards the first place in the table of the top 100. I would like to ask you What would I consider my steps to follow should be in order to climb the table towards a better position than my current one of #5. As a "street fighter" I don't care that in the group of the Top 100 whether or not they are representatives of the best Fide humans. Here in the law of anti-machine challenges, the law is different, here there is nothing that is already written, kinetic entities lack feelings and emotions, and in a millisecond they can unleash against us humans, a barrage of stronger and harder blows. That the blows that the world human chess champions could give us... I would appreciate them guiding me in my climb... for my part, I am forever grateful.
I think a possible strategy may be to beat Leela at long time control and later to try to repeat the moves at shorter time control.

I searched for games and here are some games when leela lost very fast:
I do not know what is the probability that Leela is going to play exactly the same moves but I found that I could repeat it least the same first 8 moves as one of my previous games(I did not try the strategy of memorizing my wins and repeating moves of games that I won till the end).

https://lichess.org/2gk2PfYa/black
https://lichess.org/dnbGAwqM
Uri Blass
Posts: 10632
Joined: Thu Mar 09, 2006 12:37 am
Location: Tel-Aviv Israel

Re: Amazing results of LeelaKnightOdds-Dev

Post by Uri Blass »

Here are 2 games of me that are not exact repeat because in one of them I started 1.d4 and in one of them 1.e4 but with different order of the first 2 moves I got the same first 16 moves when Leela only deviated at move 17.

https://lichess.org/JyzyxeaHXdPC

https://lichess.org/gWDXkcRgo0A2

Sacrificing my queen is objectively not best but I found it relatively easy to win after sacrificing my queen for rook and knight.
8.Bg5 is also not best and 8.Bh6 is stronger.
Father
Posts: 1660
Joined: Sun Mar 19, 2006 4:39 am
Location: Colombia
Full name: Pablo Ignacio Restrepo

Re: Amazing results of LeelaKnightOdds-Dev

Post by Father »

Uri Blass wrote: Wed Nov 27, 2024 1:28 am
Father wrote: Wed Nov 27, 2024 12:47 am
lkaufman wrote: [/pgn]1732654803 user_id=4773]
Father wrote: Tue Nov 26, 2024 2:47 pm
Marcus91 wrote: Thu Nov 21, 2024 5:12 pm
Uri Blass wrote: Thu Nov 21, 2024 4:21 pm
Father wrote: Thu Nov 21, 2024 4:01 pm
Marcus91 wrote: Thu Nov 21, 2024 3:00 pm
Brunetti wrote: Thu Nov 21, 2024 11:05 am
Marcus91 wrote: Thu Nov 21, 2024 10:40 am maybe I should add an arbitrary rule, like halve 'K' for all draws?
Doing this is like saying that a win is worth 4 draws, which means you're changing the rules of chess and, consequently, the consistency of the Elo formula. I'm not saying this because I'm currently drawing games, but because that's just how it is :)

Alex
The Elo formula remains consistent if you consider draws as half the weight. Four draws are not equal to one win, but rather one win and one loss. I'm not making any difference between players. If you draw against a bot rated higher than you, you will continue to rise, just slowly. By giving half the weight to draws the model assumes a better fit, this means that those who drew a lot were overestimated and those who drew little were under estimated, given that lichess blitz rating is the reference here
Thank you Marcus91 for your time and your message. You know that I am not a mathematician, so I would like you to explain it well and clear up my doubts. Let's take an example context: I find myself in a man versus machine challenge in which myself and the hundred best ranked human and machine players in the world participate, all fighting against the machines in the Harvard Cup style, of a group of coffee and coffee players. not federated, some of us are self-taught and others with non-school education, obtained outside of schools, others excellent representatives of schools. We all know in the competition that it is a race for knowledge, to obtain the
higher score. If in such a hypothetical 20-game competition I obtain 20 draws and Carlsen obtains 9 victories and one draw, does it mean that I would be the champion of the hypothetical tournament and the one with the highest rating given my best performance? If your answer is yes, then my duty is to continue playing, otherwise I don't see any point in doing so. Thank you in advance Marcus91. I remind everyone: "Any protectorate that closes the path to runners for the sole reason of competing without shoes in the marathon is an absurdity."

I understand that basically 50% will give you equal rating to the machine if you play enough games.
20 games are not enough to get a stable rating but I think that 200 games are enough.

The main problem is when you get less than 50%
If you get a draw and a loss in every 2 games that is 25% your rating is going to be smaller than a player who get 20% by a win and 4 losses every 5 games assuming a lot of games by both sides.

I think that it is not fair even if I get a better place in the table relative to humans by the new rule.
It's an arbitrary rule, like in soccer where a win earns three points and a draw earns one. The rationale behind this system is as follows:

1. Discourages excessively drawish playstyles

2. Promotes balanced time controls – This system pushes players toward time controls where they perform closer to their true skill level, discouraging the trend of favoring ultra-fast formats simply because they make it easier to grind out draws and rating points.

3. Improves rating accuracy – By adjusting the rating system to reflect more dynamic play, it aligns better with the Lichess blitz rating, ensuring a more accurate representation of a player's actual strength.

It's a thoughtful adjustment designed to balance the competitive landscape and encourage more engaging play, even if it might feel arbitrary
Good morning Marcus91. I hope you woke up well and continue well.
I would like you to help me with two questions, I will be very grateful in advance:
- The first question is for me to know the reason why my Elo score dropped approximately between yesterday and today by 7 points despite having reached 30 draws yesterday, most of these with my black.
- The second is if we were applying the standard chess rules on Elo scoring, at this point what would my Elo be?
- The third is that LeelaQuernOdds definitely seems to refuse to play with the black chips.
In advance I thank you and the entire LeelaOdds team. Happy day I wish you.
I'll try to answer, Marcus91 may have better answers. 1.You have played enough games so that your rating will only depend on your draw percentage (and color choice), unless you change time controls or win a game. The number of draws is not relevant, only the percentage of draws. If you improve that, your rating will go up. 2. If the draws were fully counted instead of half counted, I believe your rating would be about 120 elo higher. But if they were fully counted, we probably would have modified the settings to have Leela try harder to avoid draws. 3. Leela has played various players today as Black. Probably it just happened that when you tried to play as White, four people were already playing the bot. I would like to add though that your games suggest that the difference between playing White and Black is even more than the 100 elo we assumed. My calculations indicated that the gap was at least 150 elo in general, but we didn't yet have enough data to justify using such a high value. Maybe we'll redo the math in the near future and adjust this. It is remarkable to me how much easier it is for you to make draws when you have White than with Black. Of course in normal chess it is easier to draw with White than Black, but I wouldn't guess that it would be such a dramatic difference at queen odds! I'll have to check whether this applies to most players or just to you. Naturally the rules are based on the overall experience, not individual players.

Thank you very much Mr. Larry Kaufman for your time and important information. What I am observing is that I had undertaken to follow an exhausting and apparently wrong route on the climb route towards the summit of Mount Enverest, that is, towards the first place in the table of the top 100. I would like to ask you What would I consider my steps to follow should be in order to climb the table towards a better position than my current one of #5. As a "street fighter" I don't care that in the group of the Top 100 whether or not they are representatives of the best Fide humans. Here in the law of anti-machine challenges, the law is different, here there is nothing that is already written, kinetic entities lack feelings and emotions, and in a millisecond they can unleash against us humans, a barrage of stronger and harder blows. That the blows that the world human chess champions could give us... I would appreciate them guiding me in my climb... for my part, I am forever grateful.
I think a possible strategy may be to beat Leela at long time control and later to try to repeat the moves at shorter time control.

I searched for games and here are some games when leela lost very fast:
I do not know what is the probability that Leela is going to play exactly the same moves but I found that I could repeat it least the same first 8 moves as one of my previous games(I did not try the strategy of memorizing my wins and repeating moves of games that I won till the end).

https://lichess.org/2gk2PfYa/black
https://lichess.org/dnbGAwqM
[pgn] [Event "Casual bullet game"]
[Site "https://lichess.org/hBd7dell"]
[Date "2024.11.27"]
[White "LeelaQueenOdds"]
[Black "Catecan"]
[Result "1/2-1/2"]
[UTCDate "2024.11.27"]
[UTCTime "00:34:45"]
[WhiteElo "2000"]
[BlackElo "2030"]
[WhiteTitle "BOT"]
[Variant "From Position"]
[TimeControl "60+1"]
[ECO "?"]
[Opening "?"]
[Termination "Normal"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[SetUp "1"]
[Annotator "lichess.org"]

1. g3 d5 2. f4 e6 3. g4 f5 4. g5 g6 5. d4 Ne7 6. Nf3 Bg7 7. h4 Rg8 8. h5 Nd7 9. e3 Nf8 10. h6 Bh8 11. b4 c6 12. a4 b6 13. a5 b5 14. Nbd2 a6 15. Nb3 Bd7 16. Nc5 Qc8 17. Bb2 Ra7 18. Ne5 Ra8 19. Be2 Ra7 20. O-O-O Ra8 21. Kb1 Ra7 22. Ka2 Ra8 23. Kb1 Ra7 24. Kc1 Ra8 25. Kd2 Ra7 26. Ke1 Ra8 27. Kf2 Ra7 28. Kg3 Ra8 29. Kf2 Ra7 30. Ke1 Ra8 31. Rh2 Ra7 32. Kd2 Ra8 33. Kc1 Ra7 34. Kb1 Ra8 35. Kc1 Ra7 36. Nf3 Ra8 37. Kb1 Ra7 38. Ne5 Ra8 39. Nf3 Ra7 40. Bd3 Ra8 41. Re2 Ra7 42. Rh2 Ra8 43. Re2 Ra7 44. Rf2 Ra8 45. Rh2 Ra7 46. Re2 Ra8 47. Rf2 Ra7 48. Rg2 Ra8 49. Rh2 Ra7 50. Rg2 Ra8 51. Rc1 Ra7 52. Rd2 Ra8 53. Rdd1 Ra7 54. Re1 Ra8 55. Red1 Ra7 56. Rd2 Ra8 57. Be2 Ra7 58. Ne5 Ra8 59. Nf3 Ra7 60. Rdd1 Ra8 61. Rd3 Ra7 62. Ne5 Ra8 63. Rd2 Ra7 64. c4 bxc4 65. Bc3 Ra8 66. Bd1 Ra7 67. Ba4 Rc7 68. Nf3 Ra7 69. Ne5 Rc7 70. Nf3 Ra7 71. Bc2 Rc7 72. Bd1 Ra7 73. Ne5 Rc7 74. Nf3 Ra7 75. Rcc2 Rc7 76. Kc1 Ra7 77. Ne5 Rc7 78. Nf3 Ra7 79. Rb2 Rc7 80. Rb1 Ra7 81. Rbb2 Rc7 82. Rb1 Ra7 83. Na4 Qc7 84. Nc5 Qc8 { The game is a draw. } 1/2-1/2[/pgn]

Uri. Thank you very much for your message and advice. It's true, it would be taking the knowledge from my bullet games to slower time control games. I would believe that this could give favorable results to win, in fact, if you look at my bullet games, you can see that a winning strategy has been developing for me... the problem is that there is very little time to implement the long-term strategic plan…
I am thinking chess is in a coin.Human beings for ever playing in one face.Now I am playing in the other face:"Antichess". Computers are as a fortres where owner forgot to close a little door behind. You must enter across this door.Forget the front.
lkaufman
Posts: 6108
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA

Re: Amazing results of LeelaKnightOdds-Dev

Post by lkaufman »

Father wrote: Wed Nov 27, 2024 12:47 am
lkaufman wrote: Tue Nov 26, 2024 10:00 pm
Father wrote: Tue Nov 26, 2024 2:47 pm
Marcus91 wrote: Thu Nov 21, 2024 5:12 pm
Uri Blass wrote: Thu Nov 21, 2024 4:21 pm
Father wrote: Thu Nov 21, 2024 4:01 pm
Marcus91 wrote: Thu Nov 21, 2024 3:00 pm
Brunetti wrote: Thu Nov 21, 2024 11:05 am
Marcus91 wrote: Thu Nov 21, 2024 10:40 am maybe I should add an arbitrary rule, like halve 'K' for all draws?
Doing this is like saying that a win is worth 4 draws, which means you're changing the rules of chess and, consequently, the consistency of the Elo formula. I'm not saying this because I'm currently drawing games, but because that's just how it is :)

Alex
The Elo formula remains consistent if you consider draws as half the weight. Four draws are not equal to one win, but rather one win and one loss. I'm not making any difference between players. If you draw against a bot rated higher than you, you will continue to rise, just slowly. By giving half the weight to draws the model assumes a better fit, this means that those who drew a lot were overestimated and those who drew little were under estimated, given that lichess blitz rating is the reference here
Thank you Marcus91 for your time and your message. You know that I am not a mathematician, so I would like you to explain it well and clear up my doubts. Let's take an example context: I find myself in a man versus machine challenge in which myself and the hundred best ranked human and machine players in the world participate, all fighting against the machines in the Harvard Cup style, of a group of coffee and coffee players. not federated, some of us are self-taught and others with non-school education, obtained outside of schools, others excellent representatives of schools. We all know in the competition that it is a race for knowledge, to obtain the
higher score. If in such a hypothetical 20-game competition I obtain 20 draws and Carlsen obtains 9 victories and one draw, does it mean that I would be the champion of the hypothetical tournament and the one with the highest rating given my best performance? If your answer is yes, then my duty is to continue playing, otherwise I don't see any point in doing so. Thank you in advance Marcus91. I remind everyone: "Any protectorate that closes the path to runners for the sole reason of competing without shoes in the marathon is an absurdity."

I understand that basically 50% will give you equal rating to the machine if you play enough games.
20 games are not enough to get a stable rating but I think that 200 games are enough.

The main problem is when you get less than 50%
If you get a draw and a loss in every 2 games that is 25% your rating is going to be smaller than a player who get 20% by a win and 4 losses every 5 games assuming a lot of games by both sides.

I think that it is not fair even if I get a better place in the table relative to humans by the new rule.
It's an arbitrary rule, like in soccer where a win earns three points and a draw earns one. The rationale behind this system is as follows:

1. Discourages excessively drawish playstyles

2. Promotes balanced time controls – This system pushes players toward time controls where they perform closer to their true skill level, discouraging the trend of favoring ultra-fast formats simply because they make it easier to grind out draws and rating points.

3. Improves rating accuracy – By adjusting the rating system to reflect more dynamic play, it aligns better with the Lichess blitz rating, ensuring a more accurate representation of a player's actual strength.

It's a thoughtful adjustment designed to balance the competitive landscape and encourage more engaging play, even if it might feel arbitrary
Good morning Marcus91. I hope you woke up well and continue well.
I would like you to help me with two questions, I will be very grateful in advance:
- The first question is for me to know the reason why my Elo score dropped approximately between yesterday and today by 7 points despite having reached 30 draws yesterday, most of these with my black.
- The second is if we were applying the standard chess rules on Elo scoring, at this point what would my Elo be?
- The third is that LeelaQuernOdds definitely seems to refuse to play with the black chips.
In advance I thank you and the entire LeelaOdds team. Happy day I wish you.
I'll try to answer, Marcus91 may have better answers. 1.You have played enough games so that your rating will only depend on your draw percentage (and color choice), unless you change time controls or win a game. The number of draws is not relevant, only the percentage of draws. If you improve that, your rating will go up. 2. If the draws were fully counted instead of half counted, I believe your rating would be about 120 elo higher. But if they were fully counted, we probably would have modified the settings to have Leela try harder to avoid draws. 3. Leela has played various players today as Black. Probably it just happened that when you tried to play as White, four people were already playing the bot. I would like to add though that your games suggest that the difference between playing White and Black is even more than the 100 elo we assumed. My calculations indicated that the gap was at least 150 elo in general, but we didn't yet have enough data to justify using such a high value. Maybe we'll redo the math in the near future and adjust this. It is remarkable to me how much easier it is for you to make draws when you have White than with Black. Of course in normal chess it is easier to draw with White than Black, but I wouldn't guess that it would be such a dramatic difference at queen odds! I'll have to check whether this applies to most players or just to you. Naturally the rules are based on the overall experience, not individual players.

