182 replies to this topic

[MischiefSC]

Amsro, on 24 March 2014 - 04:06 PM, said:

This is claiming that matchmaker is working, yet new players are still stuffed into matches they don't belong in. Not working at all.

Sandpit, on 24 March 2014 - 04:10 PM, said:

Yea and NObody can explain to me how the "new" launch module the restricts group sizes stops that. But that's definitely par for the course around here unfortunately.

"Fix" everything BUT the cause of the problem and then wonder why new players still get stomped. It doesn't fit into PGI's "vision" apparently :\

You know, I've posted in dozens of threads about this. Dozens. Posted links to larger studies, done breakdowns of the statistical modeling and formula around how and why it works. In a lot of ways I feel like I'm arguing about the world being round, or evolution.

I'm not trying to offend anyone here but this isn't a new discussion.

New players start out at a semi-average Elo. Low side of average. This is important as it lets them see how regular players play the game. If you kept them segregated they'd start out with bad habits they'd have even harder times unlearning, or that they'd never unlearn. You'd literally create a point beyond which it was very difficult to improve as new players wouldn't get the experience with advanced players to see a lot of how the game is played. New players absolutely SHOULD be in matches with more experienced players.

The matchmaker works. I can't say it works perfectly but I'd say that's more an issue with population density than anything else.

Also, my understanding is that we're moving into tiers of Elo. 0-1100 (I think) is one tier, 1101 to 1500 is another, 1500 to 2800 is another. You'll drop with anyone in your tier. If there isn't a match to fit you in it'll bump you up or down a tier. Within each tier it will try to fit at close as possible while still fitting tonnage requirements.

I don't know if that's in yet. I do know I've had better matchmaking lately.

As new players start at 1400 it's quite possible they have a great first 5 matches and end up squeezing into the 1500 range and drop with more experienced players. What you're way less likely to see is golden elites matched with underhives to reach an 'average'.

You'll always see new players though. To enable them to advance they need dropped in with more experienced players, if only so they can spectate. We were all n00bs once, cut them some slack and try to show them the ropes. Someone has to.

Matchmaking is working better now than it has in a long time. Tonnage mismatches are way less common for me as are wide skill variance. Certainly on some matches one side is better than the other - that will always be the case. Sometimes it's just a matter of good people having a bad day, or bad people having a good one.

The new launch module will help with tonnage mismatch. It's actually a good solution IMO.

[Amsro]

MischiefSC, on 24 March 2014 - 05:01 PM, said:

Great post

I entirely disagree, but well written. In order to believe what you are saying I would need to assume that PGI can program a matchmaker to work properly. When gameplay tells me otherwise. (I'm also not alone in the observation)

Elo Buckets "may" help. But as is, no the matchmaker is not working, if it is working then it needs to be replaced. The Average Elo teams as it is currently doesn't make for good matches or teams. No amount of theory crafting can legitimize it.

Good thing it is being reworked soon, although I'll wait to see if it gets any better.

[MischiefSC]

Amsro, on 24 March 2014 - 06:34 PM, said:

I entirely disagree, but well written. In order to believe what you are saying I would need to assume that PGI can program a matchmaker to work properly. When gameplay tells me otherwise. (I'm also not alone in the observation)

Elo Buckets "may" help. But as is, no the matchmaker is not working, if it is working then it needs to be replaced. The Average Elo teams as it is currently doesn't make for good matches or teams. No amount of theory crafting can legitimize it.

Good thing it is being reworked soon, although I'll wait to see if it gets any better.

Okay. Please present the statistical formula that show it doesn't work. I've put up in a dozen posts close to 50 times the formula that show how and why it does. I'm curious to see it - it'd be ground breaking to show how you can't distil a single factors impact on the outcome of a 12 v 12 environment over a sufficient sample size.

*EDITED*

Sorry. didn't mean to get short with you.

PGI presented us with the formula being used to generate our Elo score and it's perfectly sound. That Elo calculations work and work effectively is something that isn't hard to affirm. Population density can be too low to populate matches well. Also no matter what you do with a matchmaker matches will both look and feel uneven - the 'perfectly balanced match' is more a myth than a reality and that's going to be the case in any environment with any degree of customization. You don't have an Elo for every mech and build, you have it by weight class. So your performance with a Victor poptart is calculated in the same bucket with your performance with an Awesome PPC boat.

If populations were bigger then a more granular Elo calculation a la TrueSkills metrics system might be useful. Currently though a general feel for someones performance is more than good enough.

Edited by MischiefSC, 24 March 2014 - 06:53 PM.

[Amsro]

MischiefSC, on 24 March 2014 - 06:40 PM, said:

Okay. Please present the statistical formula that show it doesn't work. I've put up in a dozen posts close to 50 times the formula that show how and why it does. I'm curious to see it - it'd be ground breaking to show how you can't distil a single factors impact on the outcome of a 12 v 12 environment over a sufficient sample size.

*EDITED*

Sorry. didn't mean to get short with you.

PGI presented us with the formula being used to generate our Elo score and it's perfectly sound. That Elo calculations work and work effectively is something that isn't hard to affirm. Population density can be too low to populate matches well. Also no matter what you do with a matchmaker matches will both look and feel uneven - the 'perfectly balanced match' is more a myth than a reality and that's going to be the case in any environment with any degree of customization. You don't have an Elo for every mech and build, you have it by weight class. So your performance with a Victor poptart is calculated in the same bucket with your performance with an Awesome PPC boat.

If populations were bigger then a more granular Elo calculation a la TrueSkills metrics system might be useful. Currently though a general feel for someones performance is more than good enough.

Maybe population size is the issue, either way PGI should have this info, but they refuse to show any useful data, so either the matchmaker isn't applicable to our small player base (not a good sign) or the matchmaker is somewhere broken.

I agree with your data and analysis (I've read your other posts) but it just doesn't coincide with the average match. Something is amiss. I don't have the data to tell you what unfortunately. But then none of us do.

[MischiefSC]

Amsro, on 24 March 2014 - 06:58 PM, said:

Maybe population size is the issue, either way PGI should have this info, but they refuse to show any useful data, so either the matchmaker isn't applicable to our small player base (not a good sign) or the matchmaker is somewhere broken.

I agree with your data and analysis (I've read your other posts) but it just doesn't coincide with the average match. Something is amiss. I don't have the data to tell you what unfortunately. But then none of us do.

Let me ask you this - what doesn't seem to jive? Your team, the other team? What are you expecting that's not happening?

[Amsro]

MischiefSC, on 24 March 2014 - 07:02 PM, said:

Let me ask you this - what doesn't seem to jive? Your team, the other team? What are you expecting that's not happening?

Depends on the match, but regularly having a derp squad or a great team doesn't seem to imply that everyone in the match is evenly skilled.

The fact that the matchmaker is trying to normalize your wins by predefining which team should win is laughable. Just match skill for skill.

[MischiefSC]

Amsro, on 24 March 2014 - 07:09 PM, said:

Depends on the match, but regularly having a derp squad or a great team doesn't seem to imply that everyone in the match is evenly skilled.

The fact that the matchmaker is trying to normalize your wins by predefining which team should win is laughable. Just match skill for skill.

The matchmaker is going to just match you in a tier. I do agree that matching in a band is better than high/low to a target, though high/low to a target is more a product of low population and trying to match weight with it.

Just remember; everyone else has the same handicaps you do for teams with and against. That, like all other random variables, wash out over time. Your Elo is accurate. If the current matchmaker is the best way to use it, well that's debatable. I've always said match in a band (or in this case tier) is going to be better.

[Amsro]

MischiefSC, on 24 March 2014 - 07:15 PM, said:

The matchmaker is going to just match you in a tier. I do agree that matching in a band is better than high/low to a target, though high/low to a target is more a product of low population and trying to match weight with it.

Just remember; everyone else has the same handicaps you do for teams with and against. That, like all other random variables, wash out over time. Your Elo is accurate. If the current matchmaker is the best way to use it, well that's debatable. I've always said match in a band (or in this case tier) is going to be better.

Hard for the "Elo" to be accurate when the matchmaker is trolling you. I would say its "close" at best. It's not about the handicaps it's about the sheer lunacy that goes on during a match, people asking how to target enemy mechs.! While the Lords are GGClosing the newbs on my team to death. Other times it's me carving my way through 7 of the enemy team while they are trying to aim and missing horribly.

Again once the buckets come into play this may infact get alleviated but as of right now its still a broken system.

Edit; There is no way you can at this time make me lower my standards in order to accept that we all get the same crap matchmaker so it's cool.

Edited by Amsro, 24 March 2014 - 07:22 PM.

[Deathlike]

MischiefSC, on 24 March 2014 - 07:02 PM, said:

Let me ask you this - what doesn't seem to jive? Your team, the other team? What are you expecting that's not happening?

I think it has to do with the MM more than anything else, but indirectly it may be giving higher Elo to players undeserving of it (group play) and subsequently slap in the "underhive" to "even things out".

There are flaws in PGI's system, and I can attribute many roflstomps to it. It doesn't mean every roflstomp can be categorized the same (like bad plan, or just people derping mindlessly), but there are patterns to failure... and there's a smorgasbord of issues ranging from skill to non-skill related that cause it to happen. It's not that good at all really.

The fundamental concept of Elo is sound, but the actual reality and implementation through the MM leaves a lot to be desired.

Edited by Deathlike, 24 March 2014 - 07:22 PM.

[MischiefSC]

I've been that 'Underhive' guy who does something stupid and goes down in less than 100 damage. I've also had 8 kill matches in a 1G battlemaster.

It can be easy to also take those events, the statistical outliers, and provide them more credence than they deserve.

On average most your games will be pretty well balanced. Within a couple ticks of of balanced. You don't want perfectly balanced every match - makes it tough to really test your position in Elo. You HAVE to play against better people - if you're losing to a better team your Elo won't change, when you beat them it'll bring you up a bit. If you can then keep an ~50% win/loss at the new Elo then you're clearly in a better place. You also need tested against lesser teams, for the same reason.

You don't WANT perfectly balanced in a matchmaker. It's static. You need to shake the box if you want to see what's going to come out.

The recent changes have been good though. For example I'm getting better matches than I have in forever. Better balanced. They can be stomps, sure. There are stomps in perfectly balanced matches. I've been in plenty of 12mans and they're almost always stomps one way or the other, even with two excellent teams. A stomp isn't a mismatch. Even a match where one team has little chance of winning isn't a mismatch; it's an opportunity for either side to show the matchmaker that its assessment of them was incorrect, one way or the other.

The further out from center you get the more you become the statistical outlier as well. This is where population density comes in; you could have a MM that perfectly accurately knows exactly what your skill is in the specific mech and loadout you've got - if the closest match available for the tonnage you've got in the 2 minutes or so it's going to spend trying to match you is way off your Elo, well, that's still where you're going.

I still agree that matching tiers is better - but that's what they're doing. Also the 3/3/3/3 may feel.... forced, but for tonnage matching it's a much better option. Pleasantly elegant. I'd like it to have two 'floating' spots that can shift up or down 1 slot for some variety but it's a good concept. So one team may be 4/2/4/2 and the other 5/1/3/3 or the like but still, for weight matching it's a better concept. Especially with the new value of LRM boats - almost all the serious ones are assaults, plus the odd Catapult. This fixes situations where one team has 8 assaults, 4 of them LRM boats plus 4 poptarts, then 4 light/medium mechs for spotters. Half their team will be lights/mediums, half heavy/assault. You can always count on 3 or so lights on your team so LRMs are a good gamble; you'll have spotters. Being a spotter goes the same route; you'll have at least 3 Assaults and 3 heavies so Orion/Catapult LRM boats plus the 3 Assaults means you're probably going to have someone to spot for.

Get what I mean? It's a significant improvement and I'm looking forward to it.

[Amsro]

MischiefSC, on 24 March 2014 - 07:36 PM, said:

Get what I mean? It's a significant improvement and I'm looking forward to it.

I really want to agree, trust me I do.

[Alex Warden]

PhoenixFire55, on 24 March 2014 - 06:11 AM, said:

PUG teams are still getting massacred and they still blame everyone and everything but themselves. Nothing will change.

that´s EXACTLY what i´ve been saying back when PGI announced group size limits for the first time...

Edited by Alex Warden, 24 March 2014 - 08:05 PM.

[Davers]

I never argue with MischiefSC about Elo.

But, if PGI's numbers are correct then there is a LOT of people playing MW:O. So why did PGI feel the need to extend the search range of the MM? You would think there would be more than enough people at all Elo levels that they could actually shrink the Elo parameters.

[MischiefSC]

Davers, on 24 March 2014 - 08:43 PM, said:

I never argue with MischiefSC about Elo.

But, if PGI's numbers are correct then there is a LOT of people playing MW:O. So why did PGI feel the need to extend the search range of the MM? You would think there would be more than enough people at all Elo levels that they could actually shrink the Elo parameters.

Define 'lots'.

With a larger pool of players you can drill down on details, like how well you perform with a specific mech and loadout, on certain maps, etc. It's useful because you've got enough players to fill matches based on that sort of detail.

Admittedly weight-matching puts a huge load on the Elo-side of matchmaking. I'm absolutely unsure how the 3/3/3/3 setup will impact that - making it harder or easier to match Elo. In the end though, Elo in tiers is vastly better. If you claw your way over 1500 Elo score then you're golden - no more matching you to underhivers and n00bs. You'll still see new players who've had a good 10 or 15 matches to start for example but in general it's going to be experienced players.

So based on that decision my expectation is that they forsee the 3/3/3/3 technically making matchmaking HARDER on Elo.... however, it won't then have to go find someone to offset yours. It just needs to find anyone who fits the weightclass who's in your same tier.

I expect it to be amazing.

[Deathlike]

MischiefSC, on 24 March 2014 - 09:07 PM, said:

I expect it to be amazing.

I sincerely hope you win lots and show up @ "true high Elo".

[MischiefSC]

Deathlike, on 24 March 2014 - 09:08 PM, said:

I sincerely hope you win lots and show up @ "true high Elo".

You just need >1500. That's the awesome bit. You cross that line and it's all experienced players, every match, both sides.

Just think about it....

Rainbows, puppies and kittens for everyone in every match!

[Deathlike]

MischiefSC, on 24 March 2014 - 09:33 PM, said:

You just need >1500. That's the awesome bit. You cross that line and it's all experienced players, every match, both sides.

Just think about it....

Rainbows, puppies and kittens for everyone in every match!

Depending on PGI's implementation, I could see more "kitten underhive drivers" in my possible future.

[Ordellus]

Mystere, on 20 March 2014 - 09:23 AM, said:

I doubt that is indeed the case and therefore classify it as tinfoil hat territory.

What I do think is happening is that a whole lot of players automatically blame anything and everything -- except themselves -- when they get stomped. They don't want to admit that they probably suck at the game and instead blame premades, poptarts, heavies/assaults, etc., etc. etc.

In other words, they blame the Bogeyman.

That sounds like the words from a habitual pug stomper.

Come on, everyone that has played this game for any real amount of time can tell the difference between them screwing up and them never having a chance in the first place.

And we all know the match maker sets up the latter FAR more often than it should.

Edited by Ordellus, 25 March 2014 - 01:57 AM.

[The Great Unwashed]

You can usually predict the outcome by watching the first 30 seconds of team movement; when lances scatter from the start you know the team is in trouble. When they do not: no guarantees... Then there is the light test; I've seen perfect setups disintegrate completely when half the team chases the Jenner....

Edited by The Great Unwashed, 25 March 2014 - 02:06 AM.

[Joseph Mallan]

MischiefSC, on 24 March 2014 - 09:33 PM, said:

You just need >1500. That's the awesome bit. You cross that line and it's all experienced players, every match, both sides.

Just think about it....

Rainbows, puppies and kittens for everyone in every match!

I see you used to do drugs like me!

Or

This is the best part of the trip
the trip the best part
I really like
...
...
... What'd he say?

+1 to he knows.

Edited by Joseph Mallan, 25 March 2014 - 04:16 AM.