It is currently Thu Mar 28, 2024 5:20 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 13 posts ] 
Author Message
Offline
 Post subject: Strength difference between pros in Light of AI
Post #1 Posted: Tue Sep 20, 2022 9:43 am 
Lives in sente

Posts: 724
Liked others: 1023
Was liked: 30
Rank: BGA 3 kyu
KGS: Elom, Windnwater
OGS: Elom, Elom0
Online playing schedule: The OGS data looks pretty so I'll pause for now before I change it.
It seems that the difference between female/seniors and top pros is about 2 stones


Last edited by Elom0 on Wed Sep 21, 2022 3:17 am, edited 1 time in total.
Top
 Profile  
 
Offline
 Post subject: Re: Strength difference between pros inb Light of AI
Post #2 Posted: Tue Sep 20, 2022 1:10 pm 
Lives in sente

Posts: 905
Liked others: 22
Was liked: 168
Rank: panda 5 dan
IGS: kvasir
I am not sure exactly what you mean. If you mean the strength difference between top pros and female and senior pros. Well, that has everything to do with where those pros are in terms of their tournament results and other indications of their strength instead of gender and age. For example Rui Naiwei has won a number of titles in China and South Korea and this includes some of the most important tournaments. Maybe she is or was capable of taking on the lower end of the pro spectrum successfully giving two stones.

The bulk of pros are probably closer than this to the top title holders, but I'll admit that this is simply something I'd like to believe :) I think this also depends on the time allotted and how you stretch the definition of what is a correct handicap.

Top
 Profile  
 
Offline
 Post subject: Re: Strength difference between pros inb Light of AI
Post #3 Posted: Wed Sep 21, 2022 1:41 am 
Lives in sente

Posts: 724
Liked others: 1023
Was liked: 30
Rank: BGA 3 kyu
KGS: Elom, Windnwater
OGS: Elom, Elom0
Online playing schedule: The OGS data looks pretty so I'll pause for now before I change it.
I'll admit that when I was younger, soon after I learned go around 12 or so, I'd obsess about the strength difference between pro-level players, especially how close era female pros 'catching up' to male pros or how strong Korean and Chinese female player were compared to, maybe the younger you are the more biased you are to that kind of thing, which is another reason why I think focusing on women's go in the west is the best way improve western go overall, young children can identifytheir future selves with fathers but they can identify their current selves with their mothers, so it's just much much more efficient. Obviously a one-dimensional metric of strength doesn't necessarily convey the nuances of what actually happens on the go board and above it, and isn't as important as STYLE, so I focus more on the latter if I actually want to improve, but it is fun to check a rough view of professional strength levels.

Choi Jung and Shin Jinseo are exceptions stronger than their peers by a similar winning probability, but that means that the handicap between Choi Jung and her peer is greater. The even more important thing though, is that Koreans say Choi Jung is probably stronger than Rui Naiwei at her peak, because they believe professionals have gotten stronger since the 1990's. I find it interesting, since Cho Chikun says he thinks that professional players in the 80's were not as strong as those during the older eras of Japan. It could mean that go strength got weaker in the 60's to 90's, but go stronger again. Rui Naiwei was the only pros who could beat up Lee Chango. Lee Changho's style was similar to Cho Hyeyoen This means that Cho Hyeyeon was double unlucky to have the top female pro not only be a top pro but a top oro with a style she's weak to.

It does seem that in both Korea and Japan, newly qualified shin-shodans immediately make it to the top 150, which is traditionally the range of normal top female pros. If we take the 'special age inverse bracket' to be Under 15 and Over 42, that range is exactly normal range for the best players in that profile two. Don't count Shin Jinseo, he's not normal and half a stone to a stone beyond his Korean counterparts, only the top Chinese pros or Mi Yuting can touch him. Seo Bonsoo's recent games in his special confirm his own statement that he's two stones behind the top Korean Pros. I would go out on a limb here and say in 2010's apart from Iyama the difference between top Japanese players and female players was the same as between Japanese players and top Korean pros, one stone. Now we have Iyama, Shibano, Ichiriki, but then Choi Jung and Shin Jinseo happened. The difference remains unchanged. European and American players seem about as strong as the top female Taiwanese pros, maybe, if you include Xie Yimin, and Asian countries with a strong go playing level, let's call them goAsia seem maybe half a stone stronger to the level of to Japanese female pros discounting Ueno Asami and Fujisawa Sensei. It's very interesting to me that the FeU15O42 group are two stones behind top pros and top pros are two stones behind the. How many pros like Choi Jung and Shin Jinseo erase that difference. Interestingly that two stone difference between the FeU15O42 group seems consistent even in Europe and America.

