41 replies to this topic

[Gut]

This is very true.

However,

I would like this to be based much more closely on the players skill level, not an average elo picked from the person who's been waiting in the queue the longest.

Therefore, I'm by the nature of the setup, guaranteed to be playing against people in my elo range, instead of widening the search.

[Karamarka]

I wish there were elo brackets.

Seriously,

0-500 elo

501-1000 elo

1001-1500 elo

1501-2000 elo

etc

Right now it's 0-3000 elo and both teams average out each other. Which means you have some guys expected to carry each game and the rest are dead weight to balance the MM.

Makes for stomps and just crazy stuff - You always see that guy in a trial mech or someone dying in first 60 seconds like some joke. You do 800 dmg and your team still gets rolled.

Edited by Karamarka, 16 August 2015 - 08:40 PM.

[Wintersdark]

Karamarka, on 16 August 2015 - 08:18 PM, said:

Right now it's 0-3000 elo and both teams average out each other. Which means you have some guys expected to carry each game and the rest are dead weight to balance the MM.

This is incorrect.

Teams are built as follows (solo queue):
Oldest player in queue kicks off matchmaking, his team is built around his Elo value.
Players are added based on who is closest to his Elo and in the correct weight classes, initially with very tight max spread in Elo ratings
If a team cannot be made in a couple minutes, the Elo rating bracketing is relaxed until a team is made.
Team 2 is constructed in the same manner as Team 1, being seeded with Team 1's average Elo value.
Team 2's average Elo is calculated for end of match Elo modification.

Note that this means that if there are players close to your Elo in the queue and in the correct weight classes, they are in the match with you.

Also: close team Elo ratings does not mean less stomps. In fact, closely matches teams at higher ratings are more likely to get stomp results, as more skilled players can better exploit the numerical advantage. Get up 2 mechs and you win by a landslide, typically.

So, what's wrong here isn't the matchmaker algorithm. That's clearly fine. The problem is one or more of these:

* You're in the group queue, where other extreme match!aker constraints basically fill matches with random players.
* Pilot Elo scores are wrong (too much group queue time can over inflate Elo rankings)
* There are insufficient players at the desired Elo in the queue, so the MM had to pull in more random players to !make a match happen.

[KinLuu]

The fault of the MM might also be, that elo rating brackets are relaxed too fast and too wide.

[Deathlike]

KinLuu, on 16 August 2015 - 10:21 PM, said:

The fault of the MM might also be, that elo rating brackets are relaxed too fast and too wide.

I don't think the relaxing is entirely intentional.

It assumes a lot on the distribution of players.. an even curve if you will.

However, it also assumes that there is a far amount on this curve, relative to when the match is needed...

There isn't really enough people at a particular Elo (usually high) on at the same time whenever the MM needs it.

If the game was truly "large and diverse", then this would be a non-issue... but this has always been an issue for a long time now. This isn't even restricted to tonnage/weight classes either.

[Joseph Mallan]

Karamarka, on 16 August 2015 - 08:18 PM, said:

I wish there were elo brackets.

Seriously,

0-500 elo

501-1000 elo

1001-1500 elo

1501-2000 elo

etc

Right now it's 0-3000 elo and both teams average out each other. Which means you have some guys expected to carry each game and the rest are dead weight to balance the MM.

Makes for stomps and just crazy stuff - You always see that guy in a trial mech or someone dying in first 60 seconds like some joke. You do 800 dmg and your team still gets rolled.

What I see is their Assault lance stayed together and carried the team.

[TWIAFU]

Joseph Mallan, on 17 August 2015 - 04:11 AM, said:

What I see is their Assault lance stayed together and carried the team.

The screenshot only shows the score, not what the team was doing to get it.

That end result screenshot means nothing without the context of the battle and where the loosing team was and was doing.

Were they on the offensive?

Did they run away at first sign of enemy?

They did all bundle up and wait for enemy to come and roll them?

They let enemy take high ground?

Loosing team have no ECM while enemy did?

Enemy team use teamwork at all?

Have a DC?

Screenshot does nothing but show a team got rolled and does not show what they did to get rolled.

[Mawai]

Duke Nedo, on 13 August 2015 - 10:39 PM, said:

I have a feeling that poor MM:ing in the solo queue is mainly a case of "shite in, shite out"... Russ even said that solo players generally have average Elo, the ones that have high Elo are the ones thta regularly drop as large groups in the group queue. That's the place where you can reliably build a high Elo. Solo queue only is a rollercoaster and it's hard to significantly affect your Elo as one player out of 24.

There was talk about separating Group Elo from Solo Elo, IMO that would be step 1 towards better match-making. There was talk about it, they didn't do it did they? Then, you'd probably need to help Elo migration along by in one way or the other have the match score affect it a little bit. As it is now, being only 1/12th of your team contribution you'll need somewhere in the hundreds to thousands of games to rise above the noise and really home in on your true Elo. Too many factors at play to allow your individual skill to stand out...

The actual matchmaking itself I believe would work quite OK the way it is if Elo had been an accurate description of each players contribution in the match to come. Always room for improvement for sure, but I think the main problem is Elo accuracy.

One thing that I have suggested before is a round of reshuffling the players between the teams after the 24 are selected to distribute tonnage or mech tiers or things like that between the two teams to see if teams can be improved (while this time ignoring Elo, assuming all 24 are close enough not to matter). A few iterations of that would only take milliseconds as long as it's limited to the selected 24 players only. In a way that is similar to what you suggest here, looking at tonnage after player selection. That part I think is key, fully support that!

1) "Russ even said that solo players generally have average Elo," ... this is true by definition. The Elo curve is peaked at the average ... here is a link to an old figure that shows the distribution of players at that time.

http://mwomercs.com/...ted-april-19th/

The only reason group players have a higher Elo is because large groups coordinate better and thus can win more. This is especially true when the matchmaker does not require equally matched groups on either side and doesn't distinguish between competitive and casual groups. A competitive 10 or 12 man group matched against a smaller casual group will stomp the opposition. They occasionally will be matched against another competitive similarly sized group but the matches that are unbalanced in group size are ones where they can build Elo.


2) PGI has apparently said they will split group and solo Elo. They should have done this (as many have suggested) since the time they first implemented Elo since coordination is a much bigger factor generally than individual skill in winning/losing the match.

3) It apparently does not require hundreds or thousands of games to obtain a decent estimate of Elo (based on a comment by the developer who wrote the code - you can dig through the "Paging Karl Berg" thread in off-topic if you want to find the citation). There are a lot of factors involved but the only common factor in every single one of your matches is YOU. The figure shown in (1) seems to indicate that things start settling out after about 50 games or more.

4) I agree that a tonnage normalization pass after forming the match might be quite useful if the Elo range on both teams isn't too large. It would be fast and since there are already equal number of mechs in each class it would simply be a matter of choosing which ones from each weight class to assign to each side in the match to best balance the tonnage. At least it would help reduce the number of 3 Dire Whales vs 3 Awesome match ups or 3 Stormcrow vs Cicada+Kintaro+Blackjack.

[Mawai]

Karamarka, on 16 August 2015 - 08:18 PM, said:

I wish there were elo brackets.

Seriously,

0-500 elo

501-1000 elo

1001-1500 elo

1501-2000 elo

etc

Right now it's 0-3000 elo and both teams average out each other. Which means you have some guys expected to carry each game and the rest are dead weight to balance the MM.

Makes for stomps and just crazy stuff - You always see that guy in a trial mech or someone dying in first 60 seconds like some joke. You do 800 dmg and your team still gets rolled.

You would see the same thing with brackets.

If the opposing team stays together and you split up. You lose. Even if the Elo of each individual member of your team is higher.

Maybe the players on your team all like sniping or playing support. Maybe they are really good at those roles and get lots of wins. However, they are then dropped on a team where everyone likes to play support (was that an LRM WarHawk by the way?). In that situation, they may become like a leaderless blob, each one moving to their usual support positions but without the nucleus to support. They are then picked off one by one as the opposing blob runs them over.

This has NOTHING to do with Elo. It has to do with how YOU and YOUR team mates adjust your playstyle and the use of your mechs to the folks on both your team and the opposing team.

MWO is a team game. NOT playing like a team loses far more matches than "Elo" ... and honestly, it is really a mixed bag as to whether the folks on your team play well together. Get a bunch of high Elo rambos and your side crashes and burns against anyone with even modest coordination.

I really wish that PGI would post the difference in average Elo between teams AND the Elo range on each team as part of the end of match screen ... then folks could lose the crutch of blaming Elo and figure out how to play better with the mechs and team mates they get in each random match since no matter how good YOUR individual contribution might be, you will never win without your team.

[Karamarka]

Mawai, on 17 August 2015 - 04:57 AM, said:

You would see the same thing with brackets.

If the opposing team stays together and you split up. You lose. Even if the Elo of each individual member of your team is higher.

Maybe the players on your team all like sniping or playing support. Maybe they are really good at those roles and get lots of wins. However, they are then dropped on a team where everyone likes to play support (was that an LRM WarHawk by the way?). In that situation, they may become like a leaderless blob, each one moving to their usual support positions but without the nucleus to support. They are then picked off one by one as the opposing blob runs them over.

This has NOTHING to do with Elo. It has to do with how YOU and YOUR team mates adjust your playstyle and the use of your mechs to the folks on both your team and the opposing team.

MWO is a team game. NOT playing like a team loses far more matches than "Elo" ... and honestly, it is really a mixed bag as to whether the folks on your team play well together. Get a bunch of high Elo rambos and your side crashes and burns against anyone with even modest coordination.

I really wish that PGI would post the difference in average Elo between teams AND the Elo range on each team as part of the end of match screen ... then folks could lose the crutch of blaming Elo and figure out how to play better with the mechs and team mates they get in each random match since no matter how good YOUR individual contribution might be, you will never win without your team.

Blah blah blah

Your missing the point.

If you all have equal high skill. You wont have these idiots who suicide in the first 60 seconds

You wont have these idiots who run in the water by themselves

You wont have idiots who don't even care for teamwork

People say you will get more stomps, No way.

It's just like any other team game. At low levels everyone just does there own thing, nobody really cares - CS:GO, DOTA 2. Tons of people just run off, have no idea of the game so it makes stomps.

You play CS:GO at low rank, nobody has mics and nobody even knows the maps or strats.

You play DOTA at a low rank, nobody knows the items, or the combos - and nobody teams up for ganks.

MWO new players have no idea the maps, no idea how to make their own mechs

Skilled players will know to stick with the group, because getting your team mates to tank for you is the best way to make damage.

Edited by Karamarka, 17 August 2015 - 05:07 AM.

[Wintersdark]

KinLuu, on 16 August 2015 - 10:21 PM, said:

The fault of the MM might also be, that elo rating brackets are relaxed too fast and too wide.

highly subjective. Karl posted the numbers about this a long time ago. Within a team, Elo scores were within 150pts of each other almost every match, and team vs team under 100.

Too fast? Currently, its 3 minutes to be maxed open (or was during Karl's time a few months ago)... Is that too soon? That's a judgement call, as everyone is going to have different opinions for matchmaking time vs relative Elo scores.

[Wintersdark]

Karamarka, on 17 August 2015 - 05:05 AM, said:

Blah blah blah

Your missing the point.

If you all have equal high skill. You wont have these idiots who suicide in the first 60 seconds

You wont have these idiots who run in the water by themselves

You wont have idiots who don't even care for teamwork

People say you will get more stomps, No way.

It's just like any other team game. At low levels everyone just does there own thing, nobody really cares - CS:GO, DOTA 2. Tons of people just run off, have no idea of the game so it makes stomps.

You play CS:GO at low rank, nobody has mics and nobody even knows the maps or strats.

You play DOTA at a low rank, nobody knows the items, or the combos - and nobody teams up for ganks.

MWO new players have no idea the maps, no idea how to make their own mechs

Skilled players will know to stick with the group, because getting your team mates to tank for you is the best way to make damage.

You can't compare with DOTA. All your points above assume enough skilled players not only playing concurrently but queuing upmat almost exactly the same time. There aren't enough, at higher Elo ratings, period. DOTA can because it has orders of magnitude more players.

That's the fundamental problem. There are lots of average players, and few great players. And even amoungst stronger experienced plays, many will still make a mistake/be unlucky/be running a bad "fun" build(that's what the solo queue is for!)/want to try something new/etc.

I'm getting close to 8000 drops, and am at least competent at the game. I still get randomly ganked at the start of some matches by lucky shots or due to unfortunate choices (misjudging where the enemy is, for example).

Someone has to be first to die, and it doesn't mean they're bad.

[Navid A1]

People have been bombarding Russ over on twitter, suggesting chassis based ELO for a long long time now.

That particular suggestion gets ignored every time.

[Gut]

Navid A1, on 17 August 2015 - 07:02 AM, said:

People have been bombarding Russ over on twitter, suggesting chassis based ELO for a long long time now.

That particular suggestion gets ignored every time.

There is almost no reason they should be ignoring it. It makes sense, doesn't add buckets, and would make a lot of people happy. There's almost no downside except pgi would have to code it in.

Pgi: but I don't wanna!

[Duke Nedo]

Gut, on 17 August 2015 - 08:26 AM, said:

There is almost no reason they should be ignoring it. It makes sense, doesn't add buckets, and would make a lot of people happy. There's almost no downside except pgi would have to code it in.

Pgi: but I don't wanna!

I think there is a really good reason for not doing that... Every time you buy a new chassi your elo would be at scratch. You would not have reached your supposed elo until after its mastered. Meaning most players would not be at their elo, meaning random match making... Some think that's good, not I... :-)

[Gut]

It means you can level mechs in a setting where you wouldn't automatically be matched up with high level players immediately is what it means

[Duke Nedo]

Gut, on 17 August 2015 - 09:24 AM, said:

It means you can level mechs in a setting where you wouldn't automatically be matched up with high level players immediately is what it means

I can only speak for myself but I am lightyears better now with an unskilled mech than I was 2 years ago.

[Navid A1]

Duke Nedo, on 17 August 2015 - 09:21 AM, said:

I think there is a really good reason for not doing that... Every time you buy a new chassi your elo would be at scratch. You would not have reached your supposed elo until after its mastered. Meaning most players would not be at their elo, meaning random match making... Some think that's good, not I... :-)

Easy work around: a hybrid system

Total ELO = A x (class ELO) + B x (chassis ELO) + C*
*(chassis skill level variable)

Adjust A, B and C to balance.

Edited by Navid A1, 17 August 2015 - 10:00 AM.

[Duke Nedo]

Navid A1, on 17 August 2015 - 09:59 AM, said:

Easy work around: a hybrid system

Total ELO = A x (class ELO) + B x (chassis ELO) + C*
*(chassis skill level variable)

Adjust A, B and C to balance.

That would be cool, sure! Could even put in a fraction of ("global" Elo) in there, i.e. all classes...

Edited by Duke Nedo, 17 August 2015 - 10:54 AM.

[Wintersdark]

Yeah, chassis specfici Elo is a possible thing. They've been reluctant to do it because it increases your time till good Elo (having to get 50-100 matches per chassis), though you'd end up with a more accurate score long term.

Karl favoured per chassis, and that's a huge +1 in my book. Dude was a smart guy.