Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / How would the centaurs fare?
- - By billyraybar (***) Date 2009-07-15 13:11 Edited 2009-07-15 15:56
How would the centaurs fare if they were not allowed to use Rybka 3?
Parent - - By Vempele (Silver) Date 2009-07-15 13:21
It'd be unfare.
Parent - By koroljow (*) Date 2009-07-15 14:29
Thanks!  I can't stop laughing.
Parent - - By FICGS (**) Date 2009-07-16 17:01
Good one  lol

Well, they'll use another good engine... Rybka will be catched one day, undoubtly...
Parent - - By Vempele (Silver) Date 2009-07-16 17:05

> Good one  lol

It was even better before the spelling mistake in billyraybar's post was mysteriously corrected. :lol:
Parent - By keoki010 (Silver) Date 2009-07-16 17:09
I liked the original.  faere
Parent - - By Capa (***) Date 2009-07-15 13:57
It would depend on how good the centaur is, but the really good ones would fare well.  Don't get me wrong, I don't mean that they would dominate Rybka cluster, but they wouldn't get dominated either.  Perhaps they would be just a tad better.
Parent - - By billyraybar (***) Date 2009-07-15 15:49
That's pretty impressive.  I guess that means a good centaur would faer more than okay if both computer and human used same engine
Parent - - By Uri Blass (*****) Date 2009-07-16 06:40

Capa thinks that... does not impress me.
People have a lot of wrong opinions.

Parent - By Capa (***) Date 2009-07-17 00:56
No need to be so personal---we can all err.
Parent - By sarciness (***) Date 2009-07-15 16:04
Good question. I'd guess that with other top englines they'd do pretty well.
Parent - - By Nelson Hernandez (Gold) Date 2009-07-15 22:28
I think everyone here is wrong.  It would be a pretty serious take-down.  If your opponent is surprising you time and again, and you're making computer-dependent moves that are inferior and spending a lot of time searching inferior lines it's definitely a problem.  It can only be partially offset by a superior book and high chess skill (which allows you to overrule an engine that is making a poor suggestion). 
Parent - - By Vasik Rajlich (Silver) Date 2009-07-16 04:42
Actually Harvey Williamson is planning to organize a match between the cluster and Hiarcs + GM, so we might test this.

I think Elos are more-or-less additive under these conditions, but we'll see.

Parent - - By M ANSARI (*****) Date 2009-07-16 06:19
I think it would have to go further than that, they have to choose a move that is a Hiarcs move or that is not chosen by Rybka.  Either that or verify without a doubt that only Hiarcs is actually the engine used and that no Rybka 3 somewhere is being also looked at.  I think without Rybka 3 output, even the best Centaur in the world will have a very difficult time against RC.
Parent - - By Highendman (****) Date 2009-07-16 06:51
I wouldn't mind testing that against the Cluster, until the GM+Hiarcs challange takes place.

Obviously it's a question of 'honour' if I say I'm _not_ using Rybka you'll just have to believe me ;) but I'd have no reason to lie would I?
I already have some feel on Rybka's analysis. I'm not familiar at all with any of the other engines. But I could practice with one of them for a while to get a feel, and then play a Cluster game using only that engine - I wouldn't mind which engine as long as it's one of the current top non-rybka engines out there.

Considering I held my ground against the Cluster (even if a lot had to do with lucky opening positions) then something like this would show some info, possibly.
Parent - - By Mark Mason (***) Date 2009-07-16 07:50

Please excuse my ignorance of the finer points of centaur play - but say in a given position the strongest possible move is the one found by Rybka 3, then how can a centaur not using Rybka find a stronger one ?  We know that any other engine he is using for assistance is by definition weaker than Rybka 3, so unless he has the top 5 super GMs in the world sitting next him - what resource can he use to find a stronger move ?

Parent - - By Rowlando (***) Date 2009-08-08 08:19 Edited 2009-08-08 08:22
While there's no doubt that Rybka is an amazing analysis tool, I have to take issue with your statement that Rybka always finds the strongest move.  If I've learned anything at all in my correspondence career it is this:  In complex unforced positions engines are extremely proficient at finding very good moves but they rarely ever find the best move.  I've seen evidence of this time and time again in my correspondence games against deep-ply engine monkeys.  If Rybka always found the strongest move I wouldn't be able to outplay her and yet that's clearly what's happening -- a slow, steady improvement in position over the course of the game. 

