268 replies to this topic

[Ecrof]

Matthew Craig thanks for all the feedback. I hope you will continue to talk with the community, it will keep them happy and make them feel that there thoughts are being listened to and considered. Even if you had to hire some one solely communicate on the forums a happy community equals more profit. Keep up the good work.

[Zordicron]

Matthew Craig, on 18 September 2013 - 11:54 AM, said:

Thanks for the feedback. Roland I agree more feedback from the matchmaker can make the wait more bearable, its essentially doing what is described behind the scenes currently, looking for game, couldn't find, broadening criteria etc. but all you see is a spinning dial. So something that can definitely be considered for UI 2.0

Mawai, actually no we don't have that granularity currently, being able to see wait time per weight class or per Elo bracket, something we'll likely be adding shortly. We still have the weight class restriction tighter than the Elo restriction so you may well be right, we're continuing to adjust the values today as we continue to learn more and try and find the best balance.

IMO, you should reverse this. The ELO is more important then the tonnage. Obviously, you want some type of sorting for tonnage too, but honestly, ifI am in a match in a Dragon, and they have a mid-high ELO atlas, grabbing "whoever we can find quick" in a HGn or Atlas to balance tonnage leads to MEGASTOMPS.

I would much rather see a group of team mates that can coordinate a little and have some awareness( I am in the middle somewhere I am sure on the bell curve, but i suspect towards the higher end more then the lower) and then have to fight an uphill battle against an enemy with an 100 ton advantage then be on a team with a bunch of derps because they happen to be in the tonnage we need to match the two teams up.

[HRR Insanity]

All players with high elo are experiencing very long matchmaking times (>2min on average) if not failing completely.

Matchmaker is busted.

Abort the current matchmaker ASAP or prioritize high elo players in the queue.

Edited by HRR Insanity, 18 September 2013 - 07:22 PM.

[Zordicron]

Maybe this is already sort of how it works, I dunno-
A bell curve, for ELO. Now, it might be an insurmountable task, maybe the ELo function just doesnt jive this way, but wouldnt it be better to just have plain Ol' brackets with X amount of players in them, then to have a bell curve?
Consider: the uber high and low ELO. Lets use 100 players total. They would be what, the 3-5% at each end of the curve? Maybe i read to much into the bell curve and how the Mm works, but if the MM is trying to find 23 other players to fill a team using the 3-5% of players, well, it is going to spit out a stupid amount of fails as there arent enough players in that % of ELO.

Maybe you guys addressed this in the "brackets" but seems like the uber high and low players should be included in the top and bottom brackets, and those brackkets should just be big enough to eliminate the horrendous looking for match times.

For instance, 100 players. Divide it into say, 5 brackets. 20 players each. No curve, just whatever the ELo is, you fit into that bracket and thats the pool of players used fore the match. The brackets would be populated by the 20% of the players in the respective ranges that are online at the time. So as to say, the ELO values that define the ends of the bracket are not fixed, they fluctuate based on the available players.EDIT: I had to clarify this- I dont mean available as in "currently hit launch" I mean as in "logged on". If this feels like it could be skewed somehow, shorten the logout for inactivity timer to make sure everyone online is actually online.

this would make it so even though the uber high and low ELo players might have a hard time finding a "perfect" match, they will find a match. And it wont be a match where that uber high or low is used to balance a team of the opposite end of the ELO values. At worst, a top or bottom 1% player would get put into a team at the 81% or 19% limit(for high and low respectively) so at the least, the match isnt COMPLETELY borked in regards to team skill gaps.

in order to further tune this, ELO based on chassis would be needed. For instance, me. In my master 1/1 Atlas D, "FACEWRECKZ" I am a significant force compared to my 2/8 basic Awesome PB that i cant seem to get a feel for. Both would be "assault" ELo rating, but obviously effectiveness is going to be varied significantly, and possibly enough to throw off a teams ELO balance. This would be a secondary priority however, nailing down the primary stuff so the variance isnt FUBAR is obviously number 1.

Anyway, i really dont know if the brackets can be made to function in a fluid manner like that or not, but it would make a big difference in skill disparity which is truly frustrating right now( weds night, presumably after the partial revert from tues. patch)

i would agree with others, First night after patch I had awesome fun and close games. Tonight I quit in disgust after 7 matches of horror.

Edited by Eldagore, 18 September 2013 - 07:34 PM.

[Asmosis]

^^ I'm guessing thats sortof what happens when it loosens its search conditions, but it'd probably be a lot more healthy for the community if there were brackets in the outliers at least, so that your not always fighting the same people which is probably what happens at the top/bottom.

I would hazard a guess that the top tier 3-5% or so are reasonably organised people, so quite a few at that level would migrate to 12v12 team queue once its better balanced (tonnage limit or something) which would make top level pug even harder to source players for.

heck, maybe the top elo bracket would be ok to match against 12 man premades (of any ELO level) ? would definitely work with integrated voice at least, dunno about without voice chat.

[Deathlike]

Xandre Blackheart, on 18 September 2013 - 05:00 PM, said:

I'm curious if there is indeed an Elo rating for each individual mech that a player has, or if there is a global Elo rating across all the mechs the player has?

I suppose you guys could calculate both pretty easily, maybe that's a way to get a little more granularity for matching players?

Comparing the current Elo for the players mech versus their global Elo might give a better prediction of the player's performance in a particular match.

For example a really good player who doesn't play light mechs very often would have a better chance of getting evenly matched using that type of calculation when piloting light mechs. Or mediums. Or whatever


Conversely a pilot who has terrible performance over most mechs just has the quirk of being a competent dragon pilot. If you matched him just according to his individual performance in the dragon, he wouldn't be matched evenly (he's not that good, just competent). If you compared or averaged his global Elo with his Elo from piloting his dragon he would be more likely to get a more balanced match.

I suppose that brings up issues with how you calculate global Elo for pilots who are still building their mech stable, but it should still work out. If you don't own a particular mech your win loss ratio is 50/50 which is neutral.

Maybe someone has a link to a good explication of how MWO actually does matchmaking at this point?

AFAIK, ELO is calculated between different weight classes.. so if you had a Spider+Jenner, they would be sharing the same ELO. If you got an Atlas, that would be shared between a Highlander or an Awesome....

Outside of that, it's been a while since there was a true discussion about ELO and the MM.

Also, I'm sure people would prefer ELO to be separated amongst variants (and for some people, actual builds between the 2 of the same variant with completely different loadouts, like a brawling D-DC vs a missile boat D-DC for instance). Right, I think they are struggling with just weight classes and eventually going to move to tonnage limits.

Edited by Deathlike, 18 September 2013 - 08:50 PM.

[Villz]

Still failing to find match solo queing even with the extended wait times ....

wtf is my elo ? 10 billion ?

[Kmieciu]

Deathlike, on 18 September 2013 - 01:32 PM, said:

I've never really liked that behavior... mainly because at times prior to this patch, I was generally one of the last people to get into the match... and the game always ends in a roflstomp going either way. It's almost as if the game was pulling a cruel joke on me.

Any time you end up "speeding things up towards the end", it has the same effect of "oh poop, I'll just take anyone on my team" and ignoring the consequence of that decision.

I also noticed I'm one of the last people to join a team, and the other team is already full. I have about 2.0 W/L and 3.5 K/D ratio. Most of the games I played yesterday were completely one-sided, with the loosing team only scoring 1-4 kills. I'm freaking tired of having to take the commander role each and every time, leading the charge and loosing despite scoring 4+ kills and being the last one alive on my team.

My guess is, that the match maker fills one team first, and then proceeds to fill the other. Please correct me if I'm wrong. So one side gets even ELO players (high or low) and the other gets the players from the edges of the bell curve: both pros and noobs. That's why I see lots people that do less that 50 damage during a match and few outstanding players that try to carry the team. (one time the second best player my team was a Jenner with 4 kills and 3 assists)

Edited by Kmieciu, 18 September 2013 - 11:23 PM.

[sikakraa]

I think it's alright that the matchmaker tries to average the ELO on both teams, because that way the low end players may learn from better players. A problem with that is that ELO isn't really sufficient metrics for a team-based game alone.

A suggestion:

For the matchmaking purposes, adjust the ELO of a player up or down based on the group size, matches played, win/loss & k/d ratios. The idea is to adjust the ELO system to fit to the tactical team-based nature of the game.

Some example rules for this:

- for non-12-player games, adjust a groups average ELO rating up based on player count: e.g. 2 man group is ranked 10% higher than their average. 3 man 15% and 4-man 20%.

- adjust player ELO up based on win/loss ratio. A positive win/loss ratio means, that the player usually has a positive effect on a team's changes on winning. The maximum effect of this should only be 10% though:
if (wl_ratio < 0.90) wl_ratio = 0.90;
if (wl_ratio > 1.10) wl_ratio = 1.10;
matchmaking_elo = elo*wl_ratio;

- In a similar way, adjust player's ELO up/down based on kill/death ratio, with the maximum effect of 5%.

- Adjust players ELO 15% down if they have less than 100 matches played if you are not already doing this.

