Strelka 10

  

Strelka 1.0


Back to forum


Thibault de Vassal    (2007-06-15)
Strelka 1.0

Some interesting readings in Talkchess & Rybka forums about this new engine : Strelka 1.0 x32

http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?pid=9297

http://216.25.93.108/forum/viewtopic.php?p=117821


http://computerchess.org.uk/ccrl/404/

http://www.husvankempen.de/nunn/40_40%20Rating%20List/40_40%20BestVersion/rangliste.html


Is Strelka only a joke about Rybka (and a clone of an earlier version) or the 'little arrow' (Strelka in russian) that will catch Rybka in future...

I don't know much about this one but its first ratings are "not bad" ! .. Any informations more ?


Mircea Hrubaru    (2007-06-16 14:12:43)
More on Strelka

Hello all, Well, it seems that all the fuss on wether Strelka is a clone has ended. Strela by Yuri Ostripov (St. Petersburg, Russia) is his original work and this was strongly tested by trustful persons in computerchess. Yet its status (a very materialistic engine, with very few endgame knowledge) makes this engine a very goo future hope. Yuri is currently improving Strelka so we must expect spectacular progress very soon (at least I hope that). Regards, Mircea


Thibault de Vassal    (2007-07-14 14:31:34)
Strelka : Rybka clone ?

It seems the issue on whether Strelka is a clone of Rybka 1.0 or not has not ended... An interesting thread about it in Rybka forum, comparing code of both chess engines :

http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=1655


Wayne Lowrance    (2007-07-14 18:10:06)
It is a clone

If it looks like a fish, smells like a fish, acts like a fish, it is a FISH. The topic has not gone away, not for a long time. He reversed engineered Rybka 1, period. We will watch it's progress. :) Wayne


Wayne Lowrance    (2007-07-14 18:37:07)
It is a clone 2.

The following is just one of many close observations. There are so many documentations I would not try to sum them up. :)

The main problem is identical or almost identical analysis both in evaluation and search.

Studying rybka's output in order to improve your evaluation(assuming rybka has better evaluation) make sense but even in that case I do not expect a program to have identical evaluation to rybka even in situations when it does not make sense to have evaluation that is different than 0 as King against king.

Rybka tries to hide her evaluation but not writing output at small depth and this is the reason that I am basically interested in analysis of fortress positions because in fortress positions the evaluation is partially exposed.

It is logical to learn from this experience of analysing fortress position and get evaluation that is more similiar to rybka but having evaluation that is totally identical in more than one case does not make sense and it cannot be an accident and the problem is not only identical evaluation but also identical search in many positions as evidence proved and even if the evaluation is different I can find the same patterns in the score changes.

This can happen only with copying code and it cannot happen without cloning.

The programmer of strelka did not understand rybka's code otherwise he could avoid the same bugs.

Same pattern of drop in the evaluation score when you have a queen. This pattern is also in old strelka.

Why does it happen?

Note that no other program that I know shows drop in evaluation for white by more than 3 pawns.

Strelka's score drops from +- (12.53) in depth 6 to 8.50 in depth 7

Rybka's score drops from 12.34 pawns in depth 4 to 8.37 pawns at depth 5

I find this compelling, if you understand it

Wayne