New engine from Russia

[Andre van Ark]

Hi all,

I don't manage to make it work in the CB GUI. The engine ponders but doesn't make a move. I have to press the spacebar to force a move.

Strelka only uses around 50MB for hash here.

Have a nice weekend,
André

[PauloSoare]

(A bit more endgame knowledge would have netted Strelka at least 1 point more in these games)

Marc, Take a look in the results of some engines on my computer in the Eigenmann_Endspiel_Test.

Atlhon 2 cores 3800+, HT=128Mb, without tablebases, Deep Fritz GUI:
Hiarcs 11.1 1 core Result: 54
Shredder 10 1 core Result: 61
Spike 1.2 Turim Result: 50
Rybka 2.3.1 1core Result: 57
Fritz 10 1 core Result: 64
Rybka 1.0 1 core Result: 58
Fritz 9 1 core Result: 62
Chess Tiger 2007.1 Result: 49
Loop 13.6 1 core Result: 49

And now look Strelka, I made the test manualy.

Atlhon 2 cores 3800+, HT=46Mb, without tablebases, Arena GUI:
Strelka 1.0 beta Result:47

Not bad.

Paulo Soares

[GS]

...

Please include Rylka 1.0beta in those tests I wanna see how near
they are ;-)

Regards,
Guenther

I started a 100 game match at 40/3, 32M hash (for Rybka Beta), 12-moves book on my 2400mhz pentium. My prediction is 56-44 for Rybka.

Actually Strelka plays very much like Rybka beta, almost no chess knowledge (especially endgames).
...

My prediction is 50.5 : 49.5 for either side with at least 75% draws ;)

Regards
Guenther

[Graham Banks]

Hi Graham,
a standard wb2uci.eng like the following is enough to run Strelka under a Chessbase GUI with time controls like 40 moves in 5 minutes:

[ENGINE]
Name=Strelka_10b
Author=Juri Osipov
Filename=Wb2Uci.exe

[OPTIONS]
Program=Strelka.exe

Regards Dieter

Cheers Dieter,

this is what I assumed would be the case, but it was nice to have it confirmed before I try Strelka out.

Regards, Graham.

[Graham Banks]

...

Please include Rylka 1.0beta in those tests I wanna see how near
they are

Regards,
Guenther

I started a 100 game match at 40/3, 32M hash (for Rybka Beta), 12-moves book on my 2400mhz pentium. My prediction is 56-44 for Rybka.

Actually Strelka plays very much like Rybka beta, almost no chess knowledge (especially endgames).
...

My prediction is 50.5 : 49.5 for either side with at least 75% draws

Regards
Guenther

Sounds like you believe that Strelka is a Rybka 1.0 clone Guenther?

[Uri Blass]

...

Please include Rylka 1.0beta in those tests I wanna see how near
they are

Regards,
Guenther

I started a 100 game match at 40/3, 32M hash (for Rybka Beta), 12-moves book on my 2400mhz pentium. My prediction is 56-44 for Rybka.

Actually Strelka plays very much like Rybka beta, almost no chess knowledge (especially endgames). I'm manually testing Strelka on arasan7.epd at 20sec per moves and it is currently doing 37/50 while Rybka Beta Default (very positional) had 39/50 and Rybka beta-very tactical- had 42/50. So for now it seems a bit weaker tactically.

I disagree.

Rybka beta has a lot of chess knowledge.
You cannot win games without chess knowledge and rybka is winning games.

Uri

[Marc MP]

Hi all,

I don't manage to make it work in the CB GUI. The engine ponders but doesn't make a move. I have to press the spacebar to force a move.

Strelka only uses around 50MB for hash here.

Have a nice weekend,
André

Hi André,

I did these matches in Chess Partner GUI where Strelka runs fine for X moves in Y minutes. There is a 30 day trial on lokasoft website if you want to try the new engine. Set Strelka as winboard-2 during the installation. It uses about 46-47 hash here too.

[Marc MP]

...

Please include Rylka 1.0beta in those tests I wanna see how near
they are

Regards,
Guenther

I started a 100 game match at 40/3, 32M hash (for Rybka Beta), 12-moves book on my 2400mhz pentium. My prediction is 56-44 for Rybka.

Actually Strelka plays very much like Rybka beta, almost no chess knowledge (especially endgames). I'm manually testing Strelka on arasan7.epd at 20sec per moves and it is currently doing 37/50 while Rybka Beta Default (very positional) had 39/50 and Rybka beta-very tactical- had 42/50. So for now it seems a bit weaker tactically.

I disagree.

Rybka beta has a lot of chess knowledge.
You cannot win games without chess knowledge and rybka is winning games.

Uri

It is all relative Uri. I should have said relatively little knowledge elo-wise (but still I could be wrong). I'm judging on a couple middlegame positions where for example Strelka was up a pawn but with a wrecked pawn structure and (what I think was...) an inferior position, but valued it highly.

Tord said yesterday on the other side of the board that is new beta has little knowledge yet but it is winning games. Pro Deo has more knowledge than many higher rated engines.

[Marc MP]

...

Please include Rylka 1.0beta in those tests I wanna see how near
they are

Regards,
Guenther

I started a 100 game match at 40/3, 32M hash (for Rybka Beta), 12-moves book on my 2400mhz pentium. My prediction is 56-44 for Rybka.

Actually Strelka plays very much like Rybka beta, almost no chess knowledge (especially endgames). I'm manually testing Strelka on arasan7.epd at 20sec per moves and it is currently doing 37/50 while Rybka Beta Default (very positional) had 39/50 and Rybka beta-very tactical- had 42/50. So for now it seems a bit weaker tactically.

I disagree.

Rybka beta has a lot of chess knowledge.
You cannot win games without chess knowledge and rybka is winning games.

Uri

It is all relative Uri. I should have said relatively little knowledge elo-wise (but still I could be wrong). I'm judging on a couple middlegame positions where for example Strelka was up a pawn but with a wrecked pawn structure and (what I think was...) an inferior position, but valued it highly.

Tord said yesterday on the other side of the board that is new beta has little knowledge yet but it is winning games. Pro Deo has more knowledge than many higher rated engines.

Actually Uri I don't program engines so you must know better than I do! I'm curious now: would you say that Rybka Beta has a lot of knowledge (conditional on its elo)? Like more knowledge than one would expect given the elo?

[Uri Blass]

My opinion is that the relative advantage of rybka beta is the evaluation.

Rybka beta like fruit and toga does too many extensions(it seems to extend every check and every single reply by a full ply) so I cannot believe that the relative advantage of it relative to programs like shredder is the search.

Uri