Is Mm Working?
#41
Posted 02 July 2014 - 12:58 PM
If you are solo, and it seems like you were matched against a 5-12 man group, report it. It's a bug.
#42
Posted 02 July 2014 - 01:02 PM
Gas Guzzler, on 02 July 2014 - 12:29 PM, said:
cause 1 man can make a difference against 12 when you have people in trials on your team or never played b4
#43
Posted 02 July 2014 - 01:18 PM
Karamarka, on 02 July 2014 - 01:02 PM, said:
Crap happens, I can say that is definitely not the norm for me. And if you are landing in that regularly, it might be you.
#44
Posted 02 July 2014 - 01:25 PM
stjobe, on 02 July 2014 - 11:43 AM, said:
Broken system is broken.
A 100point /player average variance in Elo score is practically negligible and generally a fair match.
12 players /team is 1200 pts of variance/team. That can logically go up, down, or balance in the middle, as would be absolutely normal for any automated system with preprogrammed degrees of variation from it`S ideal fuunctioning state.
So your point is what exactly?
Edited by Zerberus, 02 July 2014 - 01:27 PM.
#47
#48
Posted 02 July 2014 - 01:56 PM
that will give you a good indication if it is working or not. :-))
Also, I love (sarcasm) that I live in a culture where trying to do well is an insult. :-). But that's the product of mass standardized testing culture. We praise the guy who scores in the top 0.01 percent of standardized tests (a trat they owe entirely to an accident of birth) but insult the people who who put effort into things. "How dare you make my natural skill look worse with your effort." (P.S. I am the guy in the 0.01 who is the proof of how meaningless it is :-)).
It's like insulting the guys who play competitive sports for fun just because you're a rec league player :-p.
#49
Posted 02 July 2014 - 02:02 PM
Variance is used for statistical analysis the formula for discrete (what we have) set of variables is given by the formula:
Var = 1/n*sum(x_i - u)^2
u = the average
n = is the count (12)
x_i = current players elo
So say for an example of a 12 man team where everyone is within 100 points of 1000 elo:
900, 920, 940, 960, 980, 1000, 1000, 1020, 1040, 1060, 1080, 1100
The total difference from 1000 = 600
The total variance from 1000 = 3666
Its the averaged square distance.
EDIT: Wiki formulas don't paste well (http://en.wikipedia.org/wiki/Variance)
Edited by StandingInFire, 02 July 2014 - 02:06 PM.
#50
Posted 02 July 2014 - 02:33 PM
Zerberus, on 02 July 2014 - 01:25 PM, said:
A 100point /player average variance in Elo score is practically negligible and generally a fair match.
12 players /team is 1200 pts of variance/team. That can logically go up, down, or balance in the middle, as would be absolutely normal for any automated system with preprogrammed degrees of variation from it`S ideal fuunctioning state.
So your point is what exactly?
StandingInFire, on 02 July 2014 - 02:02 PM, said:
Variance is used for statistical analysis the formula for discrete (what we have) set of variables is given by the formula:
Var = 1/n*sum(x_i - u)^2
u = the average
n = is the count (12)
x_i = current players elo
So say for an example of a 12 man team where everyone is within 100 points of 1000 elo:
900, 920, 940, 960, 980, 1000, 1000, 1020, 1040, 1060, 1080, 1100
The total difference from 1000 = 600
The total variance from 1000 = 3666
Its the averaged square distance.
EDIT: Wiki formulas don't paste well (http://en.wikipedia.org/wiki/Variance)
You're both right, of course. It's painfully obvious (and more than a little embarrassing) that the 20 years since my last statistics classes is too long, and that the MM is indeed kicking off quite well-formed teams.
Thanks both of you for setting me straight.
#51
Posted 02 July 2014 - 02:46 PM
Ghost Badger, on 02 July 2014 - 12:21 PM, said:
Just out of curiosity, not to shake your tree, but what is it SUPPOSED to match you based on?
Certainly not weight class. 3/3/3/3 is NOT working.
It doesn't match you based on any form of skill separation....there are as many "elite" players on any given team as there are people in their first 25 matches.
So, what exactly is the "Matchmaker" supposed to be doing? Frankly, I don't see a difference between today and yesterday.
#52
Posted 02 July 2014 - 02:51 PM
I've had 2 hours worth of drops... while not statistically good enough... I have observed something that was clear and obvious to me.
1) Last 2 drops were in a premade, so let's exclude that data for a moment.
According to the MM, it was "consistently" reporting that Lights were needed. ETA to getting a match to in game is under 30 sec (generally under 15 sec). This is excellent. I went Lights for all solo matches.
HOWEVER, the quality of play was bad. I watched derp team after derp team after derp team. This was consistent on both sides. What I believe is happening is that the MM is "starved" for mechs is needs... and while it expands more time, the Elo buckets "completely drop", so it'll end up taking "anyone and everyone" that is soloing with that particular weight class it needs.
In essence, my matches are easier, but also lame, because I see boatloads of "underhive behavior".
The best explanation for this is that you think of it more like MM v2... 3/3/3/3 "works", but the variation in skill is MASSIVE when the solo queue is starving.
2)
I've only dropped TWICE with a premade (just 3-man)... 1 ended up in the solo queue (our team on the upper base spawn derped by going J-line towards the enemy, instead of H9 - still won with the derp)... the other ended up in the group queue.
What I saw in the group queue seemed to be a bunch of premades (somewhat verifiable), but it didn't adhere to 3/3/3/3 at all. I saw what amounted to a Steiner Lance (2 D-DCs + 1 Atlas + 1 Daishi). So what's clear to me that big groups tend to fail at the system, relaxing it because it couldn't find a proper weight match to go with it. If a 10-man team went like 3/3/3/1 Light, then it will ATTEMPT to fight a 2-man of Lights.. and lol that's not going to happen. It literally breaks down in the group queue.
This is just initial analysis of the queues and may not be entirely accurate. I'm hoping to drop more later tonight to find out the exact behavior, but this is what I'm seeing more often than not.
Edit:
"Worst case scenario" for the group queue behavior is this:
9-man takes 3/3/3 (no lights), 3-man takes 3 Assaults. The MM has a pretty good shot of paring them, so your worse case scenario is literally 6 of any weight class (so, half a Steiner Scout Company) in some combination.
Edited by Deathlike, 02 July 2014 - 02:55 PM.
#53
Posted 02 July 2014 - 06:17 PM
#54
Posted 03 July 2014 - 06:43 AM
Deathlike, on 02 July 2014 - 02:51 PM, said:
I've had 2 hours worth of drops... while not statistically good enough... I have observed something that was clear and obvious to me.
1) Last 2 drops were in a premade, so let's exclude that data for a moment.
According to the MM, it was "consistently" reporting that Lights were needed. ETA to getting a match to in game is under 30 sec (generally under 15 sec). This is excellent. I went Lights for all solo matches.
I swear, everybody forgets: your ELO is /by chassis/.
Don't ya'll think that, given the new requirements of 3/3/3/3, you're going to have lots of players strapping on chassis that they haven't played? Given that ELO starts you in the middle of the field by design, don't you think that you're going to get a ton of people, in the middle of a new matchmaker, strapping on chassis they have no clue how to run?
It astounds me, sometimes, how much people use their own experience and assume it's typical. You can run any weight class. Why do you assume everyone else can?
#55
Posted 03 July 2014 - 07:18 AM
Willard Phule, on 02 July 2014 - 02:46 PM, said:
Just out of curiosity, not to shake your tree, but what is it SUPPOSED to match you based on?
Certainly not weight class. 3/3/3/3 is NOT working.
It doesn't match you based on any form of skill separation....there are as many "elite" players on any given team as there are people in their first 25 matches.
So, what exactly is the "Matchmaker" supposed to be doing? Frankly, I don't see a difference between today and yesterday.
From my understanding, it gets you teams that are close in average ELO, though the ELO of individuals on the team may vary wildly. 3/3/3/3 is limited to each group...not necessarily the whole team. Last night we had a single match of 16 with 2 assaults, one with 2 matches with 3 assaults, and the other 11 had 4-6 assaults on one side. That said, there were an equal amount on the OTHER side.
This prevents a 12-man from ever having more than 3 assaults...honestly it seems like the "worst case scenario" for the other side is to drop a 10-man with 3 assaults, and then get paired with a 2-man of 2 assaults. This happened to us in one game. We won 12-0, but the OTHER team had 5 assaults, too (and they were heavier tonnage-wise). They just got outplayed. I do not know, however, what their group composition was.
That's how it's supposed to work (from my understanding) and it was.
Deathlike, on 02 July 2014 - 02:51 PM, said:
Sorry to hear you had derp games. I dropped with 3-10 of the Kell Hounds at a time, and we had a majority of good games last night. 16 drops. There was a variety of mech weights, with several games giving us 6 assaults on each team as the upper limit. To echo the previous statement, seems like 3/3/3/3 only applies to each group, not the entire team. This prevents a premade from stacking more than 3 assaults, but allows for some variety in what the team has as a whole, and matches those tonnage classes on the other side.
It actually gave us fun matches. One 12-0 stomp, one 12-1stomps...but those were in the minority, and sometimes you just get rolled. It happens.
I was pleasantly surprised by last night. Granted, we were dropping as a group. I did NOT solo, nor did we pay ANY attention to the Chassis %. We dropped what we felt like, and the MM got us a game within 60 seconds each time, often under 20 seconds.
Edited by Ghost Badger, 03 July 2014 - 07:22 AM.
#56
Posted 03 July 2014 - 07:26 AM
DanNashe, on 02 July 2014 - 01:56 PM, said:
that will give you a good indication if it is working or not. :-))
Also, I love (sarcasm) that I live in a culture where trying to do well is an insult. :-). But that's the product of mass standardized testing culture. We praise the guy who scores in the top 0.01 percent of standardized tests (a trat they owe entirely to an accident of birth) but insult the people who who put effort into things. "How dare you make my natural skill look worse with your effort." (P.S. I am the guy in the 0.01 who is the proof of how meaningless it is :-)).
It's like insulting the guys who play competitive sports for fun just because you're a rec league player :-p.
I asked about this specifically before. They said that they would be reviewing variance and set difference versus the actual standard deviation as a predictor of match outcome.
#57
Posted 03 July 2014 - 08:09 AM
JonahGrimm, on 03 July 2014 - 06:43 AM, said:
Don't ya'll think that, given the new requirements of 3/3/3/3, you're going to have lots of players strapping on chassis that they haven't played? Given that ELO starts you in the middle of the field by design, don't you think that you're going to get a ton of people, in the middle of a new matchmaker, strapping on chassis they have no clue how to run?
It astounds me, sometimes, how much people use their own experience and assume it's typical. You can run any weight class. Why do you assume everyone else can?
New chassis != not knowing how to target with 'r'. It's very distinctive in differentiating. Yes, I'm familiar with Elo being associated with weight class, not chassis.
Other experiences exist, the subjectivity in some cases are a little hyperbolic (although in other cases need further examination).
Ghost Badger, on 03 July 2014 - 07:18 AM, said:
It actually gave us fun matches. One 12-0 stomp, one 12-1stomps...but those were in the minority, and sometimes you just get rolled. It happens.
I was pleasantly surprised by last night. Granted, we were dropping as a group. I did NOT solo, nor did we pay ANY attention to the Chassis %. We dropped what we felt like, and the MM got us a game within 60 seconds each time, often under 20 seconds.
The derp play is STRICTLY limited to the solo queue. Very minimal problems existed in the big group queue.
Edited by Deathlike, 03 July 2014 - 08:10 AM.
#58
Posted 03 July 2014 - 08:59 AM
Edited by ztac, 03 July 2014 - 09:00 AM.
#59
Posted 03 July 2014 - 12:44 PM
ztac, on 03 July 2014 - 08:59 AM, said:
Actually the OLD MM DID have large premades. Then they removed it...now it's back. Deal with it.
Edited by Ghost Badger, 03 July 2014 - 12:44 PM.
#60
Posted 03 July 2014 - 02:08 PM
StandingInFire, on 02 July 2014 - 02:02 PM, said:
Variance is used for statistical analysis the formula for discrete (what we have) set of variables is given by the formula:
Var = 1/n*sum(x_i - u)^2
u = the average
n = is the count (12)
x_i = current players elo
So say for an example of a 12 man team where everyone is within 100 points of 1000 elo:
900, 920, 940, 960, 980, 1000, 1000, 1020, 1040, 1060, 1080, 1100
The total difference from 1000 = 600
The total variance from 1000 = 3666
Its the averaged square distance.
EDIT: Wiki formulas don't paste well (http://en.wikipedia.org/wiki/Variance)
An interesting observation, the values you used are extremely close and yet the variance number you arrived it is substantially higher than any of PGI's numbers, indicating that either the Matchmaker is extremely focused, and able to match remarkably narrow bands of Elo, or (more likely) we are seeing something else.
Edited by Agent 0 Fortune, 03 July 2014 - 02:09 PM.
6 user(s) are reading this topic
0 members, 6 guests, 0 anonymous users