More than elo I would welcome a slight change in playing style. Komodo is supposed to have a positional style, whatever that means. Lately I haven't been very impressed by K's style because it lacks 'ambition' compared to H6 for example. More often than not K manages to win by 'waiting' until the opponent makes a mistake. Elowise this might be a good achievement, but it ain't very exciting. In my view K is not realizing its full potential with this 'wait-and-see' approach.
Ron
Next Komodo vs GM match.
Moderators: hgm, Rebel, chrisw
-
- Posts: 5960
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
Re: Next Komodo vs GM match.
Higher King Safety, higher Dynamism, and higher Contempt should all make Komodo more aggressive at some Elo cost. It's been reported that super-high Dynamism settings do great on problem sets and on matching some of Alpha-zero's surprising sacrifices, but will show large elo losses in engine play.Ron Langeveld wrote:More than elo I would welcome a slight change in playing style. Komodo is supposed to have a positional style, whatever that means. Lately I haven't been very impressed by K's style because it lacks 'ambition' compared to H6 for example. More often than not K manages to win by 'waiting' until the opponent makes a mistake. Elowise this might be a good achievement, but it ain't very exciting. In my view K is not realizing its full potential with this 'wait-and-see' approach.
Ron
Komodo rules!
-
- Posts: 282
- Joined: Mon Sep 02, 2013 8:23 am
Re: Next Komodo vs GM match.
Yes, that's the reason (correctly or not) I subscribe to Komodo...if I just want ELO, Stockfish is there. But I want....more, from an engine. I am not sure I am getting it, but you spend your money and take your chances.lkaufman wrote:Higher King Safety, higher Dynamism, and higher Contempt should all make Komodo more aggressive at some Elo cost. It's been reported that super-high Dynamism settings do great on problem sets and on matching some of Alpha-zero's surprising sacrifices, but will show large elo losses in engine play.Ron Langeveld wrote:More than elo I would welcome a slight change in playing style. Komodo is supposed to have a positional style, whatever that means. Lately I haven't been very impressed by K's style because it lacks 'ambition' compared to H6 for example. More often than not K manages to win by 'waiting' until the opponent makes a mistake. Elowise this might be a good achievement, but it ain't very exciting. In my view K is not realizing its full potential with this 'wait-and-see' approach.
Ron
-
- Posts: 256
- Joined: Sun May 16, 2010 2:50 am
Re: Next Komodo vs GM match.
With a high enough contempt I believe you can win 4-0. What will be the contempt for the match?
-
- Posts: 5960
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
Re: Next Komodo vs GM match.
I haven't decided yet, perhaps somewhere around 100 or so. I haven't found it wise to go too high with Contempt, it makes Komodo play dubious moves to avoid swaps.Based on Komodo vs sKomodo bullet games, the average elo value of the four handicaps is 872 elo points.
Komodo rules!
-
- Posts: 256
- Joined: Sun May 16, 2010 2:50 am
Re: Next Komodo vs GM match.
Perhaps a higher contempt could have komodo complicate positions more like contempt of 150. Perhaps do some testings vs 2600 elo engines. Just an idea.
-
- Posts: 5960
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
Re: Next Komodo vs GM match.
Contempt of 150 or so is probably reasonable for knight odds matches, I think I used this value in some of them. But I'm pretty sure it's too high for a two pawn handicap game. The displayed evaluation of the initial handicap position should still be significantly negative or else the Contempt value is too high. The judgment is in the word "significantly".jmartus wrote:Perhaps a higher contempt could have komodo complicate positions more like contempt of 150. Perhaps do some testings vs 2600 elo engines. Just an idea.
Komodo rules!
-
- Posts: 4556
- Joined: Tue Jul 03, 2007 4:30 am
Re: Next Komodo vs GM match.
