Jump to content
  • Sign Up

Matchmaking algorithm too rigged,


Bast.7253

Recommended Posts

@Crinn.7864 said:

@Crinn.7864 said:The rating systems distributes across a curve with a mean of 1200. Gold starts at 1200, so a player that is placing Gold is merely slightly above average. Your rating placement is done with the exact same glicko2 algorithm that rates you after placements. Placements is merely a mechanic for hiding your rating change for the first 10 matches, the reason the first ten are hidden is because glicko2 has enormously high deviation in early matches, and devs don't want players flipping out over the rating swings during early matches.Thank you for quoting the rating algorithm but I still don't understand how (and if) individual skill and game performance are calculated when defining a player's rating.

It does not use individual performance. Glicko is a outcome based algorithm.

Thanks again for the good information. This is so interesting.So carrying hard or being carried in a match weighs almost the same for the rating engine.
Link to comment
Share on other sites

  • Replies 203
  • Created
  • Last Reply

Top Posters In This Topic

@Delweyn.1309 said:

@Delweyn.1309 said:The problem is the "carry" philosophy of the game. Dev think that a good or very good player can carry a group of numerous (2 ot 3 ) much lower players.So an average team of 1420 with (1700 + 1500 + 1300 + 1300 + 1300) can face a 1440 with (1500 + 1500 + 1400 + 1400 + 1400).

But not, the average lvl of team 2 will dominate most encounters against team 1, even if the 1700 player can kill any player of team 2 in 1v1. So it ends that team 2 will clean team 1 and then will jump at 3 v 1 vs the 1700 player of team 1.

But in the Anet stats, 1420 is ok vs a 1440. Systrem is working fine. Just 20 point of difference in this match. It's OK.

Fine let's see how that works with Standard Deviation(S.D.):Team 1Mean: 1420S.D.: 144

Team 2:Mean: 1440S.D.: 48

That's nearly a difference of 100 opposed to the 13 on average and 39 in the higher tiers. But let's say you said 1600 or 1500 instead of the probably exaggerated 1700 order to make a point in team 1 (also to show how S.D. behaves when you're starting to make up random team ratings)
  • With 1600 instead of 1700:Mean: 1400S.D.: 120
  • With 1500 instead of 1700:Mean: 1380S.D.: 96

None of that matches up with what the devs have told us.

Well, when you know that Sindrener got teamed with a silver player you know that I didn't exagerated the numbers.

And I would add that the numbers given are average. So if 90% of matches have a SD difference of -30, but 10% have SD difference of +100 then the global average number for SD would seems not so bad isn't it ?

Im sorry, but there's no such thing as a negative difference. And SD values (being an average of differences) is also never negative

But you're right that there will always be singular cases where you can point out the match maker doesnt function optimally. Especially in the higher rating regions.

Though even in this case, What you can say is that with teams with high standard deviations (like is the case wigh sindrerer and someone in silver) you have a team with players in very differing skill levels being matched up with a team also very differing in skill level. But the difference between standard deviations can be close to nothing while obviously having wildly differing teams. Someone earlier made a similar comment on that earlier btw.

Just so you know, I'm in no way claiming that there's no problems btw. Im just showing how standard deviation works. Even with matching standard deviation and matching average skill rating you dont automatically have an system that balances matches perfectly. (although its a good start.)

Link to comment
Share on other sites

I skipped all non-dev posts (except OP), sorry guys.I haven't played most of the last 2 seasons, because lack of time and disappointed with balance patches. But here's my 10cents:First you can isolate the OP's complaints into 2 problems:a) Matchmaker created team comps that seriously disadvantaged one of the teams;b) Matchmaker is unable to place everyone into well balanced, same rank teams.

Problem b) is caused by a conjunction of problem a) and a few more factors. As in the game doesn't have enough players playing sPvP for the matchmaker to produce proper results. This comes from a lot of factors, one of them being poor balance and one having to relog to be able to pick counter-plays, which only a slim margin of players will do, and it's up to Arena Net to work to change that (if it's at all possible to revitalize sPvP at this point).

Problem a) on the other hand is entirely fixable with a new approach to sPvP that allows for players to draft pick their classes after being matched solely by rank, which allows players to counter each other's picks, and leads to - hopefully - more balanced team comps. A lot of people's complains about sPvP, like teams relogging to pick full necro/firebrand comps or necro/warrior comps now, and other issues with team comps, people asking for class-based MMR, etc all this could be fixed with a draft pick.

Link to comment
Share on other sites

@Delweyn.1309 said:

@"Ben Phongluangtham.1065" said:I asked for numbers for this season to be pulled since people are always interested:

Average skill rating difference between teams: 11.866Average standard deviation difference between teams: 11.643Average rating difference in a match: 98.203 (min rating vs max rating across all players in the match)

One thing to keep in mind that end score difference never means that the match didn't start off even. Scores tend to snowball in our game for a number of factors. Some due to map layout/mechanic design. Some due to human nature, as people tend to tilt or give up after getting behind by a certain number. Sometimes people play above or below their potential. That's just part of human performance.

Im curious about what the average rating difference in a match is when there is a legend tier player in it? This season or last season or whatever

The sample of games with a legend player is pretty low, so I extended it down to 1700 (plat 3+). Here's the data including last season and this season:Average skill rating difference between teams: 14.16Average standard deviation difference between teams: 13.55Average rating difference in a match: 189.71Percent of games with average skill rating difference >50: 0.4%

There's a significant increase in rating range at this level, but as seen the skill rating mean and standard deviation differences between teams are pretty similar. The rating range is always going to be higher at the edges of the rating curve as a tradeoff with keeping reasonable queue times , but it doesn't stop the matcher from making fair teams which is the most important.

As some others have pointed out, the average game doesn't always tell the whole story. So I also grabbed all matches from last and this season and paired it down to the set of games with a rating range over 200 (these account for about 10% of all matches). Here's the data for that set:Average skill rating difference between teams: 19.83Average standard deviation difference between teams: 33.67Average rating difference in a match: 279.42Percent of games with average skill rating difference >50: 7.5%

While the skill rating and standard deviation differences are a bit higher in this set, these numbers aren't too bad overall.

I'm not trying to say that the matcher is perfect, but the vast majority of games are pretty balanced.

Thanks for the response, i was genuinely curious. I dont doubt the balance of the games. The only problem i really have with the games is that sometimes your influence in the game feels meaningless since regardless of how well you do your team can still kitten you over by losing while outnumbering heavily.

The problem is the "carry" philosophy of the game. Dev think that a good or very good player can carry a group of numerous (2 ot 3 ) much lower players.So an average team of 1420 with (1700 + 1500 + 1300 + 1300 + 1300) can face a 1440 with (1500 + 1500 + 1400 + 1400 + 1400).

But not, the average lvl of team 2 will dominate most encounters against team 1, even if the 1700 player can kill any player of team 2 in 1v1. So it ends that team 2 will clean team 1 and then will jump at 3 v 1 vs the 1700 player of team 1.

But in the Anet stats, 1420 is ok vs a 1440. Systrem is working fine. Just 20 point of difference in this match. It's OK.What you are saying does not even make sense. You are trying to generalize a situation that has a million variables.

But following your logic I can turn it around: The 1700 rating player shouldn't take 1v1s then, just go for teamfights, erase the lot worse players, then +1 anyone till respawn -> repeat.You can't judge a matchmaking system on theoretical anecdotes.

Link to comment
Share on other sites

Just lost a match, yet again, with a 100 point lead. People not defending caps. Downing a necro and then being bombared by a Mesmer who I get to 10%, and our team thief STILL dies to the Mesmer.

I'm just over it.

I want my placements to be reset or to be manually adjusted to bronze at this point. I'm done with even thinking I have the slightest chance of getting back to where I should be at this point because matchmaking is perpetually putting me with people who have no situational awareness. Two scourge dying to 1 because they apparently do not know how to transfer conditions back.

Just all of it. I'm done with it and would like to just be placed in the lowest possible division because that is where the matchmaking is taking me. I'm tired of the 10-12 game losing streaks over careless errors only to be followed with 2 or 3 wins.

Link to comment
Share on other sites

@"rank eleven monk.9502"

Anecdote means this:

  • a short and amusing or interesting story about a real incident or person. "told anecdotes about his job"synonyms: story, tale, narrative, incident;More - urban myth/legend;"amusing anecdotes"an account regarded as unreliable or hearsay.

Pointing at something and saying it is an "anecdote" is intrinsically just another anecdotal claim, without the evidence to support that it is or is not truth. In other words, one can claim something is anecdotal in nature, but they cannot point and say it is anecdotal in nature, without some form of evidence that it is indeed anecdotal. Sorry man, with as many times as you are using the word to debunk all claims in this thread, someone needed to point that out.

The sheer inflection in your posts would imply that no human being outside of the Arenanet staff, would possibly be capable of understanding or noticing how the algorithm's patterns work, especially not the players who take the time to post personal feedback. I don't know man, it seems unreasonable.

Link to comment
Share on other sites

A-net could remove most doubts about the matchmaking by showing the MMRs of everyone in a match. Until they do I have no reason to 100% believe them. It doesn't have to be they are lieing either they could be mistaken themselves about there own code or it is sometimes acting in ways they don't fully understand. Smarter people then there coders make mistakes all the time. Blind faith is for little children. As for the evidence being anecdotal well they refuse to give us the full information that would prove themselves 100% correct so that's all I can say about that. I feel like when the playerbase makes a request devs should fullfill that request if it is reasonable. It would take little time to add a pvp window with both teams average MMR and the MMR of every player in the match without the names attached and possibly even the win probabilty they have calculated. More information is always better.

Link to comment
Share on other sites

Same experience here. Also this season I had one team mate 1 minute afk during placement matches. The game was counted as a normal loss. Nice Anet. Very nice....You need to improve that matchmaking. There is no incentive to play solo this way.

You will continue to lose your player base. Not only are you denying players their rightful spot with your algorithm you also deny the scene its growth.

If you get more challenging matches with players smiliar in skill you will slowly see an increase in overall match quality. Right now the system let's you play with a handycap. Instead of facing more challenging opponents you try to force balance on the leaderboards by giving the players that are clearly better than their division a heavier burden to overcome. Why is it that players like me always have the same experience. We always have to carry the entire game to win. And by carrying I mean killing the entire enemy team multiple times taking all of our team mates jobs, never dying and winning every match up....

Let me paraphrase what is going on with your algorithm: It's like watching an Olympic relay run with one team consisting of Usain Bolt and 3 toddlers vs a full team of USA runners.

Your system is not working and your arrogance in denying this is annoying.

Link to comment
Share on other sites

it'll NEVER be balance unless match making can read class/elite spec being used.

Can't have team A with Fb/scoruge/mesmer /thief and warrior on a team vs Team B with double scoruge , thief, engi, warrior.

match ups like that with "rating" being equal to each other just won't mean jack if team A running more of a meta/proper team comp then team B.

Link to comment
Share on other sites

@Cal Cohen.3527 said:

@Ben Phongluangtham.1065 said:I asked for numbers for this season to be pulled since people are always interested:

Average skill rating difference between teams: 11.866Average standard deviation difference between teams: 11.643Average rating difference in a match: 98.203 (min rating vs max rating across all players in the match)

One thing to keep in mind that end score difference never means that the match didn't start off even. Scores tend to snowball in our game for a number of factors. Some due to map layout/mechanic design. Some due to human nature, as people tend to tilt or give up after getting behind by a certain number. Sometimes people play above or below their potential. That's just part of human performance.

Im curious about what the average rating difference in a match is when there is a legend tier player in it? This season or last season or whatever

The sample of games with a legend player is pretty low, so I extended it down to 1700 (plat 3+). Here's the data including last season and this season:Average skill rating difference between teams: 14.16Average standard deviation difference between teams: 13.55Average rating difference in a match: 189.71Percent of games with average skill rating difference >50: 0.4%

There's a significant increase in rating range at this level, but as seen the skill rating mean and standard deviation differences between teams are pretty similar. The rating range is always going to be higher at the edges of the rating curve as a tradeoff with keeping reasonable queue times , but it doesn't stop the matcher from making fair teams which is the most important.

As some others have pointed out, the average game doesn't always tell the whole story. So I also grabbed all matches from last and this season and paired it down to the set of games with a rating range over 200 (these account for about 10% of all matches). Here's the data for that set:Average skill rating difference between teams: 19.83Average standard deviation difference between teams: 33.67Average rating difference in a match: 279.42Percent of games with average skill rating difference >50: 7.5%

While the skill rating and standard deviation differences are a bit higher in this set, these numbers aren't too bad overall.

I'm not trying to say that the matcher is perfect, but the vast majority of games are pretty balanced.

Try to make the same math on lower rated entry ... ( bronze, silver, and also part of gold ) and u will see , i belive, i completely different situation. How many players are between legendary and plat 3 and how many are between gold and bronze ? how are the matches for the bigger part of the playerbase ?

Link to comment
Share on other sites

@"zoopop.5630" said:it'll NEVER be balance unless match making can read class/elite spec being used.

Can't have team A with Fb/scoruge/mesmer /thief and warrior on a team vs Team B with double scoruge , thief, engi, warrior.

match ups like that with "rating" being equal to each other just won't mean jack if team A running more of a meta/proper team comp then team B.

Since I cant make out if you are advocating such a system from you post alone, so im giving you the benefit of the doubt and say you don't.Cause that - unfortunately - is simply impossible to implement.

Link to comment
Share on other sites

Important for someone to say this:

Players understand & accept the presence of hiccups and general flaws in the match making that could never reasonably be fixed. They aren't complaining about that. They are complaining about elongated streaks of very unreasonable match making. When some guy reaches plat 1 for his first time, then instantly goes on an elongated lose streak for 10 games or so, that is where the heat is fueled to take the time to come into this forum, and tell a story about it.

With how many claims are stacking up in just this thread alone, that agree on experiencing the exact same pattern, I'd say it's worth looking into if not changing behavior in the algorithm. I mean at the end of the day, what's important is that players aren't getting pissed off and leaving the game because lose streaks are making them feel like the system is choosing when they win or lose.

Link to comment
Share on other sites

@BikeIsGone.8675 said:

@"zoopop.5630" said:it'll NEVER be balance unless match making can read class/elite spec being used.

Can't have team A with Fb/scoruge/mesmer /thief and warrior on a team vs Team B with double scoruge , thief, engi, warrior.

match ups like that with "rating" being equal to each other just won't mean jack if team A running more of a meta/proper team comp then team B.

Since I cant make out if you are advocating such a system from you post alone, so im giving you the benefit of the doubt and say you don't.Cause that - unfortunately - is simply impossible to implement.

and that's a reason why this game is always going to have unbalance match up.

Link to comment
Share on other sites

Here is a good old read: https://forum-en.gw2archive.eu/forum/game/pvp/High-MMR-is-punished-for-solo-que/page/1

Shows how long this has been going on. Maybe with the Glicko algorithm it cannot be fixed and Arenanet just does the best that they can with it. It also serves to demonstrate how "anecdotal" stories from 3 or 4 years ago can end up being 100% true. Pay attention to the last page in particular.

Link to comment
Share on other sites

@"Trevor Boyer.6524" said:Here is a good old read: https://forum-en.gw2archive.eu/forum/game/pvp/High-MMR-is-punished-for-solo-que/page/1

Shows how long this has been going on. Maybe with the Glicko algorithm it cannot be fixed and Arenanet just does the best that they can with it. It also serves to demonstrate how "anecdotal" stories from 3 or 4 years ago can end up being 100% true. Pay attention to the last page in particular.Wow, you are actually persistently using the same argument for 3 years. Hats off.

During this time you could have just learned to play PvP properly.

By the way, the matchmaking algorythm has been changed multiple times since 2015.

Link to comment
Share on other sites

The simple fact is that people have good days and bad days for a variety of reasons. Your opponents are only a small part. Build changes, mental and physical state, network stability etc are all big factors in the type of day you have.

Odds are if you're losing a bunch of games in a row it's probably NOT the matches that are causing it. It COULD be win trading or hacking or alt accounts, though.

Link to comment
Share on other sites

I think one thing I don't understand is why the matches need to be progressively harder the longer the win streak.

Eventually you will run into a bunch of people higher skill than you and you will lose, so does there need to be a system that makes it increasingly harder to beat people that may be of equal skill just because the system feels that you've exceeded your personal rating?

Of course it's going to be difficult in a mode like this where your "skill level" is depending on the rest of your team and sometimes strategy more than combat ability, i.e. knowing which point to rotate to in order to keep the points flowing in, but it still seems a bit unnecessary.

Is it fear of people getting too high? Because I'll admit there are plenty of people I know without a doubt would beat me everytime, like Valen the thief who seemed to be speed hacking but I guess is just extremely good and well known for it.

I just think there are enough challenges with random team draw to be stacking odds for or against someone based on a number system that may put you in a less favorable team.

You can tell people to "learn pvp" and "git good" but in the end the mode itself boils down to you and 4 other people controlling 3 points and side objectives and strategizing where to be which sometimes involves actively avoiding other players in "pvp."

Kinda silly.

And I'm not sure attacking Trevor who, even if adamantly voicing an opinion you don't agree with, or simplifying and attempting to invalidate someone's opinion with something as simple as "learn pvp" is really doing much to add to the discussion.

Anyway, that's my latest 2 cents to the discussion. Why? I guess I don't understand how the current system benefits anyone if teams truly do get more difficult to face the higher your rating instead of keeping the rating as close as possible and ensuring that it's always an even rating between both? Or why I in gold 3 would face a duo that's currently in plat 3. In what way is this a fair match-up? We had maybe one MAYBE two tier 3 plat on our team as well, but then you're expecting the ones on my team to carry me against their plat 3. How is that fair to them? Is this just an issue of low population?

Link to comment
Share on other sites

@Ben Phongluangtham.1065 said:

@"mortrialus.3062" said:Hasn't there been statements by developers stating that that average distance in player score between highest and lowest players per game is less than 50?

No. The average skill rating between teams is less than 50. The deviation is higher than that. Don't have the number off the top of my head though.

We do our best to get the standard deviation as close as we can though.

And no, I say it often, we don't rig matches to end loss or win streaks. But many people will never believe it.

I sincerely think that you should end streaks. Extended loss streaks discourage players ( like me, my personal record is approximately 68 and I've heard reports as high as 100 straight losses. This is on my alt account on which I did finally get my Ascension backpack. The grinding had me in tears at times. )

After all, ANET had no compunction in setting up deliberately tilted matches in season 2 and 3.. it was even touted as "competitive."

I suspect that those two seasons of openly tilted matches are a big part of why many people now believe the match making is rigged. I have to admit.. I don't have much confidence in the MM myself. I think it is set up to benefit high rated players at the expense of low rated.

Link to comment
Share on other sites

Three needless losses yet again. Including a thief that camps mid all match while I'm running around as apparently the only person that can even down somebody just trying to keep caps.

Why do I never get put in matches with people that carry me? This same thief was on the opposite team and just ran around temple the entire time, with somebody on my team chasing them the majority of the time.

Why should I have to be in three different places or show up and kill the scourge that's somehow wiping everybody at the same time?

This is stupid. I have 179 losses and 176 wins. So clearly from what I've learned from this and other posts is that I need to play less because the more I play the more I am expected to carry people making stupid carless decisions.

I'm not even good. I don't know why I'm expected to be the only one who cares about points the majority of the time or seemingly the only one that can even down people instead of just endlessly swinging at people, respawning, and letting the one or two that are downed get ressed every single time.

I'm sure I make a lot of mistakes, but I keep getting groups where I literally have to be the one that does everything. Capping and killing.

So next season I should just limit my games to... what? 20? Or is it too late now and I should pick up an alt account?

People far worse than me with a lot higher rating. So something isn't working or they are getting carried or paying for wins constantly.

Link to comment
Share on other sites

it is clear that you do not explain what the algorithm really does or do not know exactly how it works.

I have two accounts in one I am in Platinum 2, in a game of Platinum 3.

and in the other I am in gold 2, gold 3 all the time in a loop of losses won during 5 days.

my skill is the same, my characters the same. The pairings weighed.

I in platinum 2 do not see the games with 2 necros + 2 guard + 1 mesmers and up to 3 equal classes.

please, the scoring system is not adequate.

Link to comment
Share on other sites

@rank eleven monk.9502 said:

@"cptaylor.2670" said:People far worse than me with a lot higher rating.Maybe this seems far-fetched and really unbelievable.. But, WHAT IF they are actually better than you, that's why they are on a lot higher rating, while you are just being biased and overrating yourself, leading to blame the system instead?

Just wondering.

The problem is how you define someone is " actually better ". Most of top100 players are good so i won't argue about these, but you have many players in gold 3/platinum that are good at playing scourge or mesmer, but have litterally no idea about some capture points basics. Alternatively, you have players in gold 1 that are playing a non-meta build but have good knowledge of map awareness.

Link to comment
Share on other sites

@"Ben Phongluangtham.1065" the problem is not the algorithm, but the dead population. I can not remember the last time I actually had a real teamcomp in ranked. I am losing 500-22 to spirit weapon guardians who don't know how to play the game, because it is not possible to play the game in this matchmaking and won the next game 500-17 just now. Every single game is a complete stomp for either side on plat 2 and higher because you can't build fair teams with the players available. Only fix is to remove solo q ranked. A cardgame like Hearthstone includes less rng than gw2 ranked.

Also playing certain roles like support is a compeltely unviable strategy on plat 3+, since the game can't pair you with players on your level and the game literally turns into: who farms the noobs the fastest. Everyone runs burst builds and you cant support dead players.

Also often times teams lack roamers or duelists, causing bad matchups and ruin all type of team play. I have never seen any other game with less teamplay involved than GW2. The only time you get to play as a team is the monthly tournament, but who on earth would practise as a team to play 1 game per month ?

as you see here https://picload.org/view/dogpgawl/gw135.jpg.html I only had 1 somewhat playable ranked game that wasnt a spawncamp out of 5 games today. This happened the whole previous week while I was streaming. This system HAS to go and needs to be replaced.

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...