Top
 Profile  
 
Offline
 Post subject: Re: Strength difference between pros in Light of AI
Post #4 Posted: Wed Sep 21, 2022 3:28 am 
Lives in sente

Posts: 724
Liked others: 1023
Was liked: 30
Rank: BGA 3 kyu
KGS: Elom, Windnwater
OGS: Elom, Elom0
Online playing schedule: The OGS data looks pretty so I'll pause for now before I change it.
Unlike Chess games, in addition to win probability, we have stone handicaps to represent strength. For this reason alone go should be the football of mindsports, the beautiful game.

What I would like do is use AI in a way different to normal. AI allows us to accurately compare the go playing level of different demographics to find which area of demographic diversity are lacking in the go playing population. This is what was in my mind when I was 13/144 and made Surrounding Gardens, so I'll just continue.

Combining the two above statements, we can set the fourth strongest person in the world as having a strength of 9.00 stones, and defining your dan as equal to the next highest integer above. Anyway that will be were the material dans and performance dans align.

Top
 Profile  
 
Offline
 Post subject: Re: Strength difference between pros in Light of AI
Post #5 Posted: Wed Sep 21, 2022 4:54 am 
Gosei
User avatar

Posts: 1753
Liked others: 177
Was liked: 491
In this post https://forums.online-go.com/t/merlijn- ... ll/42659/2 gennan estimates that no komi handicap corresponds to 150 Elo difference (I don't know how he determined that number). Assuming this is correct, then a 2-stone handicap game with no komi corresponds to a difference of 450 Elo.
Excluding Shin Jinseo and Choi Jeong who are outliers, top male pros are around 3700 Elo on goratings and top females 3300, a difference of 400 Elo. This is consistent with Elom's estimate.


This post by jlt was liked by: Elom0
Top
 Profile  
 
Offline
 Post subject: Re: Strength difference between pros in Light of AI
Post #6 Posted: Wed Sep 21, 2022 5:27 am 
Lives in sente

Posts: 724
Liked others: 1023
Was liked: 30
Rank: BGA 3 kyu
KGS: Elom, Windnwater
OGS: Elom, Elom0
Online playing schedule: The OGS data looks pretty so I'll pause for now before I change it.
jlt wrote:
In this post https://forums.online-go.com/t/merlijn- ... ll/42659/2 gennan estimates that no komi handicap corresponds to 150 Elo difference (I don't know how he determined that number). Assuming this is correct, then a 2-stone handicap game with no komi corresponds to a difference of 450 Elo.
Excluding Shin Jinseo and Choi Jeong who are outliers, top male pros are around 3700 Elo on goratings and top females 3300, a difference of 400 Elo. This is consistent with Elom's estimate.


Hmm, that corroborative evidence is very interesting, of course, the elo difference per stone gets a bit bigger the stronger the players are, but if gennen is basing his estimates on the elo difference between top pros then the difference is a bit under two stones. If the Elo estimated are based on EGF 7 dan's compared to EGF 5 dans then a two stone difference from the top pros should be more than 450 Elo . . . But this implies something between 1.5 to 2 stones. Oh wait! When it comes to removing outliers (great book by the way even though everyone's probably read it), in go ranks I always use the fourth best of, Gold, Silver, Bronze and Copper. So at https://goratings.org the fourth highest rated pro, the copper pro, is Park Junghuan at 3665. The copper female is Zhou Hongyu at 3258. 407 points, now I'll check for U15fc and O42fc. Among O42's Kono Rin is at 3312, Among U15's Ma Jingyuan is at 2919 but the youth list will be more prone to the effects of outstanding individuals.

If we can compare the four rating lists of Dr Bae Tail, Xiaodai, Rémi Coulom and Mamumamu0413, we can get more answers. . ! But I don't know where to get Dr Bae Tail's international ratings. In mamumamu0413 2.00 points is 400 elo, and it seems to confirm jlt's estimate, but Xiaodai's ratings are very different for some reason.

Top
 Profile  
 
Offline
 Post subject: Re: Strength difference between pros in Light of AI
Post #7 Posted: Wed Sep 21, 2022 7:05 am 
Lives in sente

Posts: 724
Liked others: 1023
Was liked: 30
Rank: BGA 3 kyu
KGS: Elom, Windnwater
OGS: Elom, Elom0
Online playing schedule: The OGS data looks pretty so I'll pause for now before I change it.
http://www.ranka.intergofed.org/?p=1100 and http://www.ranka.intergofed.org/?p=1300 and https://www.ranka.intergofed.org/?p=1325 show the level of North Korea is capable of competing with Taiwan, but the level of the west can compete with Taiwan.

Top
 Profile  
 
Offline
 Post subject: Re: Strength difference between pros in Light of AI
Post #8 Posted: Wed Sep 21, 2022 9:15 am 
Lives in sente

Posts: 905
Liked others: 22
Was liked: 168
Rank: panda 5 dan
IGS: kvasir
jlt wrote:
In this post https://forums.online-go.com/t/merlijn- ... ll/42659/2 gennan estimates that no komi handicap corresponds to 150 Elo difference (I don't know how he determined that number). Assuming this is correct, then a 2-stone handicap game with no komi corresponds to a difference of 450 Elo.
Excluding Shin Jinseo and Choi Jeong who are outliers, top male pros are around 3700 Elo on goratings and top females 3300, a difference of 400 Elo. This is consistent with Elom's estimate.



There is a well known relationship that handicaps are off by half a stone. That is

rank difference = stones - 0.5

That is, same rank is 6.5 komi, 0.5 rank difference is 0.5 komi, 1 handicap is -5.5 komi and 1.5 rank difference is 2 stones, and so on.

In that sense 2 stone handicap is 1.5 rank difference which is 150 GoR points, if GoR is our arbitrary gold standard.

As for pro ranks, there is no relation between pro ranks and handicap at all anymore. Pro ranks in Japan used to be half-ranks compared to the handicap, then this was changed to third-ranks and about 20 years ago they abandoned their rank tournament. After that handicap, as such, no longer has anything to do with their ranks.

Of course you can multiply 150 by three to get 450 on the premise that GoR and the Oteai are gold standards. This is not the same scale as those online rating lists. Besides being a dead end it is a dubious argument.


If you look for it then there are probably many games that have been played with different handicaps. There was for example a Pro-Ama mixed tournament in 2019. The handicap was "Komi was 6.5 within each group, 2.5 for female pro against male pro, −2.5 for female ama against female pro, and −6.5 for male ama against male pro." You can double check but the women won all their games against amateurs, Shibano lost to an amateur by 12.5 pts. and all the games between male and female pros were won by the males. I don't know if one can conclude that premise was that the differences (for the players in the tournament, not in general) were 1 ama rank for female pro against male pro, 2 ama ranks for female pro against ama female, 3 ranks for male pro against male ama -- probably not!


This post by kvasir was liked by: Elom0
Top
 Profile  
 
Offline
 Post subject: Re: Strength difference between pros in Light of AI
Post #9 Posted: Wed Sep 21, 2022 1:07 pm 
Lives in sente

Posts: 724
Liked others: 1023
Was liked: 30
Rank: BGA 3 kyu
KGS: Elom, Windnwater
OGS: Elom, Elom0
Online playing schedule: The OGS data looks pretty so I'll pause for now before I change it.
Ah, I was thinking in my head of 400 as 150 times 2, my bad, haha

When I talked about two stones, what I was thinking is a player can give a one stone handicap to a player who can give a one stone handicap to you, if you define a stone handicap as any move before the opponent.

