277 replies to this topic

[JPsi]

Grits N Gravy, on 18 February 2013 - 07:07 PM, said:

False, Elo hell or the point where elo scores stagnate or the effort to raise your elo take tremendous effort and very long consecutive win streaks do exist as a function of the mathematical formulas used to create elo scores. The primary culprits are K factors and systems that clip point gains when a more skilled opponent beats a less skilled one. As is the case in our Elo formula.

It got so bad in LOL that they changed the function of their leader boards. Previously your rank on the leader board was solely based on your Elo score. They have radically changed this recently allowing players to advance up the leader boards via tournament style play. http://na.leagueofle...d.php?t=3004520

Just going to respond to that part.
A) The Elo hell thing has been debated back and forth all over those forums, some believe it some don't. It still falls under the 'Myth" category.
B )Their reasons for moving from Elo as stated, was that the Elo score felt near meaningless to the individual player. Moving up a few ranks when you are ranked against 20000+ other players didn't feel meaningful.

You've asserted that they moved away from Elo due to the Elo Hell scenario. I've yet to see where thats been said by any Riot employee. Also just another little note, they currently still have Elo, even with the leagues, its still used for their matchmaking, just not directly for their rankings. It is now hidden.
It is also used indirectly for rankings, the reduced LP earnings at about the 90 point range are influenced in severity and duration by the players Elo Score. Can go find the relevant dev posts if you'd like, just too lazy to do so now.

Edited by JPsi, 18 February 2013 - 08:27 PM.

[Mackman]

Grits N Gravy, on 18 February 2013 - 07:07 PM, said:

IMHO all the time that goes into an ELO matchmaking system is a waste, because the laws of probability get us to about the same point and Elo can't get us to a better point but could make things worse. The risk to reward doesn't justify the expenditure of time given the population size. The time would have been better spent doing a Battle value system, to ensure better mech parity in matches, creating a dynamic leader board and tournament system.

How the hell could it make things worse than a purely random distribution? How? Even if it literally just divides the player base in half, it would give us better matches than we have right now.

JPsi, on 18 February 2013 - 08:23 PM, said:

Just going to respond to that part.
A) The Elo hell thing has been debated back and forth all over those forums, some believe it some don't. It still falls under the 'Myth" category.
B )Their reasons for moving from Elo as stated, was that the Elo score felt near meaningless to the individual player. Moving up a few ranks when you are ranked against 20000+ other players didn't feel meaningful.

You've asserted that they moved away from Elo due to the Elo Hell scenario. I've yet to see where thats been said by any Riot employee. Also just another little note, they currently still have Elo, even with the leagues, its still used for their matchmaking, just not for their rankings. As such its now hidden.

Also this.

[Walk]

80Bit, on 16 February 2013 - 10:06 PM, said:

I think you are in the same boat as me. I am no elite, but I am at least above average based on my typical match score relative to my team. But I still only have a 41% win rate with the current system. Like I predicted above, things should get better for us, with fewer pre-made enemies and a much more even playing field, letting us get to 50% win rate.

Great question on the Dev hunt, did not even think of that.

Here's the thing though, with a 41% win rate, you really aren't very good. MWO is a team game, and if your win-rate is that low, it means you don't positively influence your team enough to help tip the game to your advantage, and in fact, you are a detriment to your team. As someone else who also only pub drops, my win rate is 59%. Not the best, but thats PUGS. However, my win rate means I am a huge influence for my team, and definitely a good positive influence. See, the problem is you think you are better because you are at the top of the scoreboard. That doesn't mean anything in MWO, you can easily top the scoreboard by staying back for 4 minutes then skirting around the map taking random potshots at enemies with a light mech. Your damage/kills don't mean jack squat if it was in the wrong place at the wrong time, and didn't help your team win. You can blame your team all you want, but the only common factor in all your games is YOU. Also, http://en.wikipedia....93Kruger_effect

[Targetloc]

My only real concern with ELO is how ridiculously fast people are going to die at the top.

It really bums me out when the other team knows how to aim. Stick your head up for 2 seconds to peep what they're up to and you're guaranteed to take a minimum of 30 damage.

[WVAnonymous]

Mackman, on 18 February 2013 - 05:02 PM, said:

I think you're misunderstanding Elo. If your team's Elo (which is likely going to be the average of all the individual Elo's) is higher than the other team's Elo, and you lose, then your personal Elo goes down: Likewise, if your team's Elo is lower then that of the other team, and you win, than your personal Elo goes up.

I'm not sure if the same will hold true for the opposite scenarios: If you win against a team you're "supposed" to win against (if your Elo is already higher than the team you beat), does your Elo still go up and does their Elo still go down? I don't know: In League of Legends, it goes up less if you were supposed to win (and down less if you were supposed to lose), and vise versa.

There's no "check w/l after X amount of matches to see if it's 50%." The 50% is the end result, not the mechanism by which it achieves the result.

I was unclear, so let me expand the statement a bit. "The end result will be indistinguishable from an algorithm that adjusts your rank up or down based on the trailing W/L ratio centered on 50%".

Adjusting rankings up or down based on keeping a target value of 50% is how 50% is the end result; it's a feedback loop.

[Wintersdark]

Mackman, on 18 February 2013 - 05:02 PM, said:

I think you're misunderstanding Elo. If your team's Elo (which is likely going to be the average of all the individual Elo's) is higher than the other team's Elo, and you lose, then your personal Elo goes down: Likewise, if your team's Elo is lower then that of the other team, and you win, than your personal Elo goes up.

I'm not sure if the same will hold true for the opposite scenarios: If you win against a team you're "supposed" to win against (if your Elo is already higher than the team you beat), does your Elo still go up and does their Elo still go down? I don't know: In League of Legends, it goes up less if you were supposed to win (and down less if you were supposed to lose), and vise versa.

There's no "check w/l after X amount of matches to see if it's 50%." The 50% is the end result, not the mechanism by which it achieves the result.

This is correct.

There's a Command Chair post about how it works here: http://mwomercs.com/...ost__p__1626065

[Vassago Rain]

Walk, on 18 February 2013 - 08:28 PM, said:

Here's the thing though, with a 41% win rate, you really aren't very good. MWO is a team game, and if your win-rate is that low, it means you don't positively influence your team enough to help tip the game to your advantage, and in fact, you are a detriment to your team. As someone else who also only pub drops, my win rate is 59%. Not the best, but thats PUGS. However, my win rate means I am a huge influence for my team, and definitely a good positive influence. See, the problem is you think you are better because you are at the top of the scoreboard. That doesn't mean anything in MWO, you can easily top the scoreboard by staying back for 4 minutes then skirting around the map taking random potshots at enemies with a light mech. Your damage/kills don't mean jack squat if it was in the wrong place at the wrong time, and didn't help your team win. You can blame your team all you want, but the only common factor in all your games is YOU. Also, http://en.wikipedia....93Kruger_effect

I'm sitting on a 62% win rate right now. About 8/10ths of all my games are solo griding in the ghetto, and I rarely survive matches.

Survival, however, isn't required to win. Just that you perform.

[p4r4g0n]

Walk, on 18 February 2013 - 08:28 PM, said:

Here's the thing though, with a 41% win rate, you really aren't very good. MWO is a team game, and if your win-rate is that low, it means you don't positively influence your team enough to help tip the game to your advantage, and in fact, you are a detriment to your team. As someone else who also only pub drops, my win rate is 59%. Not the best, but thats PUGS. However, my win rate means I am a huge influence for my team, and definitely a good positive influence. See, the problem is you think you are better because you are at the top of the scoreboard. That doesn't mean anything in MWO, you can easily top the scoreboard by staying back for 4 minutes then skirting around the map taking random potshots at enemies with a light mech. Your damage/kills don't mean jack squat if it was in the wrong place at the wrong time, and didn't help your team win. You can blame your team all you want, but the only common factor in all your games is YOU. Also, http://en.wikipedia....93Kruger_effect

Before you guys get into an epeen contest, could you confirm that you're both calculating win rate % as wins / (win + loss) X 100? I've seen too many people get this wrong.

Edited by p4r4g0n, 18 February 2013 - 08:46 PM.

[Mackman]

Walk, on 18 February 2013 - 08:28 PM, said:

Here's the thing though, with a 41% win rate, you really aren't very good. MWO is a team game, and if your win-rate is that low, it means you don't positively influence your team enough to help tip the game to your advantage, and in fact, you are a detriment to your team. As someone else who also only pub drops, my win rate is 59%. Not the best, but thats PUGS. However, my win rate means I am a huge influence for my team, and definitely a good positive influence. See, the problem is you think you are better because you are at the top of the scoreboard. That doesn't mean anything in MWO, you can easily top the scoreboard by staying back for 4 minutes then skirting around the map taking random potshots at enemies with a light mech. Your damage/kills don't mean jack squat if it was in the wrong place at the wrong time, and didn't help your team win. You can blame your team all you want, but the only common factor in all your games is YOU. Also, http://en.wikipedia....93Kruger_effect

Win rate is deceptive, because I, for one, absolutely tanked my W/L in my first holy crap way too many matches. I've been sloooowly inching my way back to 1-1 ever since I found some mechs that I can actually perform with. Just saying that W/L is probably going to be way off with a lot of ppl who only started playing since OB, especially given the complete lack of matchmaking and tutorials.

[Broceratops]

Targetloc, on 18 February 2013 - 08:30 PM, said:

My only real concern with ELO is how ridiculously fast people are going to die at the top.

It really bums me out when the other team knows how to aim. Stick your head up for 2 seconds to peep what they're up to and you're guaranteed to take a minimum of 30 damage.

this makes scouting important, which allows for good positioning, which allows for the correct timing of your pushes. it makes games very strategical and climatic compared to pug games where scouting is totally pointless and therefore you get none of that other stuff

[Vlad Ward]

An ideal ELO system isn't going to match you with a perfectly balanced team every team. In actuality, it ought to be matching you against teams that are some degree weaker as well as teams that are some degree stronger, with a tolerance set by the Devs.

This is what allows your ELO to change. Facing weaker teams gives those teams the opportunity to beat you and raise their ELO score (and lower your own). Conversely, facing stronger teams gives you the opportunity to beat them and raise your own ELO score (and lower theirs).

I actually just checked my own W/L and was kinda shocked to see it was up at 66%. I mostly pug and 8-man, and I have a tendency to die a whole bunch.

This will keep your W/L from staying too static. Though, the only time you'll really see super-inflated W/L ratios ought to be at the extreme ends of the ELO scale. Extremely weak players will hit the ELO minimum and may still not be able to handle their peers, resulting in a deflated W/L. Likewise, extremely strong players will hit the ELO maximum and be put up against equally rated players who may or may not actually be as skilled, resulting in an inflated W/L.

[Asmosis]

Docshifty, on 17 February 2013 - 08:36 PM, said:

And your Spider that really shouldn't be trying to damage everybody?

100 damage in a 5D is not the same thing as 100 damage in a D-DC.

which is completely different from 100 damage in a dual gauss catapult, that was my point. damage on its own isnt meaningful at all.

Walk, on 18 February 2013 - 08:28 PM, said:

Here's the thing though, with a 41% win rate, you really aren't very good. blah

Not really, all it means is you arent dropping in premades in the current system.

Edited by Asmosis, 18 February 2013 - 09:22 PM.

[Roadkill]

Vlad Ward, on 16 February 2013 - 10:34 PM, said:

the only way to reset your ELO is to make a new account that you can't transfer anything to/from.

Not really. It's trivially easy to manipulate an Elo ranking, especially in a divided environment like we have. Normally play in a group using TS? Want to reset your Elo? Just go pug for 10 games. You'll probably lose 8 of those 10 and your Elo ranking will drop like a rock.

Now PGI did say that they're going to maintain several different rankings for each player based on different circumstances, so this might not work quite as easily as that. But no matter what they're monitoring, it's easy to "reset" an Elo ranking by just deliberately losing a few matches in a row.

Of course since they've also said that you won't be able to see your Elo ranking, I don't see the point in resetting it. It's not like CW or any of the other tournaments that people run are going to be Elo based, so deliberately tanking isn't going to help anything except your next few non-competition matches.

[80Bit]

Walk, on 18 February 2013 - 08:28 PM, said:

Here's the thing though, with a 41% win rate, you really aren't very good. MWO is a team game, and if your win-rate is that low, it means you don't positively influence your team enough to help tip the game to your advantage, and in fact, you are a detriment to your team. As someone else who also only pub drops, my win rate is 59%. Not the best, but thats PUGS. However, my win rate means I am a huge influence for my team, and definitely a good positive influence. See, the problem is you think you are better because you are at the top of the scoreboard. That doesn't mean anything in MWO, you can easily top the scoreboard by staying back for 4 minutes then skirting around the map taking random potshots at enemies with a light mech. