Jump to content

Unbalanced Matches [Elo Not Working]


33 replies to this topic

#21 Insects

    Member

  • PipPipPipPipPipPipPip
  • Bridesmaid
  • Bridesmaid
  • 995 posts
  • Locationstraya

Posted 13 May 2015 - 06:06 AM

If you drop in small groups then odds are you are going to find yourself in a lot of mixed groups vs large groups.
Its not the matchmaker deciding to do that, its just odds.
There are only so many slots for 4, 3, 2 so often they will end up bunched together as 2+2+3+5

Drop in an 8 man and you will mostly be matched as 8+4, less 8+2+2
9 man will always be 9+3, 10+2, no other combinations possible.

Also mishmash faction/unit teams doesnt mean they aren't an organised teamspeak group dropping as 12 man.
Some powerful teams are formed by good leaders who hand pick and invite their friends list into groups. Can be more organized and more lethal than a unit group.
Especially when the opposition misjudged them as disorganized pugs.

Edited by joelmuzz, 13 May 2015 - 06:07 AM.


#22 Kyynele

    Member

  • PipPipPipPipPipPipPip
  • The Vicious
  • The Vicious
  • 973 posts

Posted 13 May 2015 - 06:30 AM

View PostPhoenixFire55, on 13 May 2015 - 05:28 AM, said:

What is your point? That MM is working just fine? This 12:0 roflmaostomp disagrees. This is a match from EU prime time, there are plenty people online.

As for 1:1 ratio you know well enough yourself that for most people it is just about 1:1. And it was achieved by placing them in such particular situation more or less every match, on either a bound-to-win or bound-to-lose team with equal probability.


I thought someone else claimed that there was a problem with the MM stacking all the good groups on one side. My point was that the only problem in this match was that one big group in this matchup was greatly above average and would've carried their team no matter how you rearranged these groups. If the intention of this thread was to point out that the MM is broken and puts all the best players on one side, I considered it relevant information to understand that this is not the case.

I think you might be greatly overestimating the number of groups in European prime time that are capable of winning against a comp tier group with 6+ players. Also, some of these groups will be in ongoing matches, tinkering in mechlab, or perhaps playing some CW, meaning that they're unavailable for the matchmaker despite technically being online.

The matchmaker just has to put these groups in some matches after they've queued long enough. Any screenshot that has a single good group carrying a match isn't proof matchmaking is broken, it's proof that matchmaker does what it has to do, and lets these people play some matches, even when it fails to find them a proper challenge.

You can call me naive, but I think the matchmaker does it's best to get good matches, and likely does a decent job in the midrange Elos. Problems arise when extreme high (or low) Elo players have to be put into matches that are far from their own class, when they've been in the queue long enough so that they have to be placed somewhere.

I think people think too much into the bit of matchmaker "knowing" which side is more likely to win or lose, that's just a mechanic so that the people who participate in a match like this - that had to be put together despite it not being fair - don't have their Elo affected by the result. It's not the basis for creating a match, it's a failsafe to reduce long term problems when a good match was impossible to find. You being in the higher end of the Elo range, just likely get to see a larger portion of bad matchups, since your Elo makes it much harder to get good ones. You might want to make an alt account sometime just to broaden your perspective on matchmaking of the masses?

If you're at an extreme end of the curve, you're probably not in the best position to judge how well the MM works for the average player. Which is why mentioning 1:1 W/L is sort of funny.

I think that is my point.

edit- typos

Edited by Kyynele, 13 May 2015 - 06:30 AM.


#23 Wintersdark

    Member

  • PipPipPipPipPipPipPipPipPipPipPip
  • 13,375 posts
  • Google+: Link
  • Twitter: Link
  • LocationCalgary, AB

Posted 13 May 2015 - 07:11 AM

View PostKyynele, on 13 May 2015 - 06:30 AM, said:

I think people think too much into the bit of matchmaker "knowing" which side is more likely to win or lose, that's just a mechanic so that the people who participate in a match like this - that had to be put together despite it not being fair - don't have their Elo affected by the result. It's not the basis for creating a match, it's a failsafe to reduce long term problems when a good match was impossible to find. You being in the higher end of the Elo range, just likely get to see a larger portion of bad matchups, since your Elo makes it much harder to get good ones. You might want to make an alt account sometime just to broaden your perspective on matchmaking of the masses?

If you're at an extreme end of the curve, you're probably not in the best position to judge how well the MM works for the average player. Which is why mentioning 1:1 W/L is sort of funny.

I think that is my point.

Exactly!

The matchmaker is using Elo to determine the likelyhood of each side winning, and using that to modify your Elo rating after the match. It is not building a winning team and a losing team.

The goal of the MM is to make as balanced a match as it can, but as has been said many times already in this thread, it often fails once you get very high (and very low) ranked players/groups in the queue because there aren't enough like-ranked players/groups searching for a match at that instant to make a good match.

The MM will eventually just throw people together to make a match, because a lopsided match is better than staring at a searching screen.

Then it needs to make matches of group sizes that fit together, AND don't horribly mangle the weight class slot restrictions (ie: not putting 2 groups of 3 assaults together), AND has chosen the correct game modes AND has comparable Elo ratings. It's basically impossible.

"Re coding" the matchmaker doesn't fix this. You can't right a better matchmaker when the inputs are the problem. And ultimately, highly (and very low) skilled players are always going to remain rare: the Elo distribution is a bell curve with fairly long tails and the VAST majority of players in the middle ranks.

As long as we can pick what kinds of games we play (severely fracturing the queue) and play in any size group we want (this coupled with the weight class restriction serves to basically remove ALL choice from the MM - note that I don't think it should change!) group queue matches simply WILL end up with a wide Elo range. They must. There's no way around that.

#24 PhoenixFire55

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Legendary Founder
  • Legendary Founder
  • 5,725 posts
  • LocationSt.Petersburg / Outreach

Posted 13 May 2015 - 07:21 AM

View PostKyynele, on 13 May 2015 - 06:30 AM, said:

Any screenshot that has a single good group carrying a match isn't proof matchmaking is broken, it's proof that matchmaker does what it has to do, and lets these people play some matches, even when it fails to find them a proper challenge.


Honestly ... after reading this part I can't really force myself into reading any further. If you think THAT is what Elo-MM is supposed to do then I'll tell you that the random MM we had 3 years ago did the same thing, better and faster. The % of roflstomp matches remained the same (we can even speculate that if random MM we had before had separate solo and group queues it would have probably produced better balanced matches than this Elo ****).

This whole thing is NOT about putting better players into same team all the time. It is about MM not producing a well balanced match when it has the option to. If you don't like the sceenshot the OP gave us I can dig up dozens more with Spikes/Lords/RJF/WDMC/STS on one team and a bunch of random small groups on the other.

What PGI promised is a MM that'll give balanced matches to EVERY tier of players. I don't know for anyone else, but I haven't seen balanced matches with good % ever. Promised and not delivered - usual story with them, as well as claiming that the piece of crap they gave us is actually exactly what they promised (as a founder I have an epic experience in such matters here).

As I said, we've been over this many times. Everyone can do a simple test, go and play like 50 matches, record the scores and see just how many end up 12-0 to 12-2 / 0-12 to 2-12. The % of actually balanced matches will be the same as with any MM randomly drawing two teams.

#25 PhoenixFire55

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Legendary Founder
  • Legendary Founder
  • 5,725 posts
  • LocationSt.Petersburg / Outreach

Posted 13 May 2015 - 07:24 AM

View PostKyynele, on 13 May 2015 - 06:30 AM, said:

If you're at an extreme end of the curve, you're probably not in the best position to judge how well the MM works for the average player. Which is why mentioning 1:1 W/L is sort of funny.


True, but as I said, you know exactly what I mean by 1:1. It tryes to do it, and with players below certain skill level it succedes because they can't carry. But when you start to carry harder you do tip the balance in your favor (if ever so slightly) no matter what the MM throws at you.

#26 PhoenixFire55

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Legendary Founder
  • Legendary Founder
  • 5,725 posts
  • LocationSt.Petersburg / Outreach

Posted 13 May 2015 - 07:29 AM

View PostWintersdark, on 13 May 2015 - 07:11 AM, said:

"Re coding" the matchmaker doesn't fix this. You can't right a better matchmaker when the inputs are the problem. And ultimately, highly (and very low) skilled players are always going to remain rare: the Elo distribution is a bell curve with fairly long tails and the VAST majority of players in the middle ranks.


There were many suggestions for simple changes of the already exsisting MM that would have improved the balancing without affecting wait times. All ignored. The effort to make a really good MM was never there in the first place, nor will it suddenly appear, apparently.

As for "works for most people" ... I'll call BS on excuses like this as well, because there are all sorts of people providing screenshots of horribly unbalanced matches in all kinds of "skill-tiers". Following your logic matches in middle-tier will always be balanced, because of the same bell curve. However, they are not, which means that the player population is like 200 people overall (which is nonesense of course), or the MM simply isn't working at all for anyone.

#27 PhoenixFire55

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Legendary Founder
  • Legendary Founder
  • 5,725 posts
  • LocationSt.Petersburg / Outreach

Posted 13 May 2015 - 07:33 AM

View PostKyynele, on 13 May 2015 - 06:30 AM, said:

You can call me naive, but I think the matchmaker does it's best to get good matches, and likely does a decent job in the midrange Elos. Problems arise when extreme high (or low) Elo players have to be put into matches that are far from their own class, when they've been in the queue long enough so that they have to be placed somewhere.


Problem is, this situation apparently occurs so often that the amount of non-balanced matches because of this is about same it would have been if we didn't have any MM in the first place.

I honestly don't see any difference between closed beta MM and current MM (considering separation of queues and the switch from 8v8 to 12v12). Comparing the frequency of "I was stomped by premades" kinda topics appearing on this forum now and then, I can only say that many people don't see any difference either.

#28 PhoenixFire55

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Legendary Founder
  • Legendary Founder
  • 5,725 posts
  • LocationSt.Petersburg / Outreach

Posted 13 May 2015 - 07:37 AM

View PostKyynele, on 13 May 2015 - 06:30 AM, said:

I think people think too much into the bit of matchmaker "knowing" which side is more likely to win or lose, that's just a mechanic so that the people who participate in a match like this - that had to be put together despite it not being fair - don't have their Elo affected by the result. It's not the basis for creating a match, it's a failsafe to reduce long term problems when a good match was impossible to find. You being in the higher end of the Elo range, just likely get to see a larger portion of bad matchups, since your Elo makes it much harder to get good ones. You might want to make an alt account sometime just to broaden your perspective on matchmaking of the masses?


Well, looking at the OP screenshot, I could have predicted the outcome of that match way before it started. If I can do it, then obviousely any decent computer can do it as well. Unlike me, PGI has ALL the data, so they (i.e. the MM they make) should be able to do it as well. They don't.

Done the alt account thing ages ago. Matches are easier to carry (obviousely), but no matter how good or bad I actually play it still is about 50-50 W/L with rather small overall amount of actually interesting close matches.

#29 Kyynele

    Member

  • PipPipPipPipPipPipPip
  • The Vicious
  • The Vicious
  • 973 posts

Posted 13 May 2015 - 07:49 AM

Yes, for the extreme ends of the curve, the matches of truly random matchmaker would be pretty much the same, with much reduced waiting times, because most of the time they're impossible to match properly. For the vast majority of mid-skill players, the matches would be much more random in quality than they are now.

If your opinion is that your playing experience should be increased at the cost of the average players, that's obviously an opinion you're fully entitled to. But the only real way to get more high quality matches at higher tiers, would be to increase the playerbase so much that there would be a significant amount of players even in the extreme high and low levels. And that isn't fixable by the matchmaker.

View PostPhoenixFire55, on 13 May 2015 - 07:21 AM, said:

As I said, we've been over this many times. Everyone can do a simple test, go and play like 50 matches, record the scores and see just how many end up 12-0 to 12-2 / 0-12 to 2-12. The % of actually balanced matches will be the same as with any MM randomly drawing two teams.


I have a habit of collecting all match end screens I play. For your statistics, my last 50 games had 4 12-0 games, and in addition 10 games in which the losing side scored less than 3 kills.

I don't think that's terrible, considering the way matches easily snowball after initial first kills, because of the no-respawn nature of the game. But that is of course just my opinion.

#30 PhoenixFire55

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Legendary Founder
  • Legendary Founder
  • 5,725 posts
  • LocationSt.Petersburg / Outreach

Posted 13 May 2015 - 07:53 AM

View PostKyynele, on 13 May 2015 - 07:49 AM, said:

Yes, for the extreme ends of the curve, the matches of truly random matchmaker would be pretty much the same, with much reduced waiting times, because most of the time they're impossible to match properly. For the vast majority of mid-skill players, the matches would be much more random in quality than they are now.

If your opinion is that your playing experience should be increased at the cost of the average players, that's obviously an opinion you're fully entitled to. But the only real way to get more high quality matches at higher tiers, would be to increase the playerbase so much that there would be a significant amount of players even in the extreme high and low levels. And that isn't fixable by the matchmaker.


No. Because as I pointed out multiple times, the MM we have now fails to produce a better balanced match even when it has the people to do it. Wether it does so due to poor coding or on purpose is for all intents and purposes irrelevant.

#31 Kyynele

    Member

  • PipPipPipPipPipPipPip
  • The Vicious
  • The Vicious
  • 973 posts

Posted 13 May 2015 - 08:00 AM

View PostPhoenixFire55, on 13 May 2015 - 07:29 AM, said:

As for "works for most people" ... I'll call BS on excuses like this as well, because there are all sorts of people providing screenshots of horribly unbalanced matches in all kinds of "skill-tiers". Following your logic matches in middle-tier will always be balanced, because of the same bell curve. However, they are not, which means that the player population is like 200 people overall (which is nonesense of course), or the MM simply isn't working at all for anyone.


Well, the bad matches from high and low ends of the curve have to be put somewhere, and that is among the mid-tier players, because there are lots of them.

Also, measuring the quality of the match by how even the kills were, you know it's not a good way to measure it.

Even with competitive teams, one match can end 12-0 if the other team makes a mistake or has builds that don't work at all against the tactics the other team has chosen. It doesn't automatically mean the game was meaningless and shouldn't even have been played.

Edited by Kyynele, 13 May 2015 - 08:01 AM.


#32 PhoenixFire55

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Legendary Founder
  • Legendary Founder
  • 5,725 posts
  • LocationSt.Petersburg / Outreach

Posted 14 May 2015 - 12:32 AM

View PostKyynele, on 13 May 2015 - 07:49 AM, said:

I have a habit of collecting all match end screens I play. For your statistics, my last 50 games had 4 12-0 games, and in addition 10 games in which the losing side scored less than 3 kills.

I don't think that's terrible, considering the way matches easily snowball after initial first kills, because of the no-respawn nature of the game. But that is of course just my opinion.


Was that solo queue by chance?

#33 PhoenixFire55

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Legendary Founder
  • Legendary Founder
  • 5,725 posts
  • LocationSt.Petersburg / Outreach

Posted 14 May 2015 - 12:38 AM

View PostKyynele, on 13 May 2015 - 08:00 AM, said:

Even with competitive teams, one match can end 12-0 if the other team makes a mistake or has builds that don't work at all against the tactics the other team has chosen. It doesn't automatically mean the game was meaningless and shouldn't even have been played.


That is also true although really rare. Still, most of the one sided matches you can predict the outcome by just looking at teams at the start of it. As I said, many suggestions regarding how to better balance matches including adjustments to Elo scores based on respective mech chassis and builds were made and ignored. You still often see a bunch of brawlers on one team getting torn to shreds on Alpine by LRMs and long range direct fire builds from the other team.

#34 Kyynele

    Member

  • PipPipPipPipPipPipPip
  • The Vicious
  • The Vicious
  • 973 posts

Posted 15 May 2015 - 06:36 AM

View PostPhoenixFire55, on 14 May 2015 - 12:32 AM, said:

Was that solo queue by chance?


Yes, 12 of those matches were solo, and admittedly only 1 of those was 12-1.

Solo queue naturally works better, since its matchmaking is by far less restricted than the group queue. The pool for players to choose from is likely just as large as in group queue, and the pieces the matchmaker can put together are smaller. Much fewer compromises required.

View PostPhoenixFire55, on 14 May 2015 - 12:38 AM, said:

That is also true although really rare. Still, most of the one sided matches you can predict the outcome by just looking at teams at the start of it. As I said, many suggestions regarding how to better balance matches including adjustments to Elo scores based on respective mech chassis and builds were made and ignored. You still often see a bunch of brawlers on one team getting torn to shreds on Alpine by LRMs and long range direct fire builds from the other team.


Not really arguing with that notion. I'd personally be happy if there was at least some Elo adjustment just to cover using an unelited mech. That is an obvious, measurable disadvantage that could be reliably included in the matchmaking. I'm also not claiming that the MWO MM is the pinnacle of matchmakers, and of course, small iterative improvements to it should be made. But even with that, there will be some bad matches in the group queue.

Anyway, I'm just posting this because it's somewhat relevant to the point whether the end kill count is a good or bad measure: One of the better games I had in the group queue last night was against 228. It ended 12-2 in our favor, but the game was exciting, the opposition played very aggressively, pushed in an organized manner and tore several of our mechs, including mine completely to shreds. I think we can both agree that this was a pretty decent job by the matchmaker, despite the outcome:

Posted Image

While this game (that I didn't play on) looks like a pretty bad match despite ending 12-7:

Posted Image

And yes, I am aware that these ore just singled out events. In the end, most games aren't good or bad, they're just mediocre, and those typically aren't posted anywhere.





2 user(s) are reading this topic

0 members, 2 guests, 0 anonymous users