|
Vatican City State72 Posts
There's been a lot of talk about Soulkey's recent dominance, with some people even comparing him to FlaSh and Artosis saying he could be the next Bonjwa. So I wanted to put it into perspective by comparing his current peak with, for instance, Snow's earlier this year.
I downloaded all the results for the top 20 players from eloboard, then calculated:
1. Their Elo score after each match.
2. Their Elo score by the end of each day.
3. The moving-average Elo score of the 28 previous days (measured at the end of the day). I chose a multiple of 7 so that all windows have the same number of weekends.
The results are mostly unsurprising, but I think they help put in perspective and to some extent quantify the average performance and consistency of players over time.
To no one's surprise, the results show that Snow has consistently had the best performance of all players for the past two years, even managing to keep an average Elo score above 1500 throughout more than 28 days.
When Soma was active, he and Light were roughly tied as the best-performing players after Snow. Somewhat unexpected was to see Bisu consistently as the second-best Protoss, above Mini. Also expected, Rush has been the second-best Terran for most of the last two years.
Protoss:
https://imgur.com/SekMOyv
Terran:
https://imgur.com/unqExXd
Zerg:
https://imgur.com/rrVk28t
Top two of each race:
https://imgur.com/qKKJDAK
Edit: Images are not showing, so I also included the link to them.
|
Keep in mind that elo in the current streaming era is VERY different to Kespa's elo. Which is why I once said rankings by elo at a specific point in time may not be reliable.
The main reason is every game a player plays on stream now is recorded, while in the actual progaming era, only official games in tournaments were recorded. This leads to inflation/deflation of elo depending on how active a player is at a given time because you can actively choose your opponent now.
Think Barracks playing Ggaemo 100 times on stream in a week, beating him 90 times. He'll probably be no1 in elo of that week. You couldn't do that in Kespa era.
Soulkey's elo could have been much higher had he been more active in the past two years but he's not the most "diligent" player, most of the times having the least amount of games (along with Bisu) out of the top 10-15 players. He also offraces quite a lot on the ladder, and of course plays League very often.
|
Vatican City State72 Posts
Yes, there will always be the caveat that the conclusions can only be as good as the data, but that's all we have really, unless we restrict the analysis to ASL/KSL matches, but then the sample size is too small and and the data too sparse.
On August 26 2024 04:59 TMNT wrote:
Think Barracks playing Ggaemo 100 times on stream in a week, beating him 90 times. He'll probably be no1 in elo of that week. You couldn't do that in Kespa era.
Not really. Beating a 1100 Elo player 1000 times won't get you too far, because by the point you reach 1300 (let alone 1400), the points you get by beating Ggaemo are almost negligible, and in contrast, losing one single time to a 1200 player will make you lose ~20 points.
For an actual real example of what I'm saying, you can see Soulkey's score this week. He got only 2.3 points for beating nOOb a few days ago (currently at 1,027 Elo score). Even beating Rain (currently at 1273 Elo) scored him only 6.9 points. In contrast, losing just one match to Rush today made him lose 27.3 points, an also lost 28.5 points to Shuttle. In total, he's lost 120 points in a single day. Going back to 1500 Elo beating players with 1000-1100 elo score would take weeks, assuming he doesn't lose again to any mid-level player.
Thus, farming Elo with low level players can get you only so far because of diminishing returns. And in any case, I don't think that at least any of the top 6 players of each race (on whom I focused the analysis) spend most of their time farming Elo by beating low level players. And even if they did, losing just one 1 out of 20 matches would sort of even the scores --if the difference in Elo scores is too big, a player could easily lose 30+ points per loss, and win less than 2 points per win.
|
No. That's not what I mean. I'm not talking about boosting elo (inadvertently) by farming much worse players. I'm talking about the disproportionate amount of games between players even at the same level.
For a real example, take a look at Kim Seok-hyun aka SRich - an amateur player who has never made ASL (http://eloboard.com/men/bbs/board.php?bo_table=bj_list&wr_id=127). Here's his record since July:
+ Show Spoiler +
His peak elo was on July 27 when he was ranked 14. Is there any chance he was the 14th best player at that point in time? The player behind him on that day: Light, Rain, Mind, Royal, JD, Mong, Queen,.... There's no way those players even in their slump are worse than SRich. And SRich achieved that elo by beating Yabsab 7-0, 815 4-1, HBQ 7-1, and Royal 2-1 (the results against Royal netted him 17 pts, without which he would still have ranked 18).
See the problem now: he wasn't necessarily farming weaker players, he was just beating players close to his level a lot by playing them repeatedly (think Snow beating Royal or Soulkey beating Mini). And the problem could be further compounded by players prefering to play their strong/weak matchups more. And also, they never play mirrors for spon games (the only times for mirrors are during daily proleagues).
Kespa's elo is much more accurate because you could neither choose your opponent nor matchup, and you could only play one player a limited number of times, and that applied to everyone.
|
Vatican City State72 Posts
On August 26 2024 08:03 TMNT wrote:
His peak elo was on July 27 when he was ranked 14. Is there any chance he was the 14th best player at that point in time? .
I clearly said repeatedly "28-day moving average". Moving averages are precisely used to smooth this kind of erratic data. I would give you a link to read about moving averages and how and what they are used for, but you find providing sources offensive.
In any case, I've added SRich to the plot of Terran Elo scores, and as expected, his 28-day moving average is much lower than his maximum peak.
|
Could you do one of just 2023 and one of just 2024 with shorter interv averages?
|
On August 26 2024 08:03 TMNT wrote:
No. That's not what I mean. I'm not talking about boosting elo (inadvertently) by farming much worse players. I'm talking about the disproportionate amount of games between players even at the same level.
Yes this is true but there is only a certain extent to which the ELO system can be abused this way.
If you repeatedly beat bad players, your ELO score only climbs up VERY slowly. If you repeatedly beat players with ELO scores slightly worse than yours, your ELO will climb a bit faster. But realistically speaking you should lose some games to players slightly worse than you, so your ELO score should update itself naturally.
And a player like SRich is playing many games on Ladder, I assume. Are Ladder games included on Eloboard stats if you know who the other player are? If yes, he should be losing some games to better players right? In which case his ELO score is naturally adjusted, again.
|
On August 26 2024 12:20 namkraft wrote:Show nested quote +On August 26 2024 08:03 TMNT wrote:
No. That's not what I mean. I'm not talking about boosting elo (inadvertently) by farming much worse players. I'm talking about the disproportionate amount of games between players even at the same level.
Yes this is true but there is only a certain extent to which the ELO system can be abused this way. If you repeatedly beat bad players, your ELO score only climbs up VERY slowly. If you repeatedly beat players with ELO scores slightly worse than yours, your ELO will climb a bit faster. But realistically speaking you should lose some games to players slightly worse than you, so your ELO score should update itself naturally. And a player like SRich is playing many games on Ladder, I assume. Are Ladder games included on Eloboard stats if you know who the other player are? If yes, he should be losing some games to better players right? In which case his ELO score is naturally adjusted, again.
Ladder games not included.
|
Stork and Bisu are converging hard om that ELO board D:
|
On August 26 2024 09:45 cheesehuehue wrote:Show nested quote +On August 26 2024 08:03 TMNT wrote:
His peak elo was on July 27 when he was ranked 14. Is there any chance he was the 14th best player at that point in time? . I clearly said repeatedly " 28-day moving average". Moving averages are precisely used to smooth this kind of erratic data. I would give you a link to read about moving averages and how and what they are used for, but you find providing sources offensive. In any case, I've added SRich to the plot of Terran Elo scores, and as expected, his 28-day moving average is much lower than his maximum peak.
Easy there genius. Maybe I should provide you with a link of the Dunning Kruger effect?
At no point have I said anything to contradict your graphics and at every point in my posts I stressed that elo may not be reliable if you look at it at a given point in time.
|
And to continue with the issue, look closer at the example given above from SRich:
http://eloboard.com/men/bbs/board.php?bo_table=bj_list&wr_id=127
Looking at his entire history, he gained ~ 350 points by playing ~ 120 games vs HBQ (with a win rate of 62%) and ~ 80 games vs Alen (80%). That's 200 games with just two players vs whom he had very positive win rates. The majority of the remaining players, he's only had 10-20 games each. Now you'd think if SRich beats Alen 80% of the time, at some point his wins won't gain him much, but the problem is Alen may have done the same thing as SRich (beating someone else repeatedly) to keep his elo at a respectable level.
Then this issue (with preferred practice partners) may be further accentuated by preferred matchups. To put it simply in a hypothetical example, if Bisu plays mostly PvZ, he'd have a much higher elo than that when he plays PvZ and PvT equally.
These are things you can do now but couldn't do in Kespa, as you could only play who was put in front of you.
Moving average can smoothen your data but it doesn't solve the above issue.
|
Just gonna put it out there that Snow's peak earlier this year was something special.
|
I am just happy to see that we have 3 races represented with top3 players (SK / Snow / Light).
|
Bit sad that sponbbang fell off. merging stats would be interesting but perhaps impossible to do.
|
Whatever happened to soma?
|
On August 26 2024 22:00 Simplistik wrote:
Whatever happened to soma?
military service until summer 2025. started in January.
|
Vatican City State72 Posts
On August 26 2024 10:09 RJBTVYOUTUBE wrote:
Could you do one of just 2023 and one of just 2024 with shorter interv averages?
Here they are. This time I let the Y axis have different ranges for each plot, so that the differences are more clear.
All races, 2023:
Protoss 2023:
Terran 2023:
Zerg 2023:
Top 2 per race, 2024:
Protoss 2024:
Terran 2024:
Zerg 2024:
|
This data does track pretty well.
The striking notes are the following: Snow is very consistent because he is great across matchups. This is not the case for any of the other Ps: Best is straight up bad at PvP, Mini is wildly inconsistent in any of the non-mirror matchups, Stork is streaky in both PvT and PvZ, Bisu is basically Stork but with exceptional PvZ(still).
And this actually goes for all races: Snow, Light and Soulkey all have the same profile: they're all very consistent in all matchups.
The separating factors are: Soulkey is by far the best mind the game has (except Flash), Light is probably second and after that there's a huge gap.
Below that we have outliers such as: Bisu's PvZ performance, Best's PvP, Light's PvT, Snow's PvP etc
The outliers kind of decide the margins outside of the constants described earlier.
|
It feels weird to talk about bonjwas after the kespa era imo. I know Flash just came back but i feel Soulkey would need to really edge him AND be one head above everyone else for the discussion to even start.
It’s hard today to fathom how much Nada, Boxer, or SaviOr dominated when they became bonjwas.
|
It feels natural to have a rivalry between Soulkey and Flash in the twilight years of the pro-scene. And a shame that Protoss never had a briliant mind like that in the last 15 years, except for some flashes from Rain.
|
|
|
|
|
|
EPT![[image loading]](https://i.imgur.com/SekMOyv.jpeg)
![[image loading]](https://i.imgur.com/unqExXd.jpeg)
![[image loading]](https://i.imgur.com/rrVk28t.jpeg)
![[image loading]](https://i.imgur.com/qKKJDAK.jpeg)
![[image loading]](https://i.imgur.com/etTYdj5.png)
![[image loading]](https://i.imgur.com/sAqugXR.jpeg)
![[image loading]](https://i.imgur.com/5SvNwOJ.jpeg)
![[image loading]](https://i.imgur.com/fMrU9JN.jpeg)
![[image loading]](https://i.imgur.com/bSwXHX7.jpeg)
![[image loading]](https://i.imgur.com/2QDlwRU.jpeg)
![[image loading]](https://i.imgur.com/W58vkph.jpeg)
![[image loading]](https://i.imgur.com/KzdSYRB.jpeg)
![[image loading]](https://i.imgur.com/StqgMcb.jpeg)
![[image loading]](https://i.imgur.com/L6uequ9.jpeg)
