swami wrote:I have read a lot of posts about the test between Naum and Rybka, I think Naum is not yet at Rybka's level, It's some 60-80 elo points behind but It would perform even better with the next update
Guranteed! Sure!
So we should just go to sleep, and even the programmer can. He has your gurantee it will atleast be better than Rybka 3
N4 big weakness against R3 is search ... N4 is much much slower than R3 in search ... and R3 totally destroys N4 at fast time controls because of that. The reason for that could be too much knowledge code that is hogging it down, or simply that Vas ... the king of search ... has too many tricks than nobody else has yet discovered. That is one of the reasons that N4 scores much better at longer time controls. The chances of being outsearched in a critical line are much less.
M ANSARI wrote:N4 big weakness against R3 is search ... N4 is much much slower than R3 in search ... and R3 totally destroys N4 at fast time controls because of that. The reason for that could be too much knowledge code that is hogging it down, or simply that Vas ... the king of search ... has too many tricks than nobody else has yet discovered. That is one of the reasons that N4 scores much better at longer time controls. The chances of being outsearched in a critical line are much less.
True, at fast time controls Rybka outperforms Naum by 120 points or so. On slow by 60. If Uri is right about "decisive" games, then Naum could approach Rybka, but never surpass it in very slow time controls and very fast hardware. Yet to be tested. But no doubt, Naum is the most serious competitor. Only Zappa had such improvements on long matches. You are right, Rybka is certainly faster searcher, but it is beyond me why Zappa and Naum perform better at slower time controls. More knowledge? - maybe.
I have none of these engines, but I would guess that Zappa and Naum need a slightly smaller time factor (in average) to complete one ply more than Rybka 3, which would be an advantage which grows with depth.
It's only a guess, but users who have all three engines, can test that easily with a couple of typical positions.
I would not assume that more knowledge causes the effect to be relatively stronger at longer times (at bigger depths resp.), but I am not sure about that.
M ANSARI wrote:N4 big weakness against R3 is search ... N4 is much much slower than R3 in search ... and R3 totally destroys N4 at fast time controls because of that. The reason for that could be too much knowledge code that is hogging it down, or simply that Vas ... the king of search ... has too many tricks than nobody else has yet discovered. That is one of the reasons that N4 scores much better at longer time controls. The chances of being outsearched in a critical line are much less.
True, at fast time controls Rybka outperforms Naum by 120 points or so. On slow by 60. If Uri is right about "decisive" games, then Naum could approach Rybka, but never surpass it in very slow time controls and very fast hardware. Yet to be tested. But no doubt, Naum is the most serious competitor. Only Zappa had such improvements on long matches. You are right, Rybka is certainly faster searcher, but it is beyond me why Zappa and Naum perform better at slower time controls. More knowledge? - maybe.
Kai
Yes ... Rybka is a phenomenal searcher . Vas uses some pretty amazing tricks in search ... somehow he is able to let the engine throw away the garbage and focus CPU search resources directly at important critical lines. Other engines unfortunately have not been able to figure out how to avoid going through all the garbage layers and thus need a lot more time or much faster hardware. The result is that Rybka will search deeper and that makes all the difference. At fast time controls obviously tactics play a much bigger role and thus focused search depth is much more devastating. At 1_1 time controls all Rybka 3 engines score above 80% against N4 and some go as high as 90%.
The evaluation function of both Zappa and N4 are quite good ... althoug still not as good as Rybka. You will find some situations where N4 will have a much more accurate evaluation than R3 ... but generally speaking N4 will fail to capitalize due to inferior search. I think R3 can learn a lot from N4 and can probably tune its evaluation function a little more.
2 things that N4 evaluates better are the value of bishop pair ... and also the value of 2 rooks vs. queen. Still these are not always perfect and both sides could learn from each other. N4 will sometimes misevaluate how useless a bishop pair is if the bishop is blocked ... also it will misevaluate the importance of a dynamic queen instead of 2 rooks ... something which it evaluates properly when the queen is not so dynamic. This is all coming from what I have seen from the thousands of games so far played on my computer tourneys. Most likely some games need some serious analysis for a definite conclusion. Also the fact that N4 is so much weaker in search could mask the correct evaluation by making it look that the static evaluation is wrong when it is actually poor search that does not allow it to find the correct moves.
One more thing I would like to add ... is that the default Rybka 3 has a contempt of 15. This causes Rybka 3 to overreach in many positions and loses games it shouldn't. This is especially true in long time control games. If you change the contempt of R3 to 0 ... it scores quite a bit better against N4 and avoids many of those losses. N4 is simply too strong at long time controls for R3 to have a 15 contempt.
"N4 big weakness against R3 is search ... N4 is much much slower than R3 in search". Are You serious? I thought, that Naum secret is just search.
It gets 3-4 ply more in blitz than Rybka and 1 Mnps in single CPU!
Jouni wrote:"N4 big weakness against R3 is search ... N4 is much much slower than R3 in search". Are You serious? I thought, that Naum secret is just search.
It gets 3-4 ply more in blitz than Rybka and 1 Mnps in single CPU!
Jouni
No, no, Majd is right, Rybka usually simply outsearches its opponents.
Jouni wrote:"N4 big weakness against R3 is search ... N4 is much much slower than R3 in search". Are You serious? I thought, that Naum secret is just search.
It gets 3-4 ply more in blitz than Rybka and 1 Mnps in single CPU!
Jouni
No, no, Majd is right, Rybka usually simply outsearches its opponents.
Kai
Majd is definitely right,this matter is not questionable at all
Jouni is just misleaded by the depth value which is not a direct indicator of the search function....
_No one can hit as hard as life.But it ain’t about how hard you can hit.It’s about how hard you can get hit and keep moving forward.How much you can take and keep moving forward….
Jouni wrote:"N4 big weakness against R3 is search ... N4 is much much slower than R3 in search". Are You serious? I thought, that Naum secret is just search.
It gets 3-4 ply more in blitz than Rybka and 1 Mnps in single CPU!
Jouni
I am not sure about how accurate that is. Search depth means nothing if you are searching the wrong lines or the non critical lines. Rybka 3 knows what the critical lines are and has a "focused" search. This allows Rybka to search much deeper than other engines in the CRITICAL lines ... note the emphasis on critical. I have not tried N4 in tactical suites ... but I would expect it to score very poorly in comparison to Rybka in such a test.
Last edited by M ANSARI on Sat Jan 03, 2009 3:56 pm, edited 1 time in total.
