76 replies to this topic

[Exilyth]

"For a set of players, find two disjoint sets of players with equal skill" is a variation of the generalised assignment problem: https://en.wikipedia...ignment_problem - except that each team is an agent, each player is a job/task and the skill of a player is the cost of the task (and the constraint is for both teams to have equal cost/skill).

Unfortunably, the fitness function/metric used to rate player skill in mwo is flawed (xp bar), so solutions/match ups found will be flawed too, no matter what algorithm is used.

[Deathlike]

El Bandito, on 16 April 2017 - 07:28 AM, said:

Even during the Elo system in the past we did not have the necessary amount of population. The population is low currently, but MM system is not the chief culprit of it.

The bigger complaints from people about Elo was mostly regarding "they weren't in the right place".

The most common artifact is the version where a mediocre player groups up with a bunch of good players, and his Elo gets carried up unnecessarily.

Here's a reality - if you choose to group up with higher level players, you are expected to eventually carry your weight as a consequence. Complaining about grouping with others only to succeed more (aka "fun" for some) has consequences.

Some players were "not in the right place" either... stuck at lower tiers or something.

If you believe you're "that good", you should rise in the ranks yes? You have the option to "group up" and find people that will dig you out of that hole (perceived or not) and maybe you'll learn a few things or two to dig yourself out?

I dunno... inevitably population or specifically lack thereof will cause MM mismatches. You will have to accept an SJR/EmP roflstomp once in a while, but if you prefer to complain about the wait, this is what you'll get instead. If you can accept the wait, I hope you can wait patiently above 5+ mins - there's plenty of other things to distract you on your extended circle of sleep hypnosis.

Edited by Deathlike, 16 April 2017 - 01:44 PM.

[Mystere]

CadoAzazel, on 16 April 2017 - 06:40 AM, said:

The population is so bad BECAUSE we dont have a good matchmaking system have only a skeleton of a game.

FTFY.

[PhoenixFire55]

Dee Eight, on 15 April 2017 - 10:19 PM, said:

The problem has nothing to do with the tiers. It has to do with the available total player pool. Even if they reset the tiers and re-formulated the math strictly along W/L & K/D ratios for sorting players... there simply are NOT enough players to form matches without mixing them.

We've heard this BS over and over and each time people telling us this BS ignore one simple fact ... the amount of players you need to make a balanced match is exactly ... 24.

[Marius Romanis]

I didnt actually say it was the ONLY reason its so bad, I know of those other reasons thanks for sharing, it would be " LESS BAD " if we had a good MMS.

[Shifty McSwift]

Tarl Cabot, on 16 April 2017 - 09:07 AM, said:

• Niche IP
  
• High learning curve
  
• Lack of patience and the need for instant satisfaction. Many players are used to respawn games, but when their mech is destroyed in MWO they have to start a new game.
  
• Players brought in by friends are fed to the wolves (group game) instead of assisting them through the Academy courses, then running through some private matches while occasionally running thru solo queue.
  
• The above added to playing on a system where FPS is not at a satisfactory level, partially due to Cryengine and lack of optimization (big difference between testing grounds/academy and live games).
  
• Both PGI and players' piss poor attitudes, especially when lack or extreme delay of features or updates to fix issues.
  
• Loss of critical mass when losing segments of the population such unit leadership due to many of the about issues.
  
• Other and newer games that provide that more instant gratification in a GOD mode state that is aimed more at the single player and/or is missing many of the above issues.
  
• Non-existent advertisement

Add to that a seriously daunting grind scheme for any new player that doesn't feel like splashing cash immediately into a FTP game, which is a whole lot of people.

Edited by Shifty McSwift, 17 April 2017 - 12:17 AM.

[Sjorpha]

PhoenixFire55, on 17 April 2017 - 12:02 AM, said:

We've heard this BS over and over and each time people telling us this BS ignore one simple fact ... the amount of players you need to make a balanced match is exactly ... 24.

No it's not, and there are lots of articles and research about this.

Read this for example: http://joostdevblog....s-enormous.html

[PhoenixFire55]

Sjorpha, on 17 April 2017 - 12:51 AM, said:

Read this for example: http://joostdevblog....s-enormous.html

So pray tell me why do I need to read lame excuses from lame game devs about how hard it is for them to do their job?

Sjorpha, on 17 April 2017 - 12:51 AM, said:

No it's not, and there are lots of articles and research about this.

Yes it is, and I don't need "articles" about this when I can prove it myself.

Quick example. Consider "skill" level being a value between 0 and 1000 (actual values irrelevant). Now take 24 people and assign them random "skill" values. Since you have an absolute freedom on how you divide these 24 people into two teams you can easily create matches where the difference in such "skill" level between teams is minimal (that is the result of LLN). The most imbalance you can possible achieve is an extremely unprobable scenario when 23 people would have 1000 "skill" and 24th person would have 0 "skill". In such scenario regardless of how you shuffle the teams the imbalance would be 12k vs 11k, i.e. ~8.3% of the total value. That is an absolute maximum imbalance that is even theoretically possible provided you are aiming to balance teams to begin with. The actual imbalance would be significantly less even provided your MM in absolutely random.

I've generated 24 random values in between 0 and 1000 ...
924 587 480 756 748 858 827 713 903 915 625 144 10 199 219 116 731 929 405 950 573 434 778 197

Then manually without any algorithm of any sort divided them into two teams ...
929 756 903 950 10 858 199 144 625 827 219 587 (total of 7007)
vs
924 748 915 573 434 778 197 116 731 713 480 405 (total of 7014)

The "skill" difference is 7 points, that is ~0.1%. I.e. I've taken 2 mins and manually created you a perfectly balanced match out of a completely random set of players.

That proves exactly what I've said ... "The amount of players you need to make a balanced match is exactly 24".

I can if necessery go even further and make a simple enough program that would create optimally balanced two teams of 12 out of a random group of 24, guaranteeing a minimal difference in total "skill" out of all possible combinations for any given random draw. The results will show that the imbalance wouldn't exceed ~1% for the dominant majority (~99.5%) of random draws.

Any MM is working with a certain criteria that is used to evaluate performance and thus assign a "skill" value to each player. If such criteria is faulty (which is clearly the case for current MWO tiers system) then it'll obviously lead to MM fails. However, that doesn't contradict with what I'm trying to prove here in any way, since if you have a proper criteria, then the amount of people you need for a well balanced match remains the same ... 24.

[Shifty McSwift]

But what about the idea that the system used to measure "skill" in the game is far from perfect in practice, and the idea that people don't always live up to their averaged out "skill" level no matter how accurate it is, that it is just what is likely to happen? And the idea that the conditions change consistently throughout the measuring of these averages.

We agreed in another topic on the idea that on an infinite/ongoing scale that this is likely the best way of achieving the most accurate representations of player ability, but then people start putting numbers on it, like 500 or 1000 matches being the threshold of reasonable doubt that a player should have taken full advantage of all rewards and purchased all of the meta mechs for the exact roles etc, which is just not the case across the board, and that matters when you are throwing all of those same players into the same lobbies.

[PhoenixFire55]

Shifty McSwift, on 17 April 2017 - 03:06 AM, said:

But what about the idea that the system used to measure "skill" in the game is far from perfect in practice, and the idea that people don't always live up to their averaged out "skill" level no matter how accurate it is, that it is just what is likely to happen? And the idea that the conditions change consistently throughout the measuring of these averages.

There were, are and will be plenty of ideas, I've taken part in pretty much every serious MM-related discussion on the forum over the last 3-4 years, saw some good stuff, saw some idiocy, its all there. However, as I've said, I've only proven my simple statement that player population is irrelevant given the proper MM algorithm, i.e. the excuse that PGI and any other game developer uses when their MM suxx is just that ... lame excuse for not being able to code a MM.

[Sjorpha]

PhoenixFire55, on 17 April 2017 - 02:44 AM, said:

So pray tell me why do I need to read lame excuses from lame game devs about how hard it is for them to do their job?

Yes it is, and I don't need "articles" about this when I can prove it myself.

It's cool that you don't need to check any of the previous research on a subject and can just instantly solve a problem yourself that no one before you has managed to solve, I'm very impressed.

Too bad for the rest of the gaming world you aren't working with creating matchmakers, I guess.

Edited by Sjorpha, 17 April 2017 - 04:09 AM.

[PhoenixFire55]

Sjorpha, on 17 April 2017 - 04:07 AM, said:

It's cool that you don't need to check any of the previous research on a subject and can just instantly solve a problem yourself that no one before you has managed to solve, I'm very impressed.

So much sarcasm, but sadly you've entirely missed the point of my post.

Creating a matchmaker isn't simple, nor did I "solve" it anywhere. However, saying that low population is the reason your MM isn't working is a lame excuse, which is what I've proven.

[Sjorpha]

PhoenixFire55, on 17 April 2017 - 04:15 AM, said:

Creating a matchmaker isn't simple, nor did I "solve" it anywhere. However, saying that low population is the reason your MM isn't working is a lame excuse, which is what I've proven.

No you haven't, your thought experiments prove nothing. The failure to as of yet accomplish this in the real world suggests that the problem is much harder to solve than you want to believe.

You haven't proven me wrong until you can showcase a skill based matchmaker that functions well with 12v12 teams and very low population in a real online game.

Game developers all over the gaming industry that work with this testify that they can only make skill based matchmaking work well with high populations, I have every reason to beleive they understand the issue better than you do.

[PhoenixFire55]

Sjorpha, on 17 April 2017 - 06:07 AM, said:

No you haven't, your thought experiments prove nothing. The failure to as of yet accomplish this in the real world suggests that the problem is much harder to solve than you want to believe.

You haven't proven me wrong until you can showcase a skill based matchmaker that functions well with 12v12 teams and very low population in a real online game.

Whatever. I'm not here to fix your reading comprehension fails or your math fails or other whatnot.

[Mole]

Oh, are we still blaming matchmaker for having no way to discern a player's preferred temperment and playstyle? Am I the only one who sees something wrong with the logic that it's a computer algorithm's fault for players behaving like cowards?

[KodiakGW]

Game population is low, and I have proof from just last night. Was on comms with my teammates for a bit doing other things. They are all fairly new to the game, so none are T1 yet. They sounded like they were doing well, mostly wins and a few losses. One then had to sign off for the night. So, against my better judgement from having lousy solo PUG matches earlier in the day, I sign in to play a few rounds to make it an even four man again.

Three games in a row had the same EmP four man. First one against them, second one with them, third game was against them with an -SA- eight man. The team that had them on their side won.

So, what are the odds of running into the same 4 man three games in a row? Pretty low if we have a healthy player base.

I signed off after the third one. Funny thing is the rest of my team never saw them the rest of the matches they played. They also said they never ran into them in the games prior to when I signed on. It was like the matchmaker put us in the top tier player queue when I joined. That is the main reason I signed off. I didn't want to keep my buddies from getting games against EmP and -SA-. Guess that means in order for them not to be matched against the best players in the game, I can't play with them. And since I detest playing in solo PUG...

Time to refund the Uziel and Mad Cat MKII preorders. Really wanted those decals and Uziels. PGI will never admit there is a problem, and therefore will never fix it.

[MrJeffers]

PhoenixFire55, on 17 April 2017 - 12:02 AM, said:

We've heard this BS over and over and each time people telling us this BS ignore one simple fact ... the amount of players you need to make a balanced match is exactly ... 24.

LMAO! That's the best thing I've read all day.

That is so maddeningly simple, why didn't anyone think of that before?

Oh, right. Because to pull the *right* 24 people out of a pool of a few hundred that are waiting for a match and have both the right tier and mech chassis is a remote possibility.

And of course that assumes that the right skill rating is being used to select the people in the first place, and we all know PSR isn't it.

[PhoenixFire55]

MrJeffers, on 17 April 2017 - 08:51 AM, said:

Oh, right. Because to pull the *right* 24 people out of a pool of a few hundred that are waiting for a match and have both the right tier and mech chassis is a remote possibility.

I've provided an example by perfectly balancing 24 people as is without any other players available.

MrJeffers, on 17 April 2017 - 08:51 AM, said:

And of course that assumes that the right skill rating is being used to select the people in the first place, and we all know PSR isn't it.

That is true, but has nothing to do with the fact that I've pointed out.

[MrJeffers]

PhoenixFire55, on 17 April 2017 - 09:10 AM, said:

I've provided an example by perfectly balancing 24 people as is without any other players available.



That is true, but has nothing to do with the fact that I've pointed out.

Except your example is only one criteria:skill - the match maker has *three* (Time in Queue, Ranking, Mech Chassis), which exponentially decreases the odds of making that match.

Your example only works on a one dimensional array and requires that those rankings are both correct and a relative ranking (e.g. any 500 rated player is roughly equal to any other 500 rated player) and removing the mech chassis entirely creates a new imbalance, one frequently seen in group queue where chassis type isn't a factor and instead group size is used.

Edited by MrJeffers, 17 April 2017 - 09:57 AM.

[PhoenixFire55]

MrJeffers, on 17 April 2017 - 09:52 AM, said:

Except your example is only one criteria:skill - the match maker has *three* (Time in Queue, Ranking, Mech Chassis), which exponentially decreases the odds of making that match.

Except two of the three you mentioned aren't part of any criteria and aren't part of the MM at all. And TBH I'm not sure time spent in queue is either.

MrJeffers, on 17 April 2017 - 09:52 AM, said:

Your example only works on a one dimensional array, and removing the mech chassis entirely creates a new imbalance, one frequently seen in group queue where chassis type isn't a factor and instead group size is used.

Once again, you are not reading and thus missing the entire point. I'm not building a matchmaker, I am simply demonstrating how blaming any sort of MM fails on low population is entirely false and nothing but a lame excuse.