[Funkadelic Mayhem]

Matthew Craig, on 18 September 2013 - 09:46 AM, said:

As we continue to evolve the match making algorithm it's more likely something we'd want to incorporate within the system such that it can internally reset buckets if needed to avoid this being something that needs to depend on user behaviour.
Something we've noticed with this recent test is that altering the ranges isn't having as big an impact as expected, more than anything though it's given us a lot of data that we can now work to make internal test scripts emulate such that we can feed it back into the match maker to make further adjustments to the algorithm it uses.

ELO would not be that big a deal if solo and premades dropped in diffident queues. Premades will only fight other premades and solo droppers will only fight other solo droppers.

[Doc Andrews]

Matthew Craig, on 18 September 2013 - 02:07 PM, said:

Another thought is to allow players to help determine how long they are willing to wait, so you can elect to wait 5 minutes for a good match knowing your guaranteed a tight game vs. electing for short wait times and taking the trade-off yourself that there's a chance it won't be that well balanced. At least each user can decide for themselves. Another potential was if you could ready multiple Mechs in multiple weight classes and have the one kick off into a game that finds the best match first.

Hey Matt.

I have a third option.

Instead of strictly enforcing the 12 man games (since most of the maps work just fine with 8 players), and allowing players to wait longer if they wish, how about the opposite:

If players choose the 'tighter elo' option, they opt in for drops UP TO 12 players. That means after waiting a reasonable mount of time, if there are at least 8 players per side, they drop. You could have 8 on 8, 9 on 9, 10 on 10, 11 on 11, or 12 on 12 as the result.

It adds a depth that strict numerical limits lack, and a new level of tactics that depend heavily on map and drop numbers.

It reduces overall wait times.

It allows for tighter ELO groupings.

Seems like a win all around.

[Doc Andrews]

sikakraa, on 19 September 2013 - 01:55 AM, said:

- In a similar way, adjust player's ELO up/down based on kill/death ratio, with the maximum effect of 5%.

KDR is meaningless in terms of ELO.

If you find a way to win 5 times out of 8, it doesn't matter if you ever fired a shot. It's a win. The method your contribution to the win (capping, soaking damage, distraction, non lethal damage, etc) is a non factor, as long as you can consistently replicate the result over time, and over a large number of games.

ELO measures individual contribution to the probability of a win. The method is transparent.

[Klaus]

I literally can't even get a game in one of my assaults unless it's prime time.

It is early in the morning before I head out while I am posting this, have had failed to find match 20 times now and counting.

[sikakraa]

Doc Andrews, on 19 September 2013 - 02:45 AM, said:

KDR is meaningless in terms of ELO.

If you find a way to win 5 times out of 8, it doesn't matter if you ever fired a shot. It's a win. The method your contribution to the win (capping, soaking damage, distraction, non lethal damage, etc) is a non factor, as long as you can consistently replicate the result over time, and over a large number of games.

ELO measures individual contribution to the probability of a win. The method is transparent.

Most of the games are won by killing the opposing team, especially on assault game mode. This is also enforced by the current rewards from a match, and thus I think it should affect the matchmaking too. The people with most assists, kills, component destructions etc get the best rewards.

In PUG games, the rewards matter. In those 12-man competive games, nobody cares about the rewards, only the victory matters.

BUT you are right: at least it shouldn't affect much. That's why the I proposed that effect would be only +/-5%.

The idea was to just to throw in some possible adjustments to ELO only for the matchmaking purposes that can be mined from the player statistics. I mean one could even do the adjustment based on the used mech and map and armament...

Edited by sikakraa, 19 September 2013 - 03:08 AM.

[Doc Andrews]

sikakraa, on 19 September 2013 - 02:59 AM, said:

The idea was to just to throw in some possible adjustments to ELO only for the matchmaking purposes that can be mined from the player statistics. I mean one could even do the adjustment based on the used mech and map and armament...

Could do all those things, but the Law of Large Numbers takes care of you, so no need. Keep ELO as simple as possible (like FIDE has for many, many years) and in the long term (months + hundreds of games) you'll reach a median over all maps, with all mechs, for all games types, etc etc.

If just you focus on win/loss for ELO, it also opens up the door to something everyone likes: Achievements!

[Haakon Magnusson]

Asmosis, on 18 September 2013 - 06:34 PM, said:

Question: as its filling the buckets, does it do one for each team, or does it fill a bucket of 24 then split? I'm guessing two 12 person buckets with the 2-4man teams, but if the MM could dynamically switch players from one side to the other that might help.

(just hold everyone on the "loading map" screen until buckets are sorted instead of dropping into the map loaded waiting for match screen).

This is a damn good question, as it seems that now in some matches players are added in last seconds to a team. So I guess it is per team(?)

To me it would seem if it were just trying to pull a group of 24 people first and maybe having to settle for 22-23 (or high/low elo additions) and THEN divide them up according to tonnage and elo so that things might end up more fair. To keep people involved display pool of people in the match already in UI2.0 and after finished split them up and throw them into mechs.

[Doc Andrews]

Haakon Magnusson, on 19 September 2013 - 03:44 AM, said:

This is a damn good question, as it seems that now in some matches players are added in last seconds to a team. So I guess it is per team(?)

Hmmm. Just posted this in another thread, but it seems just as appropriate here so... COPY PASTA!

---

I've submitted the approximate unified matchmaker formula used back in the 3065MUX era. They've reviewed it, but decided to go their own way for the moment. I'm sure there are a lot of factors that I don't know about, of course.

As it turns out, there are three factors to proper matchmaking that we found:

1) ELO+Level of Mech Mastery (around 45% of total matchmaker weighting)
2) Tonnage (around 35% of total matchmaker weighting)
3) % Cost Above/Below Base Price of Mech (around 20% of total matchmaker weighting)

ELO is strict win/loss calculation modified by skill of opponents.

In 3065 we used the traditional gun/pilot skill system as Mech Mastery, but here it's done by purchasing Basic, Elite, and Master skills.

Tonnage is tonnage.

% cost above/below base mech price takes into accounts putting on fancy/expensive weapons, modifications, modules, etc.


The combination of these factors with the approximate weighting seen above was key to getting balanced matches. I'm not certain the MWO even LOOKS at cost above/below base price. And I know tonnage is weighted rather lower than it should be in their formula (or it was simply broken... one or the other according to statistical results from around 3 months ago).

If they aren't looking at these three factors and properly weighting them, then the following can happen:

Master assault pilot with high ELO gets into his Jenner for a little fun. He has the Jenner at basic and is running all small lasers because he likes the way it sounds when he fires them. In order to make up for his high ELO and low tonnage, he's dropped with a Highlander pilot with low ELO. This happens a couple of times on the same side of the dop, and the match is over before it starts.

Without taking into account ELO/mastery, tonnage, and mods above/below base price, the system is making more educated guesses than it needs to. The facts that will help it make good matches are there... they just aren't being accessed.

---

Edited by Doc Andrews, 19 September 2013 - 04:07 AM.

[sikakraa]

Doc Andrews, on 19 September 2013 - 03:37 AM, said:

Could do all those things, but the Law of Large Numbers takes care of you, so no need. Keep ELO as simple as possible (like FIDE has for many, many years) and in the long term (months + hundreds of games) you'll reach a median over all maps, with all mechs, for all games types, etc etc.

If just you focus on win/loss for ELO, it also opens up the door to something everyone likes: Achievements!

Yeah. I guess I didn't explain the point well enough:

Keep ELO as it is. But don't use it as a sole mean of matchmaking. ELO just measures individual's wins and losses of an individual player, which might be intrepreted as skill if the player plays enough games in the same conditions (e.g. in chess, where the rules are always the same).

In MWO there are numerous variables that change the game conditions, where the biggest one of these conditions is the other players. It's true, that in the long run (roughly after a thousand games or so) the ELO value of a player may find a good place, but only if the player plays random games without premade groups.

Unfortunately this is not the case. ELO value for groups should be calculated based on that groups wins and losses, not the average of the group's individual ELO's. Also, the new players don't have really good starting point.

The point is that a single player's ELO rating should be used only as a part of the calculation to find a "matchmaking value" for a player, based on the player's statistics and group status.

Edit: Read your post (Doc Andrews) about the mech weight & price being a part of the calculation: I think you are right. The mech should be a part of the "matchmaking value" too.

Edited by sikakraa, 19 September 2013 - 04:16 AM.

[SchwarzerPeter]

Im sure that the mech loadout/prize will effect the outcomes of the match. But in long term everyone is upgrading their mech's so i wouldn't make it utterly complicated. Same with modules and efficiencies.

I also have seen a lot of unbalanced matches today. Many times 8-10 new players on one side, resulting in a complete stomp. It seems like the MM is focusing a lot of the tonnage.

I wouldn't mind longer waiting times if i get more feedback from the MM. And a cancel button would be nice.
Also the 3min cancel out would be unnecessary with a cancel button.

[Vassago Rain]

Villz, on 18 September 2013 - 12:12 AM, said:

Spoke with a few very known players in regards who also affirmed the same experience.

Ryan Steel and I were playing together and over a 35min period earlier today failed to find game about 12 times over that period many in a row again and again check the logs on my act please if you need verification.

Pretty jokes in a 2man group only ..... amongst 2 12man teams come on.

(Nerf Villz-Steel Mechworks Breaking Match Maker)

It's not only the top tier, but those of us edging on tier 2. I have to wait up to 3 minutes, and get some fails. Either people aren't playing at all, or something broke.