Life In 19x19
http://www.lifein19x19.com/

What's the best way to compute country ratings
http://www.lifein19x19.com/viewtopic.php?f=18&t=13214		 Page 1 of 1
Author:  macelee [ Sun May 22, 2016 4:14 am ]
Post subject:  What's the best way to compute country ratings
Now we have pretty good ratings for individual Go players. I wonder what is the best way to assess the relative strength of countries. I am particularly interesting in the comparison between Korea and China.

The algorithm is probably a weighted average of some sort. It needs to reflect recent game results. For example, Korean guest players won 5 games and lost just 1 in last round of Chinese league. So the algorithm should show that Korea has an improved position.
Author:  pookpooi [ Sun May 22, 2016 4:53 am ]
Post subject:  Re: What's the best way to compute country ratings
how about sum of all player's elo rating in one nation and divide it with the number of the player, like china has 10 + 10 + 10 = 30 and then /3 = 10, while japan has 6+6+6=18 and then /3 = 6, I think we can use elo's rating from goratings.org?
Author:  DrStraw [ Sun May 22, 2016 5:19 am ]
Post subject:  Re: What's the best way to compute country ratings
I don't think it is that simple. Which is more important to you: strength in depth, or strength at the top? If the former then a straight average does not reflect the sheer numbers of players and so underestimates; if the latter then the myriad of lower players drags it down and so underestimates but in a different way.
Author:  Elom [ Sun May 22, 2016 5:43 am ]
Post subject:  Re: What's the best way to compute country ratings
Couldn't you just weight each player's rating according to their ranking in the country? So that if there were 150 players, the top ranked player will have a weighting of her/his elo 150 times heavier than the lowest ranked player.
Author:  Elom [ Sun May 22, 2016 6:46 am ]
Post subject:  Re: What's the best way to compute country ratings
http://sports.geocities.jp/mamumamu0413 ... d2016.html
Author:  macelee [ Sun May 22, 2016 1:03 pm ]
Post subject:  Re: What's the best way to compute country ratings
Based on Elom's proposed algorithm and today's goratings data, here are the country ratings:

China: 3,250.53
Korea: 3,238.24
Japan: 3,044.74

I don't have access to goratings' full historical data, but I can perform the calculation every day and see how things change over time.

Another possibility is to give players with more games larger weight.
Author:  Bill Spight [ Sun May 22, 2016 5:08 pm ]
Post subject:  Re: What's the best way to compute country ratings
In such comparisons I would look first at the median rating for each group. I. e., how strong is the "average" player?

However, that's only one datum, and can be misleading. Once it seemed like the median club player in the US was 7 kyu. Today I walk into the same clubs and it seems like the median player is shodan. But that's because the 7 kyus of yore are the shodans of today. That is not a healthy sign, I think. Where are the 7 kyus of today?

Edit: Yes, I know that you are talking about pros, not amateurs.
Author:  dfan [ Sun May 22, 2016 5:28 pm ]
Post subject:  Re: What's the best way to compute country ratings
Bill Spight wrote:
Once it seemed like the median club player in the US was 7 kyu. Today I walk into the same clubs and it seems like the median player is shodan. But that's because the 7 kyus of yore are the shodans of today. That is not a healthy sign, I think. Where are the 7 kyus of today?

I dunno, I have the opposite reaction. I feel like back in 2000 it was a big deal to be shodan, and a 4d was a demigod, capable of writing textbooks and dispensing wisdom from the skies. Today it seems like half the players around are dan-level, and they're all significantly younger than me, and 4d just means that you're really good. Of course this is just a sample size of one geographic area.

I think a lot of the 7 kyus of today pass through 7 kyu pretty quickly. I am not that surprised, given all the material that's available online these days, including videos.

But I guess I am digressing even more.
Author:  Bill Spight [ Sun May 22, 2016 7:39 pm ]
Post subject:  Re: What's the best way to compute country ratings
dfan wrote:
Bill Spight wrote:
Once it seemed like the median club player in the US was 7 kyu. Today I walk into the same clubs and it seems like the median player is shodan. But that's because the 7 kyus of yore are the shodans of today. That is not a healthy sign, I think. Where are the 7 kyus of today?

I dunno, I have the opposite reaction. I feel like back in 2000 it was a big deal to be shodan, and a 4d was a demigod, capable of writing textbooks and dispensing wisdom from the skies. Today it seems like half the players around are dan-level, and they're all significantly younger than me,


Thanks. Glad to hear that. :D
Author:  Kirby [ Sun May 22, 2016 8:24 pm ]
Post subject:  Re: What's the best way to compute country ratings
Bill Spight wrote:
But that's because the 7 kyus of yore are the shodans of today.


Interesting. Maybe that's motivation for me to improve...

How strong do I have to be in today's ranks (e.g. on KGS, Tygem, or with the AGA) to be a "shodan of yore"? Maybe after becoming "1d", I had less motivation, but if I have a new "1d" to aim at, that could be better motivation.
Author:  Uberdude [ Sun May 22, 2016 10:21 pm ]
Post subject:  Re: What's the best way to compute country ratings
Kirby wrote:
Maybe after becoming "1d", I had less motivation, but if I have a new "1d" to aim at, that could be better motivation.

In the traditional Korean gup ranks there were no amateur dans, so one dan meant one dan pro. Go for it Kirby!
Author:  Bill Spight [ Mon May 23, 2016 12:11 am ]
Post subject:  Re: What's the best way to compute country ratings
Kirby wrote:
Bill Spight wrote:
But that's because the 7 kyus of yore are the shodans of today.


Interesting. Maybe that's motivation for me to improve...

How strong do I have to be in today's ranks (e.g. on KGS, Tygem, or with the AGA) to be a "shodan of yore"?


Oh, I think that the AGA ratings have held pretty steady over the years. :) (Japanese shodan strength has dropped about two stones, I think.)

OC, there is more than one yore. When I made shodan the top amateur ranking was 6 dan. There were a handful of amateur 7 dans, but 6 dan covered a fairly wide range of ability, overlapping that of pros. My impression is that pro shodans are stronger now by one stone, maybe more.

BTW, when I said that the 7 kyus of yore are the shodans of today I meant that literally. They are the same players, who have gotten stronger. :)
Author:  Kirby [ Mon May 23, 2016 12:30 am ]
Post subject:  Re: What's the best way to compute country ratings
Bill Spight wrote:
BTW, when I said that the 7 kyus of yore are the shodans of today I meant that literally. They are the same players, who have gotten stronger. :)


Thanks, I misinterpreted what you meant, but it makes sense now that I reread it.
Author:  emeraldemon [ Wed May 25, 2016 9:08 am ]
Post subject:  Re: What's the best way to compute country ratings
The idea of taking the average strength of all players seems very strange to me. You want a country with no weak go players? Why? How would that even happen?
Author:  mitsun [ Wed May 25, 2016 9:36 am ]
Post subject:  Re: What's the best way to compute country ratings
Whatever the algorithm is, it should satisfy the requirement that every new go player in a country increases that country's rating.
One simple example would be sum(exp(k*rating)), summed over all players. The scale factor k determines how top-weighted you want to make the overall rating.
Page 1 of 1 All times are UTC - 8 hours [ DST ]
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/