Flaw in MMR Model? (Let's do Math!)

Flaw in MMR Model? (Let's do Math!)

in PvP

Posted by: Dragon Ruler X.8512

Dragon Ruler X.8512

This write-up assumes a few things.
1) The 50/50 (win/lose) ratio is intended.
*2) The system can be “gamed” if the first assumption is true.

Let’s start fresh:
A new player walks into the PvP Lobby and has…
0 Games Played
0 Games Won
0 Games Lost

We start off kindly and let the player win.
1 Game Played
1 Game Won
0 Games Lost

The 50/50 ratio kicks in.
2 Games Played
1 Game Won
1 Game Lost

We again are kind and say they win the next round.
3 Games Played
2 Games Won
1 Game Lost

The 50/50 ratio kicks in again.
4 Games Played
2 Games Won
2 Games Lost.

Obviously this isn’t exactly what happens, but it can seem like that at times.
However – let’s take it a step further.

The same player from above has now played 100 matches:
We might expect something like…
100 Games Played
42 Games Won
58 Games Lost

However, let’s now assume that the player caught on to the 50/50 rule. The player decided to skew the games over and over by not trying very hard or just trying to be a thief who constantly stealths and waits for the enemy player to leave the point so they can decap. The player doesn’t care if they win or lose the match.

This might result in something like this…
100 Games Played
10 Games Won
90 Games Lost

Why would someone do this?

The MMR system takes into account your character’s “skill” based on how you contribute in fights, cap/decap points, and overall “how well you played the game” (or at least that’s what they are trying to measure). The system then looks at your characters rank and how many games you’ve won. It then makes decisions based on these factors and a few others.

What we’ve effectively done in this scenario is created a player with a horrible metric. By default the system should now start making the effort to match this player with better players on their teams to help that player become more likely to win that round.

This is bad.

Reason:
Being as we are now at 10% in our win/lose ratio we we’ll have this state of the game trying to “help us” for quite awhile. Realize that this is not as simple as it looks. If we now assume that the player starts trying they will have this “help” for awhile….

Old Score:
100 Games Played
10 Games Won
90 Games Lost
Ratio = 1:9 (10%) -> Goal = 1:1 (50%)

+50 Games:
150 Games Played
50 Games Won
100 Games Lost
Ratio = 1:2 (33%) -> Goal = 1:1 (50%)

Note that we’re still not at that 50/50 ratio, but we’re closing in.

However, the more interesting point to make is what just happened to this players pips/rank. In the worst case we’d try to evenly space out the losses to make this player have the shortest possible win streaks. This would mean we have 4 wins -> 1 loss. Expanding that throughout the 50 points we’d have a minimum gain of 25 pips within those 50 games. The best case would be 10 losses right at the start and then 40 wins to follow – giving a total gain of 79 pips (1st win = 2 pips from losing streak, 2nd win = 1 pip, 3rd win and up = 2 pips from win streak -> you get 39(2) + 1(1) = 79). The average would sit between the two at 52 pips gained.

This is not including for possible times where you get a loss when you actually can’t lose a pip. This is also only taking into account ~100-150 games with rather significant arbitrary values to make number crunching a bit simpler (albeit slightly less realistic – yet not entirely impossible either). However, imagine expanding this out to 200 games, 500, 1000 – the time in which you’ll be playing “catch-up” gets extended significantly if you keep your wins to a minimum.

Here lies a major flaw of the MMR system.

(edited by Dragon Ruler X.8512)

Flaw in MMR Model? (Let's do Math!)

in PvP

Posted by: Exedore.6320

Exedore.6320

TLDR from OP: If you intentionally lose a bunch of matches in a row, it lowers your MMR. You can then in theory have an easier time getting win streaks to boost your total pip count . This yields higher pip counts than playing normally because win streaks give bonus pips.


Fortunately, this isn’t how matchmaking works. There is no forced 50% win rate metric.; you should simply settle to that ratio once you reach your proper placement (in a proper system at least). The matchmaker finds teammates within your pip range who are also near your MMR. Opponents need only to fall within your pip range. By tanking your MMR, you get worse allies, so unless you’re amazing at carrying teams, you won’t gain a huge advantage by tanking MMR.

The exception is if you have a team of 5 who tank their MMR. Because you’re all skilled players in reality, you should be able to get a win streak with less trouble. However, if you’re all that good from the start, there isn’t a huge need to tank MMR in the first place.

However, this doens’t mean the league system isn’t flawed in other ways.

Kirrena Rosenkreutz

Flaw in MMR Model? (Let's do Math!)

in PvP

Posted by: Fivedawgs.4267

Fivedawgs.4267

TLDR from OP: If you intentionally lose a bunch of matches in a row, it lowers your MMR. You can then in theory have an easier time getting win streaks to boost your total pip count . This yields higher pip counts than playing normally because win streaks give bonus pips.


Fortunately, this isn’t how matchmaking works. There is no forced 50% win rate metric.; you should simply settle to that ratio once you reach your proper placement (in a proper system at least). The matchmaker finds teammates within your pip range who are also near your MMR. Opponents need only to fall within your pip range. By tanking your MMR, you get worse allies, so unless you’re amazing at carrying teams, you won’t gain a huge advantage by tanking MMR.

The exception is if you have a team of 5 who tank their MMR. Because you’re all skilled players in reality, you should be able to get a win streak with less trouble. However, if you’re all that good from the start, there isn’t a huge need to tank MMR in the first place.

However, this doens’t mean the league system isn’t flawed in other ways.

Pretty much this

Flaw in MMR Model? (Let's do Math!)

in PvP

Posted by: Crowley.8761

Crowley.8761

The logic in the OP is… backwards? I don’t know how else to describe it. If you tank your MMR you don’t get better players on your team – you get put in matches with worse players on both sides.

Flaw in MMR Model? (Let's do Math!)

in PvP

Posted by: Threat.7450

Threat.7450

If you spent as much time playing the game as you did writing that post, you would have a better win rate.

Flaw in MMR Model? (Let's do Math!)

in PvP

Posted by: Crinn.7864

Crinn.7864

The problem here is that 50/50 forced isn’t how ranked works. 50/50 is just the equilibrium point where your skill evens out.

Now if we are talking about unranked you might have a few valid points.

Sanity is for the weak minded.
YouTube

Flaw in MMR Model? (Let's do Math!)

in PvP

Posted by: Booms.3952

Booms.3952

The logic in the OP is… backwards? I don’t know how else to describe it. If you tank your MMR you don’t get better players on your team – you get put in matches with worse players on both sides.

#1 gerdian na
0 counterplay