Questions and comments related to CCRL testing study


Postby igrino » Tue Jun 20, 2006 11:37 am

I am not a chess engine expert but in the last year I started to follow the results of the different engines, etc...
Few weeks ago I was quite surprised of the result of the computer tournment in Torino won by Junior, second Shredder (not sure) and with Rybka "only" third.
I know that in few games everything can happen and therefore this result doesn't automatically mean (for example) that Junior is stronger than Rybka.
However I am wondering if there are other reasons... For example:
1] "Home preparation" like for GM games, the opening book, in this case of Rybka, has been analized to find weak spots. In this case Rybka has come out of the opening badly and then defeated.
2] Rybka looks at less positions than its opponent but its evaluation algorithm is more complex, right? Then maybe the number of positions evaluated is less important for Rybka than for other engines. I mean: if Junior (for example) see 'n' positions and has a strenght of 'm' then when it sees 2n positions it has a strenght of 2m. Rybka instead with 'n' position it has a strenght of 'm' but with 2n it has a strenght of 1.5m. In other words, for Rybka, the coefficent between number of positions and effective strenght is smaller than for other engines.

Maybe my guesses are just non sense but I would like to have the opinion of some engine expert!
Many thanks,
Posts: 2
Joined: Tue Jun 20, 2006 11:14 am

Postby Uri Blass » Tue Jun 20, 2006 12:56 pm

This is not clear that rybka is looking at less position than her opponent(I use the word her because I remember Vasik talk about rybka as she)

different programmers mean different things by the term nodes and it seems that rybka does not count part of the search as nodes unlike other programs.

If I give rybka a position of king and 9 queens against king and 9 queens it needs a long time to show an output and after showing an output the number of nodes is extremely small.

The long time may be also a problem of other engines but showing small number of nodes only happens with rybka.

I believe that the rybka simply does some search that it does not count as nodes(for example rybka may have a function to calculate if the side to move can win material by a sequence of captures when the moves in the sequence of captures are not counted as nodes).

I can add that there is no evidence that rybka earns less from time relative to other engines.

I believe that the reason for rybka's failure in WCCC is simply opening preperation.

Rybka is not perfect and has weaknesses.
Rybka was also commercial before WCCC so it was possible to prepare against it and if the shredder team could find some uncommon opening that shredder can beat the commercial rybka then it could be a good idea to try that opening in the world championship.

Uri Blass
Posts: 40
Joined: Tue Jun 20, 2006 12:44 pm

Postby igrino » Fri Jul 07, 2006 1:59 pm

Thank you for your reply Uri,
I thought a bit about your comment. My assumption that Rybka sees less position is not based on solid ground (beside the displayed number of nodes, I was thinking that the code of Rybka is larger than other engines, this extra code should be eval code and to run it the machine requires more time!).
However I am still fond of my theory and I am curious to see what will be the playing strength differences between Rybka 2 with 1 CPU, Rybka 2 with 2 CPUs and Rybka2 with 4CPUs (for example for Zap! we have 2813-->2873-->2959 (data at the 1st of July) with increments of +60 and +86).
We will see!
Thanks again,
Posts: 2
Joined: Tue Jun 20, 2006 11:14 am

Return to CCRL Public

Who is online

Users browsing this forum: No registered users and 2 guests