I completely agree. I have seen many bugs fixed here (although there is still no adequate keyboard interface). My harshness was directed to those who cannot see a bug due to their defensiveness, not toward the actual developers. I regret my lack of care with my language.
Thank you. I have just posted a bug report there.
@ImprobableBlob the decision makes so much sense to me. I think there is a 2 or 3 dan difference between my blitz level and my level in correspondence games, and thanks to the OGS system I can play interesting games at any speed.
What OGS is doing is very common in sports and other competitive activities. I have never seen a runners ranking that tries to aggregate all distances from sprint to marathon.
One solution that I have seen proposed is that the seperate rankings are all affected by games
e.g. you win a live ranked game, your live and overall rankings gets 100% of the points for winning, your blitz and corrospondence get 75% or something.
This solves the problem of having a 10k with a blitz rank of 20k while still keeping in mind that people play at different strengths depending on the time setting.
This makes things more complex and complex things are bug-prone and harder to maintain. It might be worth it, but I have my doubts. Those runners and other professionals have to participate in all kinds of qualifying tournaments, while we are a bunch of amateurs who play whenever whim takes us. It seems to me that for such player pool it’s hard to maintain even one reliable rating, and we have 4. But I might be wrong about this, that’s why seeing the stats would help.
While this proposal for mixing the separate rankings sounds better than what we have, I agree that it is complex. It is also ad hoc and unproven.
Why not research what has already been done in the field of go ranking, and implement what is simple and effective? Would it be so bad for OGS to copy the algorithm used by another online go system?
I haven’t done the research, and don’t have the time to do it, but it is the obvious path for any developer who wants to fix the current buggy system. We shouldn’t be experimenting with our players. They deserve some system of go ranking that has proven itself as reasonable in practice.
And if we insist on experimenting, why not implement the system that someone already posted, which was based on their advanced research into go ranking? I don’t recall where that proposal is, and I can’t find it using search terms, but it is right here on one of these OGS fora. It was based on interesting mathematics.
Okay. I’m going to go out on a limb and claim that rank has nothing to do with the type of game. Blitz or correspondence, a 9 dan player is going to be great and a 24 kyu player is going to be awful. I say, the grand experiment of calculating rank separately by time limit has had its day and needs to be retired. Let all games contribute to a single rank for each player. Either use ELO, or actually look at some board positions as someone suggested.
Some additional hypotheses:
Local moves can be sufficient to predict rank. While it is true that advanced players are skilled in joseki, ignoring the opening and end game is reasonable when predicting rank over the entire ranking spectrum.
Even focusing on life and death positions is sufficient. Beginners can’t even recognize an unconidtionally live or dead group. Advanced players know how to live and kill without mistake. If we want to be innovative, and analyze board positions, what positions could be better than those patterns involved in life and death?
But, again, if board analysis seems to difficult to do, then just go back to a unified rating system, like other online go websites use.
Let’s get rid of the bugs now, please.
There are not many go sites to copy from. Most other sites have a strong bias towards either live or correspondence games, while OGS tries to offer both. There are chess sites that are similar to OGS, but chess ranking, especially for correspondence chess, is strongly affected by the wide availability of GM+ level computer engines. Some chess sites do not even bother trying to rank correspondence players.
It would not be bad, it would just be less effective than what OGS is currently doing.
The current OGS system is actually closer to what official pro and amateur federations are doing. In such systems, you never mix “serious” games and blitz games in the ratings. As a reference, here are the American Go Association’s qualifications for rated games: http://www.usgo.org/qualifications-rated-games
In your case, the opponent has an invalid blitz rating because he didn’t play enough blitz games. Not having enough games will be an issue no matter the algorithm. What OGS could do is add a warning in such cases.
And I was claiming the opposite, based on my personal experience. Rating is more subtle than just great vs awful, we are talking about a 2 or 3 level difference.
If you are looking for simple and effective: no program is able to do this today.[quote=“david265, post:27, topic:9018”]
Even focusing on life and death positions is sufficient.
Unfortunately, it’s not the case. Or else go programs would be much stronger.
What’s the point of a new online server, if you just repeat the errors of your predecessors?
That would disallow correct ranking and disallow correct automatic handicaps until a person played “enough” games in a particular time category. It would also not solve the problems caused by the “overall” rank. What do you have against actually fixing the bug?
I propose that all new players be given a rank of “provisional” for each time category. There would be no “overall” category. Anyone who has played a certain number of games (say, 10) in a time category should be given a rank for that time category, calculated on board position analysis (best) or ELO (worst).
This proposal is in response to your proposal. Actually, I would like to see a rank that ignores time categories entirely.
I just played game 6266831 with player shoostr. This is another example of the bug. I think the bug is very common in 9x9 games.
shoostr is shown as 19k. But he claims to be 16k! When looking at his information, his rank is given as 15k.
I should have had a handicap.
shoostr seems to be poor at correspondence games, maybe because he never played any.
Does OGS require players to play all time categories? That makes no sense.
Come on, guys, this is a serious bug.
And what’s the point of a new online server that insists on having a serious bug in its unique ranking system?
For me the same, I just played to a 16k (I am around 12k) and had no chance at all. After the game I looked to his profil and he was around 6/7k
I hate that the system is missleading me!
I know its part of the “experimental” system, but it is so frustrating that I didn’t get all information before I accept a challenge. Sometimes I am just not in the mood for a stonger player like that.
(yes, I know i could check my opponents profil always before accepting a game, how convienent
What is wrong with showing all information in the gamegraph popup or at least on the game screen
make finaly some conclucisons from the experimental phase and make alterings to improving the system deply?
Does anyone (!?!) seriously assume, that blitz, live and correspondance ranks are soooo different, that a gap over 10 stones is realistic (only because one didn’t played a specific timing very often (or “forgot” to ranking it))
So part of the cause of this bug (a few people have admitted it is a bug) is that 9x9 games with a limit of 10 minutes are considered to be blitz. And other part is that everyone is required to play correspondence games.
When are the bugs going to be fixed? And why do I have to work so hard to say the obvious?
I like your posting very much. You present evidence in a compelling way.
sure there are qualities that help in correspondence settings such as tenacity and thoroughness and others that are required for blitz games such as intuition, focus and a steady hand. but i dont agree that it is worth the trouble.
one rank should really be enough.
this is exactly the point. because as SanDiego said any algorithm requires a fair number of games to be accurate… then why (and how?) are we supposed to achieve this in 3 categories?
Game https://online-go.com/game/6267210 is another example. The player gets two handicap stones because he/she is 23k compared to my 17k. However, his real rank is 15k, same as my 15k.
It was wrong for him to get two handicap stones (I did not have automatic handicap mode set).
Again, this is a bug!
That seems reasonable. Except, if you had bothered to read this thread, there are actually bugs due to the incorrect rank. For example, handicap stones can be automatically generated.
an overall rank would still be closer to a players “actual rank” than a rank lagging 5 to 10 grades behind as in @sTan’s example. plus this is a frequent problem.
also a bad game can happen for any number of reasons… it can also be just that, a bad game. 2-3 stones of deviation in performance is not a big deal and if you are inconsistent your rank will just be that much lower.
you are of course correct about that part :), but it shouldnt keep us from wanting to change things.