Life In 19x19
http://lifein19x19.com/

A new server is being developed: Kaya.gs
http://lifein19x19.com/viewtopic.php?f=60&t=5258		 Page 5 of 16
Author:  entropi [ Tue Aug 23, 2011 7:12 am ]
Post subject:  Re: A new server is being developed: Kaya.gs
daniel_the_smith wrote:
entropi wrote:
Seriously, the reason why I would like such a system is because I believe it would reduce the importance of counting and positional judgement, which I find the most (or only) boring aspect of Go.


I hear there's lots of old guys in Korea that will play exactly that (bangneki) with you, although that might be expensive. But I suspect you'd find it doesn't actually reduce the necessity of counting or positional judgment, just change the ways you use them. (Do I go for an almost certain 10 point win? or try for a 40 point win that might end in disaster? Conversely: Do I try to stay at a 10 point loss? Or do I try this crazy thing that might let me win, or might turn my 10 point loss into a 50 point loss?)


But these decisions rely more on intuition (or let's say a blurry reading) than on positional judgement.

In a conventional win/lose game, if you have a certain 10 points lead, you don't need to take even the smallest risk. Since a 0.5 win is the same as 50 points win, your decision whether to take the risk or not is highly affected by the positional judgement.

But in the game I propose (bangneki without money) you don't need to care about who is leading on the overall board. Your decision is based on "can I kill that group or not".
If you can, you are leading by 40 points instead of 10. If you cannot you don't lose everything, but you are behind by 20 points (still better than losing a conventional game).

No need to count before taking the decision of trying to kill or not.
Author:  tapir [ Tue Aug 23, 2011 8:07 am ]
Post subject:  Re: A new server is being developed: Kaya.gs
entropi wrote:
But in the game I propose (bangneki without money) you don't need to care about who is leading on the overall board. Your decision is based on "can I kill that group or not".
If you can, you are leading by 40 points instead of 10. If you cannot you don't lose everything, but you are behind by 20 points (still better than losing a conventional game).

No need to count before taking the decision of trying to kill or not.


You might be interested in http://senseis.xmp.net/?HahnPointingSystem. There is one tournament in Germany, that is (often/always?) played w/ this system.
Author:  daniel_the_smith [ Tue Aug 23, 2011 8:33 am ]
Post subject:  Re: A new server is being developed: Kaya.gs
entropi wrote:
But in the game I propose (bangneki without money) you don't need to care about who is leading on the overall board. Your decision is based on "can I kill that group or not".
If you can, you are leading by 40 points instead of 10. If you cannot you don't lose everything, but you are behind by 20 points (still better than losing a conventional game).

No need to count before taking the decision of trying to kill or not.


I don't completely agree, but perhaps we should stop the threadjacking now :)

+40 > +10 > -20. Trying and failing costs 30 points, trying and succeeding gains 30 points. So, with those numbers if you think you have > 50% chance of killing, you should go ahead and try to kill.

However, usually the payoff is not so even. Sometimes it's +90/+10/+8, in which case you should basically always try to kill. But sometimes it's +20/+10/-40-- failing costs 5 times as much as succeeding, so you need to be at least 80% sure you can kill or the attempt will lose points on average.

To make a good decision, you have to figure out how much a success gains, how much a failure costs, and how likely you are to succeed. The first two require counting (possibly a lot of counting), and the latter requires that you be well calibrated (do you in fact kill 8 out of 10 groups that you're 80% sure you can kill?). Oh, and figuring out how much a failure costs generally requires positional judgment. Sometimes success does, too-- sometimes, keeping a group dead can cost nearly as much as killing it gained.

I think this would actually lead to *more* counting for me, because I'd have to estimate the score both ways for many more decisions than I do currently (currently I only do that in very close games).

Finally, there will still be times when you vastly prefer a small, safe win/loss over a drastic win. E.g., you could be playing the last round of a hahn style tournament and as long as you don't lose big, there's no way for the guy in 2nd place to catch up.
Author:  Kaya.gs [ Tue Aug 23, 2011 9:34 am ]
Post subject:  Re: A new server is being developed: Kaya.gs
I have played hahn's tournament and this model i like a lot. Regardless of what we think of the effects of such game, its simply not Go. Go is about winning regardless of the difference :).


We may make it able to play like that anyway..but mixing it in with the rating system is dangerous.
I think the most complex problem presented so far is how to handle the rating system. I will make a thread soon about it.
Author:  tapir [ Tue Aug 23, 2011 10:54 am ]
Post subject:  Re: A new server is being developed: Kaya.gs
Kaya.gs wrote:
I have played hahn's tournament and this model i like a lot. Regardless of what we think of the effects of such game, its simply not Go. Go is about winning regardless of the difference :).


We may make it able to play like that anyway..but mixing it in with the rating system is dangerous.
I think the most complex problem presented so far is how to handle the rating system. I will make a thread soon about it.


Go on. But don't let the rating nerds convince you a rating system is of major importance. In fact, it would be fun to keep ratings in several rating systems simultaneously, and everyone sees the ranks in his preferred system. Guess, this would cure rating system preferences very fast.
Author:  malweth [ Tue Aug 23, 2011 11:53 am ]
Post subject:  Re: A new server is being developed: Kaya.gs
tapir wrote:
Go on. But don't let the rating nerds convince you a rating system is of major importance. In fact, it would be fun to keep ratings in several rating systems simultaneously, and everyone sees the ranks in his preferred system. Guess, this would cure rating system preferences very fast.


I'm also interested in the API that will allow clubs to interact with the server. Could this mean that clubs can implement a ratings overlay (e.g. number of handicap stones between club players). Could this also mean that if the AGA (or other national/regional association) matched up usernames on the server to actual members that they could provide their own ratings?

These are the sorts of things I'd like to see... an open format that, while the core server isn't affected by the outside, the user's experience is affected by the plug-ins or associations they choose.
Author:  Kaya.gs [ Tue Aug 23, 2011 1:01 pm ]
Post subject:  Re: A new server is being developed: Kaya.gs
malweth wrote:
tapir wrote:
Go on. But don't let the rating nerds convince you a rating system is of major importance. In fact, it would be fun to keep ratings in several rating systems simultaneously, and everyone sees the ranks in his preferred system. Guess, this would cure rating system preferences very fast.


I'm also interested in the API that will allow clubs to interact with the server. Could this mean that clubs can implement a ratings overlay (e.g. number of handicap stones between club players). Could this also mean that if the AGA (or other national/regional association) matched up usernames on the server to actual members that they could provide their own ratings?

These are the sorts of things I'd like to see... an open format that, while the core server isn't affected by the outside, the user's experience is affected by the plug-ins or associations they choose.


It will probably be possible to do "ratings overlay" if its desired, but i dont know if its a good idea. The "have every rating everyone wants" idea is not so good. Ratings get better and more accurate the more people they have.
If you have 10 systems u dont know how to keep track off, then they will all not work.
Ratings is turning out to be a more complex subject than i thought: there doesnt seem to be a system that makes everyone happy :).


But, for example, KGS teaching ladder could have their own 'karma'. So each time you play there you get a +1, and each time you recieve a game u do a -1 or the sort. So there can be some internal tools of tracking results and such.
Author:  HermanHiddema [ Tue Aug 23, 2011 2:01 pm ]
Post subject:  Re: A new server is being developed: Kaya.gs
Kaya.gs wrote:
I think the most complex problem presented so far is how to handle the rating system. I will make a thread soon about it.


Just pick a well defined modern rating system and implement it. Glicko, Sonas, WHR, whatever. Trying to design your own rating system is about as smart as trying to design your own crypto system.

You can't please everyone anyway, so I would focus on more important things.
Author:  quantumf [ Wed Aug 24, 2011 12:35 am ]
Post subject:  Re: A new server is being developed: Kaya.gs
HermanHiddema wrote:
Just pick a well defined modern rating system and implement it. Glicko, Sonas, WHR, whatever. Trying to design your own rating system is about as smart as trying to design your own crypto system.

You can't please everyone anyway, so I would focus on more important things.


Absolutely. You can even change the rating system later. Comments about rating systems, clock systems, escaping rules, etc are just tiny irrelevant noise, compared to what I believe you want to achieve, which is a completely new and fundamentally different user experience.

If your system only achieves a better rating system than KGS (if that's even possible) then I don't see why I would move.
Author:  danielm [ Wed Aug 24, 2011 1:06 am ]
Post subject:  Re: A new server is being developed: Kaya.gs
It's not tiny and irrelevant, as you can't have a superior user experience without paying attention to the fundamental details. The suggestion not to reinvent the wheel is a good one though. There is no reason to invent a new rating system, unless you happened to have studied that subject thoroughly and felt strongly that even the best existing methods could be improved.
Author:  entropi [ Wed Aug 24, 2011 3:29 am ]
Post subject:  Re: A new server is being developed: Kaya.gs
@daniel_the_smith : I see your point. One cannot get rid of counting completely, but I still do believe that a bangneki-no-money system reduces (not eliminates) the importance of positional judgement on the overall board.

Apart from counting, another advantage of bangneki-no-money system is that the games are more likely to be exciting until the very last move.

By the way, I am aware of the Hahn system but never tried it. But the system I had in mind is different than Hahn system. Hahn system rewards big wins, but still the term "win" is still more important than "big". What I had in mind is a different system.

For example I register to the server, I initially have 10,000 points. I play a game and end the game ahead by 30 points (I refrain from using the word "win"), I end up with 10,030 points and my opponent with 9970 points. New people joining the system, brings new funds. A difference of X hundred points means a handicap stone, etc etc etc.

I am sure anyone can come up with some objections on why such a system would not work. But if there is the desire one can make it work.

But what I understood from the reactions on my proposal, there is no such desire anyway :sad:

So, forget about all these and let me have a look at my dragongoserver games :)
Author:  hyperpape [ Wed Aug 24, 2011 4:39 am ]
Post subject:  Re: A new server is being developed: Kaya.gs
entropi wrote:
For example I register to the server, I initially have 10,000 points. I play a game and end the game ahead by 30 points (I refrain from using the word "win"), I end up with 10,030 points and my opponent with 9970 points. New people joining the system, brings new funds. A difference of X hundred points means a handicap stone, etc etc etc.

I am sure anyone can come up with some objections on why such a system would not work. But if there is the desire one can make it work.
Yes. Massive inflation. Maybe you could fix that by making your starting funds change over time.
Author:  entropi [ Wed Aug 24, 2011 5:31 am ]
Post subject:  Re: A new server is being developed: Kaya.gs
hyperpape wrote:
entropi wrote:
For example I register to the server, I initially have 10,000 points. I play a game and end the game ahead by 30 points (I refrain from using the word "win"), I end up with 10,030 points and my opponent with 9970 points. New people joining the system, brings new funds. A difference of X hundred points means a handicap stone, etc etc etc.

I am sure anyone can come up with some objections on why such a system would not work. But if there is the desire one can make it work.
Yes. Massive inflation. Maybe you could fix that by making your starting funds change over time.


Is this a problem? I don't see why it should create a problem but if it does, an alternative solution could be defining several leagues with limited number of players (instead of defining one league with unlimited number of players).

A further alternative could be defining a fix amount of total funds and a sharing policy.

Or, or, orrrrr orrrrrr what about putting real money in it, like a gambling site????

Another, maybe even bigger problem could be motivating people to play with stronger players. Several solutions are possible but need to be tried out.

But anyway the real big problem of this system is that it doesn't seem that there is a desire for something like that and I am way too busy to make some real marketing :) Just some ideas some potential for brain teasing :)
Author:  uPWarrior [ Wed Aug 24, 2011 10:39 am ]
Post subject:  Re: A new server is being developed: Kaya.gs
Kaya.gs wrote:
The "have every rating everyone wants" idea is not so good. Ratings get better and more accurate the more people they have.

As far as I understood, each game would count towards every rating system, unless it was "free", of course.
This seems a clever idea to me; being able to track your rating in several systems would certainly end any kind of criticism. Ideally, with a sufficient amount of games, they should converge.
Auto-match could work similarly to KGS (match me with anyone within X stones in any of these rating systems: [selection]).
Author:  daniel_the_smith [ Wed Aug 24, 2011 11:03 am ]
Post subject:  Re: A new server is being developed: Kaya.gs
uPWarrior wrote:
...
Auto-match could work similarly to KGS (match me with anyone within X stones in any of these rating systems: [selection]).


Simpler (and I think better): You just have a global option that determines which rating system you'll see.
Author:  Chew Terr [ Wed Aug 24, 2011 11:07 am ]
Post subject:  Re: A new server is being developed: Kaya.gs
uPWarrior wrote:
Kaya.gs wrote:
The "have every rating everyone wants" idea is not so good. Ratings get better and more accurate the more people they have.

As far as I understood, each game would count towards every rating system, unless it was "free", of course.
This seems a clever idea to me; being able to track your rating in several systems would certainly end any kind of criticism. Ideally, with a sufficient amount of games, they should converge.
Auto-match could work similarly to KGS (match me with anyone within X stones in any of these rating systems: [selection]).


The biggest problem I see with this is that it further splits an already smallish population of players. If I'm specifically looking to play people with about 3k AGA ratings, I won't necessarily be able to figure out a good match versus BGA, KGS, or whatever ratings. It is essentially more accurate for everything to be judged by the same scale than to have to manually convert. And if people of different systems don't play each other, it will be that much harder to get a game. Like with online console games, any online play tends to die after X amount of time (where X is how popular the game is), because there are just too few players for matchmaking to handle capably.

On the other hand, I do like what was mentioned about having additional (opt-in) scales. In addition to the GTL example, it would be cool if ASR data was integrated into the server itself. That sort of thing is exactly what this server effort seems to be trying to do, and I am interested to see how it looks as it gets closer.
Author:  Laman [ Wed Aug 24, 2011 2:21 pm ]
Post subject:  Re: A new server is being developed: Kaya.gs
Chew Terr wrote:
The biggest problem I see with this is that it further splits an already smallish population of players. If I'm specifically looking to play people with about 3k AGA ratings, I won't necessarily be able to figure out a good match versus BGA, KGS, or whatever ratings. It is essentially more accurate for everything to be judged by the same scale than to have to manually convert. And if people of different systems don't play each other, it will be that much harder to get a game. Like with online console games, any online play tends to die after X amount of time (where X is how popular the game is), because there are just too few players for matchmaking to handle capably.

just to correct probable misunderstanding, this wouldn't split the player population. simply someone would watch his and opponents' ELO ratings, someone Glicko or WHR or whatever and everyone would have a rank in all the systems. it looks interesting, at least for studying the rating systems, but it has not much meaning other than academical and in my opinion it has a low priority for starting a new server, not to mention possible confusion by several rankings used simultanously

for whatever ratings finally used, i would suggest to include option of a game with proper handicap (as it is called at DGS), ie. handicap and komi set to achieve very close to 50% winning expectance. it would make easier to get a new rank, because with traditional handicaps (like at KGS) your wins counts less and less (and loses more and more) as you are approaching the upper border of your rank, which can be very annoying and stressful and makes it harder to achieve your proper rank
Author:  Chew Terr [ Wed Aug 24, 2011 2:27 pm ]
Post subject:  Re: A new server is being developed: Kaya.gs
Laman wrote:
just to correct probable misunderstanding

Ah, now I get what you mean. Makes sense, though as you say, it's lower-priority for a new server than lots of other things.
Author:  snorri [ Wed Aug 24, 2011 2:31 pm ]
Post subject:  Re: A new server is being developed: Kaya.gs
Kaya.gs wrote:
One thing im talking with my partner about the rating system which we will try is the following: "14 victories gives you a rank up". These is something that happens with some club-rating systems like in the Nihon-kiin.
This means that we would always adjust the strength difference of any game, so if a weak 5d plays a strong 5d, there will a difference in komi. It is a more granular way to handle the rating system, and also a very predictable one. Each game you play is basically worth the same.


You have some pretty good ideas, so I think I'll contribute despite being thrown off a bit by Patricio's musketeer-style facial hair. (But hey, if we all liked the same things the world would be a pretty boring place. We're lucky enough that we all like go. :) Truthfully, you had me at Fischer time. Finally, a real-time go server with that will be a good thing.

I think you're right about the advantages and disadvantages of the KGS rating system, but getting the right solution is a bit of a challenge. Maybe only experience will show whether your system is better. As for incremental komi, I'm not fan. When IGS switched to the +/no plus system with two ratings per stone, it got more transparent (and theoretically more correct), but I would often forget that I might be giving or taking reverse komi if I wasn't paying attention to who had a '+'. Komi that might appear to the player to be any random half-integer between -7.5 and 7.5 would be even harder to think about. So if you are going to do this, two suggestions:

1. Allow players to choose to play other komi and handicap than the defaults chosen based on rating difference.
2. Make the komi and rules very visible while playing the game. Having to select a menu item is too much indirection to check these values.
3. You may want to research whether this granularity really buys you any accuracy. For example, if you are allowing rated games with more than 1 stone handicap and you are also using small komi adjustments, it would appear that you are claiming to know precise komi equivalents for handicap stones. Although there is a little data on this, the real values are not known. It's fake precision, like quoting someone's height to 6 decimal places. Even if you do gain something, you lose something else in that new players will be confused and say, "What? 2.5 points reverse komi? Who does that?"

Good luck!
Author:  Tami [ Wed Aug 24, 2011 6:43 pm ]
Post subject:  Re: A new server is being developed: Kaya.gs
snorri wrote:
You have some pretty good ideas, so I think I'll contribute despite being thrown off a bit by Patricio's musketeer-style facial hair.


LOL, I like Patricio's style! He looks cool.

Anyway, with ranking systems I think the easiest way to please most people is to make it so that

1) It doesn't change too quickly
2) It doesn't change too slowly
3) Isn't affected by somebody else's performances
4) You don't ruin your rating for the next six months by having a bad day

Let's take 6 games in a row as a kind of starting point and say 10 notches in each rank. If you think 6 games in a row justifies moving up or down, then let that be the criterion for promotion or demotion. After that, make it about percentages scored in last ten games. My suggestion would be

1) 8 wins out of ten, rank up
2) 7 wins out of ten, go up two notches within rank
3) 6 wins out of ten, go up 0.5 of a notch within rank
4) 5 wins out of ten - sorry, no change!!

This way, if people hit good form, they can get a tangible reward for it. But only being able to sustain the new level would result in a stable rank.

If you improve over time (or decline), then your rating within each rank would gradually creep up or down.

Anyway, regardless of what WMS has to say, the KGS system is obviously too stable. But, maybe better that than some mickey mouse system that was too much in flux. Definitely please don't make rank tied to other people's performances - that's just unfair - and please don't make a player's bad days hang over them for the next six months like some kind of criminal record.
Page 5 of 16 All times are UTC - 8 hours [ DST ]
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/