Jump to content


Finally a proper MM...enjoy..:)


  • Please log in to reply
206 replies to this topic

Vestrick64 #201 Posted 24 March 2015 - 11:18 AM

    Lieutenant General

  • Player
  • 11490 battles
  • 6,210
  • Member since:
    03-02-2013

View Postpussinbootsraffy, on 24 March 2015 - 11:13 AM, said:

 

But this world is not ideal, so you can't be 100% sure that will happen.

 

I'm working on railway projects on daily basis, and we always start from the ideal situation and add outside influences along the work.

 

This is because you start with a formula and modify it to a situation, and not attempt to create something that encompasses EVERYTHING from the scratch - that's inefficient and often leads to failure.

 

The simplier the basis is, the better.

 

With matchmaking in mind, you simply start with an idea that the game will divide players using a formula into tiers, and either make the tiers separated, or assure that the game mirrors players according to their tiers in the game.

 

This basis assumes 50% WR by default. For me, there is no need to further debate on how imbalance and formula errors influence that score - it goes straight to the garbage bin as the very CORE of the idea sucks.


Edited by Yuri_Yslin, 24 March 2015 - 11:20 AM.


SovietBias #202 Posted 24 March 2015 - 11:54 AM

    Second Lieutenant

  • Player
  • 37407 battles
  • 1,345
  • Member since:
    06-10-2013

View PostYuri_Yslin, on 24 March 2015 - 11:18 AM, said:

This basis assumes 50% WR by default. For me, there is no need to further debate on how imbalance and formula errors influence that score - it goes straight to the garbage bin as the very CORE of the idea sucks.

 

But you agree on the fact that there will be other meaningful stats? Like the whatever rating system they come with.

 

I'm asking because I am not sure you dismissing the SB MM just because of WR convergence or because you find that there are further issues with the enforcement of equal chances when creating a battle session out of 30 random players.

 



lord_chipmonk #203 Posted 24 March 2015 - 12:41 PM

    Field Marshal

  • Player
  • 35729 battles
  • 10,279
  • [-HOW-] -HOW-
  • Member since:
    12-23-2012

View Postpussinbootsraffy, on 24 March 2015 - 11:13 AM, said:

 

But this world is not ideal, so you can't be 100% sure that will happen.

 

Hence all the other discussion, and why I tried to suggest splitting the discussing into the ideal/principle and the implementation as they are two separate debates. In order to implement something it needs to be both desirable (from an idealised standpoint) and practical, i.e. for it to be possible to implement it in a meaningful manner. If people do not think there is the need for SBMM and all that it would, in principle, entail (for example) then there is little point discussing the practicalities of its implementation. If however you do think it is worth implementing, then you next need to assess its practicalities and whether in the `real world' such a thing should be implemented, unless this latter discussion allows you a manner of implementing the `ideal' in such a way as to maximise the positive aspects and minimise its problems.   

Vestrick64 #204 Posted 24 March 2015 - 12:50 PM

    Lieutenant General

  • Player
  • 11490 battles
  • 6,210
  • Member since:
    03-02-2013

View PostmaDNauseam, on 24 March 2015 - 11:54 AM, said:

 

But you agree on the fact that there will be other meaningful stats? Like the whatever rating system they come with.

 

I'm asking because I am not sure you dismissing the SB MM just because of WR convergence or because you find that there are further issues with the enforcement of equal chances when creating a battle session out of 30 random players.

 

 

Well, after Dota2 I don't like the core idea of skill based matchmaking.

 

But of course there are indeed further issues that would materialize should a MM like that be implemented.

For once, a game like WoT is actually drastically different for high and average based playerbase. If you got into games where the weakest player is dark green you'd know what I mean.

 

Without voice comm or just a general idea of who does what, the random team with "good" players is actually FAR FAR more susceptible to creating a horribe gaming environment. When everyone is at least good, the aspect of teamplay becomes the key, but there's this little issue - no game whatsover in the history of gaming has managed to make 15 total strangers with different (possibly random, not preselected) tools (weapons, vehicles, etc) to successfuly cooperate. Even with ingame skype there's language barrier, etc.

 

So what you get in the end is an incoherent mess of "good" players going cookie cutter tactics, well, sort of like WoT, but since you have 0 way to influence people from going to their favorite spots - fights would be more or less solved by a dice roll: check point A strenght ratio, calculate how fast it breaks through, if it's faster than point B resistance than you win, otherwise you lose.

 

In other words: EVERY game in high tier skill bracket would be solved after (at best) a minute. Uneven distribution without coordination = one chokepoint gets overwhelmed, then its game over.

 

Because you can stop reds from snowballing, but good players? not a chance.

 

So if people want to play a "chokepoint resistance simulator" and guess whether they (unanimously) choose the correct distribution among the map (impossible even for unicums without skype or teamplay) - go ahead. My fear - a quite legit one - is that the game would be even more decided "from the first minute" than randoms in WoT.

 

an example: malinovka. If team B brings more tanks to the hill than your - let's say four - tanks sent there - you get annihilated there, it's 0:4 after a minute and the game is over. But if several players have went elsewhere - you suddenly win hill and make a flanking manouver, ending the game in 2-3 mins.

 

The issue? you can not know or predict or influence in any way the players, they can go anywhere. So at that point it's just a lemmingtrain gameplay: you join the train heading into the key area, or lose. Again, sort of like Wot, eh?

 

But here's the catch: without noobs, you can't ever hope to recover from a snowball. In WoT, you can pick on weakest links and create an epic turnaround.

 

And that's a bad thing if you're just playing a role in a scenario that has already been written.

 

 

 

The main issue is that average players THINK they would do better/enjoy the game more without unicums and bads around them, but that's not the case. They are deluding themselves.


Edited by Yuri_Yslin, 24 March 2015 - 12:56 PM.


SovietBias #205 Posted 24 March 2015 - 01:36 PM

    Second Lieutenant

  • Player
  • 37407 battles
  • 1,345
  • Member since:
    06-10-2013

Generally, I share the same view. That's pretty much my experience from other games, no matter the genre. Details, optimal builds/setups, little diversity.

 

View PostYuri_Yslin, on 24 March 2015 - 12:50 PM, said:

But here's the catch: without noobs, you can't ever hope to recover from a snowball. In WoT, you can pick on weakest links and create an epic turnaround.

 

I just want to point out that you might be underestimating the effect player error. I agree that a better player is, by definition, less prone to mistakes. However, that is relative to the average level of the playing field. Maybe actions that would not be regarded as mistakes in a random could have different effects when you are playing against smart players, thus increasing the opportunities for an error to appear. 

 

It might be far fetched, yes. I can only base it on the odd turnaround seen at WGL matches where top players clash. Moreover, it is acceptable not to demand such level of effort from players for a game that requires extensive grinding. For little to no improvement in the average battle.

 

 

 

 



Trent #206 Posted 24 March 2015 - 09:01 PM

    Captain

  • Player
  • 10516 battles
  • 2,433
  • Member since:
    07-20-2011

View Postaugustocroppo, on 23 March 2015 - 12:20 PM, said:

Please, show me one single data set of any game with skill based MM where the majority of players have 50% win rate.

 

Hmm, theoretically speaking (for now):

 

Let's say a World of Tanks player who is visibly above average (when it comes to win rate) records the results of a few thousand of his random battles, along with their respective XVM win chance. Then he filters out the results to see what was his win rate in battles that had more or less "balanced" XVM win chance (let's say 45-54%).

 

Generally, we have two possibilities now. His win rate within this smaller, 45-54% sample:

- is similar to win rate across the full sample,

- is significantly closer to 50% than win rate across the full sample.

 

Assuming the results would show the latter is the case (i.e. "balanced" battles bring win rate visibly closer to 50%), would that data set be something that you had in mind?



SanyaJuutilainen #207 Posted 27 March 2015 - 10:15 AM

    Lieutenant

  • Player
  • 20820 battles
  • 1,876
  • [GUP] GUP
  • Member since:
    04-15-2011

View Postaugustocroppo, on 23 March 2015 - 09:29 PM, said:

OK, to rectify my mistake pointed out by jemster, I made another analysis of the data available from the DOTA game.

 

So, the good news are that Yuri hypothesis is partially true if he meant that the tendency from any distribution (as a result of a skill based MM) should be near the mean. However, if he meant that every and most of the players are going to have win rate with that variation, then his hypothesis is not proven true.

 

E.g.

 

http://www.dotabuff.com/players/played

 

http://pastebin.com/T0jZcsBR

 

This are 100 DOTA players filtered by most played matches. Here is the result of a basic statistical analysis:

 

| Matches Played | win rate
mean | 7674.8 | 51.41%
s.d. | 692.05 | 2.279%
min. | 6984 (row 100) | 48.89% (row 95)
median | 7469.5 | 50.94%
max. | 11038 (row 1) | 65.23% (row 88)
total | 767483 | 5141%
count | 100 | 100

 

Here is the scatter plot:

 

 

http://www.wolframalpha.com/share/clip?f=d41d8cd98f00b204e9800998ecf8427epunq4kpkif

 

This shows that after a large number of matches, players indeed achieve a win rate near 50%. However, let's observe the total number of matches among all players, 767483. Using Yuri arguments, let's assume this is the number of trials all that 100 players did to obtain either victory or defeat:

 

http://www.wolframalpha.com/input/?i=767483+trials+p%3D0.50

 

expected value | 383742
standard deviation | 438
variance | 191871

 

Let's transform that in % (per trial):

 

expected value | 383742 = 50%
standard deviation | 438 = 0.05707%

 

The standard deviation from the players sample size is greater (2.279%) than the expected sample size standard deviation (0.05707%). This indicate that not even skill based MM is able to enforce the relative lower standard deviation implied in Yuri's argument. Otherwise we should observe a similar % of standard deviation for both samples.

 

Anyway, this is just the good news, the first part. A second analysis is about to come after I have my dinner.

 

 

 

As a side note - isn't deviation in DotA / LoL different, because you play 1 hero out of 5 on your team, so your income to the battle is actually bigger? I mean, you can affect bigger percentage of the battle, skewing the outcome more.

Even in WoT, where I counted maximum possible WR to be around 70 - 75% (including platoons, abusing OP tanks and high tiers only, etc), there are people who have 80% or so WR - and they are only 1 out of 15.
Though possibly, if they ALWAYS play full platoon with people of comparable skill, they'll be in the same position as DotA/LoL (1 out in 5), but I guess we can't be sure about that.




1 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users