Strelka 5 by Yuriy Osipov

[Desperado]

Hello George,

i think "arrow" is just the translation of the russian word "strelka"...

[Daniel Shawul]

I find this translation hilarious

[Laskos]

Good, seems extremely strong for an untuned engine, for now it's next only to Houdini x32 on equal hardware. Hope he will tune it and release later the sources

Kai

Yuriy's Strelka 5 has beat Vas to the punch. .

Code: Select all

    Program                            Score      %      Elo    +   -    Draws

  1 Houdini 1.5a w32               : 1242.5/2300  54.0   3214   11  11   35.9 %
  2 Strelka 5                      : 1057.5/2300  46.0   3186   11  11   35.9 %

It rocks, only 28 +/ -11 Elo points weaker than Houdini, clearly the second strongest program, and an untuned one! Now Yuri, tune it!

Kai

[Kaj Soderberg]

I have problems installing or running the engine on my W7-32 system.

In Fritz 10 and 12 it does not install.
In ChessPartner it does not run.
In Arena it does not run.

I simply extracted the beast with 7-zip, and then installed the .exe as a UCI engine. Did i miss some important step?

My best,
Kaj

[David Dahlem]

I have problems installing or running the engine on my W7-32 system.

In Fritz 10 and 12 it does not install.
In ChessPartner it does not run.
In Arena it does not run.

I simply extracted the beast with 7-zip, and then installed the .exe as a UCI engine. Did i miss some important step?

My best,
Kaj

You need to download this dll - put it in the strelka directory:

http://www.megaupload.com/?d=W3J6V6WR

[Uri Blass]

Good, seems extremely strong for an untuned engine, for now it's next only to Houdini x32 on equal hardware. Hope he will tune it and release later the sources

Kai

Yuriy's Strelka 5 has beat Vas to the punch. .

Code: Select all

    Program                            Score      %      Elo    +   -    Draws

  1 Houdini 1.5a w32               : 1242.5/2300  54.0   3214   11  11   35.9 %
  2 Strelka 5                      : 1057.5/2300  46.0   3186   11  11   35.9 %

It rocks, only 28 +/ -11 Elo points weaker than Houdini, clearly the second strongest program, and an untuned one! Now Yuri, tune it!

Kai

I am not so sure if 17-39 elo difference from houdini1.5 make it the second best engine..

The CCRL has the following for Critter and houdini at 40/40 and I do not know if strelka5 is better or worse than Critter.

Houdini 1.5a 32-bit 3205 +23 −22
Critter 1.2 32-bit 3158 +19 −19

[Laskos]

Good, seems extremely strong for an untuned engine, for now it's next only to Houdini x32 on equal hardware. Hope he will tune it and release later the sources

Kai

Yuriy's Strelka 5 has beat Vas to the punch. .

Code: Select all

    Program                            Score      %      Elo    +   -    Draws

  1 Houdini 1.5a w32               : 1242.5/2300  54.0   3214   11  11   35.9 %
  2 Strelka 5                      : 1057.5/2300  46.0   3186   11  11   35.9 %

It rocks, only 28 +/ -11 Elo points weaker than Houdini, clearly the second strongest program, and an untuned one! Now Yuri, tune it!

Kai

I am not so sure if 17-39 elo difference from houdini1.5 make it the second best engine..

The CCRL has the following for Critter and houdini at 40/40 and I do not know if strelka5 is better or worse than Critter.

Houdini 1.5a 32-bit 3205 +23 −22
Critter 1.2 32-bit 3158 +19 −19

Well, 100% sure one cannot ever be, but if you combine all the reasonable results (64 vs 64 or 32 vs 32, only few Elo points difference), for example this one too:

CCRL 40/4

Houdini 1.5a 64-bit 3266 +13 −13
Critter 1.2 64-bit 3209 +17 −17

then Critter 1.2 is ~50 +/- 10 (95% confidence) points weaker than Houdini 1.5a. Depends a bit on time control too.

Kai

[Kaj Soderberg]

Thanks, it works

[Laskos]

LOL

It seems Yuri raped a bit Houdini 1.5. With Sim03 similarity tester I got the similarity matrix
___
sim version 3

Key:

1) Critter_1.2 (time: 100 ms scale: 1.0)
2) Houdini_15a (time: 100 ms scale: 1.0)
3) IvanHoe_47c (time: 100 ms scale: 1.0)
4) Komodo 3 (time: 100 ms scale: 1.0)
5) Rybka 3 (time: 100 ms scale: 1.0)
6) Rybka 4.1 (time: 100 ms scale: 1.0)
7) StockFish-211 (time: 100 ms scale: 1.0)
8) Strelka5 (time: 100 ms scale: 1.0)

1 2 3 4 5 6 7 8
1. ----- 61.76 60.62 50.25 58.07 55.43 51.80 62.05
2. 61.76 ----- 61.74 49.56 56.93 54.21 50.06 67.10
3. 60.62 61.74 ----- 50.19 59.53 54.72 50.47 62.05
4. 50.25 49.56 50.19 ----- 51.82 50.10 46.92 50.19
5. 58.07 56.93 59.53 51.82 ----- 58.19 51.01 55.96
6. 55.43 54.21 54.72 50.10 58.19 ----- 50.89 53.99
7. 51.80 50.06 50.47 46.92 51.01 50.89 ----- 50.06
8. 62.05 67.10 62.05 50.19 55.96 53.99 50.06 -----
___

Using IBM SPSS hierarchical clustering, the dendrogram is



The distance on the horizontal axis to the common ancestor is the degree of relatedness. Strelka 5 seems _very_ related to Houdini 1.5, much more related than Houdini is to Ippos, or than Rybka 4 is to Rybka 3, for that matter.

Soon maybe I will post a dendrogram showing circumstantially that Fritz and Naum are Strelka/Rybka clones, therefore Fruit/Crafty clones as stated by ICGA.

Kai

[mhalstern]

What does untuned mean and why is everybody claiming Strelka 5 is untuned?