http://www.goratings.org/ now has historical ratings lists

[Rémi]

Hi,

You can find the historical rating lists there:
http://www.goratings.org/history/

In 1996, #1 Lee Changho was 156 Elo point stronger than #2!
http://www.goratings.org/history/1996-01-01.html

I wanted to add this message to my previous thread, but it seems it is not accessible any more. There may be something wrong with the forum.

Rémi

[Uberdude]

I noticed the ELO of the #1 player has increased over time. Do you think this reflects a real increase in strength, inflation in the whole history rating method, a too-low initial rating (if such is used), or something else?

[Rémi]

I noticed the ELO of the #1 player has increased over time. Do you think this reflects a real increase in strength, inflation in the whole history rating method, a too-low initial rating (if such is used), or something else?

The statistical model of WHR assumes that, on average, players don't make progress. And there is a weak prior rating of 2000 Elo for newcomers. The go4go database has many more games in recent times than in the past. So if the current lists includes more weak players than in the eighties, it might explain an increase in the rating of the top player.

Comparing ratings of today with ratings of the past is not very meaningful with such statistical models, unless they can be anchored with fixed-strength players that don't change over time (a machine).

[DrStraw]

One fact I find remarkable: before 1988 all the top three were Japanese; since 1994 not a single Japanese player has been in the top 3. This is an amazingly quick turnaround.

Also interesting is that in 36 years only 5 players have been at the top.

[MP4Life]

What's amazing is that ever since international tourneament started all #1 players are koreans. Heck, even ChoChikun is Korean descent.

BTW Im not sure how all these players are rated. No doubt Dr.Baeil's rating system is the best we have right now.

[HermanHiddema]

What's amazing is that ever since international tourneament started all #1 players are koreans. Heck, even ChoChikun is Korean descent.

Kobayashi Koichi is Japanese, and was nr. 1 in these ratings between 1988 and 1990.

BTW Im not sure how all these players are rated. No doubt Dr.Baeil's rating system is the best we have right now.

This site uses WHR (as mentioned on its homepage). AFAIK, Dr Baeil's rating is a variant of Elo. Since WHR outperforms Elo, this system is probably better.

[MP4Life]

This site uses WHR (as mentioned on its homepage). AFAIK, Dr Baeil's rating is a variant of Elo. Since WHR outperforms Elo, this system is probably better.

Is that always the case? I have no idea how those systems works but I know Dr Baeil puts serious effort and time into collecting all datas necessary. He got Phd in Stanford statistics... aint that sth? IDK

[Kirby]

This site uses WHR (as mentioned on its homepage). AFAIK, Dr Baeil's rating is a variant of Elo. Since WHR outperforms Elo, this system is probably better.

Is that always the case? I have no idea how those systems works but I know Dr Baeil puts serious effort and time into collecting all datas necessary. He got Phd in Stanford statistics... aint that sth? IDK

Dr. Baeil might put effort into collecting data, but we are talking about different models for interpreting the data.

You can read Remi's paper here:
http://www.remi-coulom.fr/WHR/WHR.pdf

Basically, Remi compared several existing rating systems at the time using data from KGS games. The results suggest that WHR performed best out of the systems he compared (Elo, Glicko, Trueskill, Bayeselo, Decayed history).

The paper notes some unanswered questions, such as worse performance on the test set in his experiment vs. the training set. So there may be some unanswered questions in the study, and room for further experiments. Nonetheless, the study appears to show that WHR allows for better prediction of game results than the other methods, elo included.

This isn't to say that Elo is a bad system, or that another study might not show different results. But Remi's work seems to be well thought out, to say the least.

[Rémi]

Hi,

I updated the web site today with an interactive history graph. You can see a screenshot attached. The domination of Lee Changho was really spectacular.

[John Fairbairn]

Your graph for Yi Ch'ang-ho seems a bit skew-whiff to me.

I don't pretend to have even tried to understand the mathematics behind it, but just looking at his annual scores suggests the early part of the graph line may be suspect, especially for 1990.


1986: 1-dan, win-loss ratio: 8-3 = 72.7% (no titles)

1989: 4-dan, win loss ratio: 84-27 = 75.7.7% (1x title, 3x runner-up)

1990: win-loss ratio: 78-12 = 86.7% (3x title, 3x runner-up, and in addition he set a record 41 consecutive victories)

1991: 5-dan, win-loss ratio: 65-19 = 77.4% (6x title, 1x runner-up)

1992: 6-dan, win-loss ratio: 81-21 = 79.4% (8x, 2x)

1993: win-loss ratio: 90-19 = 82.6% (11x, 1x)

1994: 7-dan, win-loss ratio: 71-19 = 79.4% (9x, 1x)

1995: win-loss ratio: 63-13 = 82.9% (13x title)

1996: 9-dan, by recommendation; win-loss ratio: 63-30 = 67.7% (13x, 5x)

1997: win-loss ratio: 72-19 = 79.1% (11x, 3x)

1998: win-loss ratio: 49-15 = 76.6% (9x, 3x)

1999: win-loss ratio: 51-10 = 83.6% (9x, 2x)

2000: win-loss ratio: 55-11 = 83.3% (3x, 2x)

[Rémi]

Hi John,

And thanks for your feedback. I use the database of go4go.net. The database of go4go.net has many less games:
http://www.goratings.org/players/208.html
So, for 1990, I have only 11 games, and the results are 6-5. I suppose the other games were against weaker players.

I have a copy of the Winter 2011 GoGoD Database. It seems GoGoD may contain more historical game results. When I find time, I may try to produce a graph with GoGoD too.

More generally, I'd be happy to use any source of game results that is more comprehensive than go4go. Suggestions are welcome.

Rémi

[emeraldemon]

The graph is awesome. I wonder how long Park Jungwhan can hold on to the crown. With the GoGoD data it would be cool to see what WHR thought of Go Seigen.

[macelee]

For Go4Go database, I have not got the opportunity to compile games from all major Korean tournaments, in particular those from 1990 and backward. So Lee Changho's score should be underestimated. I also believe that Cho Hunhyun's score is underestimated.

[Bonobo]

Rémi, are you aware of this page? Not as many games as GoGoD, and I don’t know how it compares to Go4Go, and maybe you know it already, but I don't want to withhold this from you: http://homepages.cwi.nl/~aeb/go/games/

[Rémi]

Thanks for the link.

It is a bit too difficult for me to combine different sources of game results, because player names will always have different spellings. I don't have enough motivation to normalize all player names. So, in practice, I will probably stick to go4go.net unless I find a database of game results that is clearly better.