117 replies to this topic

[Hauser]

What mechs did you play the losing streak in btw? You know there is seperate Elo for weight classes right?

[Ronan]

Apologies if this is mentioned already in the thread, I skimmed some pages.

BUT, sometimes, even if your ELO is high, if you pug you can be lumped into a lower ELO team to bring their average up. That probably needs to be weighted some, as (has already been noted) one mech can't swing a match as well in 12v12.

Also sometimes, you get lumped in with a new breed of griefers that don't like the new CBill reward values. They drop, advance, shoot a bit, and get killed early so they can flip mechs and drop again. I don't get it (where's the fun, when do you start having fun?), I don't know how many pilots are actually doing it, and have no idea how you'd be able to tell from the server (for reporting/banning purposes), but maybe you're seeing the results? (read some posts on the forums about pilots threatening to do this, no idea if it became "popular")

Sometimes, it just can't be avoided. And the roflstomps do make the 12-11 matches just joyous to be in and spectate on!!

[Livewyr]

EDIT: Had a pick, but it was a bit too "braggy"


Let's just say MM hates me.

Edited by Livewyr, 11 August 2013 - 06:58 AM.

[Hauser]

Ronan, on 11 August 2013 - 06:31 AM, said:

Apologies if this is mentioned already in the thread, I skimmed some pages.

BUT, sometimes, even if your ELO is high, if you pug you can be lumped into a lower ELO team to bring their average up. That probably needs to be weighted some, as (has already been noted) one mech can't swing a match as well in 12v12.

I'm afraid you are mistaken. You are not being put in with lower Elo players to average you out. You may however be put in with lower Elo players because there are no other players available to make a match.

http://mwomercs.com/...-making-update/

Quote

How does the match maker compose a teams Elo rating, is it average rating or closest to a target?



It's closest to a target value, so the match maker starts trying to make a match for an Elo of say 1300 and will pull in players to those teams closest to those values; however, as mentioned earlier within growing thresholds and those curves will be tuned. Currently it may be a bit 'sloppy' about how it's filling those buckets but over time it will be tuned to be much more precise.

We need to do this carefully over time as generally the cost of precision is time to find a match we want to slowly find a very nice balance between time to find a match and the number of matches that are correctly composed.

[Hekalite]

Hauser, on 11 August 2013 - 06:17 AM, said:

What mechs did you play the losing streak in btw? You know there is seperate Elo for weight classes right?

Yes, I'm aware that the ELO is separate for each weight class, which is why I said that "I started to have reasonable results again in my medium mechs."

I can only guess at what put me in the whole. Why are you so sure that this isn't actually happening to people? I am not misinformed. I have been here long enough and read enough of the CC posts and Q&As to know how things are supposed to work. I've seen the normal pendulum swing as you rise to high and get knocked back down. I don't know how else to express that this experience is dramatically different.

[Kyle Reece]

Without knowing Elo scores, and being able to observe what goes into the black box that is the matchmaker (which we know has been flawed before) it's difficult to say why the OP specifically is seeing this behaviour.

We do know that the Match Maker does target a specific Elo score when it opens up a match, and that the overall average Elo of the current teams will affect the level of player that hits a particular spot - with further consideration to weight. What we don't know is the weightings that each of those things have in deciding which of the players who clicked ready in the last 2 minutes get to play in which matches.

We also know that your average team Elo exists and is important. This is what the final Elo movement is based on and is used to predict a win or loss which then affects the impact to you personally.

It's here that I think the main problem lies. In the use of the average team score to decide the impact to your own personal Elo score of any given match.

Basically, the only way you can positively impact your Elo in any meaningful way is to win against the odds. You either have to whip a team of lower Elo players into a cohesive unit and then beat "better" players in order to get the credit individually to improve your Elo score and get away from those lower Elo players.

12 v 12 has only made this problem worse. In 8 by 8 one man could make a difference and in fact carry a team if necessary (seen it, done it, got that t-shirt).

In 12, the impact of one of your team doing something stupid is increased (seen it, done it, got the t-shirt) and that makes the slippery slope to defeat that much steeper and easier to slide down.

How to fix it? I don't really have enough data to make a real suggestion, so I'll just go for generalisations;

1: Give me a rating based on my performance alone (say, is that a match score over there that's already a rating on my performance alone? What say we tweak that so it can be used...)
2: Use that rating to match me to people of a similar rating
3: Pick from that pool until we have 24 people split between 2 teams with a roughly similar number of 'mechs from each class (not weight)
4: Start the game so I can shoot something

[Steelgrave]

[ztac]

PGI have stated that there is a problem with the MM , but what they are proposing will be insignificant really. So you will all have to put up with it/ enjoy it for probably another year or two.

Maybe they will fix it when they are down to their last 10k players.(could be there already, who knows since they removed player counters).

But there is a lot more to a match making system than just giving players a rating, or rating + tonnage.

[DeaconW]

I just thought of something from my educational background...the MM problem looks like it may be an NP-Complete problem, hence why it is so hard to solve, even for only 24 "tasks".

http://en.wikipedia....iki/NP-complete

http://izbicki.me/pu...te-problems.pdf

Which gives me another idea...why can't the matchmaker wait until it gets 24 players, THEN use the Longest Processing Time First (LPTF) methodology (with ELO or preferably, another *appropriate* skill-based number for a player) as the "processing time". This should bin teams *evenly* more often than what is happening now IMO.

In layman's terms, teams would get picked highest to lowest ELO so, for example, in a purely lonewolf 12v12, the top two ELO's could NOT be on the same team. If there were groups, the same process could be done by treating the smaller groups as "players". Say you have 3 4-man's, one 2-man and the other 10 as singletons cued up for a match. The server calculates the highest ELO *sum* 4-man and puts it one one side, and the next highest ELO sum 4-man on the other. It then takes the two highest singleton ELO's and matches them with the 2 man to make a "4 man". This is then compared to the remaining 4 man and the two "4-man's" are split appropriately to *minimize* the difference in team ELO. The process continues with the remaining 10 players. Theoretically (and empirically given the success of LPTF in practice) this should give the fairest match possible assuming the player ELO's are valid. Heck, the more I think about it the more I think I could code the entire algorithm for this in a couple of days.

[Ronan]

Hauser, on 11 August 2013 - 06:58 AM, said:

I'm afraid you are mistaken. You are not being put in with lower Elo players to average you out. You may however be put in with lower Elo players because there are no other players available to make a match.

http://mwomercs.com/...-making-update/

Yup, I'm corrected. ELO gets averaged for a pre-made team. You can, as you mention, still get stuffed on a pug to fill out the ranks -- and you may be on a team then with an average ELO much higher or lower than yours. Either way can be fun.

[Hauser]

Deacon,

players can come in groups of of 1,2,3,4, two groups of 12 must be found such that:

• the weight difference between teams is minimal
  
• the deviation from the desired Elo level is minimal
  
• it doesn't take more then two minutes to find a match

The algorithm also has to work online, so while it is running it doesn't know what will show up in the future.



So supposing you could find 24 players near a given Elo value (to minimize deviation) LPTF would be suboptimal as it only works on one variable. It can either balance weight or Elo, not both at the same time. LPTF also can't handle groups. A knapsack algorithm might be more appropriate.

Specifically for the whole problem we'd need an online multiple constraints multiple knapsack algorithm.

Though right now it think it would be better if the current algorithm was tuned to also minimize the Elo difference between both teams and not just minimize the deviation from the target Elo. I think hat would cover your concern where the matchmaker stops caring by accident assigns all the good players to one team and the poor players to another.

Edited by Hauser, 12 August 2013 - 06:41 AM.

[Blood Skar]

Solo queue for Pugs that want it please. Thanks.

Game is a fail game when the enemy is 8 or 9 players that synch dropped and all on comms....i honestly dont know where these guys are finding enjoyment from stomping pug after pug...where is the sense of achievement *shrugs*

Basically these premades are getting to play the game exactly how they want, but everyone else has to like it and lump it. The Devs said a while back synch drops dont exist LOL riiiight ok like every 3rd game its a synch+comms. Its WORSE now its 12x12 - you know why ? because they are scared of actually losing a game if they only drop with 4...so they have to synch with 6,7 8 or more..i came from a game with 9 synchers in just now - it was 10 but 1 got put on our team by accident lol

A new player straight out the door can be wtfrolled into next week in their first game....how the f**k is that a good thing for the game???

Personally i've gone back to EVE Online/DCUO and a few other games i play and i find myself coming back now and then - but more and more infrequently.

[DeaconW]

Hauser, on 12 August 2013 - 06:24 AM, said:

Deacon,

players can come in groups of of 1,2,3,4, two groups of 12 must be found such that:

• the weight difference between teams is minimal
  
• the deviation from the desired Elo level is minimal
  
• it doesn't take more then two minutes to find a match

The algorithm also has to work online, so while it is running it doesn't know what will show up in the future.

Good post...I agree with your criteria except that online processing isn't necessary and would, in most cases, be suboptimal to can offline algorithm IMO. ..just wait to do the math when there are 24. The actual calculation time for a two variable optimization problem on 24 tasks would surely still be in the ms time domain.

Blood Skar, on 12 August 2013 - 08:58 AM, said:

Solo queue for Pugs that want it please. Thanks.

Absolutely agree with this.

[Ransack]

xenoglyph, on 10 August 2013 - 01:49 AM, said:

PGI had a stroke of genius. They realized they're murdering their player count, so they devised a way to throw skilled and unskilled players together when the queues dry up....thus lowering the wait times for us suckers that are left.

unfortunately, this is seeming more and more true

[Hauser]

DeaconW, on 12 August 2013 - 10:06 AM, said:

Good post...I agree with your criteria except that online processing isn't necessary and would, in most cases, be suboptimal to can offline algorithm IMO. ..just wait to do the math when there are 24. The actual calculation time for a two variable optimization problem on 24 tasks would surely still be in the ms time domain.

It's an online algorithm in the sense that you if you wait longer you may produce better results.

For example you'll be able to optimize much better on minimal elo deviation and weight requirements if you wait until there are 500 players waiting then if you grab the first 24.

[The6047]

I found it odd that most games I lost there would be me and 1-2 others with good scores and the rest seemed brain dead. I actually started screenshotting these losses with very odd scores. 5+ ppl <100 dmg. Am I really that bad to be in their bracket? I consider myself to be a very skilled player. I am at work but can post screens later if desired, I'm sure most don't care. I have only ever dropped solo 100% of my ~ 1k matches. I have no friends that play and dont know where to start when looking for a clan. Any takers? I can/will prove my worth.

[DeaconW]

The6047, on 12 August 2013 - 01:37 PM, said:

I found it odd that most games I lost there would be me and 1-2 others with good scores and the rest seemed brain dead. I actually started screenshotting these losses with very odd scores.

Same story here, brother.

[Agent 0 Fortune]

Ronan, on 12 August 2013 - 05:52 AM, said:

Yup, I'm corrected. ELO gets averaged for a pre-made team. You can, as you mention, still get stuffed on a pug to fill out the ranks -- and you may be on a team then with an average ELO much higher or lower than yours. Either way can be fun.

The matchmaker always places pre-made groups in the match first. So if I am added to a team where our pre-made group is of lower skill, it tells me I have been added to increase their Elo count to the minimum allowable threshold. Which mean we will be facing a pre-made group that is significantly better than the preamade group on my team.

I usually feel like I can predict the match outcome within 30 seconds of drop (but I have been surprised, usually when our premade team is a wolfpack because it is hard to judge their skill level based on opening movement).

Another trick: because the matchmaker addess premade groups to the match first, if you join a match and the other team is completely full (or maybe has 8 slots filled), and the matchmaker is slowly filling in your team, you are in for a rough ride.