I have been a little disappointed with goratings.org since it rates Ichiriki so high! E.g., he was in the top 10.
So I created my own rating system based on the last 365 days of games refitted every day.
See Welcome to Baduk Go Weiqi Ratings | baduk-go-weiqi-ratings
Comments welcome.
Why is it a problem for you that Ichiriki Ryo is rated high?
What value does a “last 1 year only” rating window have over this?
Did you make this only so that Ichiriki wouldn’t be in the top 10? Did you choose a window such that he wasn’t?
WHR is a new rating algorithm that directly computes player ratings as a function of time.
Why isn’t it a problem?
WHR clearly doesn’t work cos Ichiriki is nowhere near as high. He gets beaten by many Chinese players easily.
Hmm, makes me think of Kellyanne Conway’s “alternative facts”.
Conway later defended her choice of words, defining “alternative facts” as “additional facts and alternative information”
This was always so obviously what she meant, and the absolute IDIOCY of the entire media network and the general populace to gawk and jeer as if she’d said something weird was so painful to endure. Truly one of humanity’s most stupid moments.
Alternative transport is still transport.
Alternative facts are still facts.
Not rocket science people.
Ichiriki Ryo doesn’t “get beaten by many Chinese players easily”. I looked at his matches against Mainland Chinese players since october 2019: the list below features the ratings of his Chinese opponents and the result of each match.
3660 W
3623 L
3604 W
3598 W
3572 L
3572 L
3571 L
3564 L
3544 W
3535 W
3419 L
3283 W
3262 W
He is currently rated 3561 and #17 on goratings, this doesn’t look unrealistic.
he was top 10.
The difference between #10 and #17 is just 12 rating points. People’s rating go up and down as they win or lose games. You can’t say that the rating system is wrong if a player was a bit overrated for a few days, you have to look at longer periods of time.
What does it matter? In today’s Go there are Shin, Park, and Ke… and then everyone else. 10th? 17th? The gap between 1st and 10th is the same as the gap between 10th and 140th.
17th is way overrated for Ichiriki. Anyway, him being rated so high is just a symptom.
Look at the table above. He wins about 50% of his matches against Chinese players of similar gorating, so his rating seems very realistic to me.
not based on my model. you do your analysis like that and I will keep using my data science/machine learning skills to maintain my ranking
I am more interested on how you calculate your rating as who is the best. If you have some more time to give to us. Fot example what are the differences with gorating you put in it?
Well, you rate Ke Jie #6, I think he is way underrated in your system so I could as well jump to the conclusion that your computation method is flawed.
fair point. but I think it’s agreed that ke Jie had a pretty bad patch recently. My ratings are based on last 365 days of games, which I deem to be reflective of a player’s strength. So yeah.
gorating has their methodology detailed in a paper.
my methodology is just to take last 365 days of games, and optimize this
sum(over g logloss(logit(S_x-S_y), game_result_x_y_g)) where x and y denote the players and subscript g denotes the game where player x and y plays.
Basically just a logistic regression on the players’ strengths on the last 365 days worth of games.
I find this method has a high chance of bumping players rarely play unreasonably high, like Lee Jihyun, who just participate in one tournament last year of 6 games only, and happened to win Park Junghwan one game in tiny margin (+2.5) raise all the way to rank 33.
Someone with high rating uncertainty can go up and down that quickly just after one tournament doesn’t feel very stable and reliable to me.
Another example is Hua Chang who just entered the pro less than 2 years ago and participated only in the Longxing preliminary surpassed Zhao Chenyu after 4 games is just unreasonable.
I don’t think you’re being scientific here: you start with the biased expectation that Ichiriki Ryo should be ranked lower, and then create a new rating system that reflects your expectations. That’s not how science works…
Why would your methodology be better than that of goratings? You can’t just “claim” to be an expert and therefore your version is better 
Also, what has machine learning to do with this?!
I think I might have found the issue for players who rarely play got their unreasonable high rating
Pan Tingyu Games | baduk-go-weiqi-ratings (rank 98 which is insane)
There seem to be some rating missing or reset to 0 at some points in the records, and how is someone who only played with players who don’t seem to have records in the system get their rating set to?