Go Journey

it does seem steep… my biggest concern is moving from a complicated game where i am of a decent standard (say 1600) to one where i am not will prove disheartening. especially as an established game like go will tend to have less new players obviously available for gentler (or less one sided) games.

i’m expecting to lose a lot though so that should help.

i am finding the 2nd janice kim book helpful though as i really like the ‘try this’ followed by examples of why if the opponent attacks into it. better players wont so those lessons would not be so obvious.

As for the ‘less one-sided game opportunities’ - here’s a reasonable approximation of OGS’ playerbase:

(Source: this thread)

Your current lowest rating (based on a single game, mind you) suggests that you can play a roughly even game with about 3% of OGS’ players. Typically, at least 2000 people are online and about 150-200 are involved in a game or looking for one. Accordingly, if you create a game offer, there should be at least 4 players within your range willing to play at any point in time.

7 Likes

thats better than i thought! so thanks for the stats…

2 Likes

I added a post with some general Ratings stuff (that may be incorrect as its guess work) along with a brief first game description.

1 Like

Interesting reads. For what it’s worth, you should consider beginning by playing only on 9x9 boards, as opposed to 19x19 boards. 9x9 board games go by much faster and you’ll pick up many more useful ideas about the games, rather than floundering on a 19x19 board, with no idea of where to play.

1 Like

I’m sorry to say, but you got many things about the OGS ratings wrong.
At the moment I’ve not the time to go in the details, but I’ll come back later today for it.

I look forward to it. I have quite a fascination with ratings systems and though I have written working Glicko2 systems so understand how they are supposed to generally work I have not seen the code here so there will no doubt be many nuances it would be fascinating to know and potentially explain…

I tried for a 9x9 on that Saturday and didn’t get a game so switched to the larger board and did…

I think the 9x9 will be more useful to learn the more tactical nuances so will try and follow that excellent advise!

My advice, for what it’s worth, would be to advise you to move to bigger boards, at least 13x13 asap. Though I may be biased given as I really don’t like 9x9.

2 Likes

Here are my own two cents to the topic.
I’ve tried to reverse engineer the OGS rating system, so my main sources of knowledge about the OGS rating system are based on forum posts and my own tests.

This is rather long because I added some unnecessary comments.

The rating is normally displayed as follows

PlayerName 1150 ±350

The above is actually the rating given to new players on the server

You cannot know, but the initial rating is actually 1500±350 as the glicko2 spec suggests.
For new players (high deviation) the rating gets lowered for displaying (and hopefully for match making as well). This was proposed to give new players easier games to start with go on OGS. It was called humble rating on the forum. The rating is shown correct if the deviation is lower than 125.
I’m not sure how one would correctly show the humble rating, but 1150+700 would be more precise.

When trying to matchmake this also provides a wide range of potential opponents.

I don’t know how other sites do it, but I’m pretty sure OGS doesn’t take deviation into account for match making.

As an example let us assume the above player plays an opponent 1200 ±80

If he wins then his rating might increase 100 to 1250 and his RD (rating deviation) drop to ±340. His opponent may drop to 1190 and his deviation ±79

Just as a service: The changes are
1150±350 → 1351±252
1200±80 → 1187±80 (acually 79.728…)
or
1500±3500 → 1572±285
1200±80 → 1194±80

so is assuming that the first player was at the strong end of the band and not the weak.

Actually it’s a general wight, but for the sake of understandability it might be ok.

Player A 1800 ±80 beats player B 1600 ±100

A 1800±80 → 1809±79
B 1600±100 → 1587±98

Player C 1700 ±80 beats Player D 1800 ±80

This has more of an impact as the system thought Player D was stronger . Result

Player C 1780 ±85 Player D 1720 ±90

C 1700±80 → 1722±79
D 1800±80 → 1778±79

A single game result (always) results in a lower deviation. Glicko2 usually uses a bunch of games (up to 15 on OGS) to update the rating. Only if there are many outliers the deviation goes up (at least this is my observation). (A ?bug? on OGS causes the deviation to go up after a completed rating periode (15 games)).

glicko 2 also adds in a volatility factor

just as info: on OGS the volatility is about 0.06 and doesn’t change much.

The system also has a RD time decay included so if you don’t play for a long time then your RD increases as the system grows less certain of you.

Not sure if OGS implements this special case. I came to the conclusion it doesn’t.

GO Server system

The rating is both board (9x9, 13x13, 19x19) and game type based (i.e blitz, normal etc) so each player has a different rating for each combination. This is then combined into an overall rating.

The overall rating is the rating used everywhere on OGS. The other ratings are just “informational”.
Each of those ratings is calculated separately. The overall rating uses all games and the opponents overall rating, the 19x19 blitz uses only 19x19 blitz games and the opponents 19x19 blitz ratings and so forth.

My start rating 1150 ±350

As mentioned above, it’s actually 1500±350. You can see your true rating here.

though this should not impact a decent ratings system only the win/loss should count

that’s correct.

This is because my rating band was so wide the system is treating their win as an ‘expected’ win

I’m not sure if calling it ‘expected’ is the best phrase here. Because of the high uncertainty, the system doesn’t even know what to expect.

Note how the 1150 for all the other board/games types is increasing the 0 - 0 rating (which is the prime site rating displayed against the user name). The actual rating is the 19x19 and times 742.

See above. This are 4 independent ratings and the overall number is the actual one.

Kyu-Dan designation which drops to the alarming level of 41k!And increases to 11th dan.

OGS limits the rank to 25k as the lowest and the highest to 9d.
The upper bound is by tradition? The lower bound of 25k around the point where the separation in different ranks looses its meaning.
The Kyu and Dan ranks are the traditional way to classify player strength. A rank difference of one rank equals a one stone handicap.
On OGS there is a direct convention formula rating → rank, with rating = 850 * exp(0.032 * rank) where rank 0 is 30k.
The deviation is calculated by converting either the 1 sdt lower or higher rating. (I’m not sure which is used).

Official GO Kyus also start at 30 kyu for a total beginner whereas this system seems to drop you to 41 Kyu.

It only looks this way because the histogram applied the formula to the ratings. OGS doesn’t drop you below 25k, but your rating can drop lower.

as noted I am guessing ‘Kyu’

you guessed right.

4 Likes

I believe it does, or is meant to, but everything else sounds about right.

3 Likes

The RD increase would be ≤ 1 points for each empty rating period. This would result in an increase by about 10 points over a year of absence. The noise is much higher than this. So :man_shrugging:

1 Like

Thanks for that! I will try and tie some of those adjustments in over the weekend once I have understood what is occurring better (so rereading when not at work…)

How do I generally find actual rating as I dislike pretend or adjusted ratings quite intensely as it (to me) defeats the object entirely.

The Glicko2 volatility though is provided by the developer whether using the authors libraries or writing the algorithm yourself.

Since the adjustment only happens for players where the ratings’ uncertainty is big (we have not collected enough information about the players strength). I would argue that we cannot defeat the object of ratings if we don’t have a rating which would match the object in either case.

If the objects of using ratings are

It does this by following the assumption that players enjoy games more when they
A - are not beaten heavily repeatedly and
B - have challenging games that stick in the memory - the ‘I remember when this happened’ and I fought back from certain defeat to grasp victory.
C - are not so easy as to be instantly forgotten (people rarely boast of destroying someone who cannot ‘play’ - I,e ‘Ha I took on my four your old at chess last week. He didn’t know the rules properly and I crushed him!’

We cannot achieve B as long as we have big uncertainty of the rating. The adjustment tries to increase the probability of A over C, so player new to go don’t start by gotten beaten down to beginner ratings by several strong players in a row.

You can find the rating history of any player with

You can find the player_id in the url of the players profile

1 Like

I would just start the player at the lower ranking. Perhaps when signing up ask if ‘experienced player’ and if they are give them the standard 1500.

Solves the problem without background shenanigans.

Saying that now I know what is occurring that is perhaps less an issue.

I assume they use the ‘fake’ (or ‘adjusted’) rating for matchmaking and then adjust the actual post result?

That makes sense as the game found matched an actual 1100 ish with a fake 1100 and as I lost the actual rating dropped considerably as a 1500 had lost to an 1100. (and then the fake rating adjusted heavily downwards as well).

On consideration that actually seems quite elegant. It is is getting me to the level i need to be very quickly as someone with no skill. It (on the other hand) I joined as an experienced player (say an actual 1600) then I would be being matched frequently with lower ranked players I could beat easily (how the ratings would behave fake/otherwise in that circumstance I dont think I could guess at)

OK, I know ELO ratings and Kyu / Dan rankings can be confusing, and it’s so easy to focus on the numbers. Let me try and provide a better metaphor that will - hopefully - put things in perspective and take some of the pressure off.

In terms of reasonable expertise in the game and the quality of play you’re going to get, I would say that the rankings between 25 kyu and 18 kyu are comparable to grade school - a 25 kyu would be a 1st grader who knows their ABCs and 123s and an 18 kyu would be a 6th grader who can write complete sentences about half of the time and knows some of their multiplication tables.

The meta-point I’m trying to make here is that - while you’re just starting out - you shouldn’t worry so much whether you’re facing a 3rd grader or a 4th grader - in the grand scheme of things, the differences will be minimal.

Of course there is another matter, which is the pool of potential opponents that you have access to via the OGS matching algorithms and the limits that users put on games. Don’t be discouraged! You can always post game invites of your own, and create whatever input parameters you want (i.e. you can invite players between 25 kyu and 20 kyu etc). Alternately, you can pop on the OGS forums and request teaching games, which tends to get a good rate of response from the community.

So, let’s keep the metaphor going. If 25-18 kyu is grade school, I would say that 18-15 kyu would be junior high school, and 15-10 kyu would be high school (in terms of skill growth, time commitment, body of knowledge absorbed, etc).

10-5 kyu would be college and I’d say anything above 5 kyu would be graduate school. At the point you’re doing Dan-level study and above, we’re talking PhD level grad school, or going for multiple PhDs.

Given the long, twisty road ahead (holding many delights, pitfalls, surprises, etc) I say don’t sweat the small stuff at the beginnings. The more games you play, the more your kyu and ELO ratings will stabilize. And don’t worry about losing your first 10, 50, or even 100 games. Just lose them as quickly as you can so that you can LEARN from that School of Hard Knocks and start figuring out how to win.

Good luck!
tonyb

2 Likes

Here are 266 posts for you to read which contain all sorts of proposals (including the one you just espoused) and counter-arguments to each.

3 Likes

lol, maybe when I have a few spare hours… To be honest I am happy to just understand how this system works as I enjoy the intricacies. As a developer myself I tend to appreciate that things are done for a good reason and that those reasons often involve more detailed knowledge of the subject than I, as a recent starter, would have (dunning-kruger applies as ever)…

3 Likes

Na man you’re internetting all wrong! Meant to just wiki for 5 minutes and then rage at all the “morons” for not understanding things at your level. :sunglasses:
You can take or leave poe’s law at your own discretion.

3 Likes