Have you considering "Dynamic Contempt"? The idea here is that Komodo wants to believe that it has some advantage in the position, say, a 0.90 advantage, so, on the next move, if the previous one had 0.80 advantage, it raises contempt by 10, but if it was 1.00, it reduces it by 10.lkaufman wrote:Contempt of 150 or so is probably reasonable for knight odds matches, I think I used this value in some of them. But I'm pretty sure it's too high for a two pawn handicap game. The displayed evaluation of the initial handicap position should still be significantly negative or else the Contempt value is too high. The judgment is in the word "significantly".
The idea here is that Komodo adjusts its contempt depending on how the opponent plays, so that, if Komodo has already equalized, it now plays with Contempt 0 because it assumes it can beat the human from an equal position.
-
- Posts: 5960
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
Re: Next Komodo vs GM match.
The idea of contempt is mostly to avoid simplification. If we are playing a weaker player, we want to avoid simplifying in equal positions to minimize drawjchances. True, a draw is unlikely, but it is more likely without contempt than with it. But perhaps we could have it vary a bit with the eval, maybe one point per five or ten eval points.Ovyron wrote:Have you considering "Dynamic Contempt"? The idea here is that Komodo wants to believe that it has some advantage in the position, say, a 0.90 advantage, so, on the next move, if the previous one had 0.80 advantage, it raises contempt by 10, but if it was 1.00, it reduces it by 10.lkaufman wrote:Contempt of 150 or so is probably reasonable for knight odds matches, I think I used this value in some of them. But I'm pretty sure it's too high for a two pawn handicap game. The displayed evaluation of the initial handicap position should still be significantly negative or else the Contempt value is too high. The judgment is in the word "significantly".
The idea here is that Komodo adjusts its contempt depending on how the opponent plays, so that, if Komodo has already equalized, it now plays with Contempt 0 because it assumes it can beat the human from an equal position.
Komodo rules!
-
- Posts: 5960
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
Re: Next Komodo vs GM match.
Here is game 1. Izoria gave back one of the two pawns to reach an endgame, but he overlooked that his 22...Re2 move did not really have a threat. Later, 30...Kf8 instead of ...Kh8 might have held the draw. Next game at 5 pm Eastern time on chess.com.
[pgn]
[Event "Rapid 45 min + 15 sec"]
[Site "chess.com"]
[Date "2018.01.15"]
[Round "1"]
[White "Komodo 2000.00 64-bit"]
[Black "Izoria, Zviad"]
[Result "1-0"]
[Annotator "Kaufman,Larry"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/P1PPPP1P/RNBQKBNR w KQkq - 0 1"]
[PlyCount "71"]
[TimeControl "2100+8"]
{6144MB, Book2016.ctg, DESKTOP-16c} 1. c4 {-0.51/26 30} Nf6 {231} 2. Bb2 {
-0.54/28 78} g6 {63} 3. Nc3 {-0.80/30 78} Bg7 {43} 4. Nd5 {-0.64/27 22} O-O {56
} 5. Nxf6+ {-0.83/31 26} Bxf6 {51} 6. Bxf6 {-0.65/29 29} exf6 {14} 7. Bg2 {
-0.57/31 26} Nc6 {33} 8. e3 {-0.58/31 58} d6 {170} 9. Qb3 {-0.53/29 44} Na5 {
510} 10. Qc3 {-0.51/31 38} c5 {20} 11. d3 {-0.48/33 40} Be6 {59} 12. Ne2 {
-0.74/35 0} d5 {50} 13. cxd5 {-0.59/35 5} Bxd5 {16} 14. e4 {-0.59/34 15} Be6 {
38} 15. Qxc5 {-0.46/33 43} Nc6 {159} 16. d4 {-0.53/40 0} Qa5+ {83} 17. Qxa5 {
0.00/32 25} Nxa5 {14} 18. d5 {-0.14/31 30} Bd7 {35} 19. O-O {0.00/34 57} Rfc8 {
166} 20. f4 {0.00/34 91} Rc2 {239} 21. Nd4 {0.00/37 42} Rd2 {54} 22. Nf3 {
0.00/39 0} Re2 {42} 23. Rfc1 {0.26/33 35} Rc8 {344} 24. Rxc8+ {0.06/40 0} Bxc8
{14} 25. Rc1 {0.03/36 35} Bg4 {65} 26. e5 {0.04/38 0} fxe5 {119} 27. Nxe5 {
0.07/38 0} Bf5 {83} 28. Rc7 {0.00/41 0} Rxa2 {314} 29. Nxf7 {0.00/46 0} Rd2 {
529} 30. Nh6+ {0.00/50 0} Kh8 {160} 31. h4 {0.35/32 38} b5 {100} 32. h5 {
1.68/32 20} a6 {305} 33. Bf3 {9.36/32 43} Nc4 {52} 34. Nxf5 {10.47/33 180} gxf5
{4} 35. h6 {250.00/36 34} Kg8 {17} 36. Bh5 {250.00/41 0} 1-0
[/pgn]
[pgn]
[Event "Rapid 45 min + 15 sec"]
[Site "chess.com"]
[Date "2018.01.15"]
[Round "1"]
[White "Komodo 2000.00 64-bit"]
[Black "Izoria, Zviad"]
[Result "1-0"]
[Annotator "Kaufman,Larry"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/P1PPPP1P/RNBQKBNR w KQkq - 0 1"]
[PlyCount "71"]
[TimeControl "2100+8"]
{6144MB, Book2016.ctg, DESKTOP-16c} 1. c4 {-0.51/26 30} Nf6 {231} 2. Bb2 {
-0.54/28 78} g6 {63} 3. Nc3 {-0.80/30 78} Bg7 {43} 4. Nd5 {-0.64/27 22} O-O {56
} 5. Nxf6+ {-0.83/31 26} Bxf6 {51} 6. Bxf6 {-0.65/29 29} exf6 {14} 7. Bg2 {
-0.57/31 26} Nc6 {33} 8. e3 {-0.58/31 58} d6 {170} 9. Qb3 {-0.53/29 44} Na5 {
510} 10. Qc3 {-0.51/31 38} c5 {20} 11. d3 {-0.48/33 40} Be6 {59} 12. Ne2 {
-0.74/35 0} d5 {50} 13. cxd5 {-0.59/35 5} Bxd5 {16} 14. e4 {-0.59/34 15} Be6 {
38} 15. Qxc5 {-0.46/33 43} Nc6 {159} 16. d4 {-0.53/40 0} Qa5+ {83} 17. Qxa5 {
0.00/32 25} Nxa5 {14} 18. d5 {-0.14/31 30} Bd7 {35} 19. O-O {0.00/34 57} Rfc8 {
166} 20. f4 {0.00/34 91} Rc2 {239} 21. Nd4 {0.00/37 42} Rd2 {54} 22. Nf3 {
0.00/39 0} Re2 {42} 23. Rfc1 {0.26/33 35} Rc8 {344} 24. Rxc8+ {0.06/40 0} Bxc8
{14} 25. Rc1 {0.03/36 35} Bg4 {65} 26. e5 {0.04/38 0} fxe5 {119} 27. Nxe5 {
0.07/38 0} Bf5 {83} 28. Rc7 {0.00/41 0} Rxa2 {314} 29. Nxf7 {0.00/46 0} Rd2 {
529} 30. Nh6+ {0.00/50 0} Kh8 {160} 31. h4 {0.35/32 38} b5 {100} 32. h5 {
1.68/32 20} a6 {305} 33. Bf3 {9.36/32 43} Nc4 {52} 34. Nxf5 {10.47/33 180} gxf5
{4} 35. h6 {250.00/36 34} Kg8 {17} 36. Bh5 {250.00/41 0} 1-0
[/pgn]
Komodo rules!