Thank you very much Mr. Larry Kaufman for your time and important information. What I am observing is that I had undertaken to follow an exhausting and apparently wrong route on the climb route towards the summit of Mount Enverest, that is, towards the first place in the table of the top 100. I would like to ask you What would I consider my steps to follow should be in order to climb the table towards a better position than my current one of #5. As a "street fighter" I don't care that in the group of the Top 100 whether or not they are representatives of the best Fide humans. Here in the law of anti-machine challenges, the law is different, here there is nothing that is already written, kinetic entities lack feelings and emotions, and in a millisecond they can unleash against us humans, a barrage of stronger and harder blows. That the blows that the world human chess champions could give us... I would appreciate them guiding me in my climb... for my part, I am forever grateful.
My own strategy is to play at a time control where I am a clear favorite, though not long enough to guarantee a win which would be rather boring. In my case I do better with a large increment rather than base, but in your case probably the opposite is true, since you can play even 1 minute without increment well whereas I cannot do that. You can just keep increasing the time limit until your score reaches 60% or more. In my case I think I can do that at ten seconds per move, with the minimum one minute base, or at 5'5" or 10'0". Maybe at 5'3", probably not at 3'2" or 5'0". I haven't yet played enough games to reach a fair rating, but I expect to at least pass 2400 when I've played enough games based on present rules. I agree with Uri that if you are a queen up it is generally advisable to trade the queen for rook plus minor piece if you have the chance. I only play Black against it myself (playing White is queen and move odds), but you seem to score so much better with White that you may get a higher rating that way despite the lower Elo for Leela playing Black.
Komodo rules!
Uri Blass
Posts: 10632
Joined: Thu Mar 09, 2006 12:37 am
Location: Tel-Aviv Israel

Re: Amazing results of LeelaKnightOdds-Dev

Post by Uri Blass »

lkaufman wrote: Wed Nov 27, 2024 5:51 am
Father wrote: Wed Nov 27, 2024 12:47 am
lkaufman wrote: Tue Nov 26, 2024 10:00 pm
Father wrote: Tue Nov 26, 2024 2:47 pm
Marcus91 wrote: Thu Nov 21, 2024 5:12 pm
Uri Blass wrote: Thu Nov 21, 2024 4:21 pm
Father wrote: Thu Nov 21, 2024 4:01 pm
Marcus91 wrote: Thu Nov 21, 2024 3:00 pm
Brunetti wrote: Thu Nov 21, 2024 11:05 am
Marcus91 wrote: Thu Nov 21, 2024 10:40 am maybe I should add an arbitrary rule, like halve 'K' for all draws?
Doing this is like saying that a win is worth 4 draws, which means you're changing the rules of chess and, consequently, the consistency of the Elo formula. I'm not saying this because I'm currently drawing games, but because that's just how it is :)

Alex
The Elo formula remains consistent if you consider draws as half the weight. Four draws are not equal to one win, but rather one win and one loss. I'm not making any difference between players. If you draw against a bot rated higher than you, you will continue to rise, just slowly. By giving half the weight to draws the model assumes a better fit, this means that those who drew a lot were overestimated and those who drew little were under estimated, given that lichess blitz rating is the reference here
Thank you Marcus91 for your time and your message. You know that I am not a mathematician, so I would like you to explain it well and clear up my doubts. Let's take an example context: I find myself in a man versus machine challenge in which myself and the hundred best ranked human and machine players in the world participate, all fighting against the machines in the Harvard Cup style, of a group of coffee and coffee players. not federated, some of us are self-taught and others with non-school education, obtained outside of schools, others excellent representatives of schools. We all know in the competition that it is a race for knowledge, to obtain the
higher score. If in such a hypothetical 20-game competition I obtain 20 draws and Carlsen obtains 9 victories and one draw, does it mean that I would be the champion of the hypothetical tournament and the one with the highest rating given my best performance? If your answer is yes, then my duty is to continue playing, otherwise I don't see any point in doing so. Thank you in advance Marcus91. I remind everyone: "Any protectorate that closes the path to runners for the sole reason of competing without shoes in the marathon is an absurdity."

I understand that basically 50% will give you equal rating to the machine if you play enough games.
20 games are not enough to get a stable rating but I think that 200 games are enough.

The main problem is when you get less than 50%
If you get a draw and a loss in every 2 games that is 25% your rating is going to be smaller than a player who get 20% by a win and 4 losses every 5 games assuming a lot of games by both sides.

I think that it is not fair even if I get a better place in the table relative to humans by the new rule.
It's an arbitrary rule, like in soccer where a win earns three points and a draw earns one. The rationale behind this system is as follows:

1. Discourages excessively drawish playstyles

2. Promotes balanced time controls – This system pushes players toward time controls where they perform closer to their true skill level, discouraging the trend of favoring ultra-fast formats simply because they make it easier to grind out draws and rating points.

3. Improves rating accuracy – By adjusting the rating system to reflect more dynamic play, it aligns better with the Lichess blitz rating, ensuring a more accurate representation of a player's actual strength.

It's a thoughtful adjustment designed to balance the competitive landscape and encourage more engaging play, even if it might feel arbitrary
Good morning Marcus91. I hope you woke up well and continue well.
I would like you to help me with two questions, I will be very grateful in advance:
- The first question is for me to know the reason why my Elo score dropped approximately between yesterday and today by 7 points despite having reached 30 draws yesterday, most of these with my black.
- The second is if we were applying the standard chess rules on Elo scoring, at this point what would my Elo be?
- The third is that LeelaQuernOdds definitely seems to refuse to play with the black chips.
In advance I thank you and the entire LeelaOdds team. Happy day I wish you.
I'll try to answer, Marcus91 may have better answers. 1.You have played enough games so that your rating will only depend on your draw percentage (and color choice), unless you change time controls or win a game. The number of draws is not relevant, only the percentage of draws. If you improve that, your rating will go up. 2. If the draws were fully counted instead of half counted, I believe your rating would be about 120 elo higher. But if they were fully counted, we probably would have modified the settings to have Leela try harder to avoid draws. 3. Leela has played various players today as Black. Probably it just happened that when you tried to play as White, four people were already playing the bot. I would like to add though that your games suggest that the difference between playing White and Black is even more than the 100 elo we assumed. My calculations indicated that the gap was at least 150 elo in general, but we didn't yet have enough data to justify using such a high value. Maybe we'll redo the math in the near future and adjust this. It is remarkable to me how much easier it is for you to make draws when you have White than with Black. Of course in normal chess it is easier to draw with White than Black, but I wouldn't guess that it would be such a dramatic difference at queen odds! I'll have to check whether this applies to most players or just to you. Naturally the rules are based on the overall experience, not individual players.

Thank you very much Mr. Larry Kaufman for your time and important information. What I am observing is that I had undertaken to follow an exhausting and apparently wrong route on the climb route towards the summit of Mount Enverest, that is, towards the first place in the table of the top 100. I would like to ask you What would I consider my steps to follow should be in order to climb the table towards a better position than my current one of #5. As a "street fighter" I don't care that in the group of the Top 100 whether or not they are representatives of the best Fide humans. Here in the law of anti-machine challenges, the law is different, here there is nothing that is already written, kinetic entities lack feelings and emotions, and in a millisecond they can unleash against us humans, a barrage of stronger and harder blows. That the blows that the world human chess champions could give us... I would appreciate them guiding me in my climb... for my part, I am forever grateful.
My own strategy is to play at a time control where I am a clear favorite, though not long enough to guarantee a win which would be rather boring. In my case I do better with a large increment rather than base, but in your case probably the opposite is true, since you can play even 1 minute without increment well whereas I cannot do that. You can just keep increasing the time limit until your score reaches 60% or more. In my case I think I can do that at ten seconds per move, with the minimum one minute base, or at 5'5" or 10'0". Maybe at 5'3", probably not at 3'2" or 5'0". I haven't yet played enough games to reach a fair rating, but I expect to at least pass 2400 when I've played enough games based on present rules. I agree with Uri that if you are a queen up it is generally advisable to trade the queen for rook plus minor piece if you have the chance. I only play Black against it myself (playing White is queen and move odds), but you seem to score so much better with White that you may get a higher rating that way despite the lower Elo for Leela playing Black.
I believe the big difference between white and black is because it is easier to prepare with white.
I guess there is going to be smaller difference between white and black if the queen odd is in FRC or shuffle chess.