Topic Rybka Support & Discussion / Rybka Discussion / Rybka is definitely strong, but it should be stronger!
1
2
Poll
Rybka more needs a New (vote only one)... (Closed)
| bugfix. | 39 | 48% | |
| Stronger Version. | 43 | 52% |
![[us]](/mwf/v12/flags/us.png "United States")
Rybka should be a stronger engine! It has lost too much ground from so much as the likes of the top green and orange free engines of CCRL! The POSITION Rybka is in requires more than a bugfix!--unless it is to go the way of great chess engines of the past: Chessmaster, MChess Pro, Fritz, Hiarcs, Shredder. Paying customers will want more ELO over the next highest free engine! Right now the difference (CCRL: #1 Rybka 4 64-bit 4CPU = 3264; #2 Stockfish 1.9.1 64-bit 4CPU = 3222) is a mere 42 elo points when most would like the difference to be at least twice that amount of difference.
Right now the difference (CCRL: #1 Rybka 4 64-bit 4CPU = 3264; #2 Stockfish 1.9.1 64-bit 4CPU = 3222) is a mere 42 elo points when most would like the difference to be at least twice that amount of difference.
![[de]](/mwf/v12/flags/de.png "Germany")
Rybka is one of the strongest chess engines and it has a lot of chess knowledge compared to more tactical engines. Did you ever have a look at engine vs. engine matches? The engines move the pieces without any longterm plan unless one of the them looks deeper than the other engine and gains an advantage to win the game. I hope rybka will not be "optimized" for engine vs. engine matches because for making it look deeper, you have to get rid of most chess knowledge to make it faster. I want to play against a chess engine and analyse with it. I don't want to run thousands of engine vs. engine matches to find out which engine is the strongest. That's a nice comparison but not the main reason for developing chess engines. What do you do with a 3000+ ELO chess engine that knows nothing about chess?
das ist ein totaler widerspruch :)
Those who know a bit about chess programming get my point. Those who want their chess engine to be on the top of the list, no matter what, will never understand :p
I think that (more) chess knowledge and combinational strength are not a contradiction, rather the opposite... As far as I am able to understand these matters, a bigger eval function leads to more prunings and therefore to a smaller search tree. A simpler eval will have a bigger node rate (> kN/s), but not necessarily a bigger effective ply depth within the same calculation time.
(That is how I assume that things work but I am not a chess programmer! )
(That is how I assume that things work but I am not a chess programmer!
)
The more complex an evaluation function is, the more time it consumes to calculate the score for the position. And more different values will not lead to more prunings but to a wider search tree because alpha beta pruning works best for equal values.
it's quite a difficult subject. But I agree that one doesn't want to use a engine with crippled evaluation function (like Rybka winfinder) for analysis. The engine may be tactically strong and from elo point of view not much worse than the "normal" engine, but the quality of the suggestions is much lower.
If people only want to buy engines which are on top of the ELO rating lists, the chess programmers are forced to tweak their code for engine vs. engine matches. The mere alpha beta searchers of ~200k file size are way ahead the "serious" engines, this development is not good for computer chess. I want an engine to play entertaining chess and do some proper analysis of grandmaster games. I don't need an engine telling me that in 20 moves I will lose a pawn.
Note this is usually not the case. Any serious engine programmer will have quite a good evaluation function, so I would say that CCRL and CEGT reflect the actual analysis quality of an engine quite well. But you may be right for the engines not listed there :)
I totally disagree. Stockfish has about 200k file size (32bit version). How much chess knowledge can you put into 200k C++ code? Chess engines are not written in hand optimized machine code anymore like in the early days of chess programming. If you have a decent position evaluation your code will easily be 1 or 2 Megs and above. Have a look at the file size of Rybka 2.3.2a and you will see the difference.
![[fi]](/mwf/v12/flags/fi.png "Finland")
You just lost all your credibility.
Have a look at the file size of Rybka 2.3.2a after compressing it. Hint: 7-Zip shrinks Stockfish from 202kb to 199kb.
> Have a look at the file size of Rybka 2.3.2a and you will see the difference.
Have a look at the file size of Rybka 2.3.2a after compressing it. Hint: 7-Zip shrinks Stockfish from 202kb to 199kb.
Rybka makes heavily use of internal tables which can be highly compressed. If there are no tables, just plain code, there's not much to compress.
> Rybka is one of the strongest chess engines and it has a lot of chess knowledge compared to more tactical engines. Did you ever have a look at engine vs. engine matches? The engines move the pieces without any longterm plan unless one of the them looks deeper than the other engine and gains an advantage to win the game. I hope rybka will not be "optimized" for engine vs. engine matches because for making it look deeper, you have to get rid of most chess knowledge to make it faster. I want to play against a chess engine and analyse with it. I don't want to run thousands of engine vs. engine matches to find out which engine is the strongest. That's a nice comparison but not the main reason for developing chess engines. What do you do with a 3000+ ELO chess engine that knows nothing about chess?
+1
+1
+1
That is why I like Rybka. Hiarcs. Shredder and Junior!! (and am interested in Komodo!!)
Pure playing strength is not the only argument for buying an engine anymore. It's about reducing playing strength (not only by limiting the search depth but letting the engine do "human" mistakes), more versatile play (not only playing the best move in a specific position but also less good ones), playing style options (not only editable piece values but an easy way to change the engine's positional analysis) and accurate position evaluation (not only by tactical scores but a thorough understanding in the position's nature).
> Rybka is one of the strongest chess engines and it has a lot of chess knowledge compared to more tactical engines.
Vas had got good selling rates because his Rybka was stronger than other engines.
Not how she played but because she won against the others made many people buy her.
Let Fritz become stronger, and selling rates of Rybka will decrease heavily.
"Her style an understanding is so important" will only a buy reason for few people.
Quap
What do you need the strongest chess engine for? You cannot win against it and if it has no understanding of chess positions you can forget about its analysis results. Only few people drive a Ferrari, most have cars like Volkswagen, Dodge or whatever. But everybody wants to have the world's strongest chess engine. If you do lots of engine/engine matches, you have a ranking list based on tactical strength. If you want to find out how good the positional understanding of an engine is, you have to let it play against top human players. And you would have to analyse each game, not just look at the scores.
> What do you need the strongest chess engine for?
It's nice to have, to see, to think about.
If I only need a strong chess-opponent I never needed more than Fritz1 (as most rybka-customers too (I assume). Are you stronger than Fritz1 under serious conditions?)
I look at the scene since the eighties. I'm sure: strongnes of an engine was the main selling argument at each time!
Only very few people really would abandon to buy the strongest engine because they want a weaker one with better positional anderstanding.
And especially Rybka with her engine-built-in 'features' like missing BUP and 'wrong Bishop' shows, that people are willing to buy just the strongest, even if there are very visible uglinesses.
In my opinion "let Rybka be not the stongest Engine but that one with the best position understanding" would lead Rybke in a very short time to commercional mediocity.
And only rather few customers would honour those remaining Rybka strengths.
Quap
Those who actually play and analyse with their engine, not just perform engine matches, think differently.
What a proportion of all people, who bought a Rybka version, might think as you describe? 10%? or perhaps 15%? or only 5%?
I have to concede, that I do not really know, whether Hiarcs or Junior or Stockfish or ... have a better positional understanding than Rybka.
"Rybka with her over all strength will be good" was enough for my decision to buy Rybka.
Without tablebase her evaluations in the endgame are often bad, but overall she ist strong allthough.
Maybe you really know more clearly. OK. Perhaps you have that special personal quality. The vast majority of Rybka customers will not, I assume.
Many users might simply assume a good positional understanding of an engine because they see her winnig very often.
But wins can be produced by using different techniques. And its difficult and fault-prone to identify the main reason.
Quap
I have to concede, that I do not really know, whether Hiarcs or Junior or Stockfish or ... have a better positional understanding than Rybka.
"Rybka with her over all strength will be good" was enough for my decision to buy Rybka.
Without tablebase her evaluations in the endgame are often bad, but overall she ist strong allthough.
Maybe you really know more clearly. OK. Perhaps you have that special personal quality. The vast majority of Rybka customers will not, I assume.
Many users might simply assume a good positional understanding of an engine because they see her winnig very often.
But wins can be produced by using different techniques. And its difficult and fault-prone to identify the main reason.
Quap
I'm a chess programmer myself, that's why I know a bit about chess programming. My first program was in the 80's for the Commodore 64 (Happy Chess), the second one was for MS-DOS (Nova Chess) and my third one was for Windows (Pawn). I didn't write yet another UCI engine because my focus has always been people having fun playing against my programs and not writing the strongest engine which lets grandmasters look old. Don't know how big the engine crowd is, but I did some matches on playchess and some engine/engine tournaments myself. For me as a chessplayer the knowledge of an engine is more important than tactical strength. When most people just want to have the strongest engine for machine tournaments, programmers will address this. I hope that there will be a good balance of tactical and positional terms so that all customers get what they want.
There are many methods used in each modern engine (standard algorithms, often declared in wikipedia and other open sources).
The rather quick search basing on these technics should be implemented in each good engine. An I think, it is.
I think, basing on these methods the developer can try to do his own tunings of the search:
- maybe a better move sorting
- a better partially deepening of the search
- maybe speculative cuts
- other?
I think these improvements are important all(!) and should become standard (and they depend on position thoughts often) a little later.
The user will detect a quicker search ("only a quicker search!"). But it is more than this.
In my opinion the quality of an engine depends on positional and search strength.
And normally we will find, that such a good balance will result in a good overall performance of that engine, in good results. Serious results against strong humans are very rare, so we shoul lokk at the results aganst other engines.
An engine with a good positional understanding, which looses often, because it's search fails (and the opponents shows, that just that positional etsimation was wrong!) is not the best one in my eyes.
Quap
The rather quick search basing on these technics should be implemented in each good engine. An I think, it is.
I think, basing on these methods the developer can try to do his own tunings of the search:
- maybe a better move sorting
- a better partially deepening of the search
- maybe speculative cuts
- other?
I think these improvements are important all(!) and should become standard (and they depend on position thoughts often) a little later.
The user will detect a quicker search ("only a quicker search!"). But it is more than this.
In my opinion the quality of an engine depends on positional and search strength.
And normally we will find, that such a good balance will result in a good overall performance of that engine, in good results. Serious results against strong humans are very rare, so we shoul lokk at the results aganst other engines.
An engine with a good positional understanding, which looses often, because it's search fails (and the opponents shows, that just that positional etsimation was wrong!) is not the best one in my eyes.
Quap
Modern engines look so deep, the fear it will oversee possible threats is only valid for engine games. If an engine is optimized for machine tournaments, it may lose badly against a grandmaster because it has no understanding of longterm positional play. The basic chess algorithms you described were standard 20 years ago. Today we have dynamic evaluation, bit boards, position learning, hash tables, endgame tablebases, etc. That a good engine should have tactical and positional strength we agree on.
The reason why Bobby Fischer became world champion in 1972 is because he knew more about opening theory than anyone else.Would you want to play in a tournament and use a 2nd rate system against the Najdor Sicilian, or throw your hands up in the air and settle for a draw if your opponent chooses the Sveshnikov var.?Before going to an OTB tournament, it is important to have a well prepared arsenal at your dusposal.The stronger Rybka is, the easier it may be to find any weaknesses in your opening repertoire.I have spent many hours checking the ECO main lines for improvements and have made some startling discoveries with the aid of Deep Rybka 4 64 bit duo processor, which may then be used in OTB play.GM Grischuk once complained to Vishy Anand at Wijk aan Zee about the Sveshnikov Sicilian asking; "will chess ever be the same"? Vasik Rajlich may hold the key with a stronger Deep Rybka 5.Good luck and thank you.
Agreed, but it might be in a rental form.
![[ro]](/mwf/v12/flags/ro.png "Romania")
> might be in a rental form.
Disagreed , it might be in a free form ! S PS: I heard that the named (D) Rybka 4.1 will be an extremely buggy product !
What like you download it and use it for 2-3 years? What might be interesting is to purchase 2 years of rights to the betas the author(s) is working on. There was a Chessbase article, before the release of Rybka 4, that said there were over 3000 "Rybkas" (ie. betas) between version 3 and version 4. And during that time there were a lot of people upset with how long the wait was for the next release to come out, whereas with the rights to download betas, people could have been working on their own beta. This with the option to buy a specific beta might be interesting business (for the production-side). Perhaps the author(s) (Rajlich, Kaufman) could hire another programmer or something? And end up more productive! Would be nice if that could work anyway.
What might be interesting is to purchase 2 years of rights to the betas the author(s) is working on. There was a Chessbase article, before the release of Rybka 4, that said there were over 3000 "Rybkas" (ie. betas) between version 3 and version 4. And during that time there were a lot of people upset with how long the wait was for the next release to come out, whereas with the rights to download betas, people could have been working on their own beta. This with the option to buy a specific beta might be interesting business (for the production-side). Perhaps the author(s) (Rajlich, Kaufman) could hire another programmer or something? And end up more productive! Would be nice if that could work anyway.
