Questioning the Matchmaking system, need genuine help to understand.

Dokh·Aug 13, 2019·3 votes·1,393 views

So it happened a few time this season, and I just can't figure out what's wrong. Here's an example, yesterday I was on roll, climbed a whole rank and a half without a single loss. And then the trolls start appearing every single game for about 5 to 10 games, insanely toxic people, unranked players that obviously aren't smurfs and so on.

What's happening here ? If it happened to me like once I guess I could simply blame bad luck, but it seems pretty consistent. Go on 10 wins in a row and then endure what I stated above and the cycle starts again.

I'd like to read opinions or genuine knowledge about that. No ''RiOt DoEsn't CaRe'' Although the system is clearly not perfect, I'm pretty sure they try to have a healthy matchmaking system.

5 Comments

MordridtheBlackAug 13, 20193 votes

sadly the match making system is beyond flawed

probably even more so in ranked

now I am unsure if this is 100% accurate, so don't quote me on it

as far as I know, league uses a rather similar system for ranked that smite uses

which is pretty much just

Queue up for a ranked match (with your chosen positions)
System searches for players to fit into a game (with little to no difference in rank)
X amount of time passes unable to find enough people
System broadens the search (allowing for greater rank difference)
repeat steps 3 and 4 until enough players are found

that downside is that league also has the hidden mmr system to skew that process up

it doesn't matter what your physical rank is (or if you even have one to begin with), you are given a match making rank according to some logic only riot knows

this little bit of extra nonsense makes things difficult, because you can have people that are unranked physically but have a gold mmr, or are plat but have bronze/silver mmr, and thus the system ends up giving you a match that looks weird

what you speak of is rather common in ranked match making

and i'm unsure if there is a fix riot can implement without trashing the current system and redoing it from scratch

DokhAug 13, 20192 votes

Yeah I was speaking of ranked, haven't played any normals in a long time... It's really that cycle I was talking about that has me puzzled a bit.

But like you mentioned on your points 3, 4 and 5, it might have to do with the time of the day I play in : for 2-3 days I'll play when there's plenty of people, then I for 2-3 days after I'll play when there's less and thus I get eaten by the "broader research"... Not that I'm playing like a pro all the time, but the difference is just too obvious to blame only myself in these "cycles". Might make some kind of sense at least. And I know about the Mmr system, which is why I'm not too surprised when I see a silver on my team even though I'm high gold or plat.

Thank you very much for your explanation though.

Sailor MintAug 13, 20191 votes

Unless you're high Diamond+, it won't get past step 2.

The only arguments are:

"Are people getting their secondary position (or autofilled) facing people who got their primary position fair".

And even if you answer no to that question, you need to remember that before position selection it was 100% RNG.

Kai GuyAug 13, 20191 votes

That's why Riot decided to keep Positional MM. They only removed Positional Ques. That makes your Rank, as a less realistic measurement of skill because the gap between position assigned MMR and Rank is unknown. Adds more uncertainty.

Solution I suppose is Ranks don't unlock until players put in the minimum required # of games to be medium or low uncertainty but that would not go over well with players so I doubt Riot will do so.

on March 13 we’re ending the position ranks preview in North America and Korea and discontinuing the rollout of position ranks globally

Disclaimer. This is taken out of order from the post to highlight positional MM is still in effect to the best of my knowledge.

On the matchmaking side, we’re happy with how positional matchmaking—not to be confused with position ranks—is working. Queue times have improved for many players and off-position win rates have gone up. On top of that, primary position rates have improved, and in many cases, autofill rates have gone down.

Sause

Kai GuyAug 13, 20193 votes

On the Whole Riot adds a massive amount of extra shiny distractions to make players feel better about their rank or to encourage participation. My personal viewpoint is that they have neglected Player behavior systems to much during this and that the unnecessary features implemented to encourage popularity and participation are becoming clutter.

First up, I am one of the most active supporters of MMR systems. Most posts about them are BS and only serve to encourage making the game less fun. I will share some logic. The game is PvP. If The game had 100% flawless balance and incredibly fun gameplay potential, it would still be unfun and shit if the majority of players are awful. Because Game play Is Player Vs Player The behaviors of player directly impact how enjoyable the game is. MMR is an derivative of Arpad Elo's Work. Its not a very complex idea, in a game of skill Your ability to win is impacted by your skill. So the bigger a skill gap between two players the bigger the W/L ratio will be. There for accounting for W/L (And draws if the game has that as an option) Are how you can rate skill IF you pay attention and track who is playing who.

Rn=Ro+K(W-We) This is his Rating Adjustment formula. Your New raiting Is your Old Raiting after you add the result of the match modified by probability expectations multiplied by account and system uncertainty. MMR can have more variables, but at a minimum when your talking about a 5v5 game to port this adjustment formula over you need to make shure that you run it by individual players, not teams. Because uncertainty will wildly fluctuate by players.

I can go on a large wall for the topic but it wont address your post. Hit me up if you have curiosity on the topic in a reply thou.

So MMR. Mathematically proven to work, Very sound logic, Been around for a long time, Well documented and reviewed, basically the industry standard for sorting players by skill. Reactive, Self correcting, Pretty Easy to build thou optimization for specific games requires a good deal of work.

Riots system Is not a purist MMR system and there are some extra steps and features.

They Hide MMR and show Titles instead. This basically (duoing issues apply but that's about it) Makes no impact on MM. MM uses MMR, Titles mean nothing and have 0 representation of skill when inaccurately assigned, which is expected to be common when an account is high uncertainty. The reason being that you expect about half the players to lose points every season. Your starting MMR for all accounts is the same, so about half of the players lose out on that, then each soft reset you can see accounts sink in reflection to population size changes. Generally seeing MMR changes discourages players because the trend is that everybody thinks they are superior to their peers and don't take evidence to the contrary well.
As MM uses MMR not Titles, Duoing restrictions off Titles is inherently flawed. MM has a range to pull from, the tighter the range the more fair games are expected to be. this does not inherently impy the gameplay will feel fair or fun, just that the odds for Team A and team B are closer to 50% then if your MM randomly grabbed any players qued. The concept is Class intervals, When your probability odds for player at a raiting of A hits a level they are expected to outclass player B at a fairly dramatic level. In Chess, using Elo and Logistical distribution. A class interval in that system is 200 points, and the odds are 76% to 24%. So 3 out of 4 games. (This is what We Variable is about. The probability distribution odds.) So... Titles move slower then MMR in riots system, A gold 1 player can Que with Silver 4. Gold can have Plat MMR and Silver can have Bronze MMR so your looking at a skill gap that covers something like 80%+ of the games entire playerbase that MM is some how expected balance out. People like playing with friends but this makes for worse MM quality, even if you splash MMR into MM for folks who duo.
Riot kept Positional MMR despite the removal of positional Ques. This makes your Title even less accurate because MM looks at 5 different values when trying to build a game. I get the idea, its a smart concept on paper, but it would have been simpler to just publicly announce that players who can only play 1 role or 1 champion are inherently weaker at the game then anybody at their level who can do more. Its an attempt to offset the issue of Filled players and people who don't learn multiple roles. It makes your Rank and division a less accurate representation of skill.
Demotion protection in Division 4's like other seasons in division 5 is stupid large and a lot of folks give up when they cant climb. Probably the same exact logic behind hiding ranks, it makes people feel a little better because they don't see themselves sink.

For when you feel your seeing mismatched accounts, Why was player X in my game? Lets use your unranked Fizz as an example. There are a few steps to take. 1. look at prior season, see where he ended. 2. Look on his history and see if he participated much in Preseason, that impacts your starting MMR values. 3) Look for duoing, WE generation in response to dramatic skill gaps can put some higher Gains onto some one. This would be most dramatic early on during high uncertainty.

Players don't get mmr and fail to treat it as a balance. Majority of everybody seems to overvalue "streaks" With many people treating 3 games as a streak but streaks don't really matter all that much. MMR is a balance. Example, lets fill in some values. W W W L = W L W W = L W W W. because (1 +1 +1 -1)=(1-1+1+1)=(-1+1+1+1). So.... 2=2=2. Take a step back, don't try to prove that OMG teammates holding me back. Streak = won 3 or more games, then you lost because of a bad teammates. Any loss after your arbitrary value for a streak meets the condition of "ending your streak". Play 3 lose 1. Play 6 lose 1. Play 9 lose 1. Streak ended for all of them. Your Rank balance thou would be +2 +5 or +8. (More complicated due to K and We, just using a simplification for illustration.)

What sounds more reasonable. RITO GAVE ME BAD TEAMMATES TO FORCE A LOSS TO END MEH STREAK. Teammates who underperform increase my teams odds to lose.

If some one started to claim. "My team only lose because every player on it was better then the enemy team." Does that sound realistic? Given that basically everybody considers Winning and losing to be the marker for skill... When your team just gets RNGed and the champion matchups or player skill don't match.. why is it surprising to lose when your on the weaker team?

People, even in gold and plat MMR will seem to think the game is unfair unless every member of their team is outperforming the enemy counterpart. That's not realistic, that's not fair. But look at how salty folks get over 1 lane losing or the enemy jungler being better. Like... We won because our top mid support jungle and adc are all better then the enemy! Duh? 1-2 folks playing poorly wont kill your game, the more dramatic their failure is might snowball it. but realistically folks need to expect that the odds are very good their team wont 100% beat the enemy in every position.