61 replies to this topic

[MischiefSC]

Funkadelic Mayhem, on 16 August 2014 - 06:09 PM, said:

Le-sigh, an offspring of the extremest fox news/MSNBC mentality. We are ether with you or against you I take it? Its not black and white as you want to make it. No matter how hard you try with your straw-man/generalizing arguments. This is not other games, this is Mechwarrior, with a small playerbase.

Random, like nature is the only true fair way. Not, win some lolerstomps, loose some lolerstomps X time in a row against the same people the Elo keeps putting against you instead of taking the winners and mixing them up with the loosers and visa-versa.

ELO:
"You just lolerstomped 5 times in a row, now its time for you to loose by lolerstomp 5 times in a row. " LOL thats PC BS! Random, random is the only fair way. But I take it your ether a "now generation kinda a guy" or a "im old I give me my due" kinda guy.

I get that you don't understand how the MM works or how Elo works - even though I've explained how what you're trying to describe here isn't how it works, has never worked and would never work.

I also like your ad hominem attempt to somehow equate something you absolutely do not and could not know or understand - what news I watch or my opinions on, well, anything, as a strawman argument to attempt to attack me, as a poster, instead of my argument.

That would be because you don't have an argument. You're wrong, demonstratively wrong and the way to show that you're not wrong, giving any evidence aside from your baseless opinion like another competitive game with random matchmaking, you're avoiding.

To clarify yet again on Elo I'm going to put this in really big words for you because the last half a dozen times in this thread alone I've posted it doesn't seem to have sunk in-

Elo does not stack matches to give you a 50% win/loss rate. If you win several matches it will not make you lose several matches to make up for it. It builds teams with an average Elo, searching first to get everyone within as close a range to each other as possible. The teams are within 40-90 points of each other total - given that each team will have a score of 1000-1800 or so that's pretty close.

The team with the higher Elo average is expected to win. If they win neither team gains or loses Elo.

The team with the lower Elo average is expected to lose. If they lose neither team gains or loses Elo.

If the higher Elo team loses they will lose Elo.

If the lower Elo team wins they will gain Elo.

That's it. This is the exact post where it's described in detail exactly how it works.

[MischiefSC]

wwiiogre, on 16 August 2014 - 06:13 PM, said:

Realize that ELO matchmaker is not and never has been designed to make even matches. ELO as implemented by PGI is a win/loss ratio manager. It purposefully stacks matches to attempt to keep everyone's win/loss ration near 50%.

When a stock Atlas with a new pilot equals a tricked out Dire Wolf with modules, upgrades, etc then ELO is fatally flawed. Unless of course it never was meant to balance matches. Merely manage win/loss records.

In other words ELO does not take clan vs IS, new pilot vs experienced pilot, upgrades, modules, ecm or anything that really matters into account when making a match. It takes a few pilots with a plus win/loss average that have high elo, sticks them with a bunch of new pilots in stock mechs and on a team without elo so they have a good chance to lose. Thus lowering their elo and win/loss records.

Before you say elo ratings is a way to show skill I must call BS. Since the program intentionally stacks matches. Using stacked matches in anyway as a basis for skill is so pathetic its laughable. Yet, PGI says working as intended. Because ELO was a way to hide the fact matchmaker doesn't work at all. Plus a way to scam the system by intentionally giving games to everyone that they cannot lose, thus moving their win/loss more to average.

ELO is a system designed for ranking solo competitions where everything is level in a game between two people. Where everything is exactly the same. Take chess, where elo works. Both sides have exactly the same equipment. Have no teammates, cannot upgrade their side. And based on matches played against other ranked players elo then assigns them a number showing what skill they possess.

Where in MWO, how could that possibly work. Nothing is ever equal unless you match mech for mech, with upgrade for upgrade in every match for every player on both teams. Even then, how do you rate skill? You cannot since MWO is a team game and teamwork always trumps individual skill. That LRM boat cannot do well unless someone is spotting for him and also someone protects him from brawlers and light mech wolf packs.

So Matchmaker is a joke in MWO. Will always be a joke as long as it tries to manage win/loss by stacking matches. But that is where we find ourselves.

Good luck and try to find good close games and cherish them when you do. They are so rare in solo pug drops.

Chris

Lol. We've already had this debate and you made the same flawed arguments. Walked you through where you were wrong there and invited you to show me, mathematically, why I can't solve for 8% probability with sufficient sample size. Still waiting on that one.

That's all Elo in a team game does. It uses the law of large numbers to solve for the 8% value of your contribution. In 1v1 in chess your score settles in a handful of matches; in a game like this you're going to need about 40 matches, maybe more, to settle in each weight class. Why? Because you're only 8% of the total performance.

But we've gone over all of this and your argument comes down to you not understanding how it works or why so it can't work.

You're wrong, still. WoW and Microsoft use the TrueSkill system; which is Elo at its core but using a complex series of qualifiers to account for how you would perform in, say, CoD based off your skill at Checkers. They do this because they have millions of people playing at any given time and drilling down like that is very useful in better slicing your population. They can rank players 1-10 million in order based on skill.

We don't need that level of detail for MW:O because we don't have the population density to make it relevant.

Yet, again, we've already done this. Your argument comes down to trying to argue that statisticians, analysts, Blizzard, Microsoft and their associated nerd squads, indeed even statistical analysis itself is all wrong when it comes to 12v12 in MW:O because "it's too complex".

That's not just wrong but a raised middle finger stuffed in the face of math, science and reason.

Which is why I take the time to make perfectly crystal clear that you are, absolutely, wrong.

I posted it prior but here again just for you is the post that includes the Elo formula. Please identify for me mathematically and based on what mathematical principle that doesn't work in 12v12 and why the law of large numbers doesn't apply nor are you unable to solve for the players consistent 8% representation over a sufficient sample size.

Which you can't, so again. This is more just me pointing out that you're both absolutely wrong.

Edited by MischiefSC, 16 August 2014 - 06:39 PM.

[Funkadelic Mayhem]

MischiefSC, on 16 August 2014 - 06:30 PM, said:

Elo does not stack matches to give you a 50% win/loss rate.

yes....... it..... does......
You can run MERIC and win 5 matches, then log in solo and play 5 matches and loose, and then rage quit. The next time you long in you may loose another 5 matches then rage quit. Then you log in a week later and you will win 10 matches in a row, exc exc and onward and forward.

It totally stacks, and keeps your w/l @ 50/50

Edited by Funkadelic Mayhem, 16 August 2014 - 06:46 PM.

[CMDR Sunset Shimmer]

nehebkau, on 16 August 2014 - 09:28 AM, said:

Matchmaker has a fatal flaw that will need to be addressed, specially if we go full clan vs. IS and 10 vs 12. The simple flaw is that MM doesn't take into account what mech a player is using vs their ELO.

Let me explain:
Lets say we have two players, each with identical very high ELOs and identical skills who are dropping in the solo queue. Player 1 drops in a kitfox and player 2 drops in a Timberwolf. Red team gets player1 and blue team gets player2. Which team has gotten the advantage? Lets continue, to offset those high ELO players MM picks to other players with very low ELO, Player 3 is driving a Timberwolf and Player 4 is driving a kitfox. Player 3 gets put on the red team and player 4 gets put on the blue team. Which team has gotten the advantage?


Before we start the whole kitfox could have ECM, timber wolf could just have small lasers, kitfox could be a NARCing machine, etc.etc. that BS clouds the issue and if you partake in it, you are doing the game a disservice. Likewise starting with the battle value crap will simply cause the devs to shut-down because its too complex for them deal with at this moment. (Please remember *KISS*! Battle value may be a good idea but it would take PGI a year to get around to it)

Now, on the surface which team has the advantage from the start? Identical skill players, identical ELO. One is dropping in a 30 ton mech, the other in a 75 ton mech. I say that the team who got the High ELO player in the timberwolf has the advantage. Now the very bad players added, one in a 30 ton (blue team) and one in a 75 ton (red team).

What I think this shows is that effectively, the blue team will have a huge disadvantage because MM didn't take into account weight at drop. This is probably not a big problem in the group queus but I think MM should normalize ELO based on weight at drop (I searched the forums but I can find no mention by PGI that this is taken into consideration in this way), specially in the solo drop queue.

What you want, OP, is a battle value system that caculates the "Value" of a mech based on it's equipment. Adds in the ELO of the player, then balances the teams accordingly.

[MischiefSC]

Funkadelic Mayhem, on 16 August 2014 - 06:45 PM, said:

yes....... it..... does......
You can run MERIC and win 5 matches, then log in solo and play 5 matches and loose, and then rage quit. The next time you long in you may loose another 5 matches then rage quit. Then you log in a week later and you will win 10 matches in a row, exc exc and onward and forward.

It totally stacks, and keeps your w/l @ 50/50

Read this link.

That's what it does. That's it. That's all it does.

You are wrong. You're anthropomorphizing the matchmaker, attempting to ascribe to it an intent or quasi-sentience it doesn't have. It doesn't 'stack' anything. This is pareidolia; the attempt to ascribe a purpose or pattern where there is none.

So, again. Without question, the MM does not stack matches. It doesn't make you lose 5 because you won 5. If it had that sort of foresight and there were players always available for this sort of micro-management there wouldn't be any matchmaker issues or trouble finding perfect matches.

It would also serve absolutely no purpose - it also means there would be no 'curve' to Elo scores - we'd all be average. That's not the case, some people are way higher than others. They differentiate by winning more than they lose.

Just in case you missed it though....

You're wrong. It doesn't stack matches. Just because you've refused to read the last 3 times I've posted it here is a copy-pasta quote of the posts from the devs on exactly what Elo does and how it works.

Paul Inouye said:

Posted 19 December 2012 - 12:19 AMMatch Maker – Phase 3 – Update

Summary:
Phase 3 incorporates a variation of Elo and is undergoing testing against the telemetry data the servers have been providing us. Below is how Elo calculations are being processed and giving us an indication of how Match Making Phase 3 will match players of relatively even skill against one another.


Figure 1In Figure 1, you will notice the highest score a player can have is 2800. The lowest score a player can have is 0. When a player first starts playing MWO, they are assigned an Elo score of 1,300.

The maximum amount a player’s score can change in a single match is ±50.

One of the key calculations that needs to be done is to determine the probability of a lower ranked player beating a higher ranked player.

An example of this is, if two players of the same ranking have a 50% chance of winning each. The bigger the gap between the two player’s scores, the less probability of the lower score player beating the higher scored player.


Figure 2A player’s new rating, after a match is complete, is calculated using the probability value determined in Figure 1’s formula. In Figure 2, we take a look at a sample of how a player’s score changes between matches.

In the example, one player has a rating of 1,350. Another player has a rating of 1,410.

The first step is to calculate the probability of the 1,350 rated player beating the 1,410 rated player. The rating difference between the two players is found by subtracting the lower player score from the higher player score. In this case it’s 1410 – 1350. There is a difference of 60 points between the two players.

Filling in the formula from Figure 1, we find that the 1,350 rated player has a 0.41 probability of winning (or 41% chance). This is where Elo starts to determine the outcome of a match and the result of player scores after the match is complete.


Figure 3
We now use the probability value of 0.41 to determine the change in player ratings.


Case 1: 1350 Player WINS over 1410 Player

Since the lower rated player won despite the odds against him/her, they are rewarded a much higher change in score than the higher player is. The higher player score will actually be reduced.

One variable that is set by winning or losing is the WinFlag (as seen in Figure 3). The WinFlag is a binary (true/false) value of either 1 (Player has won) or 0 (Player has lost). In this case, the WinFlag value is 1.

Let’s begin the calculation:

Old Rating = 1350
Maximum Change Allowed = +50 for a win, -50 for a loss (as seen in Figure 1)
WinFlag = 1
Probability of Winning = 0.41

1350 Players new ranking = 1350 + 50 x (1 – 0.41)
= 1380

1410 Players new ranking = 1410 – 50 x (1 – 0.41)
= 1381

A player’s rating will only go down if they are beaten by a player who has a lower rating than theirs. In this case, if the 1350 player lost, their score would not change since the Match Maker was correct in its prediction.

So how does this affect Match Making – Phase 3?

This formula and scoring system is run against match data that the current dedicated servers are giving us. Basically it’s being tested on real world numbers generated by you, the community.

Once we get a full understanding of how accurately the Match Maker is working, we are going to add some additional parameters to the mix. These include a more defined player skill rating and a Mech weight class balancing system. More info on these when the first pass of Elo testing is done.

Flash Frame, on 16 August 2014 - 06:52 PM, said:

What you want, OP, is a battle value system that caculates the "Value" of a mech based on it's equipment. Adds in the ELO of the player, then balances the teams accordingly.

More to the point you've got an Elo score modified by how you, personally, handle a mech and equipment. L0RD FLASHHEART in a Firestarter/Ember will crush Assaults all day long every match. He's probably not nearly so good in a Dire Whale.

So the most accurate system would drill down even further; for example I do pretty good with AC20s. If I put an AC20 no a Banshee my Elo is higher. Also I do good in Banshees, so if I'm in a Banshee with an AC20 my relative Elo is even higher. Make sense? You adjust Elo based on your historical performance with a given chassis and loadout.

Again though, we don't have the population to make this very relevant.

[Almeras]

I've played since close beta so I've experienced every incarceration of MM'ing

First we had class balanced matches (but you could run 8mans)
Then we had 4mans but no class balancing
Then we had 4mans, elo and no class balancing
Now we have elo class balanced queue (but if you're in a medium or a light chances are it'll be 1/2/5/4 not 3/3/3/3 fun for you right?)

The MM has too many blind spots for it to make fun games "ELO to high? Don't worry we'll throw a trail mech user under the bus to balance it out. Fun for him fun for you right?"

IMO PGI will never be brave enough to look at ELO objectively and throw it out if it if proven to be carp and go with a more simple set up based on balanced teams and maybe a simple ranking bucket.

[IraqiWalker]

MischiefSC, on 16 August 2014 - 06:30 PM, said:

All the Elo descriptions.

Ok, at this point in time. I think you are just wasting time.with some of these people.

[MischiefSC]

IraqiWalker, on 17 August 2014 - 06:37 PM, said:

Ok, at this point in time. I think you are just wasting time.with some of these people.

Absolutely understand and agree - however, if you just leave them to spread incorrect information you end up with other people confused. So when someone says something patently false you powder up the glove and lay the smack down. Otherwise you end up 3 months down the road with everyone believing something totally incorrect.

I'm not talking to them. I'm talking to everyone else who doesn't already know better but is willing and able to accept the facts.

[IraqiWalker]

MischiefSC, on 17 August 2014 - 08:11 PM, said:

Absolutely understand and agree - however, if you just leave them to spread incorrect information you end up with other people confused. So when someone says something patently false you powder up the glove and lay the smack down. Otherwise you end up 3 months down the road with everyone believing something totally incorrect.

I'm not talking to them. I'm talking to everyone else who doesn't already know better but is willing and able to accept the facts.

I see your point, and I agree.

[BLOOD WOLF]

MischiefSC, on 17 August 2014 - 08:11 PM, said:

Absolutely understand and agree - however, if you just leave them to spread incorrect information you end up with other people confused. So when someone says something patently false you powder up the glove and lay the smack down. Otherwise you end up 3 months down the road with everyone believing something totally incorrect.

I'm not talking to them. I'm talking to everyone else who doesn't already know better but is willing and able to accept the facts.

Dude, I believe your a mathematician

[MischiefSC]

BLOOD WOLF, on 17 August 2014 - 08:26 PM, said:

Dude, I believe your a mathematician

lol. Statistical Analysis Specialist is my job title. Or Business Support Analyst. I play with MATHZ and where they interact with PEOPLEZ.

[IraqiWalker]

MischiefSC, on 17 August 2014 - 08:35 PM, said:

lol. Statistical Analysis Specialist is my job title. Or Business Support Analyst. I play with MATHZ and where they interact with PEOPLEZ.

Basically, you're over-qualified for this discussion.

Edited by IraqiWalker, 17 August 2014 - 09:09 PM.

[MischiefSC]

IraqiWalker, on 17 August 2014 - 09:08 PM, said:

Basically, you're over-qualified for this discussion.

Nah, but it's cool stuff. Helped identify the exact statistical impact on customer satisfaction using 'please' and 'thank you' in conversations with customers has. Not just that but exact word choices and sentence structures and how they change peoples opinion of the conversation.

Also I identify metrics and generate reports on the performance of a few thousand people who work in person, on the phone or via livechat with their customers. Essentially breaking out the metrics that show who is doing a good job and who isn't.

That's why this stuff tweaks me. At work I can literally tell you exactly and precisely why good performers and good employees are good - what they do and how they do it and how they get from bad to good (I also help parse metrics for managers, team leaders, leads, etc. and how the metrics of their employees change. Which ones are good at coaching and developing new leaders) and the like.

The biggest difference?

Bad/mediocre people say they got bad calls/mean customers. They're on the wrong team. The wrong shift. They got scored on a 'bad call' but all their others were great. The stats are wrong/not fair. They didn't get the support they need.

Good reps know the metrics, know the job and have their head in the game to figure out how to get better pretty much every day. So they make fat bank and get promoted.

Anyway. All off topic, sorta, but sorta-not. Same thing with the MM. It works fine and does what it's supposed to. For some people that is exactly what's wrong with it.

[BLOOD WOLF]

http://forum.warthun...should-be-done/

they do have some beautiful maps.

Edited by BLOOD WOLF, 17 August 2014 - 11:18 PM.

[John Branon]

MischiefSC, on 16 August 2014 - 06:30 PM, said:

It builds teams with an average Elo, searching first to get everyone within as close a range to each other as possible. The teams are within 40-90 points of each other total - given that each team will have a score of 1000-1800 or so that's pretty close.

Can you point me to where it says the underlined part? Because while I wish this was true, i fear it's not, and I didn't find anything on that when I skimmed your link.

If the range of Elo scores in a match is very large (and it often appears to be) an average is not as meaningful, and a difference of 50 on average can be quite significant with 12 players in a team.

Edited by John Branon, 18 August 2014 - 04:47 AM.

[Khobai]

Quote

Elo, the ranking system, works just fine. Unless you have some mathematical insight you'd like to share with Microsoft and, well, every other purveyor or ranking systems in the world. Or statisticians, or statistical analysts.

Which you don't, and we've had this discussion before.

Elo is used to rank individual players in 1v1 games like chess. Elo was NEVER meant to be used in team vs team games.

Thats not to say it cant work in team vs team games, it can, provided an individual can still consistently alter the outcome of the game. Generally 5v5 is considered the absolute extent of an Elo based system (although you can make a convincing argument that Elo doesnt work in 5v5 either). More than 5 players per team starts to make it impossible for an individual to consistently influence the outcome of the game. At that point its no longer the sum of individual skill that matters but rather the collective skill of the team as a whole.

So the statement that Elo doesnt work in 12v12 is a correct one. Because an individual has very little influence on the outcome of the game in 12v12. And if an individual cant consistently and significantly affect the outcome of the game then Elo cannot function as intended.

Although one way that Elo could work is if it actually looked at your match score rather than only looking at whether you W/L. Because in all fairness someone who loses but still has a high match score should get an Elo increase. While someone who wins but has a very low match score should not get an Elo increase.

Edited by Khobai, 18 August 2014 - 01:30 PM.

[Roland]

Quote

Likewise starting with the battle value crap will simply cause the devs to shut-down because its too complex for them deal with at this moment. (Please remember *KISS*! Battle value may be a good idea but it would take PGI a year to get around to it)

The second link in my signature presents a system which would never take a year to implement, and which would provide a fully functional BV system.

If you don't embrace a system like that, then you are gonna get the clans nerfed into the ground.

[RussianWolf]

MischiefSC, on 16 August 2014 - 01:32 PM, said:

It's important because it's an easy way to recognize who does and does not actually know what they're talking about. Someone going on and on about how terrible ELO is can pretty easily be ignored, because they don't even understand what the word is.

It's not about being pedantic. It's about separating who knows what they're talking about from those who don't. If I was trying to tell you how dumb Futball is in the US and how the rules don't make sense and clearly the game was designed poorly you'd pretty quickly realize that I wasn't in a place to make any worthwhile judgements on it, right? If I can't tell football from Futball I'm in no place to make judgements on the rules.

Same thing.

And some of us use ELO just to push the buttons of those that care.

[RussianWolf]

MischiefSC, on 16 August 2014 - 02:25 PM, said:

Okay. So how is it misapplied? Here is the post where it's defined and explained.

Here's the thing. You're ~8% of your teams performance. You can get a lot out of that 8%. Maybe it wasn't that you team was bad but that the other team was just so good. Hard to say, that's why it's taken as averages over time. Absolutely it's frustrating to lose when you did well. Sometimes your team wins when you're the first guy to die.

You'll remember the first far more than you'll remember the second.

That's not a matchmaker issue it's a byproduct of how our brains work.

As I and others have explained in the past:

System designed specifically for 1v1 is used in a team setting (strike one) where the team changes every match (strike two) and is then compounded by a matchmaker setting up predictions based on it (strike three).

ELO should not be used for this scenario, it is being misapplied.

Math doesn't lie, but if you misapply a mathematical premise, you can get bad data as a result.

[Mawai]

It is impossible to build a matchmaker that prevents stomps since the level of team work is a key deciding element in how a match turns out and this factor can not be included in any calculation by the matchmaker.

Individual player skill is incorporated by the Elo value which is a statistical average of performance over all of the games you have played in. Theoretically, the individual player contribution to each match result will average out over time giving a fairly accurate reflection of the individual contribution to team play. This contribution could be in terms of individual damage output or accuracy ... or in terms of effective support of team mates ... or in terms of just adding some cohesion and leadership ... any of these can contribute to the success of your side and if used consistently will contribute to the Elo calculation.

How many premade 12 vs 12 matches turn into stomps? Is this due to better strategy/tactics (and luck)? Perhaps someone who plays regularly in competitive 12 vs 12 matches with players of comparable skill might be willing to comment?

Edited by Mawai, 18 August 2014 - 07:17 AM.