Iyama's world ranking

[Uberdude]

Right, Iyama has been having more losses lately both domestically and in the extra international matches he's done, plus the ratings of people at the top has been increasing. Iyama's all-time high rating was something like 3586 and #3 at the end of last year (https://web.archive.org/web/20171205171 ... gs.org/en/), though if you look now at what it thinks his rating was then it's only 3534 due to the way WHR retroactively updates the past based on present information. Back then 3540 would get you world #5, now it only gets you #10, there's been a big growth in the number of players 3550+.

Lee Sedol is a top player though who has seen a bigger decline than Iyama: back in Jan 2016 he was #3 with 3553 (Iyama close behind at #5), but now he is down to #22 with 3478.

[xiayun]

Yeah, and Gu Li, Lee Sedol's main rival, has fallen all the way to #75, three years after his last international title.

[TheCannyOnion]

Iyama has dropped all the way to 39th on Mamumamu's ranking. He was ranked 15th at the beginning of 2018.

http://sports.geocities.jp/mamumamu0413/total.html

[TheCannyOnion]

Iyama has dropped all the way to 21st on Goratings.org. On Mamumamu's ranking, he is now down to 44th.

[TheCannyOnion]

Iyama is now down to 26th place on Goratings.org. If he loses to Yamashita Keigo again, he's likely out of the top 30.

[Uberdude]

<admin> This and next post moved here from viewtopic.php?f=13&t=13751</admin>

Apparently we now want to say that Go Ratings (actually past go ratings) is the 'truth' and what we actually see happening in the world around us should conform to that or be subject to criticism. However, the Go Ratings algorithm itself rejects this idea since AFAIK it retrospectively adjusts past ratings based on current results.

And no, Goratings does not retrospectively adjust past ratings based on current results. What a ridiculous thing to say! If that were the case, there would be no past ratings.

Wrong. It does. Go read the Whole History Rating paper, or use archive.org to observe this effect.

[TheCannyOnion]

And no, Goratings does not retrospectively adjust past ratings based on current results. What a ridiculous thing to say! If that were the case, there would be no past ratings.

Wrong. It does. Go read the Whole History Rating paper, or use archive.org to observe this effect.

No, you are wrong. Nowhere in the paper does it say that it "retrospectively adjusts past ratings based on current results" or something to that effect.

[Uberdude]

On the other Iyama thread and now moved to above this post, TheCannyOnion demonstrated his ignorance of the whole history rating method as used by goratings.org.

And no, Goratings does not retrospectively adjust past ratings based on current results. What a ridiculous thing to say! If that were the case, there would be no past ratings.

Wrong. It does. Go read the Whole History Rating paper, or use archive.org to observe this effect.

No, you are wrong. Nowhere in the paper does it say that it "retrospectively adjusts past ratings based on current results" or something to that effect.

Here is an attempt to remedy that.

As confirmed by Rémi Coulom, the creator of Whole History Rating and goratings.org, on this forum at https://www.lifein19x19.com/viewtopic.p ... 40#p193740: (nice explanation from ez4u why this is a good idea shortly after at https://www.lifein19x19.com/viewtopic.p ... 55#p193755).

Remi: The 2000 rating list is headed 2000-01-01. So not only should it not reflect the results for 2001, surely it shouldn't include the results for 2000 either. On that basis, go4go has just two games, for 1997, scoring 1-1. It's well known I'm not a numbers guy, so what's the trap I've fallen into?

The WHR rating algorithm has retroactive effect: the rating of one day depends on games of the past and games of the future. That is what makes it more accurate than incremental rating algorithms.

WHR paper: https://www.remi-coulom.fr/WHR/WHR.pdf

First, some clues of what to read in the WHR paper. In the intro it describes the shortcomings of incremental rating systems such as Elo:

Incremental Rating Systems. ... After each game, this data is updated for the participants in
the game. The rating of the winner is increased, and the rating of the loser is
decreased.

Incremental rating systems can handle players of time-varying strength, but
do not make optimal use of data. For instance, if two players, A and B, enter the
rating system at the same time and play many games against each other, and
none against established opponents, then their relative strength will be correctly
estimated, but not their strength with respect to the other players. If player A
then plays against established opponents, and its rating changes, then the rating
of player B should change too. But incremental rating systems would leave B’s
rating unchanged.

Then discusses more advanced approaches under "Accurate Bayesian Inference"

Authors of incremental algorithms already proposed to correct inaccuracies by
running several passes of the algorithm forward and backward in time

and then WHR:

”. Experiments presented in this paper clearly indicate that he underestimated
his idea: evaluating ratings of the past more accurately helps to evaluate
current ratings: the prediction rate obtained with WHR outperforms decayed
history and incremental algorithms.

Evidence that past ratings change:

Iyama's rating on goratings.org on Oct 15th 2016, as captured by archive.org was 3520:
https://web.archive.org/web/20161015073 ... tings.org/

Iyama's rating on the current goratings.org site (https://www.goratings.org/en/players/601.html), looking in his game history for the relevant period shows his rating is now thought to have been 3533

Code: Select all

2016-10-17	3533	White	Win	Yu Zhengqi	3364	♂	jp	View game
2016-10-12	3533	White	Win	Takao Shinji	3336	♂	jp	View game
2016-10-04	3533	Black	Win	Takao Shinji	3336	♂	jp	View game

3533 != 3520

QED.

(Edit: thanks jlt, corrected)

[jlt]

You mean 3533 != 3520.

Iyama.png (38.05 KiB) Viewed 22325 times

[dfan]

Ridiculous or not, it is absolutely the case that WHR, the algorithm used by goratings.org, changes its opinion about the rating of a given player on a given date once future results are made available to it. There are plenty of valid reasons to regard this as either a positive or a negative feature of WHR, but it is a fundamental property of the system.

(I have implemented WHR.)

[hyperpape]

While it's cool to see Cho U making a bit of a comeback, it's somewhat disappointing that the recent title matches have been less a case of new players rising up to meet Iyama so much as him falling back to the level of the other Japanese pros.

[gowan]

As for Iyama falling back, I think it is to be expected. Even the strong Chinese and Korean pros who win international tournaments don't stay at the top for ever. All of them have a run at the top and then fall back. It's a standard thing for a pro to win a lot of titles an hold them for a while but falling back, as you say, after a while. Cho U is having a comeback but he was at the top a while ago. Rin Kaiho was at the top for a while until Ishida dislodged him. Cho CHikun and Kobayashi Koichi held top titles for long periods of time but fell back. It seems to me to be a matter of being able to put out at a very high level of intensity for a long time. After all there is pressure from fans, journalists, other players, to stay at the top. And these title holders have life outside of go which eventually can make demands. Speaking of that, we don't read or hear much about the lives of our go idols away from the go board. I do recall hearing that Iyama was divorced. All the top international pros eventually fall back: Ke Jie, Yi Changho, Yi Setol, Ma Xiaochung, etc.

[hyperpape]

I see a difference in those scenarios.

Before falling back, Lee Changho had Lee Sedol nipping at his heels. Then Park Junghwan caught up to Lee Sedol while he was still on top of the world. Shin Jinseo has now caught Park, while they're both among the best in the world. Even Cho, though he fell off quite rapidly after 2012, remained quite competitive for a few years: he beat Iyama in 2011 Kisei. In contrast, none of the players taking titles from Iyama seem likely to reach his former level of play, unless they surprise us.

As a fan, it's exciting to see a new upstart make the top player sweat while they're still at the top of their game. It's also bad for Japanese go on the international scene if they can only field one top-20 player in a decade.

[TheCannyOnion]

Holy cow, Iyama has dropped all the way to 55th on Mamumanu's latest ranking.

[jlt]

Iyama is back to the 24th place on goratings.org while Shin Jinseo is 1st.