HAL9000

HAL9000
"It just isn't conceivable that you can design a program strong enough to beat players like me."

January 31, 2016

Stockfish 7 Revisited: No superiority after first round

As mentioned in a previous post, i'm currently running a relatively longer tournament between different Stockfish 7 builds in order to find out, if there's such difference, which build is better.

I must admit i won't be surprised if thousands of games reveal nothing precise. Maybe there's nothing to find.

After 1000 games played, there's no clear sign of superiority between 3 builds of Stockfish 7. Even the gap between Komodo 9.3 and Stockfish 6 doesn't tell too much.

That needs more and more games. Therefore, there will be more and more games!

The standings after playing first 50 openings of TCEC-7:

1) Stockfish 7 JA        230.0 / 400
2) Stockfish 7 DF1.60    220.0 / 400
3) Stockfish 7b2 JA      218.5 / 400
4) Stockfish 6           170.5 / 400
5) Komodo 9.3            161.0 / 400

3 comments:

Алексей Чернаков said...

Komodo 9.3 по умолчанию настроен на контроль 120+1,при контроле 300+1 в настройках komodo,следует уменьшить selectivity до показателя 117,я проводил тестирование komodo 9.3,при контроле 300+2 он stockfish 6 уверенно обыгрывает.Настройки нужно оставить default,а показатель selectivity плавно уменьшать с увеличением времени на партию.Нужно вывести кривую на графике для этого показателя.

Dariusz Domagała said...

Алексей, thx for your comment.
A lot of us using english. Please translate your comments before posting.

Алексей, спасибо за ваш комментарий.
Многие из нас используют английский язык. Пожалуйста, переводчик комментарий до его публикации. Спасибо.

I'm from Poland, chciałbyś czytać coś takiego ? ;-)

Regards, Dariusz

Gurcan Uckardes said...

@ Alexey: Ok maybe... but my rule is to use default settings except threads and hash. I don't wanna tweak Komodo and do nothing with Stockfish. It would not be fair.