Jump to content


A small theory why 'matchmaking got worse' ... but not really

matchmaking uncanny valley fizzle theory

  • This topic is locked This topic is locked
5 replies to this topic

FizzlePopBerryTwist #1 Posted 11 January 2020 - 06:59 AM

    Staff Sergeant

  • Player
  • 32412 battles
  • 395
  • [FIM] FIM
  • Member since:
    05-18-2011

This isn't a rant, but a simple theory why some consider the 'MM' to have gotten worse over the years, arguing louder with each iteration. 

Not going to sugarcoat anyone's nostalgia here (which also plays into the issue) or claim the new MM is perfect since you can always improve a 'machine' in some way...

 

Now, for the sake of argument, i will keep my examples VERY simple! I know some will try to argue "if these and that happens and then magic falls from the sky, team X might COULD win/loose during a blue moon"... for those people: i don't care. You are not my audience, but the ones that might are interesting at a more serious 'different' angle towards the issue at hand.

 

First, i want to give some broad strokes about the 'matchmaker systems' i'm going to talk about, so we get a rough idea of how things have been back then and are today. Roughly. 

 

>The 'old MM'

It had fewer unique tanks to choose from and didn't use (flexible) templates with any kind of limits. Tier 5 vs X being possible. Same for 8 SPG's per team. Fail platoons (Tier 6+4 ending up in Tier 8 matches). Also, it was possible for 1 type of vehicle to dominate the lineup -> like mostly E100's vs mostly IS7's.

 

>The 'new MM'

It has flexible templates and more different tanks to choose from with a lot shorter MM spread! Fail Platoons being impossible. Limiting how many tanks of a certain type can be stacked against another. It is less likely if not impossible to have 1 type of vehicle dominating the lineup of any given team. (not taking 3x Platoons into account, this is a player-created variable). 

 

 

Now, how comes players believe the "mm get's worse and worse"? Simple: it does not. BUT... i think there is also a 'tipping point' , which can mislead one's perception!

Let me explain or rather 'demonstrate' what i mean by that. Also! It takes the BIG PICTURE to get the overall idea i try to bring across, so cherry-picking will ruin the thought experiment.

The fact i use 'current tanks' for all my examples does not change the results since they are nothing but 'space holders' for the old counterparts... 

 

Example 1 - using the 'old MM' putting 'uneven' vehicles against another was far more likely:

Let's assume the first 10 tanks of each team are 1:1 the same. All drivers play identical well = 'average' (for the sake of simplicity).

 

Yet...

>Top 5 players of Team_A play Object 252U (aka 'Defender'. https://tanks.gg/tank/obj-252u/model?vm=visual

>Top 5 players of Team_B play VK45.02_Ausf_A https://tanks.gg/tank/vk-4502-a/model?vm=visual

 

**((the Defender's represent the 'SUM of all advantages' one team can have over another, which usually would be spread across all 30 tanks...))

 

> Most would agree 'Defender' is stronger and therefor Team_A wins.

> giving the Defender team above-average players the result won't change.

> giving the VKA team above-average players may even the odds, but there is still a chance of loss. 

 

Therefore we have 2 of 3 cases ending in a victory for Team_A with near 100% certainty.

and 1 of 3 cases COULD end in a defeat for Team_A, given the skill gap is high enough...

 

Any uneven distribution of tanksmakes those the main balancing factor.

+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -

 

Example 2 - using the 'new MM' creating 'even' teams:

Let's assume all 15 tanks of each team are 1:1 the same. All drivers play identical well = 'average' (for the sake of simplicity).

 

Most would agree both teams have 50-50 odds of winning this round (including a draw, in which case neither side wins/loses) - so far 'perfect' Matchaking(!?)

Tank type's due not influence the outcome in this scenario.

 

Yet...

> giving Team_A above-average players means it's more likely for them to win, by tipping the scale in their favor by making better plays.

> giving Team_A below-average players means it's more likely for them to lose... stacking the odds against them by deciding poorly.

 

Increasing those skill gaps, while keeping the same 'perfect' conditions for the tanks, will only increase the likeliness for these cases to occur. 

 

Since the tanks are a non-issue in this scenario, the balancing factor are suddenly the players despite not being part of the matchmaking calculation!

+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -+ - + -

 

And this is where matchmaking 'turns around' and becomes 'worse' again. 

 

I would like to compare this to the phenomenon called UNCANNY VALLEY.

[for those not familiar https://de.wikipedia.org/wiki/Uncanny_Valley + https://media.gizmodo.co.uk/wp-content/uploads/2016/12/tarkin3.jpg]  

Uncanny Valley is the 'gap' between "not real" and "real". Using the 2nd link you will see Gouverneur Tarkin looks human, but... Not really. This is the part which will freak you out. 

It's clearly not something like a CGI character, but it isn't life-like either. It doesn't fit any given drawer and keeps jumping back and forth between them. 

 

-> only making it LESS realistic or A LOT MORE realistic will solve this issue! <- and this is where we find our selves right now.

 

What the hell are you talking about? You may ask...

> The old MM allowed for 5-minute fights to happen by stacking tank types poorly against another, while skill could hardly hold against it.

> The new MM was changed to give both teams a similar line up of tanks, so much so, that suddenly we find our selves on a point at which the teams are once again NOT BALANCED.

 

The matchmaker cares so much about tanks, to the point tanks play the smallest part within any given battle, while the real reasons for 5minute defeats ain't taken into account at all.

Balancing went full circle. Removing or better said 'reducing' one factor so much... other factors are in comparison HUGE all of a sudden. They do stand out far more so than ever before.

 

> WE ARE WITHIN THE UNCANNY VALLEY of MatchMaking <

 

During the times of the old MM, people complained about "how is that fair? We had to play vs a team of IS-7's!" 

Today we reached a point when people complain about "braindead teams" while not even mentioning the tanks they were up against!

 

IF we were to scale the matchmaking back X years -up until the point when the matchmaker was "half as accurate" as it is today... you can be sure people were arguing heavily if it was a problem of the tanks OR the player driving those.

 

However. 
I'M NOT SAYING THIS IS THE SOLUTION! Turning back the clock...

Not just is it near impossible to measure 'half as accurate MM', it wouldn't work with all the new factors which have come into play. New tanks, mechanics, maps, physics, equipment and so on... -all it would do, is to once again blur the line between the factors which made you win/lose, and i for once prefer it to be a players fault since it gives YOU AND ME more control instead of facing the alternative, which would be: 'having picked the wrong tank at the wrong time'...

 

It's easy to respect skill, hard to digest chance.

 

...also... i AM saying, it will get EXPOTENTIAL HARDER to ESCAPE 'THE UNCANNY MATCHMAKING'.

 

The only points we can improve on would be everything it does NOT YET INTO ACCOUNT... and for every issue we remove, another will stand out unless we take care of ALL variables at once.

Equipment. Stock vs Elite tanks. Player Skill. Map differences (statistics about the flow of the battle). Ammo loadouts. Game modes and so on and so forth.

>> this means for each time there is not a perfect counterpart in the queue... ten or more factors would have to compensate the issue... and there you see the issue at hand.

 

How do you value each and every variable if you have 30 tanks ^ 30 players ^ 3 equipment ^ x-factors of the map ^ 3 consumables ^ camo or not ^ and so and and so forth.


Long story short: it would take a neat quantum computer and an incredibly smart and flexible AI to even attempt to predict how all those components come together in order to make the final step OUT of the Valley... and into the holyland of perfect Matchmaking. 

 

 

May the GG be with you. Always.


Edited by FizzlePopBerryTwist, 11 January 2020 - 07:12 AM.


Spurtung #2 Posted 11 January 2020 - 07:45 AM

    Lieutenant General

  • Player
  • 78684 battles
  • 7,770
  • [WG_PT] WG_PT
  • Member since:
    07-05-2013

Just...this:

http://forum.worldoftanks.eu/index.php?/topic/678354-replay-analysis-what-does-this-tell-us-about-the-mm



NUKLEAR_SLUG #3 Posted 11 January 2020 - 08:14 AM

    Major General

  • Player
  • 36415 battles
  • 5,643
  • [FISHY] FISHY
  • Member since:
    06-13-2015
That's quite a long post to say the MM is about as good as you're going to get, deal with it. 

MeetriX #4 Posted 11 January 2020 - 08:22 AM

    Brigadier

  • Player
  • 25856 battles
  • 4,443
  • [_ACE] _ACE
  • Member since:
    08-12-2012

"Fail platoons being impossible."

Tell that to Valentine II players who team up with the Luchses. 



Frostilicus #5 Posted 11 January 2020 - 10:47 AM

    Colonel

  • Clan Diplomat
  • 23727 battles
  • 3,652
  • [-ZNO-] -ZNO-
  • Member since:
    07-12-2011

 

This - analysis of a huge number of games proves that these posts are pretty much all confirmation bias



unhappy__bunny #6 Posted 11 January 2020 - 10:55 AM

    Brigadier

  • Moderator
  • 20873 battles
  • 4,239
  • [-OC-] -OC-
  • Member since:
    08-01-2012
Ok, I acknowledge there was a lot of effort put into this but..... It is still discussing MM and you know what that means... It should be in the pinned MM thread. 





Also tagged with matchmaking, uncanny valley, fizzle, theory

1 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users