Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Tripple Brain Shredder working with 2.3.2a + win finder
- - By jamerolle (**) Date 2007-08-02 21:32
My guess this combo might be better then 2.3.2a allone i realize you loses some processor speed however its been said that Rybka is lacking in tactics therefor win finder might be able to find hard wins in such position posted

Has anyone put any ratings together with Rybka working side by side with another engine , shredder interfance has such an option which might be worth looking into , this might prove to be even better then the great 2.3.2a working alone.

I know people might shoot my idea down but i say put it to the test then tell me what you think.  In this position working triple brain in shredder,  2.3.2a can not find the right answer but winfinder finds it in secounds and choses the winning move for the combo engine approach.  I will make a prediction that i wish the testers will look at playing Rybak 2.3.2a vs 2.3.2a and winfinder, my prediction the combo will win the match , however its hard to do so because , you need two computers and someing transfering moves

8/pR4pk/1b6/2p5/N1p5/8/PP1r2PP/6K1 b - - 0 1


R x B2
Parent - - By Vempele (Silver) Date 2007-08-02 21:43
Been already suggested several times. Nobody has tried it; the consensus is that the loss in search efficiency matters much than the improved performance in a tiny minority of positions.

> you need two computers and someing transfering moves


One quad is enough.
Parent - - By jamerolle (**) Date 2007-08-02 21:51
vas,  is this posible for the next GUI as your main engine is done scaning a move to be played could there be an engine transfer made " none manual" to give winfinder a scan before the move is played. Or Even more crazy could there be two engines build in one to get the best of both worlds.
Parent - - By Nelson Hernandez (Gold) Date 2007-08-03 03:56
Here's a thought.  Let's hypothetically say you had five quads identically rigged.  Each one used a different engine.  Each one had an identical amount of time to analyze a statistically-significant series of totally random positions.

In what percentage of cases would Rybka have the optimal answer, while the other four, in complete agreement, were sub-optimal?
In what percentage of cases would Rybka have a sub-optimal answer, while the other four, in complete agreement, had the optimal?

The questions are unanswerable for a lot of statistical and conceptual reasons which someone else can bother articulating if they feel like it.  Clearly Rybka wins hands-down against any other single engine.  But against an array of competitive engines that had reached a unanimous consensus?  That isn't totally clear, is it?  Surely Rybka isn't infallible (yet).
Parent - By Dhanish (***) Date 2007-08-03 10:32
When different engines recommend different moves, how do you decide which move to play? This is a common situation in correspondence chess, where finally it is a human decision.
Parent - - By Vasik Rajlich (Silver) Date 2007-08-03 12:14
If you played Rybka against Rybka, with the second Rybka's move being replaced by the choice of 4 other engines every time these four were unanimous, then I am quite sure that the pure Rybka would win unless those 4 engines had very special properties. (For example, 4 winfinders with completely non-overlapping positional styles).

Vas
Parent - - By Nelson Hernandez (Gold) Date 2007-08-03 22:36
That's quite a claim.  Maybe you're right, though.  If such a thing could be proven, that would be conclusive evidence that your engine is supernatural.  Nothing could impress me more.  That would be truly awesome.
Parent - - By Vasik Rajlich (Silver) Date 2007-08-05 13:19
Actually, I don't think it would be that impressive. In most fields, there's nothing really impressive about consensus.

Vas
Parent - - By Nelson Hernandez (Gold) Date 2007-08-05 17:07
No doubt that the greatest genius is the one that defies collective wisdom and goes his own way, proving the others wrong.  No vast fortune was ever amassed as a result of buying Suzie Orman's financial books.  Instead it takes a Sam Walton or a Bill Gates to visualize and execute something nobody else has before.  But you have to admit, the greater the deviation from consensus, the more likely the one doing the deviating is wrong or just crazy.  For every truly spectacular success there are hundreds or thousands of obscure, unknown failures or mediocrities.

My point is that the greater the consensus (using different methods and approaches) the more difficult it becomes to come to a different conclusion, and the greater the achievement when the deviation is correct.  And to do this more often than not--that's really amazing.
Parent - - By Zruty (*****) Date 2007-08-06 07:30
Suppose we have the 'council' of 19 really BAD engines (I'm not personal, so let's say I've written them myself, so they are REALLY bad) and Rybka. This 'council' will play better than Rybka alone? How can this be? Other engines' analysis may as well be just discarded, but instead it influences the play. I'm not yet talking about Rybka getting only 5% of her 'normal' computing time.

Suppose another case: we get 3 good engines and Rybka among them. They suggest 3 different moves. How do we choose the right one? And it can happen that Rybka i.e. makes 20... Nf6g4 in order to construct something like kingside attack and then Shredder makes 21... Ng4f6 to prevent some threat that only it does see. And the resulting 'council' plays worse than Shredder or Rybka alone - it loses tempo.
Parent - By Uri Blass (*****) Date 2007-08-06 10:32
In case that you have 3 good engines that include rybka then
I think that a rule that is slightly better than choosing rybka's moves can be simply to choose rybka's move unless another engine see winning score for another move or losing score for the move of rybka(and in that case you can choose the moves of this engine).

In other words choose rybka's move unless another engine's analysis strongly suggest that the move of rybka is unsound for tactical reasons or there is another winning move when it is not clear if rybka's move is winning based on rybka's score.
 
It may be possible to improve this rule but I doubt if it is possible to improve enough to compensate for 3:1 loss of speed.
I expect rybka to win even if the rule is to use a GM to choose between engines moves.

Uri
Parent - - By Uri Blass (*****) Date 2007-08-06 10:08
I diagree here.

There is no proof that the choice of the majority is better.
From my correspondence games I remember a case that I chose a move of Junior when all the other engines that I tried suggested a different move.

I believe that there is a better rule to decide which engine to choose than majority rule.
A simple rule that is probably better than majority choice is improvement in score rule.

Choose the move of the program that showed the biggest progress in score from iteration n-1 to iteration n when n is the last iteration that was searched(n can be different for different programs).

Uri
Parent - By Uri Blass (*****) Date 2007-08-06 10:19
I can add that I believe that the best automatic rule that I see for choices between rybka and other engines is simply to choose
rybka's move(there may be something slightly better but not much better and certainly not good enough to compensate for loss of speed by a factor of 2).

My guess is that
improvement in score rule can be productive relative to decide about one of the engines
only for choice between 2 engines that are near the same strength.

I also guess that for 3 different engines near the same strength majority rule may be worse choice
than choosing one of the engines regardless of the engine that you choose.

It is only a guess and I have no time to run games to check it.
Parent - By Vasik Rajlich (Silver) Date 2007-08-06 18:45
Keep in mind though that you are not taking consensus and forcing Rybka to disagree with it. In this example, she does it voluntarily.

Anyway, it shouldn't be too hard to try this.

Vas
Parent - By Uri Blass (*****) Date 2007-08-03 05:43
My prediction is that the tripple brain is going to lose for the following reasons:

1)loss of computer speed that is going to lead rybka to be outsearched and be in bad position.
2)winfinder is only good for finding wins and when there is no win it is not good for finding draws that tripple brain needs not to lose
3)tripple brain is stupid and will often choose moves of winfinder instead of superior moves of 2.3.2a

I will be very surprised if tripple brain of 2.3.2a and winfinder can score better than 35 out of 100 games against 2.3.2a.

Uri
Parent - - By Vasik Rajlich (Silver) Date 2007-08-03 12:15
The triple brain would lose badly.

Moves like 1. .. Rxb2 are cool and we all like to be amazed by them, but they don't 'pay the bills'.

Vas
Parent - - By Banned for Life (Gold) Date 2007-08-04 00:14
"The triple brain would lose badly."

Yes, but TRIPLE BRAIN is such a cool name that the results hardly matter.

Regards,
Alan
Parent - By bluemax (**) Date 2007-08-06 11:21

> Yes, but TRIPLE BRAIN is such a cool name that the results hardly matter.


Sounds like the bad guy in a James Bond film!!!
Up Topic Rybka Support & Discussion / Rybka Discussion / Tripple Brain Shredder working with 2.3.2a + win finder

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill