Hi,

I applied the WHR algorithm to go4go.net data, and produced a web site with the ranks of top active players. This is the top 14:

More details with rank graphs, etc: http://www.goratings.org/

Feedback is welcome.

Rémi

I'd be interested in a list of peak ratings too. E.g. current leader park Jungwhan has a peak of 2700, Lee Changho had 2747 at his peak.

An Jungki at #16 looks like an anomaly, there's only three of his games in the database, which is not enough data, IMO, to say anything much about his rating.

That's interesting and might help to keep in touch with what's going on these days...
And I'm glad to see that Cho Hunhyun is still alive and kicking at his age!
As for feedbacks... Maybe the graphs could be more like in KGS, with a grid or lines helping to keep track of the years and the ratings, because it's quite hard to tell which rating there is for which year.
Apart from that, I think it's all really neat.

_________________
Bored with skillful play? Watch my blunders on my YouTube Channel

Thanks, very interesting.
It would be nice to see the nationality next to the players.

I sometimes check the following page (it's in Japanese), which you may also be interested in:
http://sports.geocities.jp/mamumamu0413/total.html
The 2015-07 rating looks like this:

Thanks for pointing this out. In the future, I will indicate a confidence interval for each rating, and order players based on the lower bound of the confidence interval.

Rémi

It would be interesting for you/macelee to comment on the data used. Does it include only the games listed in go4go's tournament pages or does it also include results from preliminaries and so forth?

But in any case, thanks for this!!! Very interesting.

_________________
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21

Thanks to all for your feedback.

I updated the results with the latest games from go4go.net. Lee Sedol takes #3, and Iyama Yuta #8. From now on, my website will be updated automatically every day.

I don't know how extensive the go4go database is. My impression is that the strong players have almost all their games listed, but it is not the case for the weaker players. Go4go is a game-record database, and I guess some are difficult to get.

What time span does the rating depend on? The last five or ten years or what? I would like to see peak values as well, so that older players could be recognized as the greatness they had at one time. I know these ratings really represent the current competitiveness of players, not career achievement.

Rémi's work is based on the entire GO4Go database. I supplied a web address so that he can get real-time update. As he correctly pointed out, go4go is only a game record database - it does not cover all the game results. But it does nearly cover all major tournaments from main stage onwards. There are limited number of game records from preliminary stages. Ratings for top players should be quite accurate as their collections are nearly complete.

It would be interesting for Remi to comment here. It is not intuitively obvious to me that top player ratings would be accurate if their records are complete but those of their opponents are incomplete.

By the way, what are '个人赛' games? This seems like the biggest source of differences when I compared the go4go games to the list of games used for the Chinese ratings list for young players (e.g. Xie Ke).

_________________
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21

Of course, more game results would make the ratings more accurate. I wonder if it is possible to get them anywhere. I don't use the moves of the game record, so I don't care if I get only the final results.

The time span is the time span of go4go. All the games are used.

I was looking at the games behind the Chinese professional ratings compared to my database. I have collected the games reported for the Chinese ratings from September 2014 through June 2015 (see this post for how that works). This was a total of 3,150 games. Meanwhile I have received about 2,200 games from the weekly go4go service for the same period. The go4go service covers all the major countries not just China so we can see that there are quite a lot of games that have the results reported but not the game records. Within the 3,150 games from China a total of 460 games (14.6%) were games between Chinese pros and Korean pros. This compares to 62 games (2.2%) against Japanese pros and 92 games (2.6%) against Taiwanese pros.

_________________
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21

I wonder if it would be possible to try this with different databases, to empirically quantify the influence of different data.

_________________
Occupy Babel!

Thanks for this info. I tried to follow the link to the Chinese web site, but I could not find where the game results are located. Can you give me a direct link, please?

I could try to collect this data and combine it with go4go. It would be a little work, but it seems doable.

I am now thinking about producing historical rating lists too. I would have a rating list for every month of the past, as well as a time table with the top 3 players of every month in the past 40 years or so.

The gap is not as big as shown by the figures. I have only limited amount of time collecting games for the delivery service. But many more games are added to the database without being included in the weekly packages.

Thanks again for a great website. I was wondering, does the "Elo" reported here work the same way as traditional Elo for estimating a winner? For example Iyama Yuta and Yamashita Keigo are going to play the 4th Gosei title match in a few days. Currently they have ratings of 2472 and 2368 respectively. Using the usual formula that translates to a 64.5% chance for Iyama Yuta to win. Do you think that's a good estimate, or do these numbers represent something different?

According to igokisen Iyama leads Yamashita 28-18 in head-to-head, which is a 61% win rate. This is admittedly over all time so would likely include a young Iyama when he wasn't as strong as he is now losing against an already mature Yamashita so might understate his current superiority. Looking in go4go since 2012, it is 25-11 which is 69% win. That elo one is about in the middle of those .

There are only three games before 2010 and Iyama won 2-1 there, so the uppityness of youth applies even in Japan.

GoGoD has a score 32-19 to Iyama over 51 games. They have played 15 times so far in 2015 alone, and Iyama leads 10-5.

One possibly interesting factoid: Iyama leads only 14-10 when taking White but 18-9 as Black. As Black Iyama has varied his openings to an immense degree. Maybe that is what is troubling Yamashita.

Takao just won the playoff for Meijin challenger so we won't see yet another Yamashita/Iyama series for a bit.

Yes, they work the same as "traditional" Elo.