Having them play lots of games against each other would do the trick, wouldn’t it?
Yeah, that would probably work toward fixing their relative rankings, but it might also have some drawbacks:
- It doesn’t address the issue of random players causing needless swings in the bot ratings in the first place.
- I imagine that the bots might play more consistently than humans. By that, I mean that a match-up between two bots that are only slightly different in strength (when compared via their performance against humans) might wind up being much more one-sided (in terms of winning percentage) than expected. The stronger of two humans may win more often, but there is still the variability inherent to human nature. The Elo system is designed to take that into account when estimating the expected win rate. Playing bots against each other might quickly correct an inverted relative rating, but it might end up driving their rankings further apart than they actually should be.
- It seems like a waste of CPU cycles.
2 Likes
These are all very good points.
Perhaps there could be some sort of automatically detecting big swings in a bot’s ranking. If, say, a 20k resigns to RandomBot because something suddenly comes up that means they can’t complete the game, then this would cause a big spike in RandomBot’s ranking. If OGS detects that, then Randy could play a few games against other bots to settle things down again.
That is true. One way of dealing with this is with handicaps. If Bot A beats Bot B most of the time with 3 stones handicap, but almost never with only two stones, then the bots are between 2 and 3 ranks apart. If they play even games all the time then, yes, the ranking difference will be inflated because of the consistency of computer players. Besides, a little bit of “forced separation” in the system would not be a bad thing because the major drawback of the previous rating system was that the ranks were all bunching up.
So far so good. But there are now some challenges on the open games list that don’t make sense.
(For example, players who challenged at 29k and are now 15k after adjustment, but I can’t accept them as 12k because the rank difference is supposedly more than nine stones as it’s measured against the time they put the challenge up.)
I noticed that too. People will figure out what’s happening eventually.
Ah shoot, yeah hopefully they’ll cancel and recreate those, if not i’ll try and remember to clear those out this evening
Dumb question: difference between rating and ranking? Specifically, what is Rating?
Rating and rank are highly related. Rank refers to the integer ranks used by the kyu/dan system (e.g., 10 kyu, 5 dan, etc. are ranks). Rating is a finer grained number that the system actually works with in adjusting rating/rank. Winning or losing a game might not change your rank, but it will likely change your rating by at least a few points. Every hundred rating points corresponds to one rank, and rank is simply calculated from which 100-point interval a player’s rating is currently in.
For example, on your profile, your current rating is 399 and your rank is 18 kyu, which is given for anyone with a rating between 300-400. Hence, according to the rating system used by the site, you are right on the edge between 17k and 18k. If you lose you next game, your rating will go down, but your rank won’t actually change unless your rating falls by more than 99 points (which is probably unlikely). If you win your next game, your rating will likely go up by at least a few points and hence push you up at least one rank.
I am a 20k provisional. Would like to be accurately ranked, up or down, don’t really care that much. Right now am playing the bots (unranked games)–reasons for that. Maybe not good ones but reasons. So, here’s the question: What k players should I play to get an accurate ranking? Tho I’m getting stronger (at least against the bots) playing a 10k I’d be massacred. But playing and winning (and losing, perhaps), say, against a 20k doesn’t say much, which keeps me at 20 k which is okay. My preference now is to continue playing against the bots, ranked games, and let the chips fall where they may. Hope this isn’t a silly question. I guess bottom line is recommendation for what kyu players I should play to get an accurate (not necessarily better) ranking. Sorry for the long message. Thank you. BTW: I love OGS. Today am a supporter.
1 Like
Hi, @Pengyou. The best way to reach your accurate rank in the system is playing against players having the same rank you have right now. So, if you are 20k, just play against 20k players. Let’s say you reach 19k, then start playing 19k players as many games as you can, and so on. This way you’ll get an accurate rank eventually. I hope this answers your question.
And by the way, thank you very much for supporting OGS! 
2 Likes
Wasn’t the answer I was hoping for but okay and makes sense. Was hoping that bots would do it. Anyway, I think you guys–well, all members–are rather impressive in so many ways. Which is why I decided to commit. Thank you for the answer and I will follow your advice :<( (Prefer bots. Less judgmental I think)
@Pengyou, if you want to get out of provisional by playing bots, there is nothing wrong with it. On the old, correspondence-only, OGS playing against bots was the fastest way to get a solid rank and many people were doing just that. On a real-time server playing people can be both faster and more fun, but if your preferences swing toward robots, well, why not? 
As for the accuracy of your rating, do not worry too much. Nobody expects your starting rating to reflect your strength precisely. DDK ratings on servers are inherently inaccurate simply because many (most?) players on this level do not play consistently enough for their strength to be measured with accuracy. Once you’re out of provisional, join beginners’ tournaments, play a lot of games against similarly-ranked players and eventually you’ll have a rating as reliable as it can get.
The best way to correct your initial rank, if you messed up during sign-up, is to ask a moderator to fix it. Just tell them how strong you are and you will be set accordingly, with provisional status of course.
At least that is the advice I used to give - I don’t know why @mlopezviedma didn’t mention it. They do it for accidental 6-dans, and there is no reason that I can see why you should go through the boring chore of bashing helpless 20-kyus until you rank up. That is antithetical to the purpose of playing Go, a game for fun and entertainment 
Edit: ah, I should have read more carefully. The problem is that you don’t know your rank? Then disregard what I’ve said. ^^
2 Likes
Hi there!
I’m a huge fan of OGS and regulary play here for more then half a year.
After AlphaGo vs Lee Sedol event there seems to be an increased flow of games and users which is good for site and players as well but there’re a lot of those who set thier rank not appropriate to thier skill which is misleading and frustrating.
That statement is especially true for dan level players.
I have two suggestions:
-
Make a check option in game creation table to play only against those with solid rank (like option on IGS)
not all of us in mood to evaluate new players, it’s ok to do it from time to time but, please, let us choose when to do it. -
Let bots into the server which can help in numerous ways:
2.1 be available to play against which would be helpful for those with ? mark rank to if no one available or wanting to play against “? rank” and player wants to get a solid one (in that way suggestion 1. won’t be a problem);
2.2. compare ranks on different servers if bot is multi server one;
2.3. computer go is on the rise and it’s increased number of players so it should be good in general.
So, what do you think?
4 Likes
2.4 Bots have another great use case! Because they have exactly the same strength, no matter when or where they play, they can effectively be used to stabilize the ELO rating system. This is how: say gnugo is very well known to play at 9k across all servers. Whenever a game ends between gnugo and a human player, gnugo’s ELO rating would stay the same, while the humans ELO rating would change just as expected. That way gnugo is fixed at 9k and ELO points can flow in and out of the system in a very controlled way. No need to adjust people’s rating manually. You can do this for bots, but also for players that have reached a high level of play and don’t change ranks quickly anymore. Also, you can use this to smoothly adapt one ELO system to another rating system, say KGS or AGA. I’ve been thinking about this quite a long time already and wondering why this is not already done
A very similar thing is used already now on OGS for new players: their first 4 rated games grant them elo points to find their correct rank, but not so for their opponents. This is done to not punish the opponent for playing someone with unknown rank. However, I would increase the number of “startup” games to 10, because it’s a great way to prevent new players to suck too many elo points out of the system - which is the major cause for ELO deflation.
ELO inflation and deflation is something we don’t want to have, because then a 10k today can not be compared to a 10k one year ago. It causes the rating to drift over the time.
1 Like
Let me introduce some general scepticism to ideas that involve bots and ranking.
While their play style and depth is very consistent, the same is not true of their human opponents. There are players who play against their preferred software again and again. They become very skilled at exploiting the particular bot’s weaknesses. The bot does not learn and always makes the same mistakes.
When these players then go up against a human opponent of the same rank, they get destroyed, even though the opponent may be an equal match for the bot.
On the topic of inaccurate ranks: here on OGS we have great processes to fix ranks. You can choose your own starting point. If you’ve been inactive, you can ask a mod to adjust your rank. Anyone can report off ranks and have them fixed, e.g. if the player does not know English and is unaware of OGS etiquette.
Noone should have to fight GNU Go for rank points. That is a waste of time and will drive people away.
This is not a particular great process. I feel a little bad each time for “reporting” a player because of his score.
Anyway this is already off the topic. I was talking about how to keep the whole rating system constant over time, and how to link OGS rating system to any other system without anyone really noticing. It’s not about the rating of a single person, but about the rating of all persons participating in ogs.
You totally misunderstood the point of my post. Sorry 
in my proposal, rank points are exchanged just as we are used to, except if someone chooses to play versus a bot. In that case, humans points change, bots points don’t. Simple mechanism to stabilize the rating system as a whole.
How is that? You might ask. Let me explain. The elo system is a sum-zero system. Winner and loser exchange the same amount of points, in sum, the points stay the same. In general, this works fine to compare users at any single point of time. Anyway, the elo system has a weakness: you can not compare, say a 10k today to a 10k 1 year ago, because elo based ratings tend to drift over time. The system is exposed to inflation and deflation, very similar to our financial system. There are various reasons for this drift. One of them is that new players start with low rating and suck points out of the system as they improve. When they become inactive they effectively took points out of the big pot. Players with constant playing skill will tend to lose rank points, because in sum, they are getting rare.
So in general, it’s hard to keep elo rating system constant over time. So my post was about how to do this from a software developers perspective. Having a constant elo system is nice, but the cool thing is, you can actually link one ELO system to any other given rating system (like KGS or AGA) using this technique, if we wished to do so.
1 Like
All is relative 
On some servers, you are expected to set your rank by no means other than ranked games.
You should feel proud to be contributing to the community! I am often too lazy to investigate rankings.
Your idea of a stable system sounds nice, but does it work in practice? I don’t play against GNU Go, and I’m probably far removed from its influence. Then, if a 10k from last year is so different, doesn’t that mean that people’s idea of what 10k means also changes over time? The bot’s rank will become obsolete.
Also, OGS’ system will not automatically adapt to another system that does not use the same reference points. AGA is certainly out of the question.
I edited my post and added some more information, you might have missed
No worries, you cannot escape, except you stop playing