PSA how placements work

[Exedore.6320]

> @"Vancho.8750" said:

> > @"Exedore.6320" said:

> > You're doing a great job of illustrating my point with your uninformed opinion.

> >

> > Glicko/Elo may not be perfect, but it's been shown by basically every online game that it's still a reliable system and better than all others to date.

>

> UGH sorry for my filthy uninformed peasantry beliefs that chess single player system doesn't work on random multiplayer game without taking into consideration anything besides winning.

> I want to play the kitten game not some broken kitten outside the game.

Right. Many groups and people have done statistical analysis of Glicko2/Elo with variations for both 1v1 and single players in a team setting. The result is that they work well enough.

 

For most players - all but those at the ends of the bell curve - the system works great and anything you're seeing is your own imagination. It's best if you just accept it and focus on improving yourself.

[Master Ketsu.4569]

It's true that placements matter slightly too much and climbing out of a shitty placement is going to take some grind, but you also have to consider it from Anets perspective: They know it's extremely unfun for new players to go up against top players, and getting destroyed makes them often quit PvP / stay in casual-land open world until that gets boring and then ultimately uninstall. Players quitting is obviously not good business for Anet, they need those players to stay and get addicted to buying gems or else they will be out of a job.

 

This is because the difference in _mechanical_ skill between a 1200 rated player and a 1500 rated player tends to be much bigger than the difference between a 1500 rated player and a 1800 rated one. People in Plat III/Leg get creamed by Plat I players all the time if the build matchup is a counter, since at that level just about everyone understands the basics of how skills work. This is not the case for when a low gold/silver player fights a plat I player, in most cases they will get absolutely destroyed 1v1 in less than a few seconds, lose 1v2s in less than 20 seconds, and often even lose 1v3s. This leads to very uneven games since a single platinum level player in a game full of silvers will often just dominate the map. Oneshot everyone, cap all the nodes, and then proceed to spawn camp for the rest of the match.

 

The placements are brutal to basically ensure that you don't get placed higher than 1400 unless you really should be there. By making rating very volatile at the start, it greatly decreases the number of games that an unskilled/new player might get matched up against someone way out of their league. The "soft reset" is there for the same reason.

 

 

 

 

[Tharan.9085]

> @"Master Ketsu.4569" said:

> It's true that placements matter slightly too much and climbing out of a kitten placement is going to take some grind, but you also have to consider it from Anets perspective: They know it's extremely unfun for new players to go up against top players, and getting destroyed makes them often quit PvP / stay in casual-land open world until that gets boring and then ultimately uninstall. Players quitting is obviously not good business for Anet, they need those players to stay and get addicted to buying gems or else they will be out of a job.

>

> This is because the difference in _mechanical_ skill between a 1200 rated player and a 1500 rated player tends to be much bigger than the difference between a 1500 rated player and a 1800 rated one. People in Plat III/Leg get creamed by Plat I players all the time if the build matchup is a counter, since at that level just about everyone understands the basics of how skills work. This is not the case for when a low gold/silver player fights a plat I player, in most cases they will get absolutely destroyed 1v1 in less than a few seconds, lose 1v2s in less than 20 seconds, and often even lose 1v3s. This leads to very uneven games since a single platinum level player in a game full of silvers will often just dominate the map. Oneshot everyone, cap all the nodes, and then proceed to spawn camp for the rest of the match.

>

> The placements are brutal to basically ensure that you don't get placed higher than 1400 unless you really should be there. By making rating very volatile at the start, it greatly decreases the number of games that an unskilled/new player might get matched up against someone way out of their league. The "soft reset" is there for the same reason.

>

>

>

>

 

The difference between a plat 1 and a Legend Player is way Higher then the difference between a Silver and a Gold player

[Luthan.5236]

It it even known for sure that it works this way? I mean ArenaNet never made their secret algorithms public. I pretty much guessed something like explained in this thread ... otherwise it would not make sense. Scores/rating can only try to guess your real skill ... that's why it gets better with more info. (More matches played = less deviation.)

 

Bad players or good players (that carry you or might make you lose) ... that should be the same on average for all the players/matches played. The main problem will be the low population where it can be possible to often get paired with the same bad (or good) players. (That's why people often suggest to wait and queue at another time when you get loss streaks.)

[Exedore.6320]

> @"Luthan.5236" said:

> It it even known for sure that it works this way? I mean ArenaNet never made their secret algorithms public. I pretty much guessed something like explained in this thread ... otherwise it would not make sense. Scores/rating can only try to guess your real skill ... that's why it gets better with more info. (More matches played = less deviation.)

 

Yes, it's known for certain. Glicko2 is publically documented by its creator. ANet hasn't published some of the parameters they use, but they can be determined through analysis. For example, the range goes from 0 to 2400 instead of the more typical 0 to 3000. Analysis also shows that ANet does player vs. average team for rating adjustments.

 

Additionally, ANet posted psuedo-code for the matchmaking algorithm to the GW2 wiki. It's a little out of date, but still accurate.

[Luthan.5236]

> @"Exedore.6320" said:

> Analysis also shows that ANet does player vs. average team for rating adjustments.

 

You mean ... they use your individual rating and compare it to the average rating of the enemy team? Isn't that weird? I mean ... it would make more sense to use the average ratings of both teams and compare them. Otherwise a strong player will lose more than he actually needed to lose. (Same for rating gain and a weak player.)

[Exedore.6320]

> @"Luthan.5236" said:

> > @"Exedore.6320" said:

> > Analysis also shows that ANet does player vs. average team for rating adjustments.

>

> You mean ... they use your individual rating and compare it to the average rating of the enemy team? Isn't that weird? I mean ... it would make more sense to use the average ratings of both teams and compare them. Otherwise a strong player will lose more than he actually needed to lose. (Same for rating gain and a weak player.)

 

Yes, you understood what I said correctly.

 

There was a thesis paper done on this topic and it found that team rating vs. team rating adjustment is worse overall compared to player vs. average team. Your concern really only impacts the case when players of vastly different ratings queue together. Under most circumstances, the variation of ratings on a team isn't that large.

[Luthan.5236]

I see. Makes sense that it doesn't make much of a difference if players of a same/similar rating are in a team. As long as the population is high enough for the matchmaker to make teams like that. We often hear in the forums though from top level players that complain when they get matched with lower levels (platin with gold) ... don't know if this is true. Team rating vs. team rating could lead to less problems here.

 

There should be lock/limit on how much your own rating can be different from your own team average then. I heard matchmaker does have a fixed range and then inceases after a few minutes? (So I guess by leaving the queue if it gets too long you can prevent getting paired with lower people that might make you lose more in a case of a loss.)

 

As long as the matchmaker uses the actual difference (player a few points barely in platin still easily getting matched with players that barely did not manage to make it in platin and are gold 3 for example) I guess it is not a big problem. Only if for some reason they made the start/end of the range the matchmaker uses fixed. (Like ... not pairing the barely platin 1 with gold 3 first and looking for ages for other platin 1 cause it uses the platin as "range" to look for.)

 

There should be a FAQ with gathered infos. First time I heard a lot about the system here. I saw other threads about people complaining/wanting changes and asked stuff and no one could answer that. Thanks for the answers.

 

Edit: Also winning 500 vs 450 should result in a different rating change than winning 500 vs 50. Since it means different skill level to win by a higher margin.

[Exedore.6320]

> @"Luthan.5236" said:

> There should be lock/limit on how much your own rating can be different from your own team average then. I heard matchmaker does have a fixed range and then inceases after a few minutes? (So I guess by leaving the queue if it gets too long you can prevent getting paired with lower people that might make you lose more in a case of a loss.)

 

You are correct about the matchmaker's current operation. The absolute limit is a design conundrum: how do you balance match quality with wait time? In other words, is it better to find less optimal matches quickly so that players don't wait long; or should it wait a long time (possibly hours for off-peak and very high and low ratings) for a slightly better match?

 

ANet has favored letting people play in a reasonable time frame.

 

 

> Edit: Also winning 500 vs 450 should result in a different rating change than winning 500 vs 50. Since it means different skill level to win by a higher margin.

 

There are two components to this:

1. Does adding "tie" to the algorithm as opposed to just win-loss improve rating? Microsoft came up with a system called TrueSkill which uses close games to reinforce that players are properly matched. They showed that it works better than Elo, but Elo is admittedly weak (it was designed before the days of PCs). Glicko2 is a significant improvement on Elo. Seeing as TrueSkill doesn't have much adoption outside Microsoft, it's likely not a huge improvement.

2. Is a close score truly indicative of equally skilled players? I would argue that it's often not. When one team is significantly better, they may slack off. There are also map mechanics which can skew the score (Foefire lord). And losing a series of close fights in conquest can yield a large score differential, even those the skill difference was small.

[Luthan.5236]

Well ... I actualy have the problem with keeping myself motivated in a match where it is 200 vs 50 or so after a few minutes (me on the team with 50 only). If others already think we lost and no one really wants to fight ... it gets boring. You could actually just wait afk until the match ends. No difference on the rating outcome. (Until you manage to turn it around and win actually.)

 

Player behavior might change if the system were different here (and if this was communicated so players would know). Less "slacking off", more fighting until the end even if it is 400 to 50 or so.

 

In a 500 vs 50 win you could actually interpret it as 550 matches played with 500 won by one team. For the 2 vs. 2 death matches it makes more sense when there are actually 3 real matches (rounds) in the 1 match that gets rated.

 

I mean it's not like chess where the past stages of the match affect the later match. Capping a point with both teams at the same strength after 5 minutes is the same as capping the point at minute 1, 2, etc. (There might be individual situations when it is hardre/easier depending on how much players rush into outnumbered fights, get killed fast, etc. but the same can happen early and late.) It also isn't like soccer (I think I found that paper you mentioned that argued in favor of comparing 1 player score against average of enemy team ... they mentioned using something like "robo soccer" or something like that) where usually the amount of goals is lower.

 

We always get that big amount of points (at least 500 ... in a 500 vs 0). The system might get results of different quality depending on the actual game played. Afaik in games like LoL it is different and early mistakes make you weaker in the later game. (When chars gain levels in the match per match.)

 

Now the mechanics with lord ... yeah a bit different here. I'd actually prefer if in conquest only capture points gave score (victory) points. No 5 points from kills. Lord kill mechanics in maps with lord should mean: Only the lord kill gives you the win. (No other points besides that. Either 1:0 or 0:1. Even Stronghold where the lord kill ends the match counts other points besides that.) A tie (or close win/loss) wouldn't even be possible to measure then - if the system always makes the matches end some time. (There afaik are special mechanics on stronghold for after the timer runs out to make it end fast.)

[Vancho.8750]

> @"Exedore.6320" said:

> > @"Vancho.8750" said:

> > > @"Exedore.6320" said:

> > > You're doing a great job of illustrating my point with your uninformed opinion.

> > >

> > > Glicko/Elo may not be perfect, but it's been shown by basically every online game that it's still a reliable system and better than all others to date.

> >

> > UGH sorry for my filthy uninformed peasantry beliefs that chess single player system doesn't work on random multiplayer game without taking into consideration anything besides winning.

> > I want to play the kitten game not some broken kitten outside the game.

> Right. Many groups and people have done statistical analysis of Glicko2/Elo with variations for both 1v1 and single players in a team setting. The result is that they work well enough.

>

> For most players - all but those at the ends of the bell curve - the system works great and anything you're seeing is your own imagination. It's best if you just accept it and focus on improving yourself.

 

Keep telling yourself that, that doesn't change the quality of the games is bad. 9 out of 10 games are total blowouts win or lose. I'm still not sure if the matchmaker makes one team and then trows random people in on the other. Statistics may show that you get around 50% win rate and that looks ok on paper, but it doesn't fucking show that the games played were terrible.

[Faux Play.6104]

> @"Vancho.8750" said:

> > @"Exedore.6320" said:

> > > @"Vancho.8750" said:

> > > > @"Exedore.6320" said:

> > > > You're doing a great job of illustrating my point with your uninformed opinion.

> > > >

> > > > Glicko/Elo may not be perfect, but it's been shown by basically every online game that it's still a reliable system and better than all others to date.

> > >

> > > UGH sorry for my filthy uninformed peasantry beliefs that chess single player system doesn't work on random multiplayer game without taking into consideration anything besides winning.

> > > I want to play the kitten game not some broken kitten outside the game.

> > Right. Many groups and people have done statistical analysis of Glicko2/Elo with variations for both 1v1 and single players in a team setting. The result is that they work well enough.

> >

> > For most players - all but those at the ends of the bell curve - the system works great and anything you're seeing is your own imagination. It's best if you just accept it and focus on improving yourself.

>

> Keep telling yourself that, that doesn't change the quality of the games is bad. 9 out of 10 games are total blowouts win or lose. I'm still not sure if the matchmaker makes one team and then trows random people in on the other. Statistics may show that you get around 50% win rate and that looks ok on paper, but it doesn't kitten show that the games played were terrible.

 

Anet publishes > @"Luthan.5236" said:

> > @"Exedore.6320" said:

> > Analysis also shows that ANet does player vs. average team for rating adjustments.

>

> You mean ... they use your individual rating and compare it to the average rating of the enemy team? Isn't that weird? I mean ... it would make more sense to use the average ratings of both teams and compare them. Otherwise a strong player will lose more than he actually needed to lose. (Same for rating gain and a weak player.)

 

You are correct that it will make it harder for good players to rise and bad players to fall. However the match maker favors mixing skill levels to make two teams with the same average rating vs. making teams composed of players with the same rating. I'm assuming the thought process is it will make for more competitive matches.

[Faux Play.6104]

> @"Luthan.5236" said:

>

> Edit: Also winning 500 vs 450 should result in a different rating change than winning 500 vs 50. Since it means different skill level to win by a higher margin.

 

Problem with that is if you have a close match and both teams have highly rated players vs the average, both high rated players lose rating while the lower rated players vs the average team rating will gain rating.

 

The current system allows people that are slightly better to seperate themselves over time.

Also this is similar to any competition in life. If 10 closely matched people apply for a highly sought after job, only one person will get it.

[Luthan.5236]

There is still the problem with "bad" (winning by a huge margin or losing by a huge margin - which gets boring when one team dominates early through the whole match) matches. I don't think matchmaker does make 2 teams of the same average rating that often. (Maybe it tries but has huge flaws.)

 

Also there might be different performances depending on the map. (Players being less good at fighting can help their team a lot in maps where the side mechanics has a huge impact if they know how to play it and it can be gotten without lots of fighting. For example Temple if no player except you - a weak fighter - cares about the buff.)

 

And if course as I explained: Not wanting to play anymore after gettin dominated early (cause it does not matter for the rating if you lose 500 vs. 50 or 500 vs. 200 lots of people migh stop at 300 vs. 50 not even trying to fight back anymore) ... could make the matches more boring. (Leading to more matches that "seem" only unbalanced then the losing team could have actually fought more.)

 

---

 

For the winning 500 vs. 450 compare to 500 vs. 50 ... it could be seen as individual matches over each point. Without updating the scores after each point (only after the whole) match there should be same rating gain for winning and same for losing.

 

500 vs. 450 could be 500*x + 450*y / (500+450).

Similar for the 500 vs. 50. x for loss, y for gain. The /(500+450) to not get huge values to "normalize" it a bit.

 

There might be some expected value as well ... on how much you'd expect the team to win.

 

If the strong player gets rating loss in the end it can be interpreted as "he wasn't strong enough as his current rating suggested ... if he actually had been that strong the team would have won higher". (On the other hand shouldn't win rating then though ... hm. ... I guess comparing team scores vs. team scores would be better for this system.)

 

---

 

I found this paper https://rhetoricstudios.com/downloads/AbstractingGlicko2ForTeamGames.pdf

that suggests the method ArenaNet uses (if people here are correct). There are some things to point out though: He mainly uses the (if I understand it correctly ... I didn't look too deep into the math behind it but I did read the explanations mainly) the win percentage and the rating convergence of the top player to assess the quality of the systems.

 

We don't have things like "how close", "how good" are the matches. People with lower ratings could have more win ratio if the 100 matches he used in his simulation was not yet enough for them to climb to their actual rating or they could just have played against weaker teams more often. 25 different players and 2 teams of 5 people ... might behave different than 2 teams of 5 players with a huge pool of 1000+ (or even only 100 ) players where you get mixed with different players most of the time.

 

Edit: Also regarding "TrueSkill" mentioned here. Since it isn't available for free you can't just say that it probalby is less good because less games are using it. According to Wiki "TrueSkill is patented, and the name is trademarked, so it is limited to Microsoft projects and commercial projects that obtain a license to use the algorithm. " Shouldn't there be more the question why they are using their own system instead of using others - if TrueSkill was not that good? ;)

[BasedGod.3675]

its absolutely strange people are saying u can work your way up to plat from bronze, has it been done, sure.. the percentage is deff very low. You can be the best player in the world, does not matter, the people u get grouped with in bbronze and silver and sometimes low gold, arent there to follow the veteran players lead or listen, they dont even go to were u ping, u would literally have to controll all 3 caps and kill the entire team some games to win, alot of times peopel go into ranked the moment they hit the level req. Still used to doing whatever they want from unranked, silver and low gold is a bunch of road fighting and or idiots who watch mighty teapots youtube of the monthly tournies, see a crazy strat that works cause it cought someone off guard, then try to repeat it every ranked match they see, u dont kno how many failed raod fighting back cappers you see in gold. I would say the % of players going from silver to plat and staying plat is below 20%, i have seen with my own eyes and im sure other high level players have also, players in gold and silver who are wayyyy better then there rating, Duo queing is diff story by the way, its alot more possible to go from silver to plat duo queing nd being in voice comms, but solo queing, you do depend alot on who it matches you with. Also the ability of the team to adapt, and without communication which mos tthe time randoms in a random ranked game do not do. ONCE again it is possible to go from silver to plat, but its just very low %, The best way to fix this is to not force pve kids and weirdos to have to do ranked to get a certain mat or achievment, Must have a unranked win % of 50% or better on the class to que, and a avoid queing with player option (players who use this must accept and realize slightly slower que times, you should be able to mark someon like u would if u blocked them, that avoids you from queing with them, would be done without the other person even knowing. is it fair that an y idiot that hits level req can que, go 3-7 in prelims get gold 1 and ruin it for people, no.. but its not against any law or rule either so, its on anet to make ranked a place people go to compete, not be some weirdo win or lose i have fun person, thats what unranked is for. No one takes Ranked in any game seriously these days, theres no drawback.. and the rewards are 10x better for doing ranked... its on ANET to set the difference in unranked and ranked. And out of experience i have a friend who just started, got to the level req for ranked, qued.. went 3-7 got gold 2... if a 30% win rate gets u to gold 2, then imagine how many other new players and bad players get gold even know they went 3-7, gold should be a slightly above average placement, like 6-5 not 3-7 silver is suppoused to be the average/casual.. so just imagine how many people are in the high silver low gold area, even know they suck.. that deff does increase the difficulty for players to get out of the hole.

[Exedore.6320]

> @"BasedGod.3675" said:

> its absolutely strange people are saying u can work your way up to plat from bronze, has it been done, sure.. the percentage is deff very low

Probably because the rating system works. Few people are misplaced, so few people make large rating shifts in a short time.

[Tharan.9085]

> @"BasedGod.3675" said:

> silver is suppoused to be the average/casual..

 

Actually gold2 is the average/casual

 

 

[hotte in space.2158]

> @"BasedGod.3675" said:

>i have a friend who just started, got to the level req for ranked, qued.. went 3-7 got gold 2.

With 3 wins and 7 losses you wont get matched in gold 2

>silver and low gold is a bunch of road fighting and or idiots who watch mighty teapots youtube of the monthly tournies

>is it fair that an y idiot that hits level req can que

Stop calling other ppl idiots ! I prefer to have a lower ranked player in my team than a toxic one who just thinks he was superior

>imagine how many other new players and bad players get gold

Cant find your name in top 250 btw

 

[Khalisto.5780]

> @"BasedGod.3675" said:

> its absolutely strange people are saying u can work your way up to plat from bronze, has it been done, sure.. the percentage is deff very low. You can be the best player in the world, does not matter, the people u get grouped with in bbronze and silver and sometimes low gold, arent there to follow the veteran players lead or listen, they dont even go to were u ping, u would literally have to controll all 3 caps and kill the entire team some games to win, alot of times peopel go into ranked the moment they hit the level req. Still used to doing whatever they want from unranked, silver and low gold is a bunch of road fighting and or idiots who watch mighty teapots youtube of the monthly tournies, see a crazy strat that works cause it cought someone off guard, then try to repeat it every ranked match they see, u dont kno how many failed raod fighting back cappers you see in gold. I would say the % of players going from silver to plat and staying plat is below 20%, i have seen with my own eyes and im sure other high level players have also, players in gold and silver who are wayyyy better then there rating, Duo queing is diff story by the way, its alot more possible to go from silver to plat duo queing nd being in voice comms, but solo queing, you do depend alot on who it matches you with. Also the ability of the team to adapt, and without communication which mos tthe time randoms in a random ranked game do not do. ONCE again it is possible to go from silver to plat, but its just very low %, The best way to fix this is to not force pve kids and weirdos to have to do ranked to get a certain mat or achievment, Must have a unranked win % of 50% or better on the class to que, and a avoid queing with player option (players who use this must accept and realize slightly slower que times, you should be able to mark someon like u would if u blocked them, that avoids you from queing with them, would be done without the other person even knowing. is it fair that an y idiot that hits level req can que, go 3-7 in prelims get gold 1 and ruin it for people, no.. but its not against any law or rule either so, its on anet to make ranked a place people go to compete, not be some weirdo win or lose i have fun person, thats what unranked is for. No one takes Ranked in any game seriously these days, theres no drawback.. and the rewards are 10x better for doing ranked... its on ANET to set the difference in unranked and ranked. And out of experience i have a friend who just started, got to the level req for ranked, qued.. went 3-7 got gold 2... if a 30% win rate gets u to gold 2, then imagine how many other new players and bad players get gold even know they went 3-7, gold should be a slightly above average placement, like 6-5 not 3-7 silver is suppoused to be the average/casual.. so just imagine how many people are in the high silver low gold area, even know they suck.. that deff does increase the difficulty for players to get out of the hole.

 

you prolly can and it's not that hard, i tried once and I made it from bronze 3 to gold 2, i wasn't good enough to make to plat at the time, the problem was me not the matchmaking.

 

You talk a lot how hard is to carey your team, and you don't realize the other team is equally bad and they don't have you in their team. It surely takes a lot of matches to do it, because if you not bronze you have to lose a lot of matches and when you start winning you already in 11-13 rating per match.

 

30% ish matches you'll just lose, ppl are very toxic and often go afk even when they're winning, they run into 3 ppl die in 2 secs and type on map chat "no team gg afk" likewise you win 30% of the matches pretty much for the same reason you lost the other 30%

 

now you have 40% to work with, if you're high gold 2 you can go 3-1 easily until you reach gold 1 and that's 60 % wr.

 

How you do it? playing something really bursty or really tanky, I did it with water weaver, ran far everymatch, they have to rotate 4 to kill me, they tunnel vision you and don't look at the minimap and your team is basically 4v2 the whole game

 

Playing a bursty build is even better they como in line 1 by 1 if you manage to kill them quickly, they won't group up and most are not smart enough to avoid you

 

the other team mistakes is a reliable source for you to win, just forget about your team mistakes and profit on the enemy's