Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka 3 -- a hundred elo!
1 2 3 4 5 Previous Next  
Parent - - By dareapa (**) Date 2008-07-09 21:24
How many elos is single processor R3 over sp R2.3.2a? you may have already stated this but if you did i missed it.
Parent - By lkaufman (*****) Date 2008-07-09 22:43
I haven't tested this, but if the gain on 4 proc. is 100, the gain on 1 should be at least close to 90, as scaling wasn't such a big issue until you get to 8 processors.
Parent - - By staylor (***) Date 2008-07-10 22:31 Edited 2008-07-10 22:36
Do you think that this (beta)version looks like it is better in all aspects and positions of regular chess playing, than all existing programs known to date?
In other words, have the tactical low points (compared to some other programs, which had been discussed) now been rectified, and has long term human-like planning improved? And opening understanding out of book? Is it now the greatest attacker (when sound)? or maybe not?   Might it still play some positions weaker than some other existing program known as of now?
Parent - - By lkaufman (*****) Date 2008-07-10 22:58
     It is impossible to write a program that will play all positions better than all other engines. If you mean all easily described types of positions, that might be possible, although even in this case there are some exceptions for Rybka 3. Rybka still does not do what I would call "planning", which to me means choosing a move partly on the basis of looking at the consequences of making many consecutive moves for one side, but she will appear to be playing with a plan more often than older versions. By her very nature (highly selective, or highly variable search depth if you prefer) she simulates planning better than conventional programs. Probably some other programs are still superior when very specific knowledge of a certain endgame is required, and certainly some (at least Shredder) are superior in recognizing positions where no progress is possible due to total blockade. Tactics are improved, but whether Rybka is as good now as some other programs at spotting difficult sacrificial attacks I don't know. Opening play without book should be particularly improved, and I would be surprised if Rybka 3 is not the best in this arena now. It might be even better in the "humanized" version that will come with Rybka 3. 
Parent - - By staylor (***) Date 2008-07-11 11:48 Edited 2008-07-11 11:51
Yes, thanks, so it seems that it takes up to 300 elo points in estimated chess playing strength, to overtake another program in all aspects of chess playing strength......
Would that be about right?

Another question, is the "simulation of planning" a good step forward, and very worthy of analysis?
Parent - - By lkaufman (*****) Date 2008-07-11 15:02
To overtake other programs in all aspects of chess playing strength requires working on lots of rare issues; a good example is the total blockade draw, which apparently Shredder's author has spent time solving. We prefer to spend time on issues that come up more often. So I would say that being better in all aspects of chess and improving the rating have very little to do with each other.
Parent - By 8lrr8 (***) Date 2008-07-11 17:18
right.  u dont have to be #1 in every category to be clearly #1.  does woods lead every statistical category the PGA tracks?  does jordan do the same for all the categories that the NBA tracks?  no.  are they the greatest players of all-time in their respective sports?

no question about it.
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka 3 -- a hundred elo!
1 2 3 4 5 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill