Statistical Bonjwa Theory: the History of BW

d_ijk_stra

United States36 Posts

July 31 2012 06:19 GMT

-1. Introduction

I guess every StarCraft: Brood War (BW) fan is looking forward to watch JangBi vs. Fantasy final match this weekend. But after that game, sadly that's it! We will not be having OSL (Ongamenet StarLeague) in BW anymore. As I am a huge SC2 fan as well, it is exciting to have OSL in SC2. But man, I have watched BW progaming for more than 10 years and it is torturing me to finally see its end. I couldn't let it go like this; I had to do something.

Because I am a Statistics student I did some statistical analysis, mainly to answer the following question: 'Who is the fifth bonjwa?' It is indisputable that Lim-Lee-Choi-Ma (BoxeR, NaDa, iloveoov, and sAviOr) were bonjwas, who showed dominating performance in their respective peak era. But who succeeds sAviOr, the fourth bonjwa, has been a hot issue in BW community. And I thought this should probably be the best time to attempt to answer that question, as we only have 3~5 matches left. (although proleague is still going on...)

Before we start, I wrote the Korean version of this article first. Without question my Korean is much smoother than my English, so those who can speak Korean may refer to here

0. Statistical Model and Data Collection

First of all, let me provide you some background about how it is possible to use Statistics to answer this 'bonjwa question'. For those who hate math & stat, you may skip to the next section. But don't worry too much, as I wouldn't go into technical detail here. I just want to make it clear how I came up with those numbers, so that if you are suspicious you can study yourself and criticize it if you feel so.

In games like Chess and Go, which two players challenge against each other to determine who is the winner, probably the most famous rating system is the Elo rating system which is named after the Physicist Arpad Elo. This Elo system is actually also used in TLPD to rank both BW and SC2 gamers. But the system I used is the extension of Elo, called TrueSkill Through Time (TTT). Compared to the Elo system, this TTT system models how the skill of a player changes over time more explicitly. Actually, it was proposed by researchers in Microsoft to investigate the history of Chess; and as a SC nerd, I thought it would be fun to apply it to BW matchup records.

I used this TTT system with almost no modification, so you may refer to the linked paper about assumptions I had to make for the analysis. I have to admit that the model relies on heavy assumptions. (So again there will be people claiming 'Statisticians are liars!') But using a statistical model such as Elo or TTT is often more useful than just plotting descriptive statistics such as win-rate each month, as the statistical model fully utilizes the information in our data while descriptive statistics may exaggerate certain aspect of data, possibly on purpose. At least in my opinion this TTT model is the standard way a Statistician may formulate the problem.

I used Team Liquid Progaming Database (TLPD) BW database from 1999 September 25th PKO matches to 2012 July 24th Proleague matches for analysis. The original TTT paper uses each year as a time window; as BW has much shorter history than Chess, I used each month as a time window.

The TTT model evaluates the skill of each player at each month as a quantified number. Let me just call it rating. Actually the system gives an estimate of mean and standard deviation of the rating, but if I plot both of them simultaneously the graph becomes messy. So I just defined rating as rating=mean - standard deviation * 3, which is a conservative estimate. When a gamer retires and we do not have any record of him anymore, then the uncertainty about our information on his skill increases, so the rating gets lower.

Also, this TTT model assumes the skill of a player changes slowly, so when we find a surprisingly strong rookie the system thinks he has secretly honed his skill behind the scene. Therefore early outstanding players such as Grrrr... is overshadowed by later gamers such as BoxeR. However, this is just because we don't have enough records in the beginning, and as we accumulate more and more matches the model stabilizes quite quickly.

Enough about statistical mumbo-jumbos! Let's talk about the game now!

1. The Emperor rises, but the Genius takes him over

So, let's see TTT ratings of top 5 players from 2000 to 2003! (of course there are more than 5, as the ranking surely fluctuates)

[image loading]

'Emperor' BoxeR is increasing his skill very rapidly, and 'Kong' YellOw is right behind him! Just at the moment YellOw overcomes BoxeR, however, 'Genius' NaDa suddenly appears and takes BoxeR's crown instead. (What a Kong!) We can find good ol' gamers such as ChRh, TheMarine who is now working in OGN, Sync... and actually H is H.O.T Forever! Sorry for the mistake in the legend.

2. Oh my god! It's a Monster!

[image loading]

Then the 'Monster' iloveoov takes over NaDa's throne. Interestingly, the period of time iloveoov dominated the scene wasn't that long compared to my perception. His skill starts to decline quite soon after the peak and he catches up NaDa's slow declination. Maybe this shows how outstanding gamer NaDa was, in terms of the length of time he remained as a top player.

In this era, there were too many outstanding gamers, so the graph becomes quite messy. But I just wanted to include all of them, as many of you would want to see them as well. I am sorry I couldn't include more...

3. "He-Who-Must-Not-Be-Named"

Note that I am not using a consistent color/style scheme; the color for each player changes over time. I am sorry, but I am a lazy fellow...

[image loading]

The next Bonjwa is the one "He-Who-Must-Not-Be-Named", but this is too lengthy to type so let me just call him sAviOr. For a brief amount of time Midas becomes the rank 1 gamer, but sAviOr suddenly rises in this era of Terran-dominance and shows unbelievable performance. He was really great at that point of time... Too bad he made a wrong choice. Actually I thought Bisu had no chance against sAviOr, and didn't even watch the match, missing the revolutionary 3.3 moment.

4. The era of TBLS

[image loading]

After his striking 0:3 loss in 2007 March 3rd, the rating of sAviOr plummets and TBLS (Taek-Bang-LeeSsang, or Bisu-Stork-Flash-Jaedong) starts to dominate the scene. From here it is easy to plot top 5 rank players, as there is hardly other. Only Sea was capable of competing with TBLS, but after 2008 even he is no longer able to continue to do so. Jaedong actually continues to be ranked 1st for quite a long time, but Flash follows him right after, therefore it is hard to say whether Jaedong was a bonjwa or not. Maybe this ambiguity was the reason why we discussed so much about who was the fifth bonjwa.

5. The heyday of 'The Last Weapon'

[image loading]

While the performance of Jaedong declines, Flash rises further up. It looks his rating was as outstanding as sAviOr's bonjwa days... Bisu is keep torturing us, by giving us some hope but then crushing it again and again... Fantasy is slowly rising and JangBi is probably experiencing 'Do you know how to play game?' era...

6. And... the end

[image loading]

So this is the last graph, plotting until 2012 July. It looks like recently Flash has not been as dominating as he was before. And finally in the last two months Fantasy achieves the rank 1! Slow and steady paid at last, but maybe it was a little bit too slow; he could be the next bonjwa; who knows? But he is not given the chance...

7. Conclusion

And at this point, you may ask me 'So who is the fifth bonjwa? It's of course Jaedong, right?' Well... I showed you numbers and plots! :-D I am a Statistician, after all, and I always avoid responsibility! It is up to you how to define what bonjwa is and determine who should be the fifth bonjwa.

I am sorry I could not include more players, as I focused my analysis on top 5 rank players for each era. But we should not forget those who made us laugh and cry, showing incredible matches and giving us unforgettable memories. Thank you BW gamers!

Also, thank Team Liquid for the maintenance of perfect database system. Without TLPD I could not even imagine of conducting this analysis. I hope I am not abusing your system here... Teamliquid Hwaiting!

Thank Josh Menke for helpful comments and discussions. Actually Josh did a very similar analysis long before this, and because he showed me his wonderful results I was not afraid of investing my time on it. While the credit of this work should be given to him, all the blame should be given to me because I did not ask him to verify this post.

Thank you very much, you BW fans, for enduring this horrible broken-English writing, and supporting the BW progaming with me for quite a long time!

Lastly, every opinion expressed here is my personal one and has nothing to do with the official position of Blizzard Entertainment. Still, why don't we also thank Blizzard for developing this wonderful game, following up with constant patches and maintaining the Battle.net server for more than 10 years without any further fee after the box sales? I believe SC2 and its expansions will give us unforgettable wonderful memories as well.

p.s. "Bonjwa" is not a common word; it is never used in day to day Korean. It is actually mostly used in certain fantasy genre in Korea, as a first-person pronoun to brag himself. 본좌 is written 本座 in Chinese letters. 本 denotes 'me' or 'myself', and '座' means 'seat' and its meaning can be extended to 'status'. So 본좌(本座) means 'a person like me with such a status!' So using 'Bonjwa' to refer to a third person is actually abusing the original meaning of the word, but I would not oppose it as I guess it acquired a new meaning in BW community.

8. More plots

Lim-Lee-Choi-Ma + Jaedong, Flash lifetime ratings

[image loading]

TBLS

Six Dragons

[image loading]

krndandaman

Mozambique16569 Posts

July 31 2012 07:42 GMT

--- Nuked ---

Ciryandor

United States3735 Posts

July 31 2012 07:51 GMT

Interesting how Bisu's consistency in PL made him much more relevant as a player than his mediocre individual league results after winning his MSLs; and how Stork's performance has only suffered since his 0-3 vs Fantasy.

xxpack09

United States2160 Posts

July 31 2012 08:21 GMT

On July 31 2012 16:42 krndandaman wrote:
nice write up !
imo the only reason flash and jaedong aren't called bonjwa's is because both of them were too damn good at the same time.
if they reached their respective peaks during different time periods both of them would probably have been named bonjwa's.

Uh, I'm pretty sure most people call Flash a bonjwa--there was definitely a long period of time where he was the undisputed #1

Drazzzt

Germany999 Posts

July 31 2012 08:55 GMT

Thx for this nice study and your effort.
The problem with such a statistical approach, however, is that it only takes into account the win rates and not the ability of a player to win important matches and, thus, titles. If you compare the curves of Boxer/Nada and Jaedong/Flash (Plot 8a) they look very similar regarding their rising flanks (of course jaedong/flash is much higher). Boxer and Nada are of course Bonjwa, jaedong/flash are not undisputed.
So, the difference might be that
1.) winning titles is more important than win rates (or flash or even jaedong would easily be bonjwa)
and
2.) winning titles without another bonjwa candidate winning titles in the same year (or period of time) is also very important (jaedong won 2007 and 2x 2009, Flash won 2008 and 2009 and 2010). [i was only referring to OSLs]

However, in my opinion jaedong and flash were both bonjwas, similar to boxer and nada.

endy

Switzerland8970 Posts

July 31 2012 09:03 GMT

Great article, thanks !
Very interesting perspective.
It's really a pity to see BW ending this way, I wouldn't have been surprised to see Fantasy become the 6th Bonjwa. Or maybe Jangbi is gonna win on Saturday, and he would have maybe won even more OSLs after... Damn, it's impossible not to be sad recently.

I totally agree. If any one of them could have played without the other in its way he would have been the biggest bonjwa ever. They were just both so so good.

lichter

1001 YEARS KESPAJAIL22272 Posts

July 31 2012 09:12 GMT

An interesting addition to your graphs might be to mark when players win a major championship or place 2nd. It'll be interesting to also see whether players' successes occur during their rise, while at their peak without significant competition, at their peak with significant competition, or even during a slump or decline.

I'll try to study the Trueskill Through Time algorithm if I have time and see if I can think of a better way to apply it to Starcraft (operations research background)

Baozer

Sweden159 Posts

July 31 2012 09:13 GMT

Well according your graphs Flash had a total of three years where he were in total dominance, which is clearly more then anyone else... That should _statisticly_ qualify his as a bonjwa?

Kleinmuuhg

Vanuatu4091 Posts

July 31 2012 09:17 GMT

Just because JD's peak was during the time I joined BW I am 100 % convinced he deseveres to be called bonjwa.

Plexa

Aotearoa39261 Posts

July 31 2012 09:39 GMT

#10

Keep in mind that Boxer's TLPD doesn't take into consideration a good number of smaller korean tournaments that he crushed (iirc they weren't on TV though).

No0n

United States355 Posts

July 31 2012 09:41 GMT

#11

Just looking at the statistics, Flash SHOULD be the fifth bonjwa, but I think the problem is the term "bonjwa" started and ended with sAviOr. Before sAviOr the term bonjwa wasn't used to describe Boxer, NaDa, or Oov. After sAviOr, people looked back and decided that those three should be labelled Bonjwa as well. In all honesty, Flash should be, but I think it'd be better to leave the Era of the Bonjwas behind and follow the koreans with calling Flash something other than Bonjwa as Broodwar comes to an end.

krndandaman

Mozambique16569 Posts

July 31 2012 09:43 GMT

#12

--- Nuked ---

IMLyte

Canada714 Posts

July 31 2012 11:03 GMT

#13

I completely agree with dandaman here Flash and Jaedong stats are incredible but they kept butting heads with each other making it impossible for one or the other to attain complete dominance and bonjwa status.

Great Write up btw!

rift

1819 Posts

July 31 2012 12:48 GMT

#14

This uses Glicko-2, which Trueskill is based on, and is the best rating system. Source is every game in TLPD.

For those saying "but this is just based on win rates!", you are incorrect. Points fluctuate based on the score of the player beaten; i.e. OSL wins are typically of more value than beating up Proleague scrubs. Se how Bisu is the best player in the first half of 2007 despite having a worse record than Stork or Jaedong.

GLICKO-2 also improves on Elo with the introduction of a reliability rating, which assesses the accuracy of a player's score depending how recently they've played games.

greenelve

Germany1392 Posts

July 31 2012 12:50 GMT

#15

Im surpised how long Jaedong was a top tier broodwar player.

and great comparism, i like statistics too ^^x

Riquiz

Netherlands400 Posts

July 31 2012 13:33 GMT

#16

Really cool writeup!
I have never been into BW, but this was def. worth spending my time on, really cool to read about these legends!

NicksonReyes

Philippines4431 Posts

July 31 2012 14:08 GMT

#17

On July 31 2012 15:19 d_ijk_stra wrote:
Lim-Lee-Choi-Ma + Jaedong, Flash lifetime ratings

[image loading]

Damn, JD performed way better than any of the first 4 Bonjwas. Too bad the very definition of the word didn't allow him to be one because of Flash.

Xiphos

Canada7507 Posts

July 31 2012 14:11 GMT

#18

On July 31 2012 23:08 NicksonReyes wrote:

Show nested quote +

Damn, JD performed way better than any of the first 4 Bonjwas. Too bad the very definition of the word didn't allow him to be one because of Flash.

JD just had to choke on those finals vs Flash though.

His fault really.

mathemagician1986

Germany549 Posts

July 31 2012 14:13 GMT

#19

On July 31 2012 21:48 rift wrote:
[image loading]

that graph doesn't tell me anything to be honest. too many lines.

Signet

United States1718 Posts

July 31 2012 14:25 GMT

#20

Great writeup, you obviously put a lot of effort into this. It's cool that this rating captures how a top player like Boxer reaches his peak and plateaus, where he is surpassed by a later generation of more talented players... rather than what it feels like at the time, ie that he is getting worse.

In fact when Savior was in his first slump and making a slow (ultimately unsuccessful) comeback, he made the comment that he thought he was actually playing better than when he was winning everything, but that other players had simply become much better.

Also interesting that Bisu was behind 4 other players, including Flash and Jaedong who at the time hadn't won a Starleague, when he won back to back MSLs... then later when he is getting knocked out of the individual leagues he was statistically the 3rd-best. But again that doesn't feel soo wrong considering how it played out. I felt at the time like a lot of those individual league exits came down to some bad luck and losing to cheeses, even though he was really separating himself from everyone except Flash/Jaedong in terms of mechanics and multitask. (Stork said he was amazed by how fast Bisu's hands were after watching him play in a tournament held in China I think.. this was well after the MSLs but when Bisu was doing really well in proleague)

1 2 3 4 5 6 Next All

Please or register to reply.

Statistical Bonjwa Theory: the History of BW

Completed

Ongoing

Upcoming