However according to what you seem to be saying, whether you have sente after the handicap stone is important, after all, if you give up sente after playing it it's not a free stone, in other words, a handicap stone is a free stone that doesn't give up sente, which is obviously true.

Just thinking out loud here; true correct Komi is a integer. So any fractional points would throw off the value of a stone. The extra point in area rules is simply from black playing last so the Komi is universally considered 6.5 in essence. Which means true correct Komi is either six or seven.

So giving the second player six or seven points makes the game the same as if each turn players sealed a move and revealed it at the same time.

I don't really pay attention to pro ranks

Top
 Profile  
 
Offline
 Post subject: Re: Strength difference between pros in Light of AI
Post #10 Posted: Thu Sep 22, 2022 4:37 am 
Lives in sente

Posts: 724
Liked others: 1023
Was liked: 30
Rank: BGA 3 kyu
KGS: Elom, Windnwater
OGS: Elom, Elom0
Online playing schedule: The OGS data looks pretty so I'll pause for now before I change it.
2 Sentes is equal to one stone, because each handicap stone has both the value of sente and the value of being an extra stone on the board except the stone before white plays, which only has the value of sente. In chess games, letting a player make two moves in a row to develop their pieces better is doesn't put any extra pieces on the board, so that's a two-sente handicap which is one stone.

This means a 1 stone difference is when your even with someone with first move advantage who themselves is even with someone with first move advantage. In other words, double sente. And double sente represents different things in Chess, Shogi, Elephant Chess and IGo. Draughts/Checkers may not even have first move advantage.

So one stone is like one note of music, and sente is a half note, and there're seven notes, 14 half-notes, and there're seven dans 14 sentes.

In a rating system that uses board points and winning probabilities, the difference still remains, since some games are inherently more 'drawry' than others, skewing the gaps between players to make it smaller.

So a rating system that's wants to balance it out needs to take into account this drawryness. a few months ago I thought of multiplying the rating difference by the tendency for a person's games results to approach 0.5

Because go can then be the standard we use to study the FeU15O42 difference in all ganes

My estimate was based on Seo Bongsoo's performance in handicap games against the top Korean pros excluding Shin Jinseo, but given that actually 1.5 Dan handicap, then perhaps we could assume the average elo difference per half Dan between the fourth best pro and 450 elo below is about just under 150; the elo difference getting slightly smaller the further down one goes.

However in Xiaodai's rating the elo difference between pros is a lot bigger than in goratings and mamumamu.

Top
 Profile  
 
Offline
 Post subject: Re: Strength difference between pros in Light of AI
Post #11 Posted: Fri Sep 23, 2022 7:41 pm 
Lives in sente

Posts: 724
Liked others: 1023
Was liked: 30
Rank: BGA 3 kyu
KGS: Elom, Windnwater
OGS: Elom, Elom0
Online playing schedule: The OGS data looks pretty so I'll pause for now before I change it.
http://www.wbaduk.com/news/news_view.as ... div=column http://www.wbaduk.com/news/news_view.as ... div=column

This remains one of my favourite articles.

I believe most of the Korean players were high-level insei. As I said before, new Korean pros are generally at around Seo Bongsoo's ranking, a bit outside the top 100. That's 1.5 Dan difference, 3 sentes, so between 18 to 21 points.

We can add a quarter of a stone to account for the fact that they didn't make pro and another quarter to account for them playing in Europe rather than Korean-level competition. This puts them exactly two stones behind the fourth best player. European 7 Dan's results against them is consistent with a 211 or so elo difference.

150 elo is for a no-komi game then it represents half a stone, and 211 three quarters of a stone,

The thing is, according to https://europeangodatabase.eu/EGD/winning_stats.php https://wismuth.com/elo/calculator.html#elo_diff=-150

The difference between EGF 5 Dan and EGF 6 Dan is 150 elo.

Also, if by 'pro level' Gennan means 7 Dan, then at top pro level the elo shift per stone would be significantly bigger, not by much but significant.

Of course, I expect that the gap between European players and Korean 7 Dan's has closed down to half a stone since 12 years ago, otherwise what's the point of all the CEGO stuff.

According to the article, the elo difference between European 6 Dan's versus European 7 Dan's is 23%, 211 elo. The EGF database also gives 24.7% for EGF 6 v 7 Dan.

We should expect the percentage to be a bit lower next stone up, and the EGF 7 Dan to Korea 7 Dan difference falls exactly in line with 225 elo. If we assume a percentage drop of 3.5%, the next stone up from rusty top Korean insei level, itself close to FeU15O42shinshodan level, is 18% so 262 elo. This covers a free handicap stone in Seo Bongsoo's special games. This leaves about 140 elo left. Well, if we go up 15% that 296, half of that is 148. So it seems 1.5 Dan difference between top pros and new pros in Korea is accurate.

Top
 Profile  
 
Offline
 Post subject: Re: Strength difference between pros in Light of AI
Post #12 Posted: Tue Sep 27, 2022 9:06 pm 
Lives in sente

Posts: 724
Liked others: 1023
Was liked: 30
Rank: BGA 3 kyu
KGS: Elom, Windnwater
OGS: Elom, Elom0
Online playing schedule: The OGS data looks pretty so I'll pause for now before I change it.
I've wondered for a few years about Choi Jung has been having an Iyama Yuta effect on Korean Women pros. This would explain two major discrepancies:

1-Ueno Asami's and Fujisawa Rina's high ranking compared to Korean pros.
2-The fact that the rankings of Korean men players who play in the GGAuction cup are significantly higher than the Women players, yet there results in the GGAuction cup are more or less even

The theory is that the rating system over estimates the probability of the weaker player being able to win. Or rather, there is a certain skill difference where within it, it under estimates the weaker players ability to win, but above it, it over estimates the weaker players ability to win. That's why whenever you have a standout player their rating suddenly shoots up. This leads me to the belief that maybe Choi Jung has been overranked in a similar way Iyama Yuta was when he had no competition in Japan, since in both cases they played a significant proportion of their games against players who are weaklings compared to them, and perhaps Magnus Carlsen could be added to this effect until recently. And perhaps these kind of discrepancies in general are due to maybe not using a pure logistical distribution? The rating system Xiaodai proposed seems more immune to this kind of effect, and mamumamu0413 seems almost as good. Although Choi Jung hasn't been playing as well recently, so this effect is disappearing a bit.

Top
 Profile  
 
Offline
 Post subject: Re: Strength difference between pros in Light of AI
Post #13 Posted: Wed Nov 30, 2022 4:15 pm 
Lives in sente

Posts: 724
Liked others: 1023
Was liked: 30
Rank: BGA 3 kyu
KGS: Elom, Windnwater
OGS: Elom, Elom0
Online playing schedule: The OGS data looks pretty so I'll pause for now before I change it.
Okay so obviously 최 정's recent return to form means that aspect is obsolete

From what I understand, with some clarification from Gennan on OGS, is currently that Komi is six points, and a difference of one stone is equivalent to 12 points. 촤헤옌 said that 최정 was above than Rui Naiwei's level of 3 stones ahead of top female pros, which is a surprisingly large difference to me, since as I said earlier that's the difference between 서본수 and the fourth strongest pro in the world so I'm not sure how that's possible.

Assuming that she's about a stone behind the fourth best pro in the world, and the fourth best placed women player in the world is about as good as a Korean 1p's that qualify through the open exam, then that implies a difference of about three stones between an average Korean shodan and the fourth best pro in the world which to me seems a bit much but maybe the gap became that big after AI? The Seo Bonsoo based estimate is two stones which seems more realistic. Let's just go for 2.5 stones. According to the Wbaduk article by Hwang In-Seong Korean 1p's are probably usually around EGF 2800, so a mid level EGF 8 dan. So the fourth best pro in the world is EGF 3150?

So the correct one stone handicap is giving black 5.5 points reverse Komi in Japanese rules or 4.5 points reverse Komi in Chinese rules?

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 13 posts ] 

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: Reasonablepower, TegaiS and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group