![[mx]](/mwf/v12/flags/mx.png "Mexico")
bugfix.
There's a point where adding more elo doesn't actually significantly change the move choice of the engine to a point where it's obvious the newer version is better, it may actually slow me down as I have to use both versions for a while to check if the new version is actually better.
For strength, apparently Rybka 3 suffices me, but that's just for features like Persistent Hash that are so useful I can find better moves by just interacting with it than by using an stronger engine. Stuff like getting PH in Rybka 4 and fixing it would be a lot more useful to me than 40 elo more than Rybka 4, something I can't even notice if the output moves aren't obviously better most of the time (like it was with the jump from Rybka 2.3.2a to Rybka 3, 232a was clearly obsoleted and of no use anymore, unlike Rybka 3 that still outperforms everything with a warm persistent hash).
There's a point where adding more elo doesn't actually significantly change the move choice of the engine to a point where it's obvious the newer version is better, it may actually slow me down as I have to use both versions for a while to check if the new version is actually better.
For strength, apparently Rybka 3 suffices me, but that's just for features like Persistent Hash that are so useful I can find better moves by just interacting with it than by using an stronger engine. Stuff like getting PH in Rybka 4 and fixing it would be a lot more useful to me than 40 elo more than Rybka 4, something I can't even notice if the output moves aren't obviously better most of the time (like it was with the jump from Rybka 2.3.2a to Rybka 3, 232a was clearly obsoleted and of no use anymore, unlike Rybka 3 that still outperforms everything with a warm persistent hash).
Sorry Vytron but we disagree here. How many new customers do you expect to get from a bugfix of Rybka 4? Something like zero. Marketing: "Get Rybka 4....now with the bugs fixed!!"? Absurd. Absurd.
It's more like "we care about our customers and therefore we deliver an update if severe bugs have been reported". The race for the prime spot on the rating lists is over because simple alpha-beta-searchers will outperform "serious" engines by high margins. Maybe Vas will code a special light version for all engine/engine freaks but this has nothing to do with what chess is all about.
I kind of like your perspective, but I'm curious what engines you are thinking of....perhaps Junior, Hiarcs, Shredder? Those are programs that have touted their "knowledge". I think Junior had a non-elo approach to its marketing--focusing on computer vs human type of play, etc. Many seem to be talking about Komodo the way you seem to like. As I recall Kaufman said he and Dailey were trying "different" things.
What alpha-beta searchers? SOS for Arena?
IMO he would if he "could".
> simple alpha-beta-searchers will outperform "serious" engines by high margins
What alpha-beta searchers? SOS for Arena?
> Maybe Vas will code a special light version for all engine/engine freaks but this has nothing to do with what chess is all about.
IMO he would if he "could".
> What alpha-beta searchers?
Engines like Stockfish, Critter, Komodo, Spark, Gull, etc., (and maybe the socalled clones), are alpha-beta-searchers because they got all the chess algorithms implemented but have only rudimentary chess knowledge.
> simple alpha-beta-searchers will outperform "serious" engines by high margins.
An engine which gets outperformed by a simple alpha-beta-searcher is not 'serious' in my eyes.
Such an serious engine has to have a good search and a good positional evaluaten.
And this rather good balance is what Rybka makes so strong.
In an Interview some years ago Vas declared, that Rybka search is as good as that of her opponents, but her positional understanding is better.
Naturally:
positional understanding of engines is much worse than human understanding is.
But search is much better.
Just as one can say: "engine's chess understanding is rudimentary!"
we could also say: "human chess search is rudimentary!" :-)
relatively.
Quap
> Just as one can say: "engine's chess understanding is rudimentary!"
> we could also say: "human chess search is rudimentary!" :-)
That's a good one.
If you have a look at the sources of Fruit, Stockfish, Critter, Gull, Crafty, etc. every lousy wannabe could hack together some strong engine that rivals the "serious" engines. But it would only have a state of the art search with all the fancy pruning and extension stuff without any chess logic. It takes weeks to code an alpha-beta-searcher but it takes years to build in well balanced chess knowledge.
> ow many new customers do you expect to get from a bugfix of Rybka 4?
There were members that claimed they didn't buy Rybka 4 because Rybka 3+ (3's bugfix) was never delivered. So, those...
People will complain either way, and Vas wasn't going to do anything about the strength anyway (I think the plan was "Do you want strength? Then go rental Rybka!")
Vas should focus on fixing the bugs, which will result in a better program.
I hope that Vas pushes his way like he has done in the past.
The most important thing about an engine is that it provides good moves and evalutions,
all this kiddy-crap-engine-tournaments (1+0 and that stuff, how dumb are those
people to judge an engines strengh with such time-limits?) are not important.
Maybe Rybka will come in two versions/parameter-sets in the future,
for analysis/longterm-timeframes and kiddy-tournaments.
Still a bugfix for the multi-pv-problem is highly appreciated, its a shame
that it takes so long with such less feedback to the paying customers.
Ok, its the standard Rybka-policy we got used to over the years...
Alex
The most important thing about an engine is that it provides good moves and evalutions,
all this kiddy-crap-engine-tournaments (1+0 and that stuff, how dumb are those
people to judge an engines strengh with such time-limits?) are not important.
Maybe Rybka will come in two versions/parameter-sets in the future,
for analysis/longterm-timeframes and kiddy-tournaments.
Still a bugfix for the multi-pv-problem is highly appreciated, its a shame
that it takes so long with such less feedback to the paying customers.
Ok, its the standard Rybka-policy we got used to over the years...
Alex
> I hope that Vas pushes his way like he has done in the past.
I'm afraid Vas is not interested in Engine development any more as he has done in the past. :-(
Quap
haha, good one :)
just wait a bit and you will see you are wrong :)
just wait a bit and you will see you are wrong :)
I guess Quapsel is not to far away from truth...
As history showed Vas doesn't even care about solving serious bugs (and breaking in multi-pv is for sure a serious one, as is/was hash-problems and so on) from paying customers, but more about fancy tournaments/cluster-project & politics/religions against better engines, in friendly handshake with the CB-monopoly....
Even the free engines have way more "customer" service!
As history showed Vas doesn't even care about solving serious bugs (and breaking in multi-pv is for sure a serious one, as is/was hash-problems and so on) from paying customers, but more about fancy tournaments/cluster-project & politics/religions against better engines, in friendly handshake with the CB-monopoly....
Even the free engines have way more "customer" service!
Vas "does" care about bugs. Or how do you explain he's working on the Rybka 4 update?
About tournaments: They are just some good reason to have a look at the engine's play and are extra motivation to improve the engine. Nothing fancy about it. The Cluster project will be available for everyone at some point, so it's for everyone. Vas didn't spend much time on the clone stuff.
About tournaments: They are just some good reason to have a look at the engine's play and are extra motivation to improve the engine. Nothing fancy about it. The Cluster project will be available for everyone at some point, so it's for everyone. Vas didn't spend much time on the clone stuff.
Simple question: Where is the free bugfix/update for the users who paid for Rybka 3?
How long has it been now, ~2,5-3 years?
How long has it been now, ~2,5-3 years?
that is another topic :) at least he does it better for Rybka 4.
Pffff, for R4 bugfix we will wait more than 7 months, maybe 8, 9 .... ?
Depending on the number of bugs, testing takes a while. The more complex a system is, the more time it takes to test it thoroughly. We programmers have a saying for all impatient people: "It's done when it's done!". Maybe you've heard of it...
![[gb]](/mwf/v12/flags/gb.png "United Kingdom")
Yep, unfortunately, its never done, there are always bugs, as a programmer this is something you may also recognise!
and since there is no bug free sofware then it is never done. What I do not understand, why is he not releasing "intermediate" versions with some of bugs fixed. That way we could use less buggy software.
First of all it's hard work, alpha testing, beta testing, takes a lot of time. You have to be sure that you have found as many bugs as you can, otherwise you will be doing updates regularly. If you release dozens of patches, people get the impression your software is buggy. Then the distribution thing, Rybka is not only sold as an UCI engine for download. But I agree, the multi-pv bug should have been fixed quick by some patch, because many people are using this feature for analysis.
> Pffff, for R4 bugfix we will wait more than 7 months, maybe 8, 9 .... ?
Which would still be better than having nothing in 9 months.
> just wait a bit and you will see you are wrong :)
> just wait a bit and you will see you are wrong
Yes, I really want to learn, that I was wrong in this point. It would be nice.
Can you give a hint about the time needed for 'wait a bit'?
Quap
Topic Rybka Support & Discussion / Rybka Discussion / Rybka is definitely strong, but it should be stronger!
1
2
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill