How can overall total ranking and overall ranking for live games be different (in this case)?

[AllanFelipe]

Suppose I have only ever played live 9x9 and 13x3 games (no blitz or correspondence games. By the way, this was before the recent update. Now some blitz games appeared and I can no longer show this question with numbers).

My overall rating for 9x9 games [1] is equal to the overall rating for live 9x9 games [2], because this is the only type of game I played (the same applies to 13x13 games, meaning [3] = [4]).

How can my overall rating of all sizes [5] (whatever type of average it may be) differ from my overall rating for live games of all sizes [6]?

[GreenAsJade]

I think you need to think about this: what was the “live overall” rating of your opponent?

You will find that IF your opponents also had all only played 9x9 and 13x13 then you and they might have the same overall rating as well.

However, since at least one of your opponents had played 19x19, their overall live rating was different to their 9x9 rating, and so your overall rating was affected differently by the game you played with them compared to yout 9x9 rating.

The “overall” rating is not any kind of “average of the other ratings”, rather it is a rating that is calculated on every game you played irrespective of board size, and it’s calculated using that same rating from your opponent.

So your overall live rating is affected by the fact that (for example) the opponent you defeated in 9x9 sucks at 19x19 so they have a lower overall live rating than their 9x9 rating: so your overall rating impact is less advanced when you defeated them.

[shinuito]

@flovo

We changed the breakdown ratings to compute against you opponent’s overall rating, so the ratings should be on comparable scales.

From what I remember reading each individual rating is updated by using the overall rating of the opponent.

So if you play a 13x13 live game, your overall, overall 13x13, overall live and live 13x13 should all be updated using the opponents overall rank.

I think it’s relatively easy to understand why the overall rank differs from any one rank if you’re playing multiple settings, because the overall rank gets updated with all game information but the individuals only use a subset.

However if you only play one category exclusively, for example only 19x19 (maybe multiple time settings) or only live (maybe multiple board sizes) then you’d expect that that specific overall rating should be the same as the overall rating since they both use all the information because of the focus on a certain game type.

@AllanFelipe

So I would expect 5 & 6 to be the same provided

@AllanFelipe

Now some blitz games appeared and I can no longer show this question with numbers).

which would cause the overall to deviate from the live.

[shinuito]

I guess we can look into it further with some new accounts. The first one I looked at that seemed to play only one type of time/board setting

which seems a bit strange.

Firstly the games in the game history are using a live symbol but the updated ratings are blitz.

The time control I’m seeing is

Time Control
Fischer: Clock starts with 2 minutes and increments by 5 seconds per move up to a maximum of 20 minutes. Pauses on weekends.

although I don’t understand why it says “pauses on weekends” XD.

So I’m guessing A) there’s a slight bug since the game history symbol doesn’t match the table entry being updated B) that these ratings should all be the same as the overall, and the three of them are but not the overall.

Now flovo also mentioned something before

@flovo

I don’t know, why the numbers are slightly different but 4 rating points are nothing to bother to much about. My guess is some annulled games caused the numbers to diverge slightly.

so it would be interesting to know if rating refunds affect only the overall for example and that’s what cause the ratings to diverge in a case like this. E.g. the first game played timed out but it was only like 4 moves long. So if you lost and all ratings decreased but then it was autoannulled and only the overall went back to its original state for example, that could explain it.

---

Let’s check another new user.

same thing with a first game annulled. Interestingly the game annulled for a timeout was 86 moves long. I’ve no idea why.

---

One more user where they don’t seem to have an annulled game

This one I can’t understand, because there’s no annulled game, they seem to be all the same time control and board size.

This user also only played a bot which in the same time controls looks like

so if it was really updated with like ratings 9x9 updates 9x9, you’d expect all four numbers to be different for the user, but rather 3 of the 4 numbers are the same which is expected if you update each with only the overall rank of the opponent, but again the overall is off by like 4 points for some reason.

[shinuito]

For the sake of fully exploring the problem I made a new account and played one 9x9 game

Rating9x9test vs. rickcheese321

and for no reason I can understand, the overall rating is one rating point different.

Relevant part of opponents table

It could be some rounding error somewhere? Then maybe that rounding error propagates.

Any ideas @GreenAsJade

[GreenAsJade]

I don’t have an idea - I described what I believed to be the intent.

If evidence is showing us it’s not as I described then either that’s a bug, or I simply was wrong about what is supposed to happen.

I believe that we expect overall rating to be different to specific ratings, for the reason I gave.

Rounding errors could also contribute.

However, overall rank is not supposed to be “some kind of calculated average from the other ranks”, it is supposed to be “the rank that you get when you include all games”.

[shinuito]

This would be the relevant ratings update thread/quote I suppose

@anoek

Speed and size specific ratings

It’s been a known issue that you can’t really compare the different ratings on your profile. Your overall rating had pretty much nothing to do with your live 13x13 rating for instance. This is because they were computed using completely independent rating pools.

@flovo came up with the great idea of using your opponents overall rating when updating a sub-rating, this has the effect of the rating pools converging into the same scale as the overall rating pool, so the individual ratings should be somewhat comparable now.

@GreenAsJade

I believe that we expect overall rating to be different to specific ratings, for the reason I gave.

Sure, but not for one game like in the example I provided, and likely not when a player has only played one category, one boardsize with one timesetting only.

@GreenAsJade

However, overall rank is not supposed to be “some kind of calculated average from the other ranks”, it is supposed to be “the rank that you get when you include all games”.

Of course, I’ve also commented something to this effect in many threads, and that’s not what I’m commenting on or disagreeing with in any way.

[GreenAsJade]

I haven’t had a chance to look at the specific example you created yet, but…

@shinuito

Sure, but not for one game like in the example I provided, and likely not when a player has only played one category, one boardsize with one timesetting only.

I’m not so sure about that, in general.

If you have only played one game but your opponent has played many, then their overall rating will be different to their specific game type rating, so it’s effect on your overall rating is in fact expected to be different to their specific game rating effect on yours, even for that one game.

[shinuito]

@GreenAsJade

so it’s effect on your overall rating is in fact expected to be different to their specific game rating effect on yours, even for that one game.

But why?

@anoek

@flovo came up with the great idea of using your opponents overall rating when updating a sub-rating,

Let’s say Overall, Live Overall, 9x9 Overall and Live 9x9 on a new account all start on 1500±350.

All four of these after a live 9x9 game get updated only using the opponents overall rating. (Unless this isn’t the case?)

They all start the same, they get given the same numbers, opponents overall rating and deviation and volatility, when being updated, and I assume they all use the same rating update function, so the output should be the same?

As long as you don’t play a different time setting or board size, all the games are 9x9 live for example, then at each step the numbers come out the same and so should stay the same.

In practice the numbers almost come out the same except as I mentioned the overall seems to deviate by approximately one point - which over time seems to accumulate to 3, 4 or more points.

[GreenAsJade]

@shinuito

But why?

The example I’m thinking of is that a person plays one, only, game against another person who’s played a lot.

That other person has played all types of games, so their overall rating is different to their 9x9 rating.

The new person starts with a 9x9 rating of 1000 and a overal rating of 1000.

Their opponent has a 9x9 rating of 1500 but an overall rating of 1000.

Clearly the impact on the new person’s 9x9 rating - where they lost to a much higher rated person - is less than the impact to on their overall rating, where they lost to someone who was the same rating as them.

The bigger the rating difference between the players in a loss, the less the impact of the loss…

Right?

[shinuito]

@GreenAsJade

Their opponent has a 9x9 rating of 1500 but an overall rating of 1000.

Clearly the impact on the new person’s 9x9 rating - where they lost to a much higher rated person - is less than the impact to on their overall rating, where they lost to someone who was the same rating as them.

@GreenAsJade

Right?

But no, not if only the overall rating is being fed into the exact same rating functions. If opponents 9x9 isn’t used, then it doesn’t matter at all. When the player’s 9x9 rating and overall are the same, the deviations are the same, then the result of the update should be the same?

And this is the point that I quoted above twice, and now for the third time

@anoek

@flovo came up with the great idea of using your opponents overall rating when updating a sub-rating,

@GreenAsJade

Clearly the impact on the new person’s 9x9 rating - where they lost to a much higher rated person - is less than the impact to on their overall rating, where they lost to someone who was the same rating as them.

So why “clearly”, if you never actually care about the opponents subratings when updating your own subrating?

It does say

using your opponents overall rating when updating a sub-rating

right?

It doesn’t say

using your opponents sub-rating when updating a sub-rating

right?

---

Let’s assume we have this setup as you said, and because it’s glicko2 there’s volatilities v1 and v2 for each player.

p1 overall	p1 9x9	p2 overall	p2 9x9
1000 ±d1	1000 ±d1	1000 ±d2	1500 ±d3

Now if you only ever use the function f to update the ratings in the following way

f(1000,d1,v1, 1000, d2, v2) which updates the overall rank using opponents overall rank and

f(1000,d1,v1, 1000, d2, v2) which updates the 9x9 rank again using opponents overall rank

why do you expect different results? Are we using some non-deterministic algorithm? Does it depend on what day and time the game finished?

[shinuito]

Just for emphasis:

@flovo

We changed the breakdown ratings to compute against you opponent’s overall rating, so the ratings should be on comparable scales.

@anoek

Speed and size specific ratings

It’s been a known issue that you can’t really compare the different ratings on your profile. Your overall rating had pretty much nothing to do with your live 13x13 rating for instance. This is because they were computed using completely independent rating pools.

@flovo came up with the great idea of using your opponents overall rating when updating a sub-rating, this has the effect of the rating pools converging into the same scale as the overall rating pool, so the individual ratings should be somewhat comparable now.

@anoek

@flovo came up with the great idea of using your opponents overall rating when updating a sub-rating,

@shinuito

@anoek

@flovo came up with the great idea of using your opponents overall rating when updating a sub-rating,

If it’s changed since then, like we reverted back to the old system of using a sub rating to update sub ratings, or some other random notion instead, I am not aware of it - I’m not just playing stupid for the sake of it, which I feel like this very much implies…

@GreenAsJade

The bigger the rating difference between the players in a loss, the less the impact of the loss…

Right?

[GreenAsJade]

Ah - I had totally missed that, thanks for making it clear.

It sounds just buggy then

@shinuito

I’m not just playing stupid for the sake of it, which I feel like this very much implies…

Implications are such a pain - I wasn’t intending to imply you were stupid! " " says “hmm I wonder what is going on here”, not “Hmm this person is stupid”

Clearly I’m the “stupid” one in this case anyhow …but I was _genuinely_asking “right?”, needing the clarification that you provided, because my assumptions were so firmly set in what I thought I knew.

[shinuito]

@GreenAsJade

Clearly I’m the “stupid” one in this case anyhow

No I don’t think so. You obviously have more information than I do on these things generally, but I was just trying to figure out where we were getting our wires crossed.

[GreenAsJade]

It is “stupid” not to read replies carefully, which I clearly did not do.

It’s a trap easist to fall into when you do indeed often have more information!

Anyhow, I’m glad we uncrossed the wires, I’m humbled and educated and none-the-wiser how they can diverge in the current situation!