Strelka - Rybka ponder hit

[Kirill Kryukov]

Hi everyone! As Graham quoted me saying to someone else earlier, I found that Strelka and Rybka 1.0 play almost identical chess. This is based on CCRL 40/4 blitz games. Ponder hit table:

Strelka - Rybka ponder hit

For comparison, here you can see what ponder hit looks like normally for independent engines:

Ponder hit table for free best single-CPU engines

Also, here is

List of most similar pairs, different families only

These results were enough for me to conclude that Strelka was a clone a few days after it was first released. The ponder hit numbers are huge, considering it is blitz time control. Rybka versions are more different from each other than Strelka from Rybka 1.0. Also, please note that I ran double number of games (64) in each of the pairs, compared to my normal match length of 32 games. This, together with the fact that two Strelka versions and 4 Rybka versions were used, allows us to say that correlation is significant.

I have to repeat again, that ponder hit stats don't give me any answer about Sterlka legality. I don't know if Strelka is legal or not, and I'll consider it a normal engine (like Toga) until I am convinced that it is illegal.

It is funny that here on CCC I was several times attacked for "supporting" Strelka (which I don't), while in Russian forum I had to oppositely argue with pro-Strelka people. Being a pro- or anti- something at least gives you home in one camp, but being objective you are left alone.

Best,
Kirill

[Vempele]

These results were enough for me to conclude that Strelka was a clone a few days after it was first released. The ponder hit numbers are huge, considering it is blitz time control. Rybka versions are more different from each other than Strelka from Rybka 1.0.

In that case, you should conclude that Loop is a clone, too. In your third link:

Code: Select all

#	Pair	Ponder hit	Moves counted
1	Loop 13.6 32-bit – Toga II 1.3x1	73.1	465
2	Loop M1-T 64-bit 2CPU – Toga II 1.2.1a 32-bit	70.6	1485
3	Loop 13.6 32-bit – Toga II 1.2.1a 32-bit	70.5	562
4	Loop 13.6 64-bit 2CPU – Toga II 1.2.1a 32-bit	70.2	1334
5	Loop M1-T 64-bit 2CPU – Toga II 1.3x4b egbb	70.2	1152
6	LoopMP 12.32 2CPU – Toga II 1.3x1	70.1	1964
7	Loop 13.5 64-bit 2CPU – Toga II 1.2.1a 32-bit	70.0	1202
8	Loop 13.5 32-bit – Toga II 1.2.1a 32-bit	69.7	1504
9	Rybka 1.0 Beta 32-bit – Strelka 1.0b 32-bit	69.4	3572

And so on; 21 out of 30 pairs are Toga-Loop.

[Dr.Wael Deeb]

These results were enough for me to conclude that Strelka was a clone a few days after it was first released. The ponder hit numbers are huge, considering it is blitz time control. Rybka versions are more different from each other than Strelka from Rybka 1.0.

In that case, you should conclude that Loop is a clone, too. In your third link:

Code: Select all

#	Pair	Ponder hit	Moves counted
1	Loop 13.6 32-bit – Toga II 1.3x1	73.1	465
2	Loop M1-T 64-bit 2CPU – Toga II 1.2.1a 32-bit	70.6	1485
3	Loop 13.6 32-bit – Toga II 1.2.1a 32-bit	70.5	562
4	Loop 13.6 64-bit 2CPU – Toga II 1.2.1a 32-bit	70.2	1334
5	Loop M1-T 64-bit 2CPU – Toga II 1.3x4b egbb	70.2	1152
6	LoopMP 12.32 2CPU – Toga II 1.3x1	70.1	1964
7	Loop 13.5 64-bit 2CPU – Toga II 1.2.1a 32-bit	70.0	1202
8	Loop 13.5 32-bit – Toga II 1.2.1a 32-bit	69.7	1504
9	Rybka 1.0 Beta 32-bit – Strelka 1.0b 32-bit	69.4	3572

And so on; 21 out of 30 pairs are Toga-Loop.

Recently there are a lot of statements regarding the relationship between Loop and Toga
Can someone spot more light on this issue.....

[Kirill Kryukov]

These results were enough for me to conclude that Strelka was a clone a few days after it was first released. The ponder hit numbers are huge, considering it is blitz time control. Rybka versions are more different from each other than Strelka from Rybka 1.0.

In that case, you should conclude that Loop is a clone, too. In your third link:

Code: Select all

#	Pair	Ponder hit	Moves counted
1	Loop 13.6 32-bit – Toga II 1.3x1	73.1	465
2	Loop M1-T 64-bit 2CPU – Toga II 1.2.1a 32-bit	70.6	1485
3	Loop 13.6 32-bit – Toga II 1.2.1a 32-bit	70.5	562
4	Loop 13.6 64-bit 2CPU – Toga II 1.2.1a 32-bit	70.2	1334
5	Loop M1-T 64-bit 2CPU – Toga II 1.3x4b egbb	70.2	1152
6	LoopMP 12.32 2CPU – Toga II 1.3x1	70.1	1964
7	Loop 13.5 64-bit 2CPU – Toga II 1.2.1a 32-bit	70.0	1202
8	Loop 13.5 32-bit – Toga II 1.2.1a 32-bit	69.7	1504
9	Rybka 1.0 Beta 32-bit – Strelka 1.0b 32-bit	69.4	3572

And so on; 21 out of 30 pairs are Toga-Loop.

It is obvious that Loop and Toga play very similar chess, to say the least. I am not totally sure about the explanation, so I'll reserve my judgement for the moment.

More games are needed to complete the picture, many important pairs are untested. For example it's not clear since which version Loop started to be like Toga - was it sudden or gradual change, etc.. I don't have Loop so I can't help here.

Best,
Kirill

[Chuck Wilson]

This is a very curious possibility supported by the Ponder Hit statistics during CCRL testing. Whether it is true and to what extent this matters remains for the computer chess community to decide. But the fact does still remain that the correlation found between Strelka and Rybka by Ponder Hit statistics is startling to the say the least!

Cheers,

Chuck

[GenoM]

These results were enough for me to conclude that Strelka was a clone a few days after it was first released. The ponder hit numbers are huge, considering it is blitz time control. Rybka versions are more different from each other than Strelka from Rybka 1.0.

In that case, you should conclude that Loop is a clone, too. In your third link:

Code: Select all

#	Pair	Ponder hit	Moves counted
1	Loop 13.6 32-bit – Toga II 1.3x1	73.1	465
2	Loop M1-T 64-bit 2CPU – Toga II 1.2.1a 32-bit	70.6	1485
3	Loop 13.6 32-bit – Toga II 1.2.1a 32-bit	70.5	562
4	Loop 13.6 64-bit 2CPU – Toga II 1.2.1a 32-bit	70.2	1334
5	Loop M1-T 64-bit 2CPU – Toga II 1.3x4b egbb	70.2	1152
6	LoopMP 12.32 2CPU – Toga II 1.3x1	70.1	1964
7	Loop 13.5 64-bit 2CPU – Toga II 1.2.1a 32-bit	70.0	1202
8	Loop 13.5 32-bit – Toga II 1.2.1a 32-bit	69.7	1504
9	Rybka 1.0 Beta 32-bit – Strelka 1.0b 32-bit	69.4	3572

And so on; 21 out of 30 pairs are Toga-Loop.

It is obvious that Loop and Toga play very similar chess, to say the least. I am not totally sure about the explanation, so I'll reserve my judgement for the moment.

More games are needed to complete the picture, many important pairs are untested. For example it's not clear since which version Loop started to be like Toga - was it sudden or gradual change, etc.. I don't have Loop so I can't help here.

Best,
Kirill

Kirill,
its strange that you can not draw any conclusions on the basis of CCRL list of most similar pairs about Loop and Toga but you do draw conclusions on the same basis when it comes to Strelka.

[Zlaire]

Kirill,
its strange that you can not draw any conclusions on the basis of CCRL list of most similar pairs about Loop and Toga but you do draw conclusions on the same basis when it comes to Strelka.

He gave a perfectly good explanation in his post.

[Sylwy]

Hi !

Some years ago was a nice cloning scandal .The very first List (Loop) was considered to be a Crafty clone ! O tempora !
Only God knows exactly the number of "pure " chess engines.Seems often we payed for "grafts".
Another curious problem: Anthony Cozzie are retired,Fabien Letouzey are retired,Ryan Benitez are retired.Why ? Both engines ZapChess &Fruit are very original & strong works !

Regards,
Sylwy

[Henrik Dinesen]

Fabien Letouzey are retired,Ryan Benitez are retired.

Ohh, really!? Where do you get that from??
Ryan has posted here recently, and even if he hadn't, it doesn't mean retirement!
Fabien? Very persistant rumours in here, informs us that he's working on new engine! What kind of retirement is that?

[swami]

Ohh, really!? Where do you get that from??

He assumed it because Fabien only released 2 years older betas.

Fabien? Very persistant rumours in here, informs us that he's working on new engine! What kind of retirement is that?

Yes,Edsel said it and he said he had heard it somewhere and I heard it from him!
Maybe Ryan or Jaochim can tell us something about what Fabien is working on.