Opening ideas for TCEC 9

[Henk]

Engines should choose there own opening lines.

[supersharp77]

I don't think you should go just by eval. If you do that you may just include inferior opening lines that won't lead to more interesting games, just more one-sided games.

I think if you want to make it more interesting you should go for complex positions, regardless of eval.

Here is one for example (Najdorf, B97):

1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Bg5 e6 7. f4 Qb6 8. Qd2 Qxb2 9. Rb1 Qa3 10. Bxf6 gxf6 11. e5

11. e5 is probably inferior to 11. f5 but leads to very complex play.

--Jon

----------------------------------------------------------------------
Well my experience shows that opening books should be allowed (for interesting and sharp play) and all opening lines (except the obviously inferior ones) in to be played (gui books discard bad lines for the most part) with the gui opening books limited to
8-10 moves max.....computers quicky go "out of book" early for the most part....most engine books 90% useless.....a opening book will not "win" a engine game against a strong program alone...(waste of time).....
AR

[carldaman]

I don't think you should go just by eval. If you do that you may just include inferior opening lines that won't lead to more interesting games, just more one-sided games.

I think if you want to make it more interesting you should go for complex positions, regardless of eval.

Here is one for example (Najdorf, B97):

1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Bg5 e6 7. f4 Qb6 8. Qd2 Qxb2 9. Rb1 Qa3 10. Bxf6 gxf6 11. e5

11. e5 is probably inferior to 11. f5 but leads to very complex play.

--Jon

Precisely. The more complex, the likelier will be for engines to go wrong somehow, leading to more decisive games, or at least highly entertaining games, if drawn.

[jdart]

I like "own book" tournaments myself (like the YAT tournament Ed Schroder is running now) but with engines of the same general strength you are then measuring both the quality of the engine and the quality of the book. While I agree a good book can't make a winner out of a weak engine, for strong engines competing with each other a good book makes a substantial difference. If you want to see this look at the games of account "Polyphemus" on ICC, which is using a strong private opening book (1337Chess Pro - http://1337chess.blogspot.com/p/blog-page.html).

--Jon

[carldaman]

In tcec 8 there was a distinction between "balanced" and "unbalanced" opening positions. "Balanced" openings occurred in 66% of the games.

Continuing that approach requires two different sets of openings. But first, I would like to see unambiguous definitions of "balanced" and "unbalanced".

My understanding is:

"balanced" is when the sets of pieces are exactly the same or will be the same after an obvious recapture on the next move. No distinction is made for bishops if they are on different colors.

"unbalanced" is not the above.

If I'm missing something, please let me know.

Balanced/unbalanced refers to the engines' evaluation score when exiting the book. A drawish eval score will be considered "balanced", whereas a more lopsided score, such as .70-.80 for White, will be taken to be "unbalanced".

This is a flawed concept in my opinion. We are asking the engines to give us an evaluation of opening positions, when this is often the weakest part of their games, and then we take that score for granted, as if absolutely correct.

CL

[carldaman]

I would like to see more King's Indians, especially the Classical variation. Even though the engines consider the latter as being -0.75 for Black, in my private tests between top engines, Black scores very well. This makes the initial, out-of-book eval score likely to be incorrect.

Slowly but surely, engine play is improving even in the type of closed positions they used to have major trouble with.

Cheers,
CL

There have been discussions about how to make TCEC 9's superfinal more interesting than TCEC8's, and an opening exit eval of about .80 has been suggested by Kai Laskos as ideal. In this thread I'll post some openings I've found that come close to that, and others can post their openings as well. Perhaps we can work on a list of openings that the organizers of TCEC may consider for TCEC9.

Evals from Komodo 9.3:

1) -.79 d29 1. e4 c6 2. d4 d5 3. Nc3 dxe4 4. f3 exf3 5. Nxf3 Nf6 6. Bc4 Bf5 7.Ne5 e6 8. O-O


2) .75 d31 1. f4 Nc6 2. Nf3 d6 3. e4 g6 4. d4 e6 5. c3
Nf6 6. Bd3 Bd7

3) -.75 d31
1. e4 e5 2. f4 Bc5 3. Qg4 Nf6

4) .89 d34 1. d4 d6 2. c4 g6 3. Nf3 Bg7 4. g3 c6 5.
Nc3 e5 6. Bg2 Bg4

[Leto]

There have been discussions about how to make TCEC 9's superfinal more interesting than TCEC8's, and an opening exit eval of about .80 has been suggested by Kai Laskos as ideal. In this thread I'll post some openings I've found that come close to that, and others can post their openings as well. Perhaps we can work on a list of openings that the organizers of TCEC may consider for TCEC9.

Evals from Komodo 9.3:

1) -.79 d29 1. e4 c6 2. d4 d5 3. Nc3 dxe4 4. f3 exf3 5. Nxf3 Nf6 6. Bc4 Bf5 7.Ne5 e6 8. O-O


2) .75 d31 1. f4 Nc6 2. Nf3 d6 3. e4 g6 4. d4 e6 5. c3
Nf6 6. Bd3 Bd7

3) -.75 d31
1. e4 e5 2. f4 Bc5 3. Qg4 Nf6

4) .89 d34 1. d4 d6 2. c4 g6 3. Nf3 Bg7 4. g3 c6 5.
Nc3 e5 6. Bg2 Bg4

I tested the first opening at 3 minute per 40 moves between Komodo 9.3 and Houdini 4 both with 12 cores, black won all ten games despite H4 being about 150 elo weaker.

I then did 10 games of Komodo 9.3 with 12 cores vs Critter 1.6 with 4 cores, K93 scored 80% managing to win 3 games with white and drew a game with white and a game with black and won the rest of the games with black. Keep in mind K93 with 12 cores is about 300 elo stronger than C16 with 4.

So this shows that you have to be very careful when selecting unbalanced openings.

[Hai]

20 possible opening moves.
Draw rate ~70%
Games with opening phase and not without.

[Leto]

I don't think you should go just by eval. If you do that you may just include inferior opening lines that won't lead to more interesting games, just more one-sided games.

I think if you want to make it more interesting you should go for complex positions, regardless of eval.

Here is one for example (Najdorf, B97):

1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Bg5 e6 7. f4 Qb6 8. Qd2 Qxb2 9. Rb1 Qa3 10. Bxf6 gxf6 11. e5

11. e5 is probably inferior to 11. f5 but leads to very complex play.

--Jon

This one is probably busted for white. Critter 1.6 x64 4CPU playing black managed to beat Komodo 9.3 x64 8CPU despite K93 being about 300 elo stronger.

[Event "Jon Dart Najdorf, 3m/40"]
[Site "?"]
[Date "2015.12.05"]
[Round "1.1"]
[White "Komodo 9.3 64-bit 8"]
[Black "Critter 1.6 64-bit 4"]
[Result "0-1"]
[ECO "B97"]
[Annotator "-2.10;-1.71"]
[PlyCount "84"]
[EventDate "2015.12.05"]
[EventType "tourn"]
[Source "Atreides"]

{Intel(R) Xeon(R) CPU E5620 @ 2.40GHz 2393 MHz W=19.5 plies; 5,
527kN/s; 7,910,403 TBAs B=19.0 plies; 6,334kN/s} 1. e4 c5 2. Nf3 d6 3. d4 cxd4
4. Nxd4 Nf6 5. Nc3 a6 6. Bg5 e6 7. f4 Qb6 8. Qd2 Qxb2 9. Rb1 Qa3 10. Bxf6 gxf6
11. e5 fxe5 {-1.71/19 19} 12. fxe5 {-2.10/22 12} Bh6 {-1.64/20 23 (dxe5)} 13.
Qxh6 {-1.97/21 5} Qxc3+ {-1.64/18 0} 14. Qd2 {-1.99/22 3} Qxd2+ {-1.73/19 9}
15. Kxd2 {0.01/5 0} dxe5 {-1.73/19 0} 16. Nf3 {-1.96/22 9 (Nb3)} Nd7 {-1.75/18
4} 17. Ke3 {-2.16/22 13 (Be2)} f5 {-1.84/19 9 (Ke7)} 18. Be2 {-2.13/23 14 (g3)}
Ke7 {-1.89/18 4} 19. Rhd1 {-2.25/23 11 (g3)} e4 {-1.96/17 4} 20. Nd4 {-2.30/23
10 (Nd2)} Nf6 {-2.09/18 5} 21. c4 {-2.23/23 8} e5 {-2.22/18 4 (h5)} 22. Nxf5+ {
-2.17/17 3} Bxf5 {-2.22/17 0} 23. Rxb7+ {-2.08/18 4} Bd7 {-2.16/18 4} 24. Rb6 {
-2.33/20 7 (g4)} Rhf8 {-2.28/16 0} 25. g4 {-2.61/20 8 (Rb7)} Nxg4+ {-2.73/19 4
(Be6)} 26. Bxg4 {-2.23/18 3} Bxg4 {-2.73/18 0} 27. Rd5 {-2.33/19 4} Rf3+ {
-2.79/19 7 (Rf6)} 28. Kxe4 {-1.88/15 2} Rf4+ {-2.91/20 7 (Rf2)} 29. Kxe5 {
-1.79/16 2} Rxc4 {-2.91/19 0} 30. Rb7+ {-1.74/17 2} Kf8 {-3.02/21 7} 31. Rxh7 {
-1.81/19 4} Kg8 {-2.90/20 7} 32. Re7 {-1.90/20 16} Rf8 {-3.08/20 7} 33. Rd4 {
-1.96/18 6 (Kd6)} Rf5+ {-3.17/19 6 (Rxd4)} 34. Ke4 {-1.94/18 2} Rxd4+ {-3.17/
18 0} 35. Kxd4 {-1.97/19 2} Ra5 {-3.17/23 16 (Rf2)} 36. Ke3 {-1.95/18 5 (Re4)}
Rxa2 {-3.60/19 6} 37. Kf4 {-2.26/18 3 (Re5)} Bd1 {-3.83/19 6 (Bh3)} 38. Rd7 {
-2.52/16 8 (Re8+)} Bb3 {-3.83/19 5} 39. Rd3 {-3.53/18 6 (Ke3)} Bf7 {-4.11/22 7}
40. h3 {-6.81/20 8 (Ke5)} a5 {-5.19/18 7} 41. Ke5 {-250.00/23 8} a4 {-6.31/19 6
} 42. Kf6 {-250.00/32 23 (Rd1)} Kf8 {-6.39/19 2 (Bb3)} 0-1

[MikeB]

Not that I have a vote , but I actually liked the openings selected this year. I like the process where they take the most favored openings current openings been played by GMs. I also ok with a limited number of openings ( say no more than 15%) that explore gambits where one side side +.8. Also believe some lines should come through the play by the current chess correspondent leaders.