Sting SF 4.8.4 correction

[lech]

[d]2k5/8/8/8/1p6/1p6/pP2pB2/K6n b - -
for a derivative position you given me Stockfish 6 is able to see equality evaluation in 10mn,
Houdini 4 in 2h but Sting sf 4.8.4 is not able to decrease his evaluation.

Why Sting should do this?

The derivative position is
[d]1k6/3p4/1B6/4Pp1p/1p5R/1p4p1/pP3n2/K6n w - -
RxNh1
only StockFish 6 is able to see it in 1h .

A very good news. I hope you are sure and Stockfish (without tricks) can solve this position.
Sting is done to solve positions which other engines are not able to solve ever. This position is very difficult but maybe some high depths can compensate this subtle sacrifice.

[fishpov]

For sure Sting sf is able to solve positions which any other engines are able to find. I really like this engine and I hope it will become on the 3 top engines soon !
before I swore only by Houdini now Sting sf 4.8.4 is my new Hero !
(just still remove some infinite loop bug in long time analysis)

[fishpov]

I want to say Sting is able to solve positions that any other engines are NOT able to find !

[lech]

While testing engines on drawn positions with big material disadvantage and fortress positions, this version (win 64 and 32bit) crashed when analyzing the following position, I used 1 thread.
[d]8/Bk6/1P6/2K5/8/8/8/8 w - - 0 1

...
I tested my compilation of the parent (for Sting) version of Stockfish 2.1.1.
It happens too!
...

It is a next (second in short time) bug found in the Stockfish 2.1.1 code.
I corrected it.
Looking at both the bugs, some next are expected., but not welcome!
Thanks for the reaport!
BTW. Sting sf 5 should be soon!

[fishpov]

I have a little tested.

[d]r2qrb1k/1p1b2p1/p2ppn1p/8/3NP3/1BN5/PPP3QP/1K3RR1 w
Here Sting sf 4.8.4 takes 2h30 to find e5.
Houdini 4 is finding in 10s and StockFish 6 in 1mn.

This is kind of progress which should have to do Sting.

[Dicaste]

I have a little tested.

[d]r2qrb1k/1p1b2p1/p2ppn1p/8/3NP3/1BN5/PPP3QP/1K3RR1 w
Here Sting sf 4.8.4 takes 2h30 to find e5.
Houdini 4 is finding in 10s and StockFish 6 in 1mn.

This is kind of progress which should have to do Sting.

Sting's main point is endgame without tablebases. Marek believes engines doesn't need tablebases. A good coded chess engine can solve endgame positions. Anyway this is not an endgame position.

[michiguel]

I have a little tested.

[d]r2qrb1k/1p1b2p1/p2ppn1p/8/3NP3/1BN5/PPP3QP/1K3RR1 w
Here Sting sf 4.8.4 takes 2h30 to find e5.
Houdini 4 is finding in 10s and StockFish 6 in 1mn.

This is kind of progress which should have to do Sting.

Not so bad for Gaviota, ~16 min in a slow 3.4 Ghz AMD quad. This position from the Nolot suite used to be unsolvable for engines. We have made a lot of progress...

Miguel

Code: Select all

 858701918  24:    553.2    +1.83  1.Nf3 Bc6 2.Nh4 g5 3.Ng6+ Kh7 4.Nf4 Bg7
                                   5.Nxe6 Rxe6 6.Bxe6 Qb6 7.Qh3 Nxe4
                                   8.Bf5+ Kg8 9.Bxe4 Bxe4 10.Qe6+ Kh8
                                   11.Qxe4 Bxc3 12.b3 Bg7 13.Rg4 [>]
1128543400  25     719.9    +1.81  1.Nf3 Bc6 2.Nh4 g5 3.Ng6+ Kh7 4.Nf4 Bg7
                                   5.Nxe6 Rxe6 6.Bxe6 Qb6 7.Qh3 Nxe4
                                   8.Bf5+ Kg8 9.Bxe4 Bxe4 10.Qe6+ Kh8
                                   11.Qxe4 Bxc3 12.b3 Bg7 13.Rg4 Rf8
                                   14.Rxf8+ Bxf8 15.Qe8
1563679565  25     991.2      :-)  1.e5
1952849433  25    1228.6    +2.39  1.e5 dxe5 2.Ne4 Nh5 3.Nf3 Bb5 4.c4 Nf4
                                   5.Qc2 Bxc4 6.Bxc4 Qc7 7.Nh4 Rac8 8.Rxf4
                                   exf4 9.Ng6+ Kg8 10.Nf6+ Kf7 11.Nd7 Bd6
                                   12.Qf5+ Kg8 13.Bxe6+ Rxe6 14.Qxe6+ Kh7
                                   15.Qf5
2273125582  25:   1414.4    +2.39  1.e5 dxe5 2.Ne4 Nh5 3.Nf3 Bb5 4.c4 Nf4
                                   5.Qc2 Bxc4 6.Bxc4 Qc7 7.Nh4 Rac8 8.Rxf4
                                   exf4 9.Ng6+ Kg8 10.Nf6+ Kf7 11.Nd7 Bd6
                                   12.Qf5+ Kg8 13.Bxe6+ Rxe6 14.Qxe6+ Kh7
                                   15.Qf5
3787657889  26    2305.2      :-)  1.e5
5885797488  26    3561.8      :-)  1.e5
14414458331  26    8814.5    +4.47  1.e5 dxe5 2.Ne4 Nh5 3.Qg6 Nf4 4.Rxf4
                                   exf4 5.Nf3 Qb6 6.Rg5 Qd8 7.Ne5 Qe7
                                   8.Nf7+ Qxf7 9.Qxf7 Re7 10.Qh5 Be8
                                   11.Qh3 Rd8 12.c3 b6 13.Rg4 e5 14.Qf3
14465867300  26:   8844.8    +4.47  1.e5 dxe5 2.Ne4 Nh5 3.Qg6 Nf4 4.Rxf4
                                   exf4 5.Nf3 Qb6 6.Rg5 Qd8 7.Ne5 Qe7
                                   8.Nf7+ Qxf7 9.Qxf7 Re7 10.Qh5 Be8
                                   11.Qh3 Rd8 12.c3 b6 13.Rg4 e5 14.Qf3

[lech]

I have a little tested.

[d]r2qrb1k/1p1b2p1/p2ppn1p/8/3NP3/1BN5/PPP3QP/1K3RR1 w
Here Sting sf 4.8.4 takes 2h30 to find e5.
Houdini 4 is finding in 10s and StockFish 6 in 1mn.

This is kind of progress which should have to do Sting.

Project Sting is only to prove that chess engines are able to solve difficult chess positions without tricks (e.g. disabling Null Move Reduction) and without some important lost of ELO. For this reason positions which are solved by other engines in default settings don't interest me. More, any ELO gain is not my goal too. I test Sting against its parent Stockfish 2.1.1 and I try to keep similar strength ONLY..
I want to find a solution for the horizion effect (mainly happens in endgames). I selected a few branches and I think my work is near to finish.