OGS has a new Glicko-2 based rating system!


#61

I don’t see how this would be fair. If you have gone to the trouble of beating someone, then they run away by timeout, how is it fair that you don’t get the ranking for that?

I actually don’t understand the underlying problem. What is the difference between whether someone lost by losing or because they knew they were going to lose so they timed out?

In each case, the winner is going to get the same ranking boost. So either that ranking boost is correct or its not. The fact that is was a timeout doesn’t come into it.


#62

Well, it’s not the case that everyone timeouts because they’re losing. I’ve had a few games where my opponent was clearly beating me, but I won due to timeout.

That’s the underlying problem.


#63

Yeah, I’ve seen those before. It does seem wierd at first glance.

In the case of real-time, it could be that even though they were winning, they literally ran out of time to think about what next. Or it could be that some real time interruption occurred for them. EG they got called to dinner or whatever. I don’t know what you can do about this. Maybe the fact is that if you time out you lost. As simple as that, It’s part of “how good you are” IE your rank is not only how good you are at placing stones, but how good you are at finishing games.

I see now why someone might get an unexpected boost though: if you are playing someone who outranks you, and they time out, then you will get quite a boost even though you might not have been able to actually beat them.

But still - to be fair in this case you’d have to assess whether the lesser player actually was winning and thus the cause of the timeout was the higher ranked player running away. If that was the cause, the lesser player would deserve the boost.

This all gets too hard: it’s much easier to treat a timeout as an actual loss, but also punish timing out more than we do here in OGS. Most conversations about timeout seem to end in the OGS concensus being that it is not a sin. But look at the trouble it causes…

GaJ


#64

i think it’s great what you guys are doing! cheers!

what does exp stand for?
thanks!


#65

exp is the exponential function e (aprox 2.718) to the power of x (in this case rank* 0.032).


#66

I think correspondence time outs are rare enough that any rank in inflation will be mitigated by the overwhelming majority of finished games.
To create a separate use case for such a rare occurrence doesn’t make sense to me.

I agree with the old T system preventing one IRL emergency from resulting in 20 rank losses but can’t really see a valid reason to not treat a win as a win for the players who have opponents time out against them.

As GaJ says


#67

I like the change in general, but maybe the dan ranks are inflated quite a bit? It’s interesting to see that a few players surpass 9d, even 11d:

FrancisCol (12d)
sine (11d)
pares (11d)
ThunderGod (10d)
MagariSimoku (10d)
Inven (10d)
peecatchoo (10d)
vlin (10d)
Simba (10d)
krnzmb (10d)

BTW, I believe there’s solution that doesn’t explicitly define an absolute limit of the highest rank you can achieve, while ensuring that ranks above 7d cannot be achieved too easily.

EDIT: @arnaudgo You’re right, just found a 12d!


#68

I actually agree with both of you. I was merely responding to GaJ’s post:

My point was that there is a difference because some people DO time out even if they’re winning. Regardless of how minute the number is.


#69

Hi,

you even have at least one player 12d ! :slight_smile:

Arnaud


#70

Yay for a new rating system, the way you describe it, it sounds a lot better than the old elo system.However… it appears I have gotten a rather big rating change: from 14 kyu down to 21 kyu :open_mouth:

I guess the weight of the different speeds has been changed? I nearly only play correspondence (I know I really have to change this…) so perhaps that’s to blame. I don’t really mind the ranking change, but I do not have the feeling that all my opponents have undergone the same change. Looking at someone I regularly play, who also was around 14 kyu, he remained there. But I’m pretty sure that the difference between us is not 6-7 stones :slight_smile:

If I look at my rating graph in the old system, it has been steadily increasing over the last year. And that is in line with what I feel I know about the game. However if I look at my overall rating graph in the new system, it has been slightly declining for a year now. Exactly the opposite. I don’t really know what to do with this information; I probably do not have the time (work / small kids) to start playing lots of live games to compensate for that declining factor from mostly playing correspondence.

I’m not asking for a ranking change btw, but I do wanted to mention this here. Anyone with some more information and perhaps some insight in my particular situation? Is everything still being calculated and do I just need to wait it out? :slight_smile:


#71

In the search window to find a player I cannot see the Glicko-2 rating but the classical GO rating (10K, 3K… instead of 1900 Glicko 2300 Glicko…)…

How can I choose the Glicko-2 rating I would like to follow in all windows instead of the classical GO one ?


#72

That was a seriousy polite enquiry about the new rank!

I don’t know that the explanation is undue weight to live games: the Glicko2 rating you have for correspondence games in itself is remarkably low, compared to your old system record.

Will be interesting to see what happened there!


#73

Guys, come on I hate it!!! I was already afraid of playing against 1 dans, and now I have to play against 3 dans?? That’s just mean!


#74

You do understand that no number changes your ability, their abiliyy, or the stones on the board, right?


#75

The new rating makes more sense to me. I was rated 15k - lightyears to high. Now I am 20k which looks much more realistic. Thanks for all this work!


#76

Sure I understand. I’m just scared.


#77

I looked at your ranking diagram and noticed that some losses are not visible in the bars of the histogram in the lower side: no purple at all since july 2016 until may 2017!
Maybe there’s a bug in the histogram itself… or maybe the bug affects also the ranking calculation… (even though losing some losses should reasonably cause a higher rank instead of a lower one).


Wins/losses chart
#78

Look on the bright side, if you now meet a 1d it is only a 2k “you can beat it!”

@anoek
except my own rank I see all the ranks as very very specific numbers with 10 decimal places

like: 13.30435034088 k

@anoek
for the table with rank and board size, I wished I could switch the Glicko-numbers to rank numbers. As an not-programmer I am not used to the glicko numbers and therefore the table is only to readable for me, when have the power to translate them in to rank numbers.

(at least a a conversion chart as pop up?)

…and thank you for the energy that you put in the website.


#79

Where are you seeing this? Also, what browser are you using?

The reason why we don’t provide ranks for the breakout ratings is because they are all different rating pools, so we can’t just take the same rating -> ranking mapping and apply it to each (I tried, the results are humorous :slight_smile: ).


Show Kyu/Dan instead of Glicko Rating on Player Profile
Show Kyu/Dan instead of Glicko Rating on Player Profile
#80

I looked it up, it is only on the playing site and in the gamegraph. Eveywhere else it is fine.

for the table:

I looked it up too. and realized I didnt understood it at all.
I can identify the boardsize vs timestting, but what does 1492 ^= 158 mean ?

(Is there an explanation pop up? can you make one?)