Hannibal1.2 released

[Jouni]

In test match vs. Deep Fritz 12 Hannibal won , but in this game it loses even with +19,94 evaluation !?

[Event "Hannibal"]
[Site "?"]
[Date "2012.05.01"]
[Round "5"]
[White "Hannibal 1.2"]
[Black "Deep Fritz 12"]
[Result "0-1"]
[ECO "C19"]
[Annotator "4.35;0.47"]
[PlyCount "66"]
[TimeControl "60/60:0/0:0/0"]

1. e4 e6 2. d4 d5 3. Nc3 Bb4 4. e5 c5 5. a3
Bxc3+ 6. bxc3 Qc7 7. Nf3 Ne7 8. a4 b6 9. Bb5+ Bd7 10. Bd3 Nbc6 11. O-O {Both
last book move} O-O {0.47/14 2} 12. Bxh7+ {4.35/17 3 (Ba3)} Kxh7 {1.72/17 5}
13. Ng5+ {4.84/14 0} Kg6 {2.07/17 4} 14. h4 {5.12/15 1} Qc8 {2.28/16 3 (Qd8)}
15. h5+ {9.17/15 5} Kh6 {2.31/14 1} 16. Qd2 {5.73/16 1 (Qd3)} Qa6 {0.59/12 1
(f6)} 17. g4 {10.43/13 7 (a5)} Qc4 {-0.06/10 1 (f6)} 18. a5 {13.10/13 1 (Kg2)}
cxd4 {0.12/12 2 (f6)} 19. cxd4 {6.93/12 2} bxa5 {-0.02/13 1} 20. Rxa5 {15.46/
17 10 (Ra3)} Nxa5 {-2.01/14 1 (f6)} 21. Kg2 {19.94/17 6 (Qxa5)} Nac6 {-5.28/13
1 (f6)} 22. Rh1 {7.92/14 2} Qxd4 {-5.43/14 1 (f6)} 23. Qxd4 {-4.96/16 2} Nxd4 {
-5.44/14 0} 24. Nxe6+ {-4.34/14 1} Kh7 {-5.42/17 1} 25. Nxd4 {-4.76/13 0} Bxg4
{-5.53/16 1 (Rfc8)} 26. Ba3 {-4.50/15 0} Rfe8 {-5.46/17 1} 27. Nb5 {-5.21/16 4}
Nc6 {-5.49/16 2 (Rab8)} 28. Nc7 {-4.64/15 0} Nxe5 {-5.67/17 3} 29. Kg3 {-4.69/
15 0} Bd7 {-5.70/16 2 (Bf5)} 30. Nxe8 {-4.60/13 0} Rxe8 {-5.74/15 0} 31. Rd1 {
-4.75/14 0} Be6 {-6.03/15 1} 32. f3 {-4.94/14 1 (Bb4)} a5 {-6.58/16 1 (Nc4)}
33. Bc5 {-5.13/15 0} Nc4 {-7.04/16 1} 0-1

[BubbaTough]

In test match vs. Deep Fritz 12 Hannibal won , but in this game it loses even with +19,94 evaluation !?

Looking at this game, it appears that the large score is purely due to king safety. Basically, given the position, it just assumed it would end up mating somehow, even though it did not see one. Perhaps we should reign in the degree of speculation in the evaluation function in future versions .

-Sam

[BubbaTough]

Will you go to Leyden?

Not unless Edsel wants to. Between work and the kids, I am not sure I could attend a tournament across town, let alone across the globe.

-Sam

[fern]

But maybe someone over there could play the program.... It would be a good occasion for you to see how is going this Hannibal...

Fern

[fern]

impressive indeed....
Fern

[gleperlier]

One characteristic whose effect I am curious to watch is the new "piece coordination" code in the evaluation. It is ported from a human-like engine I was trying to put together (that project is currently on hold). LearningLemming had an earlier version of this created some very nice upsets (sacrificing to uncoordinated opponent pieces) and some spectacular losses (crazy sacrifices for no reason). Hopefully the more mature version of this concept in Hannibal will result in more of the former than the latter. Still, I would guess the code makes Hannibal both more likely to upset stronger engines, and more likely to be upset by weaker engines.
-Sam

The long Time Control first game versus Delphil 2.9a seems a great example of your "piece coordination" code in the evalutation.

Will post the 2 games when finished.

Gab

[BubbaTough]

But maybe someone over there could play the program.... It would be a good occasion for you to see how is going this Hannibal...

Fern

Well if someone wanted to enter Hannibal on our behalf and it is within the rules, we would be happy to supply them with a slightly improved version with whatever support necessary for an in-person tournament .

-Sam

[fern]

May I advice you to contac any officer in charge of Leyden and ask him if eome fan over there could do the job?

Fern

[BubbaTough]

Enjoy. Impressive!

Thanks. I looked at some games from the Junior Antics tournament (which I think is the source of what you just posted) and watched one of them live (and interesting battle of Rook and pawns vs. 2 knights and pawns). It looks like Hannibal is grinding out a lot of endgame victories, rather than earning its victories through crazy attacks as is the Hannibal tradition. I am not sure if that is a fluke from the small sample size, or just the natural progression as a chess player gets older and wiser.

-Sam

[gladius]

In test match vs. Deep Fritz 12 Hannibal won , but in this game it loses even with +19,94 evaluation !?

Looking at this game, it appears that the large score is purely due to king safety. Basically, given the position, it just assumed it would end up mating somehow, even though it did not see one. Perhaps we should reign in the degree of speculation in the evaluation function in future versions .

-Sam

Some very nice attacks from Hannibal!

In this game, 20. Rxa5 seems to be the losing move. Perhaps a bug here? Even with a big score for king safety, not sure how sacking the rook is good.