Topic Rybka Support & Discussion / Rybka Discussion / Rybka4 database of games where Rybka doesn't play good endga
![[pl]](/mwf/v12/flags/pl.png "Poland")
> There is a fallacy in your thinking. If I am happy with my present hardware, and I am, then subscribing to the remote version represents 100% incremental cost to me.
Obviously, there will be some sort of transition period. It's like Model-T customers who owned horses. Maybe some of them even waited for their horses to get old before upgrading.
> At the end of which, I have nothing of tangible value that I can point to and say "that's mine".
Yes, I realize this. When remote email (like yahoo, gmail, etc) first started to take hold, people also had this uneasy sense about not "having" their email on their computer. It will pass.
> It simply does not work for me and never could. But, never mind that, I still want accuracy in those things I can afford.
As you just said above, it's not about affording, it's about physically owning. Anyway, accuracy was also important for Rybka 4 and will remain important for Rybka 5, you don't have to worry that I'll just turn it into a toy.
Vas
![[us]](/mwf/v12/flags/us.png "United States")
I can't see paying for any service. The world just does not allow me be service-free, but to the extent that it does...I shall try. I want to be off the electrical grid and plan to make my own panels, solar water heater and solar air heater. Heck I want my own underground tank to catch all the rainwater to water the plants growing my own food. If they started charging to walk on the sidewalk I would look into digging a tunnel ;) I am just wired that way. I guess I am more against utilities...I hate monopolies. I have to pay for internet...grrr. I do pay for ICC, and have seriously considered a paying membership at chesstempo.com, but online position calculation, I can't imagine a position I would pay to see a better eval on. I want something that does not evaporate. I even did a huge genealogy thing without spending a dime even though it took 10 times more time.
Perhaps if anything someone recorded on the site they could use at any date even after their membership expired, it would make people like me consider it. And perhaps adding something to sweeten it like giving everyone who signs up permanent access to full 6-man tables and the use of the area to record games and analysis so long as the server stays up. I can see a lot of people trying the service for 6 months provided they can continue to access their findings, games and whatnot. And I suspect if it is done well, many people will stay as paying members. And it might also be good to allow former members the possibility to pay for one position to be checked to some depth for a fixed amount without any additional commitment.
Perhaps if anything someone recorded on the site they could use at any date even after their membership expired, it would make people like me consider it. And perhaps adding something to sweeten it like giving everyone who signs up permanent access to full 6-man tables and the use of the area to record games and analysis so long as the server stays up. I can see a lot of people trying the service for 6 months provided they can continue to access their findings, games and whatnot. And I suspect if it is done well, many people will stay as paying members. And it might also be good to allow former members the possibility to pay for one position to be checked to some depth for a fixed amount without any additional commitment.
Another thing you could try would be to allow people to allow you to use their processing power for other people's analysis and reduce their cost based on how much was used. This is a win-win. The users who have spent a ton on hardware don't have to pay much if anything perhaps get payed and the number of users you can support goes up as well as the quality of the analysis. You also have less hardware cost and less electric bills. Of course there is the small risk that someone will be able to reverse engineer the program but I think it is easier to safeguard with the software having to be on a computer connected to the internet and self-destruct time codes and other protections like an ID that makes tracking the source easy and having it only in RAM in a virtual RAM drive if necessary. The actual engine would have to be downloaded each time and is unusable after it disconnects from the server and is corrupted in the process. You went from no protections to hyper protection...I think there is a reasonably safe middle ground.
![[gb]](/mwf/v12/flags/gb.png "United Kingdom")
As you just said above, it's not about affording, it's about physically owning. Anyway, accuracy was also important for Rybka 4 and will remain important for Rybka 5, you don't have to worry that I'll just turn it into a toy.VR.
i like a number of others will never use cloud.i like to believe my hardware upgrades will counteract any temp improvements in Rybka.of course this costs a lot of bread but this is part of the hobby.no matter how strong the cloud is,i do not believe it will overule big hardware armed with a good program(R4 in this case) and human imaginination.you cannot let the cloud get to far ahead as the opposition is catching up fast.in a few months i would love to play at corr time controls against your cloud.maybe someone will give me a few games.maybe even yourself,if you wish to prove your case.i will be first to apologise if i am wrong,but would be prepared to play for a reasonable amount of money to pay for your time if you are correct.
regards Paul.
i like a number of others will never use cloud.i like to believe my hardware upgrades will counteract any temp improvements in Rybka.of course this costs a lot of bread but this is part of the hobby.no matter how strong the cloud is,i do not believe it will overule big hardware armed with a good program(R4 in this case) and human imaginination.you cannot let the cloud get to far ahead as the opposition is catching up fast.in a few months i would love to play at corr time controls against your cloud.maybe someone will give me a few games.maybe even yourself,if you wish to prove your case.i will be first to apologise if i am wrong,but would be prepared to play for a reasonable amount of money to pay for your time if you are correct.
regards Paul.
NNNN, l would love to watch; I am plus 1 with Paul. I want my own programs on my own systems.
>instead of assembling some complex hardware in your home as is currently the case.
Assembling hardware isn't really complex if you can follow instructions and have time to read, even a teenager can do it nowadays. And it is actually much cheaper since you can choose your hardware depending on your needs, eg. If you are an engineroom player probably a overclockable i7 is best suited but if you are a corr player probably a farm of cheap AMD 4-6 cores are probably the best choice. Plus it isn't hard to sell the hardware once you don't want it anymore.
"coolness" equates different things to Vas and yourself Nelson hehehe.
I should have used different terms. Substitute "elite users" for "expert users" and "expert users" for "normal users".
This reminds me of a problem that a Japanese condom maker was having twenty five years ago. They sold their condoms in three sizes, small, medium, and large, but nobody wanted to buy small condoms, so there were frequent failures. This problem was solved by renaming the categories: large, gigantic, and gargantuan. Mission accomplished!
This reminds me of a problem that a Japanese condom maker was having twenty five years ago. They sold their condoms in three sizes, small, medium, and large, but nobody wanted to buy small condoms, so there were frequent failures. This problem was solved by renaming the categories: large, gigantic, and gargantuan. Mission accomplished!
Ok, but in my case the terms are accurate.
Vas
Vas
Looks to me like the Lake Wobegon effect, where there is a claim that nearly all of a group is above average...
Ok, but it's not just random flattering - it's specializing.
Vas
Vas
since rybka 4 tends to steer games in to draws even when she is much better, how can i stop this besides draw never option? any thing else i can do?
>Decades - I'll just set it up and let it roll.
Does that mean you expect the state of computer chess 10 years from now the same as today? And what about the depreciation rate of your remote service, how many years do you expect it to be ahead in terms of accuracy and strength compared to the best available public engine?
![[kw]](/mwf/v12/flags/kw.png "Kuwait")
I will guess the remote version will have a dramatic difference in how it values rook end games. I was looking at an endgame book yesterday and noticed that almost 75% of the book was only about rook end games. Rook end games happen in at least 10% to 15% of chess games, so an improvement there is a great improvement in the overall chess engine. If you managed to get almost an entire endgame book of rook endgame heuristics in the remote package, that would be a great achievement. I have a feeling that Rybka 4 overly high evaluation in drawn rook endgames a pawn up is due to these missing heuristics .
.
but, while that makes it a pain in the ass for analysis purposes, does it effect elo much? i'd think (& hope) vas's primary goal in limiting strength was to reduce elo, not make rybka stupid. i'm sure you've seen many more r4 games than i have -- does she steer winning middle or late middlegames into drawing endgames because of this bug? if so -- couldn't simply lowering RES help performance quite a bit?
![[nl]](/mwf/v12/flags/nl.png "Netherlands")
> does she steer winning middle or late middlegames into drawing endgames because of this bug? if so -- couldn't simply lowering RES help performance quite a bit?
Yes and yes, a value of 60 was suggested.
yeah, but i think that may have been me, for analysis purposes -- or did someone test with 60??
I recall that Kapaun did it.
Actually I'm kind of surprised that this problem is worse in Rybka 4 than Rybka 3, that wasn't my intention (although it's certainly possible).
Vas
Vas
As far as I am concerned there are only two really nettlesome problems with Rybka 4. There may be other issues but they are trivial in comparison.
1. Rook endgames
2. Time management
If you improved these two issues in 4.1 there would be a significant ELO boost. I don't know why you don't commit to improve them unless it is your policy to hold back strength deliberately.
1. Rook endgames
2. Time management
If you improved these two issues in 4.1 there would be a significant ELO boost. I don't know why you don't commit to improve them unless it is your policy to hold back strength deliberately.
![[mx]](/mwf/v12/flags/mx.png "Mexico")
There are two ways to go about 2):
a) Find the optimal NM and MM values for every possible time control and make Rybka switch to the default on time control change, the parameters would become offsets from optimal.
b) Rewrite the thing.
B includes borrowing sophisticated time control from the clones, as a vengeance, I'd expect instant elo boost and Rybka looking at normal chess like she looks at FRC (where the time management seems to have been self-optimized without needing to do anything, as in, no better settings have been found regarding TM).
I think the testers that can play thousands of games haven't focused in what really matters, for instance, they haven't played Rybka 4 core against Rybka 8 core at time handicap, so that they should score the same. If Rybka 8core wins by a significant margin it means there's some serious problem with multicore implementation that fixes itself with enough cores, and this can't be ruled out yet. I wonder if that will get tested at all.
a) Find the optimal NM and MM values for every possible time control and make Rybka switch to the default on time control change, the parameters would become offsets from optimal.
b) Rewrite the thing.
B includes borrowing sophisticated time control from the clones, as a vengeance, I'd expect instant elo boost and Rybka looking at normal chess like she looks at FRC (where the time management seems to have been self-optimized without needing to do anything, as in, no better settings have been found regarding TM).
I think the testers that can play thousands of games haven't focused in what really matters, for instance, they haven't played Rybka 4 core against Rybka 8 core at time handicap, so that they should score the same. If Rybka 8core wins by a significant margin it means there's some serious problem with multicore implementation that fixes itself with enough cores, and this can't be ruled out yet. I wonder if that will get tested at all.
I think the best for Rybka 4.1 is to just optimize the parameter values for the mechanism we have and leave the rest for Rybka 5. Ditto for rook endgames - they will be better in Rybka 5 than in Rybka 4.
The long term is what matters here. For the past four or five years, I've put a really large number of hours into Rybka's playing strength. I probably won't be able to keep this up, and can't have a situation where the moment I slow down the business comes to a stop.
Vas
The long term is what matters here. For the past four or five years, I've put a really large number of hours into Rybka's playing strength. I probably won't be able to keep this up, and can't have a situation where the moment I slow down the business comes to a stop.
Vas
> I probably won't be able to keep this up, and can't have a situation where the moment I slow down the business comes to a stop.
A very honest remark. Can I make it simpler?
"Better for me to hold back all the best strength improvements for Rybka 5 otherwise there goes my cash-flow".
Actually this looks bad at first reading, but I urge people to consider what you would do in his position, taking everything into account. When is that baby due, Vas?
> but I urge people to consider what you would do in his position, taking everything into account.
I would have withheld the improvements of course, but instead of the remote model, I'd have gone with a monthly release, for 10 elo improvements, at full price. The people that only buy the engine every 18 months would have bought it anyway (one of the versions), it's the people that want to live at the edge of strength that are interesting, those that upgrade their hardware every few months and overclock their machines to the limit, those are going to buy the engine every single month.
Also, with so frequent releases the community finds and reports bugs every time, so they're instantly gone the next month, and I wouldn't have broken time management due to this. Even the pirate that releases pirate Rybka to the internet has to buy it each month.
I'd have done this since Rybka 2.3.2a, probably having to run the risk of skipping some months where 10 elo wasn't possible, damn the logistics. Vas is going for an even more frequent release model (I think he has planned weekly updates or just giving the people that go with remote Rybka the option of using the latest beta), but the engine does never belong to the customer.
This is a nightmare for users and doesn't really address the "barriers to entry for competitors" issue. The first time I go let's say two years without an improvement, I'd be toast.
Vas
Vas
> This is a nightmare for users
For users that opt in, the rest can just buy the engine every year and ignore the other releases.
> The first time I go let's say two years without an improvement, I'd be toast.
I think remote Rybka has the same problem. People predict that an engine will top Rybka 4 in 5 months (probably Stockfish), it's unclear how much time Remote Rybka will remain useful.
Anyway, I'm just saying how would I have done things if I had an engine 200 elo stronger than the next competitor in my hands, if Remote Rybka is a big success I'll be the first retracting all my negative comments about the model
I don't believe that if a program, let's say Stockfish, surpassed Rybka 4 sometime this year or next that there would be a collapse in demand for Remote Rybka.
First, you have to consider that demand for Remote Rybka will not be large at all and Vas's paying customer base for that product at least some period of time will number in the dozens or at most the low hundreds. In fact I feel pretty sure that Vas would view 200 paying customers as a success from a financial standpoint, because among those would be an elite handful of plutocrats that would be paying top dollar for the very best available strength.
Second, you have to consider that for many of the users that will emerge the idea is not to replace their PCs but rather to permit portability. If you are an OTB player or commentator at a tournament what a difference it would make to have a laptop with a wireless connection able to tap a high-end machine on demand.
Third, Remote Rybka offers the possibility to scale up to cluster strength, which Stockfish would not be able to do.
Fourth, it would be a fallacy to think of Remote Rybka being the same as Rybka 4. As we have been told in a variety of ways, it is much stronger. It will never be rated on CCRL/CEGT because it has no hardware peers.
First, you have to consider that demand for Remote Rybka will not be large at all and Vas's paying customer base for that product at least some period of time will number in the dozens or at most the low hundreds. In fact I feel pretty sure that Vas would view 200 paying customers as a success from a financial standpoint, because among those would be an elite handful of plutocrats that would be paying top dollar for the very best available strength.
Second, you have to consider that for many of the users that will emerge the idea is not to replace their PCs but rather to permit portability. If you are an OTB player or commentator at a tournament what a difference it would make to have a laptop with a wireless connection able to tap a high-end machine on demand.
Third, Remote Rybka offers the possibility to scale up to cluster strength, which Stockfish would not be able to do.
Fourth, it would be a fallacy to think of Remote Rybka being the same as Rybka 4. As we have been told in a variety of ways, it is much stronger. It will never be rated on CCRL/CEGT because it has no hardware peers.
I can see a number of folks that play games in the Correspondence Chess sub-forum using remote Rybka from time to time to get an edge.
And I can see them being disappointed with the results (like, maybe Remote Rybka will point to them something critical they were missing 1 out of 6 times, but they also paid for the other 5 where the Remote Rybka doesn't tell them anything they didn't already know).
1 out of 6 games is a huge improvement, are you kidding me? If you're playing for keeps 1 in 50 is a bargain.
> 1 out of 6 games is a huge improvement
I meant 1 out of 6 critical positions (the other 5 times Remote Rybka suggests a move that loses/doesn't win), I don't know how this translates to real games, I get critical positions on my games very often, the terrain is really hard and far away from perfect, if Remote Rybka failed me 5 out of 6 times, I wouldn't feel it was worth it.
What are your expectations on Remote Rybka and critical positions? I doubt they're lower than "I see a winning/saving move 1 out of 50 times", so you'd be expected to be a Remote Rybka customer.
I am definitely not a Remote Rybka customer. I am in the wrong tax bracket. I'm thinking I might consider it as a toy if I had an annual income of, say, a half-mil a year. Unfortunately I am not a bankster.
While I'm in the same boat with you I'm confused about your comment of it being a bargain.
Simple: if you are a really serious competitor and it is a real wonder-weapon then sure, it's a bargain. For such a person winning is the chief objective, not cost.
> I don't believe that if a program, let's say Stockfish, surpassed Rybka 4 sometime this year or next that there would be a collapse in demand for Remote Rybka.
I agree. Stockfish is a prime (legal) example of an engine that outperforms it's spotty evaluation by being a kickass beancounter. Unfortunately, for the vast majority of positions it's a sub par analysis engine. The eval is downright hostile to human evaluation and worse yet, it's extremely unstable and fluctuates constantly when analyzing complex positions. Trying to get it to remember a position that was analyzed only seconds prior is an exercise in frustration - sometimes it works, most of the time it doesn't. At the end of the day, Stockfish is an engine that's much more useful in the engine room than it is for any kind of serious analysis and I don't think that's going to change anytime soon, even if it somehow manages to overtake Rybka on the ratings lists. (Sounds familiar, doesn't it?)
![[ch]](/mwf/v12/flags/ch.png "Switzerland")
> The first time I go let's say two years without an improvement, I'd be toast.
>
This is life !
One can not rest for 2 Years unless you win at lotery.
regards
> When is that baby due, Vas?
In a week (meaning, any time)!
In computer chess, programmers tend to peter out after a decade or so. I'm kind of different so maybe it will be different for me, who knows!
Vas
Nah, you'll peter out too. It's not your fault; it's the cycle of life. (Maybe you could extend your productive programming life with regular testosterone pills and a rigorous workout regimen?) Once you have children running around it becomes almost impossible to concentrate for extended periods of time on anything.
Here's hoping the baby is born July 11. (My first-born's birthday and wedding day!)
Here's hoping the baby is born July 11. (My first-born's birthday and wedding day!)
> July 11
and my birthday!
I already wish you all the best for you and your family !
How about an 'accurate' commerical version with guidelines on how to set parameters for coolness and novelty? The remote version will be stronger as it develops over time -- this is fine and fair -- but there are a lot of serious, expert people who want to analyze at home with an accurate engine. Accuracy should always come first and be available to advanced users, including in the commercial version.
My two cents.
My two cents.
![[th]](/mwf/v12/flags/th.png "Thailand")
I must say I'm surprised that Rybka doesn't have basic KPK knowledge (e.g.
6k1/8/8/8/8/6K1/7P/8 w - -
), which means without tablebases it is not good for analysis of some positions.
I think Rybka 4 has thrown out a lot of code that can be covered with at least 5 piece EGTB's. These take up about 7GB of disk space, which today with the smallest hard disk you can buy at about 320GB is not so much. This has been done to keep search faster and thus increase overall strength.
This is fair enough - but perhaps there should be a warning on the packet!!
Something like "Warning - weak in endgames!"? I guess that would be up to ChessBase and Convekta, I wonder what they think about it!
Vas
Vas
Better would be: "Warning - No flowers in the windows!". Then when someone writes in to complain, you can send them this classic reply:
First, you put down the foundation. Then the structure. Then the walls. Finally, you put flowers in the windows.
If you like flowers, you'll just have to be a little patient.
First, you put down the foundation. Then the structure. Then the walls. Finally, you put flowers in the windows.
If you like flowers, you'll just have to be a little patient.
This sounds like a great marketing idea. Maybe someone from ChessBase or Convekta is reading here!
Vas
Vas
![[de]](/mwf/v12/flags/de.png "Germany")
I asked Victor to add at least the 3 and 4 men to the Rybka Aquarium DVD, but apparently he either forgot it or has secret reasons not to do that :)
i played rybka 4 in an endgame where it considered the position much better for her, it was against one of the character gms on the game chess master for ps2 and so far rybka 4 on 20 ply and 24 ply can only draw yet rybka 3 wins both times. this happened more then once on different end games with similar results. this is very disappointing!
i suggest all who are upset to respond and hopefully vas will release a rybka 4.1 or 5 with in the next 6 months.
rybka 4 is a major let down. vas, dont do this to people who believe in you just because you have money.
chess is the only joy in many peoples lives..........
i suggest all who are upset to respond and hopefully vas will release a rybka 4.1 or 5 with in the next 6 months.
rybka 4 is a major let down. vas, dont do this to people who believe in you just because you have money.
chess is the only joy in many peoples lives..........
Actually private Rybka version also doesn't understand this. If I ever add this knowledge, it will probably be only to the public version (and not the private versions), ironically enough.
Vas
Vas
Topic Rybka Support & Discussion / Rybka Discussion / Rybka4 database of games where Rybka doesn't play good endga
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill