An idea for new Handicap games for dragon

[Marcus9]

Anyway I wanted to share a lc0 j92-330 defeat with 10 nodes/move, which should simulate blitz 2900 elo FIDE or 2600 elo rapid FIDE (!!)
However, I don't think lc0 simulated the human well here:

[pgn]
[Event "?"]
[Site "?"]
[Date "2020.11.28"]
[Round "?"]
[White "Player"]
[Black "Player"]
[Result "1-0"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - 0 1"]
[SetUp "1"]

1. Nf3 d5 2. d4 e6 3. e3 Nf6 4. Bd2 Ne4 5. Bd3 Nd7 6. c4 a5 7. Rc1 Bd6 8. Qc2
f5 9. c5 Be7 10. O-O c6 11. Be1 O-O 12. a3 a4 13. Bb4 g5 14. Ne1 g4 15. f3 Bg5
16. Qe2 Nef6 17. Nc2 Rf7 18. Kh1 Qf8 19. Be1 h5 20. Nb4 h4 21. Kg1 Rh7 22. f4
Bh6 23. Bxh4 Bg7 24. Bg5 Kf7 25. Bxf6 Nxf6 26. Rc3 Qh8 27. g3 Bd7 28. Qc2 Rh3
29. Rd1 Qh6 30. Rd2 Rh8 31. Rg2 Nh5 32. Bf1 Nxg3 33. hxg3 Rxg3 34. Rxg3 Qh1+
35. Kf2 Rh2+ 36. Rg2 Bf6 37. Rxh2 Qxh2+ 38. Ke1 Bh4+ 39. Kd1 Bf2 40. Nd3 g3 41.
Qe2 Be8 42. Nxf2 gxf2 43. Rc2 Ke7 44. Qxf2 Qxf2 45. Rxf2 1-0
[/pgn]

[lkaufman]

Anyway I wanted to share a lc0 j92-330 defeat with 10 nodes/move, which should simulate blitz 2900 elo FIDE or 2600 elo rapid FIDE (!!)
However, I don't think lc0 simulated the human well here:

[pgn]
[Event "?"]
[Site "?"]
[Date "2020.11.28"]
[Round "?"]
[White "Player"]
[Black "Player"]
[Result "1-0"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - 0 1"]
[SetUp "1"]

1. Nf3 d5 2. d4 e6 3. e3 Nf6 4. Bd2 Ne4 5. Bd3 Nd7 6. c4 a5 7. Rc1 Bd6 8. Qc2
f5 9. c5 Be7 10. O-O c6 11. Be1 O-O 12. a3 a4 13. Bb4 g5 14. Ne1 g4 15. f3 Bg5
16. Qe2 Nef6 17. Nc2 Rf7 18. Kh1 Qf8 19. Be1 h5 20. Nb4 h4 21. Kg1 Rh7 22. f4
Bh6 23. Bxh4 Bg7 24. Bg5 Kf7 25. Bxf6 Nxf6 26. Rc3 Qh8 27. g3 Bd7 28. Qc2 Rh3
29. Rd1 Qh6 30. Rd2 Rh8 31. Rg2 Nh5 32. Bf1 Nxg3 33. hxg3 Rxg3 34. Rxg3 Qh1+
35. Kf2 Rh2+ 36. Rg2 Bf6 37. Rxh2 Qxh2+ 38. Ke1 Bh4+ 39. Kd1 Bf2 40. Nd3 g3 41.
Qe2 Be8 42. Nxf2 gxf2 43. Rc2 Ke7 44. Qxf2 Qxf2 45. Rxf2 1-0
[/pgn]

That is really bizarre. First question, who (or what) was playing White? Black totally outplayed White, then decided to play giveaway chess. Is this normal behavior for Lc0 with few nodes?

[Marcus9]

Anyway I wanted to share a lc0 j92-330 defeat with 10 nodes/move, which should simulate blitz 2900 elo FIDE or 2600 elo rapid FIDE (!!)
However, I don't think lc0 simulated the human well here:

[pgn]
[Event "?"]
[Site "?"]
[Date "2020.11.28"]
[Round "?"]
[White "Player"]
[Black "Player"]
[Result "1-0"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - 0 1"]
[SetUp "1"]

1. Nf3 d5 2. d4 e6 3. e3 Nf6 4. Bd2 Ne4 5. Bd3 Nd7 6. c4 a5 7. Rc1 Bd6 8. Qc2
f5 9. c5 Be7 10. O-O c6 11. Be1 O-O 12. a3 a4 13. Bb4 g5 14. Ne1 g4 15. f3 Bg5
16. Qe2 Nef6 17. Nc2 Rf7 18. Kh1 Qf8 19. Be1 h5 20. Nb4 h4 21. Kg1 Rh7 22. f4
Bh6 23. Bxh4 Bg7 24. Bg5 Kf7 25. Bxf6 Nxf6 26. Rc3 Qh8 27. g3 Bd7 28. Qc2 Rh3
29. Rd1 Qh6 30. Rd2 Rh8 31. Rg2 Nh5 32. Bf1 Nxg3 33. hxg3 Rxg3 34. Rxg3 Qh1+
35. Kf2 Rh2+ 36. Rg2 Bf6 37. Rxh2 Qxh2+ 38. Ke1 Bh4+ 39. Kd1 Bf2 40. Nd3 g3 41.
Qe2 Be8 42. Nxf2 gxf2 43. Rc2 Ke7 44. Qxf2 Qxf2 45. Rxf2 1-0
[/pgn]

That is really bizarre. First question, who (or what) was playing White? Black totally outplayed White, then decided to play giveaway chess. Is this normal behavior for Lc0 with few nodes?

It is probably fair to say that I cheated here, it's an idea based on how useful it can be to know your opponents.
Here's how to proceed:
- Find an engine that simulates the opponent as much as possible (extreme case: the same opponent)
- Select some better moves with komodo MCTS mulpti pv as it is the best at this
- Play all better moves and see the simulated opponent's response
- Analyze each final position with normal Komodo, and choose the move with the highest cp

I don't know if this idea can be really effective

[JJJ]

How about this handicap for human ? :

[d]rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1QQKBNR w KQkq - 0 1

Edit : Stockfish 12 dev evaluate Knight handicap at -5.07. But it does evaluate Queens for 2 pieces at -4.25. Maybe it would be the best handicap to propose instead of a 3 pawns handicap or knight handicap ?

[Chessqueen]

The idea is the following:

Everybody who want to compete pay some fixed money for the competition(let say 100$ for a game) and choose handicap that he likes to try in controlled conditions to prevent cheating and time control of 15+10.

Dragon is white but play without part of the pieces or the pawns based on the choice of the opponent.

The person who won against dragon with the smallest handicap win the competition and get all the money that humans paid in the year.

For example
If after a year we have 20 games of people who tried to play with different handicap(the same human may play more than one game) than the winner is the human who won with the samllest handicap and get 2000$.(if he played 2 games then of course the total money neto he won is practically 2000-200=1800$)

If more than one person won with the same handicap then the winner is the human who played first.

The question is if the dragon team is interested in this type of competition.
The question is also how to decide what is the smallest handicap (for examples if pawns a2,b2,c2,d2 is smaller than knight b1.

A possible way is simply to give some fixed chess engine with 1 core and 1024 mbytes hash to analyze and use the score of it after 1,000,000,000 nodes to decide what is the bigger handicap.

Aside from details, the idea of having strong players compete to see who can win at the smallest handicap is appealing. The biggest problem is that there is just an enormous gap between two pawn handicap and knight handicap. As we've seen, at two (White) pawn handicaps, the biggest ones except f2/g2, even Hikaru Nakamura lost badly (6.5 to 1.5 in Rapid, 15' + 10"). But at knight odds, even players in the 2200 to 2300 range such as both myself and my son Ray are quite competitive with Dragon at Rapid or even at slow Blitz levels. Two Black pawns is an intermediate step, especially c7 + f7, but even that is far from knight odds. Three (White) pawn handicaps are possible but the positions are really getting far away from normal chess. Knight for pawn as we did with GM Lenderman is an option, as is Exchange plus "a" pawn (with "a" rook and "b" knight off). Also the ordering of the handicaps may depend on which engine is chosen for the task, they each have their preferences, which may differ from what humans excel at.

I have a question to Mr. Kaufman knowing the difference in strength between Komodo Dragon and Nakamura and Dragon beat Naka in the 2nd day 4 to 0, if Naka was to play against a weaker engine what does the engine rating has to be in rapid or Blitz in order for GM Nakamura to give the weaker engine the same two pawns Odds ? Would you sy at least 2400 rating for the engine and what about if he was to play against another weaker human ?

[lkaufman]

How about this handicap for human ? :

[d]rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1QQKBNR w KQkq - 0 1

Edit : Stockfish 12 dev evaluate Knight handicap at -5.07. But it does evaluate Queens for 2 pieces at -4.25. Maybe it would be the best handicap to propose instead of a 3 pawns handicap or knight handicap ?

The problem with this or similar handicaps is that they are not legal chess positions, and many GUIs (including ChessBase/Fritz) won't allow them. That doesn't rule them out completely, but it means that not only will the humans have no experience with similar material, they might not even be able to prepare for games without switching to unfamiliar GUIs. We played a game with Naka once with a third knight replacing the f7 pawn, which is not a "legal" chess position but was not obviously illegal to the GUI as the Fritz GUI does allow this sort of position.

[lkaufman]

The idea is the following:

Everybody who want to compete pay some fixed money for the competition(let say 100$ for a game) and choose handicap that he likes to try in controlled conditions to prevent cheating and time control of 15+10.

Dragon is white but play without part of the pieces or the pawns based on the choice of the opponent.

The person who won against dragon with the smallest handicap win the competition and get all the money that humans paid in the year.

For example
If after a year we have 20 games of people who tried to play with different handicap(the same human may play more than one game) than the winner is the human who won with the samllest handicap and get 2000$.(if he played 2 games then of course the total money neto he won is practically 2000-200=1800$)

If more than one person won with the same handicap then the winner is the human who played first.

The question is if the dragon team is interested in this type of competition.
The question is also how to decide what is the smallest handicap (for examples if pawns a2,b2,c2,d2 is smaller than knight b1.

A possible way is simply to give some fixed chess engine with 1 core and 1024 mbytes hash to analyze and use the score of it after 1,000,000,000 nodes to decide what is the bigger handicap.

Aside from details, the idea of having strong players compete to see who can win at the smallest handicap is appealing. The biggest problem is that there is just an enormous gap between two pawn handicap and knight handicap. As we've seen, at two (White) pawn handicaps, the biggest ones except f2/g2, even Hikaru Nakamura lost badly (6.5 to 1.5 in Rapid, 15' + 10"). But at knight odds, even players in the 2200 to 2300 range such as both myself and my son Ray are quite competitive with Dragon at Rapid or even at slow Blitz levels. Two Black pawns is an intermediate step, especially c7 + f7, but even that is far from knight odds. Three (White) pawn handicaps are possible but the positions are really getting far away from normal chess. Knight for pawn as we did with GM Lenderman is an option, as is Exchange plus "a" pawn (with "a" rook and "b" knight off). Also the ordering of the handicaps may depend on which engine is chosen for the task, they each have their preferences, which may differ from what humans excel at.

I have a question to Mr. Kaufman knowing the difference in strength between Komodo Dragon and Nakamura and Dragon beat Naka in the 2nd day 4 to 0, if Naka was to play against a weaker engine what does the engine rating has to be in rapid or Blitz in order for GM Nakamura to give the weaker engine the same two pawns Odds ? Would you sy at least 2400 rating for the engine and what about if he was to play against another weaker human ?

As for weaker engines, it seems that normal A/B engines are rather poor at utilizing handicaps, they need to be much stronger than humans to beat top engines with the same handicap. But the Lc0 and similar engines set to low NPS seem to excel at this. I found that latest Lc0 set to just 0.4 nps was able to split with Dragon (on 7 fast threads) at two Black pawn handicaps at 15' + 10", but needs more that 2 nps with two White pawns. So the answer depends greatly on whether you mean classical A/B engine or Lc0 crippled engine. As for human level that would be fair opponent for Naka at this handicap, it depends mostly on time limit. There were some games at two pawn handicap (f2+g2 I think) a year or so ago in which Carlsen and MVL each gave this handicap to IM Lawrence Trent successfully, but I think those were 3 min no inc games. At 15' + 10" I would think that a fair opponent for Naka at the Dragon two pawn handicaps would be around 2300 FIDE or so; he should try it on his shows!

[Chessqueen]

The idea is the following:

Everybody who want to compete pay some fixed money for the competition(let say 100$ for a game) and choose handicap that he likes to try in controlled conditions to prevent cheating and time control of 15+10.

Dragon is white but play without part of the pieces or the pawns based on the choice of the opponent.

The person who won against dragon with the smallest handicap win the competition and get all the money that humans paid in the year.

For example
If after a year we have 20 games of people who tried to play with different handicap(the same human may play more than one game) than the winner is the human who won with the samllest handicap and get 2000$.(if he played 2 games then of course the total money neto he won is practically 2000-200=1800$)

If more than one person won with the same handicap then the winner is the human who played first.

The question is if the dragon team is interested in this type of competition.
The question is also how to decide what is the smallest handicap (for examples if pawns a2,b2,c2,d2 is smaller than knight b1.

A possible way is simply to give some fixed chess engine with 1 core and 1024 mbytes hash to analyze and use the score of it after 1,000,000,000 nodes to decide what is the bigger handicap.

Aside from details, the idea of having strong players compete to see who can win at the smallest handicap is appealing. The biggest problem is that there is just an enormous gap between two pawn handicap and knight handicap. As we've seen, at two (White) pawn handicaps, the biggest ones except f2/g2, even Hikaru Nakamura lost badly (6.5 to 1.5 in Rapid, 15' + 10"). But at knight odds, even players in the 2200 to 2300 range such as both myself and my son Ray are quite competitive with Dragon at Rapid or even at slow Blitz levels. Two Black pawns is an intermediate step, especially c7 + f7, but even that is far from knight odds. Three (White) pawn handicaps are possible but the positions are really getting far away from normal chess. Knight for pawn as we did with GM Lenderman is an option, as is Exchange plus "a" pawn (with "a" rook and "b" knight off). Also the ordering of the handicaps may depend on which engine is chosen for the task, they each have their preferences, which may differ from what humans excel at.

I have a question to Mr. Kaufman knowing the difference in strength between Komodo Dragon and Nakamura and Dragon beat Naka in the 2nd day 4 to 0, if Naka was to play against a weaker engine what does the engine rating has to be in rapid or Blitz in order for GM Nakamura to give the weaker engine the same two pawns Odds ? Would you sy at least 2400 rating for the engine and what about if he was to play against another weaker human ?

As for weaker engines, it seems that normal A/B engines are rather poor at utilizing handicaps, they need to be much stronger than humans to beat top engines with the same handicap. But the Lc0 and similar engines set to low NPS seem to excel at this. I found that latest Lc0 set to just 0.4 nps was able to split with Dragon (on 7 fast threads) at two Black pawn handicaps at 15' + 10", but needs more that 2 nps with two White pawns. So the answer depends greatly on whether you mean classical A/B engine or Lc0 crippled engine. As for human level that would be fair opponent for Naka at this handicap, it depends mostly on time limit. There were some games at two pawn handicap (f2+g2 I think) a year or so ago in which Carlsen and MVL each gave this handicap to IM Lawrence Trent successfully, but I think those were 3 min no inc games. At 15' + 10" I would think that a fair opponent for Naka at the Dragon two pawn handicaps would be around 2300 FIDE or so; he should try it on his shows!

My next question is can your son Ray draw 4 games versus Naka or GM SO at At 15' + 10" at the Dragon two pawn handicaps on his shows?

[Nordlandia]

JJJ: i rather go for Dragon evaluation in this odds configuration, the reason is said below.

About +3.70 for white if human starts first

[d]rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1QQKBNR w KQkq - 0 1

About -3 for black if computer starts.

[d]r1qqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1

If the human player play black pieces, the odds is approximately equivalent minor piece odds. Another important thing to point out here is that Dragon/Komodo know that multiple queens tends to be redundant on a crowded board. Dragon/Komodo may want to avoid queen exchanges to exploit this redundancy as long as possible.

As long the extra illegal material is within reasonable limits i think Dragon may play without any issues Larry.

[Uri Blass]

JJJ: i rather go for Dragon evaluation in this odds configuration, the reason is said below.

About +3.70 for white if human starts first

[d]rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/R1QQKBNR w KQkq - 0 1

About -3 for black if computer starts.

[d]r1qqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1

If the human player play black pieces, the odds is approximately equivalent minor piece odds. Another important thing to point out here is that Dragon/Komodo know that multiple queens tends to be redundant on a crowded board. Dragon/Komodo may want to avoid queen exchanges to exploit this redundancy as long as possible.

As long the extra illegal material is within reasonable limits i think Dragon may play without any issues Larry.

I do not like these illegal positions
humans do not play chess in similiar conditions.

I think that bigger handicap can be also interesting so you can play games against weaker humans.
I am sure that it is possible to win most humans who play in chess.com without a queen even at rapid time control(most are with rating below 1000 in rapid) and I think that it may be interesting to see what is the rating that computers can get in chess.com if they start every game without queen d1
at different time controls and if it is possible to improve their playing strength in these conditions.