|
Public Service AnnouncementUse the Champion threads whenever appropriate. Don't use General Discussion simply out of ease. ===== If you want to whine about server lag, use the QQ thread. We all suffer alike when Riot servers kaput. No need to make a post about it in GD. |
On January 27 2012 00:06 Blyf wrote:Show nested quote +On January 26 2012 23:45 Craton wrote:
Except you have a linear rating scale when the grouping of people is parabolic. It can be displayed more effectively than with with the existing system. The gouping of people is parabolic? And why would you want to seperate the middle section more when they are of, you know, equal skill?
Because they aren't equal skill. You're talking about a substantial portion of the playerbase, akin to silver plus gold plus platinum of sc2. Each of those are relatively broad areas, but are nonetheless much representative of a player's ability than the Elo range of say 1000-1400. Right now if you're in an Elo like that, it's just a clusterfuck of "yeah you're bad," even though you're often well above average. The problem is the current system is that it doesn't paint a very clear picture. Additionally, the community has attached a relatively poor stigma to the rating system. Overhauling it would also help to pull away from that. Yes, something new would partially come in its place, but overall it would do good.
Additional points:
We have no way of knowing the actual breakdown of Elo because such a small fraction is even available. Further, because of how far the ends (notably the top end) stretches, it makes people in the average Elo's look that much worse.
Right now the Elo system feels more like the MMR system of other games, where you often streak in one direction or the other, but your overall skill level seems relatively the same. Because you actually see this movement, you're inclined to think that you're a e.g. 1500 player if you peak at 1500 but always fluctate around the 1400 mark. With the SC2 system you don't actually jump a level until you have consistently better play, which makes it more representative.
|
On January 27 2012 00:23 Craton wrote:Show nested quote +On January 27 2012 00:06 Blyf wrote:On January 26 2012 23:45 Craton wrote:
Except you have a linear rating scale when the grouping of people is parabolic. It can be displayed more effectively than with with the existing system. The gouping of people is parabolic? And why would you want to seperate the middle section more when they are of, you know, equal skill? Because they aren't equal skill. You're talking about a substantial portion of the playerbase, akin to silver plus gold plus platinum of sc2. Each of those are relatively broad areas, but are nonetheless much representative of a player's ability than the Elo range of say 1000-1400. Right now if you're in an Elo like that, it's just a clusterfuck of "yeah you're bad," even though you're often well above average. The problem is the current system is that it doesn't paint a very clear picture. Additionally, the community has attached a relatively poor stigma to the rating system. Overhauling it would also help to pull away from that. Yes, something new would partially come in its place, but overall it would do good.
I think a League based system would be good for an entirely unrelated reason; it would dempathize individual games in favor of overall aptitude, which reduces the pressure of each game and hopefully mellows the ranked community out a bit. Imagine what SC2 ladder would be like if Blizzard put in a "You must win x more games for Diamond League" counter.
|
Double WotA AD Kennen sounds imba
|
On January 26 2012 23:45 Craton wrote:
Except you have a linear rating scale when the grouping of people is parabolic. It can be displayed more effectively than with with the existing system.
Same rule applies to IQ.
|
Not sure if the ELO system needs a tweak or not but at my ELO level (around 1200) the individual player skill level seems very inconsistent. Some players are incredibly reliable (had a jungle lee sin that basically wiped the enemy clean off the face of the planet) while others can't seem to do anything well (supporting ad carry with 60 cs at the 20 minute mark). I do think that having a linearly progressing ELO system on an individual in a team game like League of Legends feels rather suspect.
|
And yet, people don't follow such an idea until its shown in tournament play.
|
On January 27 2012 01:37 Kaniol wrote:Show nested quote +On January 26 2012 23:45 Craton wrote:
Except you have a linear rating scale when the grouping of people is parabolic. It can be displayed more effectively than with with the existing system. Same rule applies to IQ.
The IQ system has a wide host of problems, too, you know.
|
On January 27 2012 01:37 Nos- wrote:
Not sure if the ELO system needs a tweak or not but at my ELO level (around 1200) the individual player skill level seems very inconsistent. Some players are incredibly reliable (had a jungle lee sin that basically wiped the enemy clean off the face of the planet) while others can't seem to do anything well (supporting ad carry with 60 cs at the 20 minute mark). I do think that having a linearly progressing ELO system on an individual in a team game like League of Legends feels rather suspect.
Which is why the idea of ELO hell exists.
But really balancing an ELO system based off a game where you have a team of 5 vs another team of 5 makes it kind of a game of luck. Hopefully you get good teammates and they get bad. All in all you should be doing pretty well too in your lane.
|
OddOne raging hard on EU. Got to carry somehow ^^
|
Anyone know the actual curve that the Elo system produces? I'm not sure it's normally distributed. IQ is, since it's basically a z score, but Elo is based on the probability you beat someone else vs the actual result. Wikipedia has a random link to a Hubbert curve which I assume is the distribution you get, and makes more sense since it's the pdf of the logistic function which pops up in stuff like this.
|
On January 27 2012 01:52 starfries wrote:
Anyone know the actual curve that the Elo system produces? I'm not sure it's normally distributed. IQ is, since it's basically a z score, but Elo is based on the probability you beat someone else vs the actual result. Wikipedia has a random link to a Hubbert curve which I assume is the distribution you get, and makes more sense since it's the pdf of the logistic function which pops up in stuff like this.
Not really a curve, but i saw those stats on reddit a few days ago:
+ Show Spoiler +
http://www.reddit.com/r/leagueoflegends/comments/oo05k/elo_distribution_statistics_week_3_2012_useuweune/
|
Tbh the ranked elo system is pretty flooded with people who've only played like ~10 ranked games, and then just stopped. or only one or two in the new season, and had their old elo carry over.
The elo system has been flooded with a lot of players, that's why I'm not all that impressed with elo until they start hitting the 1700+ ratings.
|
There cant really be a comparison with IQ because if you are dumb i dont get dumber. While being a teamgame my ELo drops because you go derpmode and feed some akali.
Thats why the gap in skill of 1200 players is so high, there is a lot of players who would have a 50% winrate vs 15-1600 players in 1v1 but arnt good enough to carry there bad teammates and so lose a lot of games.
|
On January 27 2012 02:15 Gorsameth wrote:
There cant really be a comparison with IQ because if you are dumb i dont get dumber. While being a teamgame my ELo drops because you go derpmode and feed some akali.
Thats why the gap in skill of 1200 players is so high, there is a lot of players who would have a 50% winrate vs 15-1600 players in 1v1 but arnt good enough to carry there bad teammates and so lose a lot of games.
I hate when people blame bad teammates, If you play enough ranked games, that variable doesn't matter anymore. It's similar to playing poker. 1 hand/game is based on luck, a million is based on statistics.
The fact is, everyone gets around the same level of teammates over a lot of games.
|
if lol was like poker and I could just fold a shitty team preflop that might have some relevance
|
The really surprising thing to me is how compressed the lower bits of the ladder are but how spread out the higher part is. Like I just hit 1708 and am only the 4,500th best lol player in NA (/humblebrag). But when 1500 I would jump that much by winning a game.
|
Oh, you're just looking for more distinction within the ~1100-1400 range? I think there are a multitude of problems causing that that entire 300 point range to be essentially the same skill level, while the other ranges have more clear-cut "skillcaps" to them.
One is that it's the starting point for new ranked players, who gain accelerated Elo and supposedly play games with other players with <10 ranked games played. Assuming all of these new players are of around equal skill, some of them are going to have inflated Elo and some are going to have their Elo deflated because someone has to win and someone has to lose. Then when they enter the pool of all ranked players, you have similarly-skilled players at both around 1100 to 1400, depending on how MM went during their accelerated Elo period. They screw with the general skill level of that entire range, and there is a pretty consistent influx of them.
Not only that, but the players of that range really ARE basically the same skill level. There may be slight difference in skill, but it isn't enough to cause a 1400 player to make sufficiently less positioning/teamfight mistakes to change the flow of the game enough. That kind of understanding doesn't pop up until the 1500-1600 range, so games are basically a tossup of who fucks up the least, which they all do without even knowing.
IMO the Bronze/Silver range should all just become Silver, and Bronze should be 800-1200. It's stupid to have the ranges where the most discrepancy in skill occurs (Gold/Plat) have like 300-400 Elo ranges, while the area where everyone is basically the same skill has multiple badge cutoff points that are completely non-indicative of skill. Literally nothing seperates a Bronze player from a Silver one except for the direction of the streak he happens to be on, while there is a clear distinction between Gold/Silver or Gold/Plat.
tl;dr Riot fucked up the badges. Need to lower the cap for unranked because 1100-1400 is a big blob of players with the same skill.
|
Russian Federation4235 Posts
On a side note, does anyone know a reliable streamer who plays ranged AD? Popular streamers I know are either top, mid or jungle, or plain awful.
|
Chaox, Yellowpete, Entenzwerg.
|
On January 27 2012 00:20 ManyCookies wrote:
At this point, there isn't really much Riot can do to improve the Elo system. The only "real" way is to increase the data relevency of individual games, accounting for more factors than Win/Loss. Unfortunately, it's rather difficult to do that without creating behavioral bias within the game; team gold ratio, for instance, could be abused by intentionally prolonging games.
The problem with Elo in LoL: You have to adjust 10 people's Elo but you only have 1 binary variable: Win/Loss. For 2 people that variable and some norms is enough. For 10 people, it's a bit thin.
Your approach is to gather more variables per game. However, this is not viable because there is no way to objectively and automatedly judge someone's performance (until now at least).
I would instead try to gather information over the course of multiple games (and assign players a temporary dummy rating increase/decrease for single games) and calculate a rating with more information available. Certainly not all that easy to get this right though, since you have more than 10 players total.
|
|
|
|
|
|
EPT![[image loading]](http://i.imgur.com/OHlSd.png)
