Life In 19x19

ez4u wrote:Wow! With that kind of flexibility on display I am worried you might be outgrowing L19.

Nah, I've always been flexible. You can just use this discussion as a datapoint to reassess your view of my past disposition

Thanks for the go rating data feed provided by Remi, I am now able to present the rating data inside Go4Go.net, together with the photos of the top ranked players. This is an ongoing work as I am still compiling and editing the photos which is quite time-consuming. And apparently there's a bug in my responsive theme somehow the page layout isn't right in some browsers. But I am making it available now anyway because this is a very interesting work.

http://www.go4go.net/go/players/rank

Or find it from the Go Players > Top-ranked Players menu.

I predict that Ke Jie holds onto #1 for good this time. The next 3-4 years at least.

Remi, can you please check why a rather strong player Lee Yeongkyu is missing from your list? He has 454 games in Go4Go database.

macelee wrote:Remi, can you please check why a rather strong player Lee Yeongkyu is missing from your list? He has 454 games in Go4Go database.

http://www.goratings.org/players/303.html
He does not appear in the current rating list because he has played no game in more than one year. He still appears in the historical ratings lists in the years when he was active.

OK. He might be in his mandatory service in the army as many have to do.

macelee wrote:OK. He might be in his mandatory service in the army as many have to do.

From here his profile (https://www.baduk.or.kr/info/player_vie ... e=10000261):

12.26 논산훈련소 입소 - >공익근무요원으로 복무

So I guess he's serving as a public service worker. I don't know about his particular situation, but typically you have to serve for 2 to 3 years. Since he entered at the end of 2013, maybe we can look forward to more games from him in the near future.

I think strong players are very unlikely to just go away and stop playing. They return sooner or later. Maybe it is sensible to just keep them in the main list but noting the date of his last available game.

How about an option or different page that includes all players?

macelee wrote:I think strong players are very unlikely to just go away and stop playing. They return sooner or later. Maybe it is sensible to just keep them in the main list but noting the date of his last available game.

Of course strong players (and weak players) routinely go away and stop playing. It is called getting old ( ) and happens far more often than players stopping and then coming back. An 'active' list makes sense to me. The Chinese pro rating list operates that way although I have no idea what parameters are used.

ez4u wrote:

macelee wrote:I think strong players are very unlikely to just go away and stop playing. They return sooner or later. Maybe it is sensible to just keep them in the main list but noting the date of his last available game.

Of course strong players (and weak players) routinely go away and stop playing. It is called getting old ( ) and happens far more often than players stopping and then coming back. An 'active' list makes sense to me. The Chinese pro rating list operates that way although I have no idea what parameters are used.

The Chinese system are simple - players need to register every year with the Go Association if then intend to keep playing. If they don't, they are consider inactive and move to another list.

There is another factor to consider. The Chinese list is calculated in such as way that new games won't affect old ratings. This isn't the case in Remi's algorithm.

I increased the inactivity threshold to 1000 days, so he is back into the list.

Hi Rémi, I wonder if you are able to conduct an experiment using your algorithm.

Now Iyama Yuta is in red-hot form, winning all his games over the past three months. This put him at No. 5 on your list (in fact he temporarily went to No. 4 above Shi Yue two days ago). I wonder how reliable your algorithm reflects his real strength. After all his games are mostly against opponents significantly weaker than him. And each year he only plays very few games against strong Korean and Chinese players.

My data does not support Iyama being at top 5. His score against the remaining 19 players in top 20 is a pathetic 12-21 (36.3%). It is also quite surprising that he never played any games again 6 of the top 20 players.

I believe your rating list quite realistically reflects the relative strength among Korean and Chinese players. They so frequently play each other, not just during international events, but also the Chinese league. With large sample size, the dynamics of your algorithm should work really well. But because there are much fewer games connecting Japanese players with the rest, the Japanese ratings are harder to justify.

I propose you run the following experiment. Create a number of (say 50 or 100) fictional games between top 10 Japanese players (not including Iyama) and Korean/Chinese opponents of similar ratings. Randomly generate the outcome. The purpose of this is to make more links between Japanese players and the rest. Then recalculate everything to see if this has any impact on Iyama's rating.

macelee wrote:Hi Rémi, I wonder if you are able to conduct an experiment using your algorithm.

Now Iyama Yuta is in red-hot form, winning all his games over the past three months. This put him at No. 5 on your list (in fact he temporarily went to No. 4 above Shi Yue two days ago). I wonder how reliable your algorithm reflects his real strength. After all his games are mostly against opponents significantly weaker than him. And each year he only plays very few games against strong Korean and Chinese players.

My data does not support Iyama being at top 5. His score against the remaining 19 players in top 20 is a pathetic 12-21 (36.3%). It is also quite surprising that he never played any games again 6 of the top 20 players.

I believe your rating list quite realistically reflects the relative strength among Korean and Chinese players. They so frequently play each other, not just during international events, but also the Chinese league. With large sample size, the dynamics of your algorithm should work really well. But because there are much fewer games connecting Japanese players with the rest, the Japanese ratings are harder to justify.

I propose you run the following experiment. Create a number of (say 50 or 100) fictional games between top 10 Japanese players (not including Iyama) and Korean/Chinese opponents of similar ratings. Randomly generate the outcome. The purpose of this is to make more links between Japanese players and the rest. Then recalculate everything to see if this has any impact on Iyama's rating.

Hi,

WHR is designed in such a way that these additional games would not change the ratings if the results are generated according to the current ratings. Iyama's rating is based mainly on his games against Japanese players. If the ratings of the Japanese players don't change, then his rating won't change either.

What I could do is display the likelihood of superiority matrix.

You can see such a matrix at the bottom of that page:
http://www.computerchess.org.uk/ccrl/4040/

Likelihood of superiority of Japanese over non-Japanese should rarely be close to 100%, because they have so few international games.

I am busy with much more important projects now, so it won't be in the short term.

WHR is described in public--I even believe there might be open source implementations. I've often thought it would be interesting to do simulations on separated populations that play each other infrequently--how will your predictions do? In particular, what happens to a player like Iyama who dominates his local competition?

But you'd have to, y'know...actually run those simulations, and I haven't done that.

Another test would be to see the accuracy of predictions with regard to international matches. Is it within the normal range?

P.S. Surely there's a timing problem with citing your data. Both goratings and Dr. Bae Taeil's data show Iyama getting even stronger in the past year or two. But his international record reflects more than 5 years of play.