Guild Wars 2

Nice and simple, works well for top 1000 players.

Points/game is ok, but if you extend it out of the top 1000 you’re going to have a ton of 1-0 or 2-0 players.

Ideally, you want some kind of weighted formula that cares about # of games a little bit, especially at super low numbers of games.

Wow. You did a better job at this than the guy who actually does it for a living. Well done.

Is this another way to show highest mmr?
Whats stopping a guy from not playing once he hits a high index like16 – 1? This is your classic Old Leaderboard system where playing a few lucky games will yield you top rank. Once there, all they have to do is go afk.

Make a system where 100 w/ 0 l = 500w/430l
You have to cater high mmr players with players who has high amounts of games played. I tried making a post about that here.
https://forum-en.gw2archive.eu/forum/game/pvp/Leaderboard-Idea-Please-give-input/first#post4817150

It’s going to take a lot of math and ingenuity to find the right equation where MMR = High amounts of games played. I think this is a route that we need to look into… but i’m no mathematician so i cant offer anything more than an idea and a simplistic outline.

Very tryhard Acandis! Very tryhard!

Wahoo! Bye frands!

1.In your system the best player would by default not get to top spots because he would have the highest mmr and therefore almost no chance of getting more than 1 pt per win.
2.As per another post someone who had a few good games can just stop playing and stay #1

Of course this kind of system again only really works if you split up team and solo queues. The amount of rng in winning a solo queue match is ridiculous when compared to a team queue. The times i play with my friends my win ratio is about 70-80% when i play on my own it drops way lower.

Saiyan.1704:

Is this another way to show highest mmr?
Whats stopping a guy from not playing once he hits a high index like16 – 1? This is your classic Old Leaderboard system where playing a few lucky games will yield you top rank. Once there, all they have to do is go afk.

Make a system where 100 w/ 0 l = 500w/430l
You have to cater high mmr players with players who has high amounts of games played. I tried making a post about that here

It’s going to take a lot of math and ingenuity to find the right equation where MMR = High amounts of games played. I think this is a route that we need to look into… but i’m no mathematician so i cant offer anything more than an idea and a simplistic outline.

I think you’re confusing w/l ratio with MMR. Parsing through the ladder data and re organizing it with points/total games doesn’t truly reflect MMR (more of a reflection on “deserved” w/l ratio).
There isn’t a whole lot Acandis can do to actually create a better leaderboard using data from a bad one, mathematically or otherwise, because he’s fundamentally working with bad data. The games played issue isn’t even a problem you’re going to approach with mathematical formulas, it should be a conditional in the MMR algorithm itself (rating decay “if statement”).
The problem we face with this current leaderboard is that loosing games seems to have little to no negative impact on “ranked points” value, meaning you can just spam games and as long as you get lucky and win a few you’ll climb the ladder. This is bad because it defeats the purpose of actually having ranked points in the first place; you’re essentially just sorting by total weighted wins, which basically means the “top player” is whoever can log the most hours in game.
Everyone has just been asking for a ratings ladder from the begining, idk why they don’t just do that and show ppl ratings. They could have 1 ladder for solo and 1 for team queue tied to guilds.
They have some “psuedo code” (python syntax) for their MMR algorithm and how they (used to?) sort the ladder that can be found here.

I applaud the effort, Acandis, but at this stage all you’ve done is ordered players by win rate. That’s no better than the current system – arguably worse, as it still fails to account for skill.

First, I’m assuming that the current leaderboard is using a similar algorithm to what’s posted on the wiki, where points awarded are based on predicted outcome vs. actual outcome. Further, a finish near where you’re expected to finish will always result in 0 points.

If you look at an ideal system where players have mostly correct ratings and are predominately placed in matches with ~50% chance to win, then the “quality rating” you’re proposing for any player would be 0.50, i.e. a 50% win rate. Even if you expand this to look at players getting a distribution of matches such the total magnitude of expected win and expected loss is the same, you still have a 50% win rate.

So who achieves a higher win rate than 50%? The immediately response may be “better players” but that’s not directly true. The people with higher win rates are:

1. Players who have a proper rating, but are matched such that their win rate is higher than 50% i.e. they’re put in matches they should win more than matches they should lose.
2. Players with a rating below their proper rating. Because their rating is lower than it should be, the prediction on win-loss will be off, and they’ll perform better than expected until their actual rating nears their true rating.

The reason top players tend to have a win rate which is larger than 50% by a statistically significant amount is due to #1. Because there are few people above them and many more below them in the potential match pool, the top players will be in more matches where they have a favorable chance to win compared to matches where they have a favorable chance to lose.

But with a smaller player pool from which to match, it’s hard to achieve an average chance to win of 50% across any portion of the leaderboard. In addition, if an element of the prediction algorithm isn’t accurate, then players who take that element to an extreme will skew their win rate. So what win rate or “Quality Index” really shows is which players have played more games which are in their favor or were incorrectly predicted.

Compared to the current leaderboard, which is based predominately on time spent, the “Quality Index” proposal takes control away from the player completely and turns ranking into luck of the draw.

TLDR: “Quality Index” is roughly equivalent to win ratio. Win ratio indicates imbalance in matchmaking (and prediction in this case) more than skill. Time spent (current leaderboards) is at least something players can control.

Zodian.6597 said it best: There isn’t a whole lot Acandis can do [..] he’s fundamentally working with bad data.

Zodian.6597:

Everyone has just been asking for a ratings ladder from the begining, idk why they don’t just do that

Up until December 2014, leaderboards were based on rating, with a decay filter applied. They had problems, notably:

1. There was (maybe still is) a flaw in the rating calculation or the parameters, as evidenced by someone that was 2-11 being rank1 for a few days. Similarly, you had people who were 11-0 or 10-1 making it to the top of the leaderboards, often with troll names like “fix leaderboard”.
2. The decay was too extreme on the leaderboards and did not affect actual rating. As a result, players who went inactive for a couple days would drop hundreds of spots on the boards. But when they would play one game, they would suddenly jump up a few hundred. Ironically, the decay was intended to filter the 11 game troll accounts off the top of the leaderboard.

The perceived problem with that system was that you could get to the top of the boards with only a handful of games and a handful of wins, so quantity should matter more. ANet took that too far to the extreme and came up with a leaderboard system where rating doesn’t matter; in fact, high rating is now a detriment (see Chaith’s thread).

The original leaderboard could have been improved by better tuning the Glicko2 rating algirthm and implementing a better decay algorithm.

But going forward, the main question that needs to be answered is whether time played should matter at all, after some initial amount, or if skill (rating) should be the only factor. Obviously skill should be the primary factor. The argument for putting some weight on time played is that it gives good (but not excellent) players a route to obtain end-of-season rewards. Having a small group of players monopolize the end-of-season rewards isn’t a good way to grow the PvP playerbase.

Wonderful Job!
The only way the pvp community gets an accurate leaderboard is obviously, if we do it ourselves, since A-net proven to be complete incompetent in this.

To all the stupid haters above, there is one obvious prove that this leaderboard is better than the existing one: known top gw2 players are on the top spots! A leaderboard algorithm should be proven on this ground not on some purely mathematical ground (which presumes a perfect world among other unrealistic assumptions).

Exedore.6320:

Up until December 2014, leaderboards were based on rating, with a decay filter applied. They had problems, notably:

1. There was (maybe still is) a flaw in the rating calculation or the parameters, as evidenced by someone that was 2-11 being rank1 for a few days. Similarly, you had people who were 11-0 or 10-1 making it to the top of the leaderboards, often with troll names like “fix leaderboard”.
2. The decay was too extreme on the leaderboards and did not affect actual rating. As a result, players who went inactive for a couple days would drop hundreds of spots on the boards. But when they would play one game, they would suddenly jump up a few hundred. Ironically, the decay was intended to filter the 11 game troll accounts off the top of the leaderboard.

This kind of sounds like a player population issue, where you’d only have a few players queuing at non-prime time hours and the rating spread of each match maxing out to accommodate the queue timers. Decay should kick in after a week or two of inactivity (like LoL’s old S2 elo system). I think with rating you already do have games played being a factor, since rating is built up in such linearity (and by winning). You’ll eventually reach this point where it takes a certain number of games to push the rating threshold.

The problem with Gw2 has always been player population, there simply hasn’t been enough ppl queueing ranked at all hours to create matchups within reasonable MMR spreads. So the highest MMR guy looses to some low MMR scrub on an unlucky game at 3:00am and gets his rating tanked before he can push the threshold to make games played matter. Meanwhile, some low MMR scrub just got propelled to the top of the ladder, which will cause his new teamates to loose and propel some other noob. It’s just a cycle. When morning hits and everyone loggs back in, the ladder has been shuffled and no one is where they should be.

Like, I’m currently 95% on the ladder and I queued into Noob Engi (who is ranked #1 in north america) like 3 times after midnight tonight.

Zodian.6597:

Exedore.6320:

Up until December 2014, leaderboards were based on rating, with a decay filter applied. They had problems, notably:

1. There was (maybe still is) a flaw in the rating calculation or the parameters, as evidenced by someone that was 2-11 being rank1 for a few days. Similarly, you had people who were 11-0 or 10-1 making it to the top of the leaderboards, often with troll names like “fix leaderboard”.
2. The decay was too extreme on the leaderboards and did not affect actual rating. As a result, players who went inactive for a couple days would drop hundreds of spots on the boards. But when they would play one game, they would suddenly jump up a few hundred. Ironically, the decay was intended to filter the 11 game troll accounts off the top of the leaderboard.

This kind of sounds like a player population issue, where you’d only have a few players queuing at non-prime time hours and the rating spread of each match maxing out to accommodate the queue timers. Decay should kick in after a week or two of inactivity (like LoL’s old S2 elo system). I think with rating you already do have games played being a factor, since rating is built up in such linearity (and by winning). You’ll eventually reach this point where it takes a certain number of games to push the rating threshold.

The problem with Gw2 has always been player population, there simply hasn’t been enough ppl queueing ranked at all hours to create matchups within reasonable MMR spreads. So the highest MMR guy looses to some low MMR scrub on an unlucky game at 3:00am and gets his rating tanked before he can push the threshold to make games played matter. Meanwhile, some low MMR scrub just got propelled to the top of the ladder, which will cause his new teamates to loose and propel some other noob. It’s just a cycle. When morning hits and everyone loggs back in, the ladder has been shuffled and no one is where they should be.

Like, I’m currently 95% on the ladder and I queued into Noob Engi (who is ranked #1 in north america) like 3 times after midnight tonight.

Population issue shouldn’t be an argument. Parameters can be set so that these same players are impossible to play with or against. A “team shuffle” for example; intead of that player being on your team twice in a row, you’re on opposite teams instead. That can be governed by whether or not you lost (or won) with that person. Thresholds like score differences can be set. It’s really quite simple.

That’s one of many solutions to the problem but Anet don’t want to put in the extra programing to make that happen… it is a population issue but that shouldn’t matter if proper algorithms were in place.

While it’s easy to claim that this system is less than perfect. I challenge anyone to claim (logically) that it’s not a vast improvement.

As MMR stabilizes and as the sample size increases, I’d expect this to become an increasingly more accurate representation of skill.

MMR itself was also a fairly good representation of skill. How quick the decay started was the only thing that was broken about it.

Saiyan.1704:

Population issue shouldn’t be an argument. Parameters can be set so that these same players are impossible to play with or against. A “team shuffle” for example; intead of that player being on your team twice in a row, you’re on opposite teams instead. That can be governed by whether or not you lost (or won) with that person. Thresholds like score differences can be set. It’s really quite simple.

If you wanted to enforce that, some people would sit in queue for 30+ minutes, especially at off-hours.

The reality is that there are times of the night where the PvP poulation is low. Under completely ideal conditions of a 10min game and grouping all within ±50 rating from 700 to 2300 rating and 10 minute average games, you would need 340 players minimum (best rating distributions) in order to maintain 5min queue times. And since that reality doesn’t happen, you’d probably need double that pool.

That’s not happening at off hours, especially once you escape the middle of the distribution. So above average players who play at off hours will always have better odds of winning and will win more matches. It goes up even more if they form partial groups.

Now we need the community to start supporting this system and attach rewards to it like gems or something probably thinking too much already but I just don’t like the current leaderboards I rather support this.

Exedore.6320:

snip

you understand this subject.

its refreshing.

thank you.

Good job Acandis!

HAHAHAH ARENA NET HIRE NEW EMPLOYEES PLEASEEEEEEEE!!!!!

Thanks for all the feedback so far!

I’d just like to say that I have exams the next few days so I’m not actively working.

That being said, I have read all the suggestions and I will have an improved version by the end of next week with at least the following things addressed:

• More emphasis on games played.
  • Ex, A player with QI of (x) and (y) games played will be below a player with QI of (x +/- epsilon) and (y + delta) games played.
• Comparing Rank, Wins, Losses from last update.

As data isn’t changing too radically every hour, I’ll still be updating this every few hours when I’m awake.

Again, thanks for all the input so far. I will continue to monitor this thread for continued feedback.

^yeah, I think that’s as good as you can do with what you got

One suggestion for that: You can do a sigmoid function based on wins as a multiplier for QI that rolls off for a high number of wins (never will exceed 1x multiplier).

So for example:
QInew=QI * 1/(1+EXP(-Wins/C))

C= is a coefficient for how fast the wins multiplier will rolloff. Set to whatever. At C # of wins, you’ll have 73% multiplier, 88% at 2C, 98% at 4C.

Everyone here forgets the game’s genre. It’s an MMO! That means grinding! Players who grind are rewarded more than skilled players. That’s the whole idea of any mmo, and the leaderboards represent that! You think that just because we are getting a moba map you should be ranked the same way? If you want to be rewarded for skillful play, stop playing an MMO!

I’m fine with the current system. In fact, I think rank should be solely determined by the total number of games played, win or lose!

cyyrix.6105:

I think rank should be solely determined by the total number of games played, win or lose!

But, then we end up with an obscene grind. Just look at the top of the ladder now. It’s at such an unhealthy number of hours spent sitting on ones butt. Seriously, they are racking up 80 hours plus in 6 days. The Dev’s should think about this and take some responsibility for what there design’s create.

cyyrix.6105:

Everyone here forgets the game’s genre. It’s an MMO! That means grinding! Players who grind are rewarded more than skilled players. That’s the whole idea of any mmo, and the leaderboards represent that! You think that just because we are getting a moba map you should be ranked the same way? If you want to be rewarded for skillful play, stop playing an MMO!

I’m fine with the current system. In fact, I think rank should be solely determined by the total number of games played, win or lose!

much troll