DiscoCheck 4.3 is out

[Mike S.]

Maybe there are only ~20 Elo to go. After a very short test, I came to the conclusion that DiscoCheck 4.3 could be ~2695 (as per CEGT/Blitz level):

http://forum.computerschach.de/cgi-bin/ ... ?pid=62161

But that were 60 1+0.5 games only (and involve one loss of Crafty in a winning position; I have stability problems with Crafty 23.6 in Arena + Win8). But it is not statistically reliable anyway, only to show a potential.

I would delay Multi-PV as long as your engine is not top-20; ponder mode and finding some more Elos seems the obvious goal for now.

Thanks for a fine engine. An example (no technical issues in the game):
[pgn]
[Event "Disco43"]
[Site "Wien"]
[Date "2013.07.15"]
[Round "9"]
[White "DiscoCheck 4.3"]
[Black "Crafty 23.6"]
[Result "1-0"]
[BlackElo "2611"]
[ECO "A07"]
[Opening "Reti Opening"]
[TimeControl "60+0"]
[Termination "adjudication"]
[PlyCount "87"]

1. g3 d5 2. Nf3 Bg4 3. Bg2 Nd7 4. c4 e6 5. d3 Ngf6 {+0.09/14 6} 6. cxd5
{+0.08/12 2} exd5 {-0.01/14 2} 7. Nc3 {-0.08/13 2} Bd6 {+0.02/16 5} 8. Nd4
{+0.03/14 4} Be5 {+0.01/15 4} 9. Nf3 {+0.01/13 2} Bd6 {+0.01/20 10} 10. Nd4
{+0.03/15 2} Be5 {+0.01/19 7} 11. Nb3 {-0.04/13 2} c6 {+0.36/13 0} 12. d4
{+0.10/15 2} Bd6 {+0.41/13 0} 13. f3 {+0.30/15 2} Bh5 {+0.45/13 0} 14. g4
{+0.23/14 2} Bg6 {+0.34/13 0} 15. e4 {+0.25/14 1} dxe4 {+0.22/12 0} 16.
fxe4 {+0.36/12 0} Bb4 {+0.24/12 0} 17. O-O {+0.47/14 1} Bxc3 {+0.27/13 0}
18. bxc3 {+0.71/13 0} h5 {-0.11/13 1} 19. g5 {+0.94/13 2} Nh7 {-0.93/13 0}
20. h4 {+0.88/13 1} f6 {-0.93/12 0} 21. gxf6 {+1.66/12 1} gxf6 {-0.98/12 0}
22. Bf4 {+1.69/12 1} Rg8 {-0.96/13 0} 23. Nd2 {+1.64/12 1} Qe7 {-1.10/13 1}
24. Kh1 {+1.55/12 1} Bf7 {-0.68/11 0} 25. Bh3 {+1.32/12 1} f5 {-0.91/12 1}
26. Bxf5 {+1.54/12 0} Qxh4+ {-1.03/13 0} 27. Bh2 {+1.50/13 2} Rg7 {-0.87/13
0} 28. Rb1 {+2.55/11 1} Nb6 {-0.98/12 0} 29. a4 {+2.36/11 1} Qe7 {-1.66/12
1} 30. a5 {+2.78/13 1} Nd5 {-2.13/12 1} 31. Qf3 {+2.20/12 1} Ndf6 {-1.99/12
0} 32. Be5 {+2.90/11 1} Ba2 {-2.28/12 1} 33. Rb2 {+3.18/12 1} Be6 {-2.53/12
1} 34. Rfb1 {+2.45/11 1} Bxf5 {-2.61/12 0} 35. Qxf5 {+2.14/10 0} Qd7
{-3.05/13 0} 36. Qxd7+ {+2.66/13 1} Rxd7 {-2.75/14 0} 37. Rxb7 {+2.66/12 0}
h4 {-2.92/14 1} 38. a6 {+3.07/13 1} Rad8 {-3.35/13 0} 39. Rg1 {+3.18/13 1}
Rf7 {-3.86/13 1} 40. Nc4 {+3.96/15 1} Rfd7 {-4.83/14 1} 41. Bc7 {+4.91/14
1} Ra8 {-6.22/13 0} 42. e5 {+6.79/13 1} Ne4 {-7.27/14 1} 43. e6 {+6.95/11
1} Nhf6 {-8.44/13 1} 44. exd7+ {+7.09/11 0} ... {1-0 Arena Adjudication}
1-0
[/pgn]

[lucasart]

But I still think that DiscoCheck is too weak to compete in the IPON rating list.

Hey man !

DiscoCheck 4.3 is a very good chess engine. Nemo 1.0.1 is the number 104 in IPON rating list and DC 4.3 is very-very-very close (tomorrow nigh my test versus Nemo will be posted in CCC) !

Good luck !

SilvianR

Your test might simply show a lucky run for DiscoCheck.

Nemo 1.01 is 2850 on CCRL 40/4, and DC 4.1 is 2744
http://www.computerchess.org.uk/ccrl/404/

Based on my estimations, DC 4.3 could be 50 ELO better than DC 4.1, but in the past, I have often overstated these estimations (mostly because they are based on self-play and at very fast time control, like 10"+0.1" or similar).

So I reckon it could reach 2800 CCRL, but that's still significantly below Nemo, and significantly below Toga 3.0 (2884) which is the minimum level to enter the IPON.

So DC is not quite there yet, ELO-wise.

I like its playing style though. Really a mix of unsound, antipositional, unorthodox, materialistic. Quite fun to watch sometimes

[lucasart]

I would delay Multi-PV as long as your engine is not top-20; ponder mode and finding some more Elos seems the obvious goal for now.

Multi-PV is really a great feature, I think. When I analyze a game, I'm more interested in the list of tactically acceptable moves, rather than the one that the engine would choose, based on its defective evaluation. I don't trust the positional judgment too much, and use it more to see if a move holds tactically or not. Multi-PV is really nice and you can easily identify the list of acceptable moves to explore.

On the other hand, why use an inferior engine like DiscoCheck to analyze, when some much stronger free alternatives exist. For me the reason is obvious: sentimental, it's my brain child. But for anyone else, it makes much more sense to use Critter, Stockfish or IvanHoe (or Komodo or Houdini if you have them).

[lucasart]

[pgn]
[Event "Disco43"]
[Site "Wien"]
[Date "2013.07.15"]
[Round "9"]
[White "DiscoCheck 4.3"]
[Black "Crafty 23.6"]
[Result "1-0"]
[BlackElo "2611"]
[ECO "A07"]
[Opening "Reti Opening"]
[TimeControl "60+0"]
[Termination "adjudication"]
[PlyCount "87"]

1. g3 d5 2. Nf3 Bg4 3. Bg2 Nd7 4. c4 e6 5. d3 Ngf6 {+0.09/14 6} 6. cxd5
{+0.08/12 2} exd5 {-0.01/14 2} 7. Nc3 {-0.08/13 2} Bd6 {+0.02/16 5} 8. Nd4
{+0.03/14 4} Be5 {+0.01/15 4} 9. Nf3 {+0.01/13 2} Bd6 {+0.01/20 10} 10. Nd4
{+0.03/15 2} Be5 {+0.01/19 7} 11. Nb3 {-0.04/13 2} c6 {+0.36/13 0} 12. d4
{+0.10/15 2} Bd6 {+0.41/13 0} 13. f3 {+0.30/15 2} Bh5 {+0.45/13 0} 14. g4
{+0.23/14 2} Bg6 {+0.34/13 0} 15. e4 {+0.25/14 1} dxe4 {+0.22/12 0} 16.
fxe4 {+0.36/12 0} Bb4 {+0.24/12 0} 17. O-O {+0.47/14 1} Bxc3 {+0.27/13 0}
18. bxc3 {+0.71/13 0} h5 {-0.11/13 1} 19. g5 {+0.94/13 2} Nh7 {-0.93/13 0}
20. h4 {+0.88/13 1} f6 {-0.93/12 0} 21. gxf6 {+1.66/12 1} gxf6 {-0.98/12 0}
22. Bf4 {+1.69/12 1} Rg8 {-0.96/13 0} 23. Nd2 {+1.64/12 1} Qe7 {-1.10/13 1}
24. Kh1 {+1.55/12 1} Bf7 {-0.68/11 0} 25. Bh3 {+1.32/12 1} f5 {-0.91/12 1}
26. Bxf5 {+1.54/12 0} Qxh4+ {-1.03/13 0} 27. Bh2 {+1.50/13 2} Rg7 {-0.87/13
0} 28. Rb1 {+2.55/11 1} Nb6 {-0.98/12 0} 29. a4 {+2.36/11 1} Qe7 {-1.66/12
1} 30. a5 {+2.78/13 1} Nd5 {-2.13/12 1} 31. Qf3 {+2.20/12 1} Ndf6 {-1.99/12
0} 32. Be5 {+2.90/11 1} Ba2 {-2.28/12 1} 33. Rb2 {+3.18/12 1} Be6 {-2.53/12
1} 34. Rfb1 {+2.45/11 1} Bxf5 {-2.61/12 0} 35. Qxf5 {+2.14/10 0} Qd7
{-3.05/13 0} 36. Qxd7+ {+2.66/13 1} Rxd7 {-2.75/14 0} 37. Rxb7 {+2.66/12 0}
h4 {-2.92/14 1} 38. a6 {+3.07/13 1} Rad8 {-3.35/13 0} 39. Rg1 {+3.18/13 1}
Rf7 {-3.86/13 1} 40. Nc4 {+3.96/15 1} Rfd7 {-4.83/14 1} 41. Bc7 {+4.91/14
1} Ra8 {-6.22/13 0} 42. e5 {+6.79/13 1} Ne4 {-7.27/14 1} 43. e6 {+6.95/11
1} Nhf6 {-8.44/13 1} 44. exd7+ {+7.09/11 0} ... {1-0 Arena Adjudication}
1-0
[/pgn]

As a side note, this game demonstrated that the new Contempt option works as intended. By default Contempt = 25cp, meaning that the engine thinks that a draw (by chess rules, 3-rep, 50-move, stalemate, insufficient material) is worth -25cp from its point of view, and +25cp from the point of view of the opponent.

Here DC playes 11. Nb3 showing a negative score of -4cp. So it decided to refuse the 3-repetition (worth -25cp instead of 0) and thereby "sacrifice" 4cp.

I did a study on the impact of this:
http://open-chess.org/viewtopic.php?f=5&t=2321
Basically it doesn't cost (or gain) any ELO, but it reduces significantly the draw rate, especially those unfought 3-repetition draws that are so pathetic to see in computer matches.

[hgm]

Do you stop thinking when its your opponents move? Without pondering the engine is playing a chess variant, but not proper chess

You have a very funny notion of what constitutes a Chess variant. So a GM giving a simul exhibition is not playing proper Chess?

[hgm]

Multi-PV is really a great feature, I think. When I analyze a game, I'm more interested in the list of tactically acceptable moves, rather than the one that the engine would choose, based on its defective evaluation. I don't trust the positional judgment too much, and use it more to see if a move holds tactically or not. Multi-PV is really nice and you can easily identify the list of acceptable moves to explore.

Multi-PV is completely trivial to implement; even Fairy-Max has it, in its 150-line code. Just increase alpha to bestScore - margin, in stead of to bestScore, each time you increase bestScore in the root, and that does the trick.

[Mike S.]

Not being a programmer, I thought your contempt factor of 25 cp. is quite big. But if it works for you in terms of Elo, great.

Nevertheless, imagine you have a 0.00 position against Houdini or the likes, would it be wise to sac up to 25 cp...? That will often turn the draw into a loss. And probably never into a win, as long as DiscoCheck is not top-6.

But it is just a matter of time!

[IWB]

Do you stop thinking when its your opponents move? Without pondering the engine is playing a chess variant, but not proper chess

You have a very funny notion of what constitutes a Chess variant. So a GM giving a simul exhibition is not playing proper Chess?

You have a very funny way of nitpicking ... and yes he is playing a variant which he himself is not considering as valuable as a full one on one game!

Bye
Ingo

[JVMerlino]

Do you stop thinking when its your opponents move? Without pondering the engine is playing a chess variant, but not proper chess

You have a very funny notion of what constitutes a Chess variant. So a GM giving a simul exhibition is not playing proper Chess?

You have a very funny way of nitpicking ... and yes he is playing a variant which he himself is not considering as valuable as a full one on one game!

Bye
Ingo

"Proper chess" has no consistent meaning from person to person. But I'm pretty sure there is nothing in the rules of what you are calling "proper chess" that says you MUST think while you are not on the clock.

jm

[IWB]

.