Firebird 1.0 and 1.01: 180 games.
Moderators: hgm, Rebel, chrisw
-
- Posts: 5960
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
Re: Firebird 1.0 and 1.01: 180 games.
Comparing unrelated engines at fixed depth doesn't tell us much. Rybka and Robbo are highly selective, so they may look bad against some program that does not prune a lot of moves. Perhaps Hiarcs is such a program. It may also have a more sophisticated eval, but this test doesn't prove that. Fritz 12 I believe does prune highly, so here the fixed depth result probably means that the eval of Rybka and Robbo are better than Fritz. The only reason that the fixed depth matchup of Rybka and Robbo is interesting is that I know that Robbo is a derivative of Rybka with a simpler eval, and that three plies are subtracted from Rybka. The test doesn't tell us whether Robbo has a worse eval or is just more selective than Rybka 3; however based on everything I've heard from those who have compared the code the answer is "both".
-
- Posts: 221
- Joined: Thu Jan 28, 2010 5:47 pm
Re: Firebird 1.0 and 1.01: 180 games.
Thanks for your reply. But I am afraid I don't understand. You say that "Comparing unrelated engines at fixed depth doesn't tell us much." and that you "know that Robbo is a derivative of Rybka" making the matchup interesting.lkaufman wrote:Comparing unrelated engines at fixed depth doesn't tell us much. Rybka and Robbo are highly selective, so they may look bad against some program that does not prune a lot of moves. Perhaps Hiarcs is such a program. It may also have a more sophisticated eval, but this test doesn't prove that. Fritz 12 I believe does prune highly, so here the fixed depth result probably means that the eval of Rybka and Robbo are better than Fritz. The only reason that the fixed depth matchup of Rybka and Robbo is interesting is that I know that Robbo is a derivative of Rybka with a simpler eval, and that three plies are subtracted from Rybka. The test doesn't tell us whether Robbo has a worse eval or is just more selective than Rybka 3; however based on everything I've heard from those who have compared the code the answer is "both".
But if I know nothing of the relationship of Robbo to Rybka what is this test proving at all?
By your logic of
Rybka and Robbo are highly selective, so they may look bad against some program that does not prune a lot of moves.
then does not Robbo prune even more moves than Rybka so it would look a little worse against Rybka at fixed depth?
-
- Posts: 5960
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
Re: Firebird 1.0 and 1.01: 180 games.
That's right, the test doesn't tell us whether Robbo loses at fixed depth due to more pruning or to less chess knowledge. However I know that is does more pruning and has less chess knowledge, so it's reasonable to assume that both contribute to the result. Which is the dominant factor I don't know.
-
- Posts: 2801
- Joined: Mon Feb 11, 2008 3:53 pm
- Location: Denmark
- Full name: Damir Desevac
Re: Firebird 1.0 and 1.01: 180 games.
Hi Larry
Try and look in the task manager and compare Robbo's seize to Rybka's...
Try and look in the task manager and compare Robbo's seize to Rybka's...
-
- Posts: 5960
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
Re: Firebird 1.0 and 1.01: 180 games.
I'm not familiar with the term "seize" in connection with Task Manager, but as far as I can tell both programs use about 25% of my quad when it is their turn (running single-core time games). Anyway this is unrelated to performance at fixed depth. But I do notice that Robbo seems to use more time shortly out of book, and I rather strongly suspect that the generally superior results for Robbo over Rybka 3 are primarily due to superior time control algorithms, since this was a rather neglected area in Rybka 3 and one that was surely not "cloned". If that's so it is a much less important improvement than any improvement in the search or eval, since it's of no use for anyone using the program for any purpose other than playing timed games against other engines.
Please clarify what point you are trying to make.
Please clarify what point you are trying to make.
-
- Posts: 221
- Joined: Thu Jan 28, 2010 5:47 pm
Re: Firebird 1.0 and 1.01: 180 games.
Isn't the search completely different though? I thought Vas said that it was Rybka without the search?lkaufman wrote:I'm not familiar with the term "seize" in connection with Task Manager, but as far as I can tell both programs use about 25% of my quad when it is their turn (running single-core time games). Anyway this is unrelated to performance at fixed depth. But I do notice that Robbo seems to use more time shortly out of book, and I rather strongly suspect that the generally superior results for Robbo over Rybka 3 are primarily due to superior time control algorithms, since this was a rather neglected area in Rybka 3 and one that was surely not "cloned". If that's so it is a much less important improvement than any improvement in the search or eval, since it's of no use for anyone using the program for any purpose other than playing timed games against other engines.
Please clarify what point you are trying to make.
-
- Posts: 5960
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
Re: Firebird 1.0 and 1.01: 180 games.
I have the opposite impression, that it is more like Rybka 3 without the evaluation. Are you sure you are quoting him correctly? If so please tell me where you saw this quote. It makes no sense to me.
-
- Posts: 221
- Joined: Thu Jan 28, 2010 5:47 pm
Re: Firebird 1.0 and 1.01: 180 games.
I don't remember where i saw it - I think someone else quoted him on this forum so I may be wrong.
So are you saying it searches deeper and more positions because some of the evaluation was taken out?
So are you saying it searches deeper and more positions because some of the evaluation was taken out?
-
- Posts: 5960
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
-
- Posts: 2129
- Joined: Thu May 29, 2008 10:43 am
Re: Firebird 1.0 and 1.01: 180 games.
Larry-lkaufman wrote: i know that Robbo is a derivative of Rybka with a simpler eval
is this a toned down version of 'anyone with half a brain knows it's a clone' argument?
man, that's really getting old.
with all due respect, your conflict of interest here is huge...
you've recently announced that you're teaming with Don Dailey on a 'commercial' engine - Komodo.
if the Ippolit engines are banned, is it not to your direct benefit?
wouldn't you be much better off spending all your energy, experience, and knowledge working on Komodo...instead of posting here in what appears to be an effort to discredit other 'stronger' engines?
because of this 'conflict of interest', i respectfully request that you you withdraw from these discusions, and instead concentrate on making Komodo the #1 engine in the world.
Norm