You can also see evidence of this when using IDeA which does a great job of exploring IA's peripheral vision (i.e. prematurely pruned).  In my opinion, IDeA's single greatest contribution to my chess analysis is that it taught me that IA isn't the word of god -- there are always improvements waiting to be found for those that are patient/skilled enough to do the finding.  Fortunately for those of us who enjoy that kind of "work", it still takes a human to do the serious digging.
Parent - By Uri Blass (*****) Date 2009-08-08 12:31
There are not many deep-ply engines monkeys.
Correspondence games prove that big majority of the correspondence players are clearly weaker than deep ply engine monkeys.

I guess that more than 90% of the ICCF correspondence players are one of the following:

1)Players who do not use a chess engine
2)Small ply engine monkeys
3)players who trust too much their own brain and use rybka only to check for tactical errors(or maybe not only for it but they too much tend to reject rybka's move and play an inferior move).
4)players who use other engines and do not use rybka.
Parent - - By M ANSARI (*****) Date 2009-07-16 09:04
I think you will find that things will be quite a bit more difficult.  In a game you will get a few instances where Rybka 3 evaluates stuff a little differently, and more often than not it will be right.  I think the best non Rybka engine to try against the cluster would be Naum 4 and then try and keep both bishops on board.  Still it would be a very difficult task to beat RC without aid of another Rybka.
Parent - - By Uri Blass (*****) Date 2009-07-16 13:10
In the game that the cluster lost with white is was the cluster that kept the 2 bishops on the board

I remember that you claimed that one of the weaknesses of rybka is underevaluating pair of bishops but it does not seem to be the case based on my experience in correspondence games and it seems to me that rybka can overevaluate pair of bishops so it may be better to be the side that does not have the bishop pair against rybka.
Parent - - By Banned for Life (Gold) Date 2009-07-16 13:23
The method used to test Rybka assures that "on average" the bishop pair will have the proper value (i.e. either raising or lowering the bishop pair bonus will reduce playing strength). This means that if the bishop pair is overvalued in some instances, it is undervalued in others. I'm not convinced that Rybka has more of a problem with this than Naum. It is more likely that Rybka wrongly evaluates different cases.

If the percentage of cases where there is a significant misevaluation is high, using another engine may be a good strategy to get some decisive games (but not necessarily to improve win percentage).
Parent - - By Uri Blass (*****) Date 2009-07-16 13:27
I disagree

Rybka was tested at fast time control
It is clearly possible that things are different at slow time control and you need to change the value of pair of bishops to get better results at slow time control.
Parent - - By Banned for Life (Gold) Date 2009-07-16 14:54
Of course the issue of feature weights being sensitive to analysis time has never gone away, but there is no evidence of a significant shift in weights either (at least in the range of testability allowing a statistically significant number of games to be played).

Note that even if the optimal weights do shift with analysis time, the optimization methodology for Rybka 3, and presumably other engines as well, focuses on getting the best weights for a given feature set on a large set of undifferentiated positions. This will inevitably lead to finding a "centroid" weight which overvalues the weight in some positions and undervalues it in others. There is no reason to expect the set of positions to be the same in different engines, so it makes perfect sense to look for types of positions where engine a plays more accurately than engine b.

I don't have enough experience with the bishop pair bonus to make any statement about Rybka vs other engines (as M. Ansari has), but I don't use Rybka in Rook-Pawn endgames for this reason.
Parent - By Vasik Rajlich (Silver) Date 2009-07-17 05:23
It can be a matter of different engines having different "centering". It can be a matter of different engines having different heuristics. It can also be a matter of all sorts of irrationality - for example a testing bug could make an engine play better in some class of positions.

That's why this is all so much fun. :)

Parent - - By M ANSARI (*****) Date 2009-07-16 19:07
I think you are very wrong.  One of the main reasons the RC lost that game is because it did not try to keep the bishop pair.  If white had taken on h3 immediately  black response would have been to push center pawn and would have kept bishop pair in an open position and if anything white would have the edge.  I did several hundred games against N4 and I am convinced that in some positions N4's exagerated score for the bishop pair wins a lot of games and draws a lot of games with a pawn dawn convincingly against R3.  Ofcourse N4's high bishop pair score sometimes backfires and is probably a bit too high, but i am absolutely 100% sure that Rybka 3's evaluation can be improved by increasing by about .2 to .4 centipawn.  The problem is programming the correct knowledge on when to score high and when to keep it more toned down.  As it is Rybka 3 now has an evaluation that works best all around, most likely this is due to the fact that in when the evaluation was tested it was tested at very quick time controls.  A bishop pair advantage is sometimes hard to calculate all the way through to a tactical advantage, but usually if you keep the pair long enough and you have an open position, eventually something good will come out of it.  This strategic advantage is very hard for an engine to quantify because an elevated evaluation that does not show a tactical advantage by brute force will confuse it.
Parent - - By Uri Blass (*****) Date 2009-07-16 19:35
I have experience in correspondence games when I had the bishop pair and could do nothing with it(I did not play that game well
4.Nc4 is probably wrong and I was affected by the fact that rybka played this move in WCCC and maybe 12.cxd5 is also not a good way to play for a win but the bishop pair clearly did not help me).

Peter zebra one of my opponents also could do nothing with the bishop pair with an opponent that let him to get it.

I also drew with him but did not play 13...Bg4 but 13...Ne4 because my fear of the bishop pair after reading your posts.
It seems that my fear had no basis.

Here is our game

I have other games in progress when I have the bishop pair and I do not see the score go up with deep search inspite of your words
so I am afraid the bishop pair worth nothing.
Parent - - By M ANSARI (*****) Date 2009-07-16 20:57
These are terrible examples of where a bishop pair can dominate.  A bishop pair is not useful in blocked positions or where their scope is limited.  As a correspondence player I am surprised that you think that simply having 2 bishops should mean that you already have a strategically won game ... if only chess could be that simple!  Ofcourse simply having 2 bishops is an advantage but this advantage is only realized if you eventually are able to get the position which makes the pair strong.  Usually that is in the late middle game or early endgame where there are a lot less pawns.  The ideal situation would be to have such a position facing a bishop and a knight.  Your opponent will obviously try to stop that from happening and if he plays well he will be able to neutralize your bishop pair.  Obviously it makes no sense to totally compromise your position stragegically simply to keep your bishop pair ... after all it is worth at the most between .3 to .9 centipawns depending on the position.  If you are interested I can look up some of the games where N4 used its exagerated high score for the bishop pair to win convincingly against R3.  The funny thing is that in many games between Rybka 3 and N4, R3 will eventually see that tactically it is in big trouble but N4 search can't see it yet.  This is why I tend to believe that if R3 had a slightly higher evaluation for the bishop pair its superior search would be a devastating combination.

By the way I used to also be very sceptical on the usefullness of the bishop pair and somehow thought that it only works when a stronger player is playing against a weaker player because the weaker player will not notice the danger of a budding initiative.  But after looking through many games that R3 lost I have since changed my mind.
Parent - By Uri Blass (*****) Date 2009-07-17 01:45
I did not think that I have a won game because of a bishop pair but I prefered lines that lead to bishop pair relative to lines that do not lead to bishop pair when the evaluation of rybka was similiar after your posts and it seemed not to help(I did not play moves that are positional blunders based on rybka and my human choice in book moves was only when the difference in evaluation was less than 0.1 pawns).

Parent - By yanquis1972 (****) Date 2009-07-16 19:19
imo the ideal combo would be something like Rybka 3, Rybka 3 Dynamic, & Naum 4, each running on 8 cores or so. i think a good centaur would have great chances with that setup although highendman is doing more than ok with less.
Parent - By Roland Rösler (****) Date 2009-07-16 10:19
I´m a little bit worried about Rybka 4 beta (CR software) improvements  more than 11 months after R3 release.
Scenario: You have to play with Rybka 2.3.2a on your Nehalem-EP against current cluster from Lukas with Rybka 3 (with cluster part). How are your chances?
My vote: There would be no chances on your side.
Hiarcs 12 is about 150 Elo worse than R2.3.2a and 250 Elo worse than R3 (CEGT 40/120 Quad).
Parent - - By Nelson Hernandez (Gold) Date 2009-07-16 10:56
You're too modest when you ascribe your success to "lucky opening positions".  You played the openings as well as any top centaur.  That isn't luck, it is skill.
Parent - By Highendman (****) Date 2009-07-16 12:01
The luck part was the book the Cluster used, I guess ;)
Parent - - By Capa (***) Date 2009-07-16 11:47
Would love to see you do this, Highendman.  You've proved already that Goliath isn't invulnerable and, yes, depending on how you perform, it would answer a lot of questions.  Have you ever entered a freestyle tournament or played in any capacity as a centaur previously?
Parent - - By Highendman (****) Date 2009-07-16 11:57
No. The Cluster game was my first attempt at playing as a 'centaur' and last night I had my first two games ever against a human centaur (Eros).
Parent - - By Capa (***) Date 2009-07-16 12:04
I find it simply unbelievable that you have no elo rating and that you have no experience as a centaur and yet you beat Rybka cluster 3 times!  How did you fare against Eros?
Parent - - By Highendman (****) Date 2009-07-16 12:19
Draw in the first game (I was black) and won the second game (I was white).
Games here:
Parent - - By Capa (***) Date 2009-07-16 12:22
Are you sure you aren't a super GM in disguise? ;)
Parent - By InspectorGadget (*****) Date 2009-07-17 10:43


> Are you sure you aren't a super GM in disguise? <IMG alt=";-)" src="/mwf/epx.png" class="sic sic_sml_wnk"/>

Maybe he is Nigel Short :-D
Parent - By Capa (***) Date 2009-07-16 09:40 Edited 2009-07-16 09:50
In this given scenario (different program than Rybka) the human aspect of the centaur would definitely have to be a very strong chess player.  A lot are assuming that the centaur would continue to draw from inferior lines and Rybka cluster from superior lines---but this isn't necessarily so.  Any program is capable of a "stupid" move, some just more likely than others.  A very good chess player armed with a very good program, armed with very good hardware, would fare well.  He could help avoid those inferior lines and possibly find a better line than Rybka (I know, blasphemy).  I think a super GM armed with like weapons would defeat Rybka cluster in a 10 game match.  I think Highendman has shown that Rybka cluster is not invincible with his 3 wins, 1 loss, and 15 draws or so.  True, Rybka made some "stupid" moves in the opening of those games and Highendman was armed with Rybka 3, nonetheless it showed the chink in the armor.  I found Highendman's comments very interesting in the last game that he won where he essentially states that a couple of times he was able to see deeper than Rybka cluster and find plans faster.  Not bad for someone with no elo.  In fact Highendman later stated: "One of my considerations, when I was thinking for a long time over a given position, was to attempt to NOT play what I thought the Cluster was expecting, so it will lose it's ponder hits and time.
Of course, so long as the alternate move made sense and was in line with my plan."  I find this quote very fascinating (to borrow a term from Spock) because he did what so many would call "not the best move" from Rybka cluster's perspective and yet won.
Parent - - By InspectorGadget (*****) Date 2009-07-17 10:35

> Actually Harvey Williamson is planning to organize a match between the cluster and Hiarcs + GM

I have a feeling that the GM will win a 10 game match. Highendman is not a GM but it looks like he is dragging the cluster on a gravel road. :-D
Parent - By Gaмßito (****) Date 2009-07-17 11:54

> Actually Harvey Williamson is planning to organize a match between the cluster and Hiarcs + GM

I don't know if this match will be organized, but I think it's quite difficult for a GM to win a match against Cluster Rybka (running the Cluster with a strong opening book) and him without Rybka 3.

It will be very nice to see this match but somebody knows what happened? I didn't listen anything about that again.

Parent - - By Uly (Gold) Date 2009-07-17 21:28
I think Highendman could sweep the floor with the Hiarcs + GM.
Parent - By keoki010 (Silver) Date 2009-07-17 22:39
I agree.  Would be an interesting match????? Anyone up for the challenge?
Up Topic Rybka Support & Discussion / Rybka Discussion / How would the centaurs fare?

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill