Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / rybka at very long time controls
- - By chessmaster15 (**) Date 2011-08-08 19:51
is rybka still above other programs at slow postal chess ect . still beats programs like hiarcs at say 3 days per move or does the rating diff shrink quite a bit . im guessing it shrinks alot , if there are any postal players let me know .
Parent - - By turbojuice1122 (Gold) Date 2011-08-08 19:54
It should shrink simply because of approaching the draw boundary of chess, but Rybka should still be above other programs, and is probably above Houdini in this regime, especially Rybka 4.1 (I'm confident enough in this that I use Rybka 4.1 instead of Houdini for my long-running IDeA openings project).
Parent - - By chessmaster15 (**) Date 2011-08-08 20:22
what kind of project you doing sounds very interesting ?
Parent - By turbojuice1122 (Gold) Date 2011-08-09 03:12
Basically analysis of many various openings using IDeA.
Parent - - By leavenfish (***) Date 2011-08-14 02:38
"...because of approaching the draw boundary of chess..."

I'm sorry, I don't think you know what you are talking about there. It's about evaluation several moves in at the moment and that is all. Computer programs do not really 'plan'. They react according to tactical calculation and some still fairly rudimentary evaluation...and play into a void where say 8 to 10 half moves later how they evaluate the position that arises with the inevitable horrizon effect forever just in front of them.
Parent - By turbojuice1122 (Gold) Date 2011-08-22 15:31
If what I said had no validity, then the draw percentage wouldn't increase as significantly as it does when you look at rating lists going from blitz to longer time controls.  In other words, your explanation is only part of the issue--also significant is all of the information that the program takes in and evaluates on the way toward reaching its final conclusion, typically about 30 half-moves (in the case of correspondence chess) or 20-24 half-moves (in the case of tournament time controls--both very far removed from your 8-10 half-moves).  The quantity of this information and how it changes is very relevant in how the program reaches the final positions that are to be evaluated and whether this evaluation represents "the truth" of the current position.  Of course, the final evaluation won't necessarily be of a position reached with the most effective possible play, but we are in a situation where the stuff that has been discounted has been "weighed" enough that the conclusion itself is generally quite reliable with the strongest programs.
Parent - - By Razor (****) Date 2011-08-08 20:04
In the various TCEC long-time control competitions between H1.5a and R4.1, H1.5a always had the edge - also worth noting from the same long-time control events H1.5a appeared to gain larger advantages against other chess engines such as Critter, Stockfish, etc., than R4.1 did.
Parent - - By NATIONAL12 (Gold) Date 2011-08-08 22:50
i and a number of others have found Rybka 4.1 better than Houdini in WBCCC tournament(on Rybka forum).In a couple of games of mine i would have lost if i had followed Houdini's ideas.to give an example Houdini was quite happy with my going into an ending with 2 Knights against 2 Bishops,both my opponent and i realised this was a lost line althogh it may have taken 100 moves to win.At corr level with big hardware i am more than happy with Rybka.

needless to say i do not always play either engines reccommended moves.:smile:
Parent - By Uly (Gold) Date 2011-08-08 23:43
+1.

My example deals with my Halloween challenge against Moz, where there were plenty of lost positions in some lines that Houdini was scoring at 0.60. Houdini has been useful in that game, but if I relied on it I would have already lost.
Parent - - By Capa (***) Date 2011-08-09 00:11
+1
The main problem with the theory that Rybka might be stronger than Houdini in very, very long time controls is the fact that in every test engine site, whether testing with short time controls or long ones, Houdini is substantially stronger, greater than the difference between Rybka 3 vs Rybka 4. Even National used top notch hardware and longer time controls and Houdini was about 30 elo stronger, which is quite significant. These are the very sites that Vas trusts to judge elo for his engines. If Vas trusts these sites, and if all of them have Houdini significantly ahead, then that should be the end of the story. If I can be brutally honest (sorry), this theory smells like a conspiracy hatched up by Rybka shills. All the objective evidence points to Houdini. Just because there were several games here and there where Rybka MIGHT have been stronger proves nothing. Perhaps this will inspire Vas to give Rybka 5 a substantial boost.
Parent - By Razor (****) Date 2011-08-09 04:28
Totally agree with your first point; I would also add that there isn't one engine that always plays the best move in every position encountered - anyone hoping for this will have to wait a little longer.

I believe the TCEC event {sadly missed BTW} was a good testing ground for all the best public domain engines and certainly from the results available H1.5a had the edge, meaning in my view, it must have been getting it right more times than its opponents, including, R4.1.  All of this of course was played on 6-core machines {can't remember the benchmark but I suspect it was significantly higher than my i7 920! :sad:} and whilst one can get more cores than this I suspect that most people will have 6 or less cores in their kit at home.
Parent - - By M ANSARI (*****) Date 2011-08-09 08:22 Edited 2011-08-09 08:25
The reason Rybka 4.1 is strong at LTC is not because of Houdini, but rather because it has just a horrible horrible time management setup which really hurts it at shorter time controls.  By that I don't only mean the time algo, but also the efficiency of how it initializes the engine and runs on a PC ... it just doesn't seem optimized for a PC OS.  There is no doubt that Rybka 4.1 has a better static evaluation and theoretically be stronger as time controls go up (you can see that in fixed depth testing)... but how much time should you give an engine?  I mean most games are at a certain time control, and here Rybka 4.1 is weaker.  One thing to add is that Houdini loses a lot of games due to a few poor static evaluation setups ... and those few weaknesses would be incredibly easy to fix, and thus would make the engine much stronger.  On the same token, Rybka 4.1 also has an incredible weakness with its time management and optimized PC porting, and that also should be very easy to fix.  Whatever happens, you can be sure that the next release of Rybka or Houdini will be at least 40 ELO stronger in the rating lists as they have some very simple "target rich" areas they can improve on.
Parent - - By Uly (Gold) Date 2011-08-09 23:17

> Whatever happens, you can be sure that the next release of Rybka


I'm not so sure. Remember Vas has to weaken Cluster Rybka software to make Rybka 5 UCI, so leaving those issues in there would save effort on the weakening.
Parent - - By M ANSARI (*****) Date 2011-08-10 06:13
I don't know, I think that would counter productive.  Rybka Cluster can always have the edge simply due to overwhelming hardware.  I think I know Vas pretty well and he doesn't like being second best.  He will want to have the strongest UCI engine and Rybka 4.1 is clearly not that at the moment.
Parent - - By Uly (Gold) Date 2011-08-10 07:55
I don't know if Vas is even aware that Rybka is no longer top 1 at games.
Parent - - By Lukas Cimiotti (Bronze) Date 2011-08-10 10:11

>Rybka is no longer top 1 at games


Depends on the Rybka version :wink:
And yes, Vas knows Houdini is strong.
Parent - By InspectorGadget (*****) Date 2011-08-10 11:29

> Depends on the Rybka version :wink:


So does that mean there are other UCI engines stronger than Houdini in other parts of the world like Germany and Poland for example? :smile:
Parent - - By Uly (Gold) Date 2011-08-10 12:17

> Depends on the Rybka version :wink:


And time control :wink:

(Referencing that Critter destroys Cluster Rybka at 1 0 time control due to lag)

> And yes, Vas knows Houdini is strong.


I think Rybka UCI would be stronger than Houdini if Vas rewrote the time management to something that worked...
Parent - By InspectorGadget (*****) Date 2011-08-10 13:25

> I think Rybka UCI would be stronger than Houdini if Vas rewrote the time management to something that worked...


Make it work like it did in Rybka 3 :cool:
Parent - - By Dragon Mist (****) Date 2011-08-10 14:43

> Depends on the Rybka version <img title="wink" class="fsm fsm_wink" alt=":wink:" src="/mwf/v4/epx.png" />
>


Lukas, I read this as "there is already an UCI Rybka 5 beta that performs stronger than Houdini on popular time controls". And it makes my heart go berserk. :cool:
Parent - By Lukas Cimiotti (Bronze) Date 2011-08-10 14:47
No, there is no Rybka 5.
Parent - - By Capa (***) Date 2011-08-10 20:58
I believe that he was referring to the cluster version.
Parent - - By Dragon Mist (****) Date 2011-08-10 23:21
Well, I wasn't referring to R5, but R5 beta, and also if he referred to cluster his statement "depending of the version" makes no sense. (For instance, who says Houdini on super duper double Xeon something isn't as good as R cluster?)
Parent - By Regularuser (***) Date 2011-08-11 07:13
Yeah, we are still waiting for like-for-like core results for the cluster software, say 4 cores of cluster software vs 4 cores of Houdini, or for that matter any other program.    All we have is Werewolf's testing which is not like-for-like in terms of cores.
Parent - By AWRIST (****) Date 2011-08-10 11:08
It somewhat makes me wonder what alleged friends of Vas are pondering. Just two details that make me think. In a German forum people are asking questions such as when will the "professionals" come out with a new version or update whatever. Here I read such "questions" if Vas knows that someone else is "stronger". So, I assume that computerchess testers and lovers are a bit in a schizo mode.

Let me sum up. Some undemocratic movement in computerchess has singled out Vas for something done not ecxactly according to some "rules" while "professionals" could keep their closed rituals with copying but nobody neither in this "movement" nor elsewhere seems to care about the wrongdoing of these programmers. Actually they are all waiting for a new Rybka so that they can continue the copying and even Vasik's closest "friends" are hoping that their favorit quickly becomes first again. The friends dont care if the new Rybka will be ripped apart for its use in competing programs. The friends dont care about exploitation, they want to enjoy the advantages of a new Rybka for a couple of weeks before worldwide the experts of crime do their disassembling plus RE.

Did ever someone observing all this nonsense had an idea how Vas should continue to deliver? Is he the milkcow for the rest of the chess programmers?
- By activethinker (**) Date 2011-08-13 17:10
I think NATWARLAL is strongest chess engine at long time contol like 40 moves in 100 years.:lol:
Up Topic Rybka Support & Discussion / Rybka Discussion / rybka at very long time controls

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill