Weird Weakness in Wasp 1.01

[BrendanJNorman]

In my excitement in getting John's new engine, I have started a blitz tournament with all of my favourite engines and one weird result (after several beautiful, human-like Wasp games) was the following anomaly.

[pgn]
[Event "Blitz Tournament 1"]
[Site "WIN-2R0B0OJBL5R"]
[Date "2016.06.24"]
[Round "1"]
[White "Wasp_101-w32"]
[Black "DeepSjeng27 Cognac"]
[Result "0-1"]
[ECO "A90"]
[WhiteElo "2200"]
[BlackElo "2200"]
[PlyCount "76"]
[EventDate "2016.??.??"]
[EventType "blitz"]
1. d4 f5 2. Nf3 Nf6 3. g3 e6 4. Bg2 d5 5. O-O Bd6 6. c4 c6 7. Nc3 O-O 8. Qc2
Qe8 9. c5 Bc7 10. Rd1 Nbd7 11. b4 e5 12. Rb1 e4 13. Ng5 h6 14. Nh3 Rf7 15. b5
Nf8 16. a4 Ne6 17. Ba3 f4 18. e3 f3 19. Bf1 Rf8 20. b6 Bd8 21. Nf4 Nxf4 22.
exf4 e3 23. fxe3 Ng4 24. e4 Bf5 25. bxa7 Rxa7 26. Bc1 dxe4 27. h3 e3 28. Qa2+
Kh8 29. hxg4 Bxg4 30. Rb3 f2+ 31. Kg2 Rxf4 32. Be2 Rxa4 33. Qb2 Raxd4 34. Rxd4
Rxd4 35. Rxb7 Qe6 36. Bxg4 Rxg4 37. Qe2 Rxg3+ 38. Kxg3 Qg6+ 0-1[/pgn]

Very strange game for such a positionally strong engine.

It looks like black is a strong human applying anti-computer strategy, but actually its just another of my crazy "cognac" personalities haha.

Anyway...

Allowing black to play ...e5 in this Dutch structure is a serious mistake and I think somewhere earlier white should have played Bf4 to trade those bishops and clamp down on the e5 square.

This plan with 11.b4 and Rb1 etc, is really slow and bad.

Perhaps John can have a look at this "closed position" weakness for the next release?!

Anyway, the other games Wasp played were very nice, but I thought I'd share this one because its reveals a big weakness, but also because the game is very beautiful tactically haha.

Brendan

[carldaman]

In my excitement in getting John's new engine, I have started a blitz tournament with all of my favourite engines and one weird result (after several beautiful, human-like Wasp games) was the following anomaly.

[pgn]
[Event "Blitz Tournament 1"]
[Site "WIN-2R0B0OJBL5R"]
[Date "2016.06.24"]
[Round "1"]
[White "Wasp_101-w32"]
[Black "DeepSjeng27 Cognac"]
[Result "0-1"]
[ECO "A90"]
[WhiteElo "2200"]
[BlackElo "2200"]
[PlyCount "76"]
[EventDate "2016.??.??"]
[EventType "blitz"]
1. d4 f5 2. Nf3 Nf6 3. g3 e6 4. Bg2 d5 5. O-O Bd6 6. c4 c6 7. Nc3 O-O 8. Qc2
Qe8 9. c5 Bc7 10. Rd1 Nbd7 11. b4 e5 12. Rb1 e4 13. Ng5 h6 14. Nh3 Rf7 15. b5
Nf8 16. a4 Ne6 17. Ba3 f4 18. e3 f3 19. Bf1 Rf8 20. b6 Bd8 21. Nf4 Nxf4 22.
exf4 e3 23. fxe3 Ng4 24. e4 Bf5 25. bxa7 Rxa7 26. Bc1 dxe4 27. h3 e3 28. Qa2+
Kh8 29. hxg4 Bxg4 30. Rb3 f2+ 31. Kg2 Rxf4 32. Be2 Rxa4 33. Qb2 Raxd4 34. Rxd4
Rxd4 35. Rxb7 Qe6 36. Bxg4 Rxg4 37. Qe2 Rxg3+ 38. Kxg3 Qg6+ 0-1[/pgn]

Very strange game for such a positionally strong engine.

It looks like black is a strong human applying anti-computer strategy, but actually its just another of my crazy "cognac" personalities haha.

Anyway...

Allowing black to play ...e5 in this Dutch structure is a serious mistake and I think somewhere earlier white should have played Bf4 to trade those bishops and clamp down on the e5 square.

This plan with 11.b4 and Rb1 etc, is really slow and bad.

Perhaps John can have a look at this "closed position" weakness for the next release?!

Anyway, the other games Wasp played were very nice, but I thought I'd share this one because its reveals a big weakness, but also because the game is very beautiful tactically haha.

Brendan

Yes, this is a well-known weakness many engines have, to play c5 (or c4 with colors reversed) against a Stonewall set-up, by wrongly extending the pawn chain towards the Q-side (a dead-end) while allowing the opponent to break thru the middle or down the K-side flank with strong attacking play.

Check out this long thread from last year that deals with this subject matter:

http://www.talkchess.com/forum/viewtopi ... 22&t=55399

Cheers,
CL

[BrendanJNorman]

Thanks Carl.

I was aware that this pawn chain issue was common for computers (like the old Crafty going ...c4 in Stonewall), but assumed it had been dealt with in modern engines.

I also thought a modern engine would understand the importance of PREVENTING black's ...e5 break in that structure (via Bf4, Bxf4, gxf4).

I read your linked article and its very interesting, even for a non-programmer like me.

Thanks for the link.

[AdminX]

Wasp vs Sjeng is a pretty good match up. I need to play more games between the two. Data from a short 5 minute match. (1 Core Used for Sjeng)

[Lyudmil Tsvetkov]

The game is not lost until later on.
Both engines seem to be making second-hand moves all too often.

A critical position arises after black's 14-th move ...Rf7(g5 was winning).

[d]r1b1q1k1/ppbn1rp1/2p2n1p/2Pp1p2/1P1Pp3/2N3PN/P1Q1PPBP/1RBR2K1 w - - 2 15

Here white has a single drawing option: f4. Seemingly, no engine sees it.
Bf4, as Komodo 8 would play, is insufficient. f3 fails to g5.

[BrendanJNorman]

The game is not lost until later on.
Both engines seem to be making second-hand moves all too often.

A critical position arises after black's 14-th move ...Rf7(g5 was winning).

[d]r1b1q1k1/ppbn1rp1/2p2n1p/2Pp1p2/1P1Pp3/2N3PN/P1Q1PPBP/1RBR2K1 w - - 2 15

Here white has a single drawing option: f4. Seemingly, no engine sees it.
Bf4, as Komodo 8 would play, is insufficient. f3 fails to g5.

The Deep Sjeng personality playing black was designed by me to play "second hand" moves and in a certain way.

I was aiming to create an interesting training partner, more than a strong analysis engine...so don't judge him too harshly.

I certainly wouldn't play training games with Komodo 8 haha.

Shared the game to give John some feedback on Wasp, that's all.

[Lyudmil Tsvetkov]

The game is not lost until later on.
Both engines seem to be making second-hand moves all too often.

A critical position arises after black's 14-th move ...Rf7(g5 was winning).

[d]r1b1q1k1/ppbn1rp1/2p2n1p/2Pp1p2/1P1Pp3/2N3PN/P1Q1PPBP/1RBR2K1 w - - 2 15

Here white has a single drawing option: f4. Seemingly, no engine sees it.
Bf4, as Komodo 8 would play, is insufficient. f3 fails to g5.

The Deep Sjeng personality playing black was designed by me to play "second hand" moves and in a certain way.

I was aiming to create an interesting training partner, more than a strong analysis engine...so don't judge him too harshly.

I certainly wouldn't play training games with Komodo 8 haha.

Shared the game to give John some feedback on Wasp, that's all.

Of course, you should not play training games against Komodo 8: better pick a stronger engine.

[PK]

It's funny that I'm just testing a piece of code that tries to deal with the pawn chain pointing the wrong side. I penalise it with, say, 20 centipawns, if there are white pawns on d4 and c5 blocked by black pawns on d5 and c4. But this is not the end of the story. Right chain or wrong chain, it still defines the direction of play, so I add a bonus for pushing the b pawn forward. by the time it reaches b5, penalty drops to 5 centipawns. This seems to be the missing piece of knowledge in the usual "anti-c5" code.