|
So I've been looking at the rankings a lot lately, and I'm really glad that Bisu's first. He's actually first on elo and kespa rankings. I've heard from a lot of people that elo is crap and that kespa is a much more accurate system of ranking. Now, I have no idea how either of them work, but I have always assumed that kespa was more "accurate" based on the skill level of the opponent. Something recently led me to oppose that. Recently, Leta, Jangbi, and Bisu have been beast. On february kespa rankings, this was ther order.
1 (P)Bisu SK Telecom T1 1849.5 + 2 + 99.0 2 (T)Flash KTF MagicNs 1824.3 - - 82.7 3 (P)Stork Samsung KHAN 1811.3 - 2 - 213 4 (Z)Jaedong Hwaseung Oz 1708.5 + 1 + 146.2 5 (P)BeSt SK Telecom T1 1598.3 - 1 - 123 6 (P)JangBi
What the heck?? stork and best have been suffering lately and they're still up there. on elo its this
(P)Bisu 김택용 P SK Telecom T1 2314 pts 2353 pts (Z)Jaedong 이제동 Z Hwaseung Oz 2278 pts 2309 pts (T)Flash 이영호 T KTF MagicNs 2276 pts 2336 pts (P)JangBi 허영무 P Samsung KHAN 2249 pts 2290 pts (T)Leta 신상문 T OnGameNet SPARKYZ 2190 pts 2259 pts
which is more of what I expected. Maybe kespa is a more long term ranking system. But then they woudlnt have the rankings for each month. I dont know how either of them work, but I think that elo is more dependable according to how the players are doing right now.
|
Valhalla18444 Posts
KeSPA rankings are calculated solely based on points assigned by winning at different stages of tournaments. Every win is a gain, every loss is merely missed points. It calculates a 3 month period and adds them up. It has nothing to do with anything else, it is just an assignment of points based on how KeSPA rates the worth of the different stages in leagues.
ELO is pretty simple, every player starts with 2000. A win gives you points, a loss makes you lose points. Winning against a player with a higher ELO than you gives you more points, losing to a player with lower ELO than you makes you lose more points. Winning against a player with lower ELO gives you less, losing against a higher ELO means you lose fewer points. Pretty simple.
That said, both of them are entirely results-based and have no concrete reflection on the players themselves and how they'll perform on any given day. They're just statistics.
|
afaik, KeSPA takes into account results over the last 12 months. that'll explain much of it
EDIT: Hm it's only 3 months now?
|
infinity21
Canada6683 Posts
KeSPA takes into account for the past year or so of records, I believe. And ELO can drop drastically if you go on a losing streak.
|
KeSPA ranks are definitely more long term. They value the last couple months at 100%, then reduce by 10% the value of each of the months proceeding, so even if people start sucking the current month, they have lots of other months of good play.
|
Valhalla18444 Posts
Someone confirm that KeSPA is only 3 months, please. It's definitely not a year, that's ancient history.
|
When sAviOr began to slump in around 2008 he was still ranked really high on Kespa for MONTHS. Because of his era of dominance. While his ELO dropped rather quickly at least in comparison. KeSPA rankings are more solid I think, and if you earn a lot of points its harder to come farther down, I can't remember how KeSPA is scored however. But ELO effects players more drastically for more games in a short period, so I would say for determining who is hot at the moment, or a rising star, ELO is where to look. When someone starts performing better they get more points, while people at the top are getting less points. So a player like Leta was able to storm the ELO rankings.
For a good example, NaDa's ELO is inconsistent, and pretty low barring a few random sparks of his former glory, so he hasn't really been top 30 ELO in a while, however someone recently made a thread celebrating NaDa being top 30 in KeSPA for 8 years. Keep in mind that he had a major slump, which would of shot him out of ELO ranking, what kept him high in KeSPA would of been his prior level of dominance, which earned him a lot of points. While that level of dominance would of made his ELO drop drastically.
|
Valhalla18444 Posts
if you want an accurate reflection of who is playing the best right now, read the Power Rank. A new one comes out today.
|
How has Bisu been suffering lately D:. He's smashing the crap outta everyone in WL.
|
United States10327 Posts
On March 04 2009 13:37 FakeSteve[TPR] wrote: if you want an accurate reflection of who is playing the best right now, read the Power Rank. A new one comes out today.
<3 fakesteve (also posted at 13:37 gives ++points)
|
I think ELO takes into account all games, while Kespa is the last year but points decay as time goes on so Flash's OSL win last March/April (can't remember which one) counts much less than Bisu's MSL win (after adjusting for 1 being an OSL and 1 an MSL). The reason Kespa reacts slower is as FS explained you don't lose points for a loss, ELO you do. Another part if it is our ELO updates within a day or so of the games that were played, so if Ganzi (lowest rated player) got an all kill tonight against Flash, Bisu, Jaedong, and Leta (forget they are on different teams) tomorrow Ganzi's ELO rank would have shot up ~60 points while it would take a month to see any effect on the Kespa rank.
|
On March 04 2009 13:37 FakeSteve[TPR] wrote: if you want an accurate reflection of who is playing the best right now, read the Power Rank. A new one comes out today. Considering the top 5 in the current power rank are the same as the current top 5 elo (albeit in a somewhat different order), I'm not sure there's a huge difference, although the comments/analysis on power rank are nice.
|
lol @ steve and the self-promotion. haha
i agree that they're just stats. doesnt really matter. just like a 70% win rate doesnt mean you'll win EVERY match
|
Valhalla18444 Posts
On March 04 2009 13:49 gravity wrote:Show nested quote +On March 04 2009 13:37 FakeSteve[TPR] wrote: if you want an accurate reflection of who is playing the best right now, read the Power Rank. A new one comes out today. Considering the top 5 in the current power rank are the same as the current top 5 elo (albeit in a somewhat different order), I'm not sure there's a huge difference, although the comments/analysis on power rank are nice.
The focus is entirely different. The order is too. There's correlation but no causation.
|
Why are they starting ELO rating at 2000 when in all other ladders i've seen (such as the original bnet one) it is 1000?
|
@OP That's interesting, because I've always felt that ELO was the more "accurate" system. The difference between ELO and KeSPA is that ELO is a skill based system (within assumptions), while KeSPA is an achievement based system. A player's ELO depends on how good are the opponents the player beats (and loses to), where "how good the opponent is" is defined recursively as how many good opponent that opponent has beaten. KeSPA points are gained according to how far one goes through the individual leagues, with some points gained through proleague performance. So KeSPA ranking is about recent achievement, and a finals game matters a lot more than a qualifying game. Contrast that to ELO where each game counts equally regardless of the context. The assumptions mentioned earlier for ELO-- one main assumption is that every player has an inherent skill level that stays constant over the player's lifetime. This is why ELO is often called a skill based system. Like Fakesteve said, the assumption is not realistic and both ELO and KeSPA says nothing about how a player will perform on a given day. While it is unrealistic, it doesn't come from nowhere. A player who is this good at a certain time is generally about more or less good some other time. (More about the skill assumption: if every player has a certain inherent skill level, and we postulate that so and so skill level A always has a X% chance of beating skill level B-- for example a 2250 has a 70% chance of beating a 2130-- then we see that games are just a reflection of the players true skill. The games and who wins are the data from which we try to guess the hidden variable of skill. The more games in total-- the more data-- the closer we are to calculating the player's "true" skill level. This is why ELO is mathematically awesome.)
|
On March 04 2009 14:18 Eatme wrote: Why are they starting ELO rating at 2000 when in all other ladders i've seen (such as the original bnet one) it is 1000?
To highlight how much better the progamers are than the standard scrubs that populate all the other ladders.
Actually I have no clue, but that sounds reasonable to me!
|
In terms of accuracy I'd say ELO > Power Rank > Kespa.
|
On March 04 2009 14:22 huameng wrote:Show nested quote +On March 04 2009 14:18 Eatme wrote: Why are they starting ELO rating at 2000 when in all other ladders i've seen (such as the original bnet one) it is 1000? To highlight how much better the progamers are than the standard scrubs that populate all the other ladders. Actually I have no clue, but that sounds reasonable to me! Well top progamers would have about 2000 on a normal ladder I guess.
|
On March 04 2009 14:18 Eatme wrote: Why are they starting ELO rating at 2000 when in all other ladders i've seen (such as the original bnet one) it is 1000?
starting value has no effect on elo systems (within reason) and was chosen arbitrarily in the case of tl i believe. the only reason 1000 is common is because people view it as a nice round number.
|
On March 04 2009 13:37 FakeSteve[TPR] wrote: if you want an accurate reflection of who is playing the best right now, read the Power Rank. A new one comes out today.
Yay finally!
Anyway, Kespa rankings are way more forgiving than ELO. If a player like Bisu edges a 2-1 against a really low-ranked progamer, he'd still get some points for the Kespa rankings, while the ELO might decrease of not modify at all.
|
On March 04 2009 14:26 Eatme wrote:Show nested quote +On March 04 2009 14:22 huameng wrote:On March 04 2009 14:18 Eatme wrote: Why are they starting ELO rating at 2000 when in all other ladders i've seen (such as the original bnet one) it is 1000? To highlight how much better the progamers are than the standard scrubs that populate all the other ladders. Actually I have no clue, but that sounds reasonable to me! Well top progamers would have about 2000 on a normal ladder I guess.
Their ELO is also a lot more scaled then others.
Iccup a win gets you anywhere from like, 50 to 130 points. I think the max people really jump in one game is like 12-15 points.
|
On March 04 2009 13:37 FakeSteve[TPR] wrote: if you want an accurate reflection of who is playing the best right now, read the Power Rank. A new one comes out today.
I think the top one right now is....
+ Show Spoiler +
|
ELO is not an accurate reflectin of skill. It is, however, a comparison of player's strength against another opponent.
The problems encountered with using ELO for SC are as follows:
1. Not enough games at k value of 16. 2. "Skill" and/or "dominance" fluctuates fairly wildly within short periods of time
The fact of these two things combined means large fluctuation in ELO values in a short period of time. Whereas in a game like chess where skill improvement is fairly static (like most turn based games as well such as go), in RTS games there are more things that can "go wrong" for the superior as much as they can "go right." Inevitably, there tends to be more "freak losses" just like reducing the amounts of time in a chess game from 2 hours + 1 hrs at move 40 down to blitz where you have 5 minutes per game tends to increase the amount of mistakes players make.
The k value for a game like chess changes as you move up levels to ensure fairly "static" display of skill at the top. Since the k value for SC is fairly high and constant, more skilled players tend to lose more points with a loss than they otherwise would with a lower k-value. Of course, the opposite is true where hot players will move up quite fast through the ranks. Increasing or decreasing the k-value from where it is is problematic because the winrates of the players vs. each other can fluctuate wildly if they get hot even though mechanics and a lot of other "skill" factors remain the same.
Basically, since ELO compares playing strengths of others it doesn't necessarily mean an opponent will/can beat another. If two players are playing "standard" one opponents style (Jaedong) may tend to counter another players (Bisu) even though Bisu's ELO may be higher than Jaedong's. This is the same in all cases where ELO is used whether in chess, go, etc.
What we can say with ELO values is how strong a person is compared to the rest of the players. So, for example, if we picked a random opponent below the Leta's ELO such as Iris, then Iris would probably be the underdog. If we picked a random opponent with an ELO above Leta's such as Jaedong then that player (Leta) would probably be the underdog. This is what ELO can be used for.
Of course, since the k value of TLPD is high a "hot" player may have a higher ELO than his actual skill level. This is one of the difficulties just straight up comparing ELO points to determine favorites and underdogs.
The one thing I like about RTS games is that there's a massive amount of unpredictability especially given the map, different races AND build orders you can select to play. This makes it all the more amazing when a player can consistently win 60.. even 70.. or 80% of the time against other players. This is also why ELO can often be wrong in prediction of when a player should and shouldn't win.
Anyway, hoped that helped someone understand ELO a bit better.
|
Germany / USA16648 Posts
On March 04 2009 13:36 FakeSteve[TPR] wrote: Someone confirm that KeSPA is only 3 months, please. It's definitely not a year, that's ancient history. It was changed several times. First it was an all-time ranking (Boxer's prime), then a 12 months ranking and then it was changed again, but I can't fully recall to what end. Winning a major league is what? 500 pts I think? So it can't be 3 months, I'm pretty sure it's still a 12 months ranking, but that the months are weighted differently (decreasingly from most recent to least recent obviously).
|
On March 04 2009 14:07 FakeSteve[TPR] wrote:Show nested quote +On March 04 2009 13:49 gravity wrote:On March 04 2009 13:37 FakeSteve[TPR] wrote: if you want an accurate reflection of who is playing the best right now, read the Power Rank. A new one comes out today. Considering the top 5 in the current power rank are the same as the current top 5 elo (albeit in a somewhat different order), I'm not sure there's a huge difference, although the comments/analysis on power rank are nice. The focus is entirely different. The order is too. There's correlation but no causation. My point wasn't that you copied the elo ratings, but rather that the elo ratings are good enough at showing who's doing well that an intelligent human-chosen list isn't going to be wildly different most of the time.
|
Katowice25012 Posts
On March 04 2009 15:04 Carnac wrote:Show nested quote +On March 04 2009 13:36 FakeSteve[TPR] wrote: Someone confirm that KeSPA is only 3 months, please. It's definitely not a year, that's ancient history. It was changed several times. First it was an all-time ranking (Boxer's prime), then a 12 months ranking and then it was changed again, but I can't fully recall to what end. Winning a major league is what? 500 pts I think? So it can't be 3 months, I'm pretty sure it's still a 12 months ranking, but that the months are weighted differently (decreasingly from most recent to least recent obviously).
The general consensus I found was that it was changed to a 12 month system, with the last 3 months being heavily weighted. I don't have a source, other than how low it takes for a player to drop off (Stork should have fallen more sharply if it was 3 yes?) so take that how you want.
|
Cayman Islands24199 Posts
elo tells you more about strength than kespa, but kespa gives more information about presence on the scene. winning matters to one's 'prominence' of course, but the stage of the win also matters a great deal. elo does not take that into consideration
|
On March 04 2009 16:00 heyoka wrote:Show nested quote +On March 04 2009 15:04 Carnac wrote:On March 04 2009 13:36 FakeSteve[TPR] wrote: Someone confirm that KeSPA is only 3 months, please. It's definitely not a year, that's ancient history. It was changed several times. First it was an all-time ranking (Boxer's prime), then a 12 months ranking and then it was changed again, but I can't fully recall to what end. Winning a major league is what? 500 pts I think? So it can't be 3 months, I'm pretty sure it's still a 12 months ranking, but that the months are weighted differently (decreasingly from most recent to least recent obviously). The general consensus I found was that it was changed to a 12 month system, with the last 3 months being heavily weighted. I don't have a source, other than how low it takes for a player to drop off (Stork should have fallen more sharply if it was 3 yes?) so take that how you want.
First 3 months are at 100%, then 10% less for each month after that, so even if a win was half a year ago it will grant 70% of the points, which is still pretty significant.
|
United States20661 Posts
On March 04 2009 16:28 lololol wrote:Show nested quote +On March 04 2009 16:00 heyoka wrote:On March 04 2009 15:04 Carnac wrote:On March 04 2009 13:36 FakeSteve[TPR] wrote: Someone confirm that KeSPA is only 3 months, please. It's definitely not a year, that's ancient history. It was changed several times. First it was an all-time ranking (Boxer's prime), then a 12 months ranking and then it was changed again, but I can't fully recall to what end. Winning a major league is what? 500 pts I think? So it can't be 3 months, I'm pretty sure it's still a 12 months ranking, but that the months are weighted differently (decreasingly from most recent to least recent obviously). The general consensus I found was that it was changed to a 12 month system, with the last 3 months being heavily weighted. I don't have a source, other than how low it takes for a player to drop off (Stork should have fallen more sharply if it was 3 yes?) so take that how you want. First 3 months are at 100%, then 10% less for each month after that, so even if a win was half a year ago it will grant 70% of the points, which is still pretty significant.
This.
|
On March 04 2009 15:04 eshlow wrote: ELO is not an accurate reflectin of skill. It is, however, a comparison of player's strength against another opponent.
The problems encountered with using ELO for SC are as follows:
1. Not enough games at k value of 16. 2. "Skill" and/or "dominance" fluctuates fairly wildly within short periods of time
3. Three different match ups for each player.
You could look at match up specific ELO, but that would make problem #1 even worse.
|
ELO is really quite good, I think. It almost always conforms fairly well to what I think the Power Rankings should be.
KESPA has it's own value as it's less volatile. It has Stork at #3 which I think more accurately describes his place in the top-ten over a longer period. Yeah, he might not be playing like a #3 lately, but he deserves it based on play over more than just the last starleague season.
|
+ Show Spoiler +Depends on what equation set up they are trying to use. Mostly weather or not the k will float because a k that isn't constant for everyone = inflation.
I find KeSPA current set up make people who are a real slump hold on to the top for too long.
ELO vs KeSPA main difference is that ELO will change faster KeSPA is slower to change only way to have a fast change in from a very inactive player become very active.
|
On March 04 2009 16:28 lololol wrote:
First 3 months are at 100%, then 10% less for each month after that, so even if a win was half a year ago it will grant 70% of the points, which is still pretty significant.
ok - this is the first time I see a good explanation of Kespa rankings - but still it would be nice to have a complete knowledge of the rating system they use. How many points are the matches wieghted for each competion / stage and opponent. Also, is there a bonus for winning a SL or reaching higher phases of tournament...
|
On March 04 2009 14:39 vx70GTOJudgexv wrote:Show nested quote +On March 04 2009 14:26 Eatme wrote:On March 04 2009 14:22 huameng wrote:On March 04 2009 14:18 Eatme wrote: Why are they starting ELO rating at 2000 when in all other ladders i've seen (such as the original bnet one) it is 1000? To highlight how much better the progamers are than the standard scrubs that populate all the other ladders. Actually I have no clue, but that sounds reasonable to me! Well top progamers would have about 2000 on a normal ladder I guess. Their ELO is also a lot more scaled then others. Iccup a win gets you anywhere from like, 50 to 130 points. I think the max people really jump in one game is like 12-15 points. I was talking from experience with ladders such as gamei and neogamei ect. You cant really compare with iccup.
|
Russian Federation1208 Posts
ELO and KeSPA rankings are just two models of player skill.
What is a model? Model is a simplifyed description of the real object, i.e ANY MODEL IS NOT adequate to the real object. Thus, KeSPA and ELO are both inadequate. And Power Rank is not adequate by the same reason.
However, different models simplify different properties of the modeled object, and you can judge a model as more adequate than another model, if you think that "those" properties of the object are more important than "these" properties of the object.
If you think that winning leagues is more important, than KeSPA is more adequate for you. If you think that winning of better players is more important, than ELO is more adequate for you. If you think that ideal performance during games is more important, than PowerRank is more adequate for you.
|
|
|
|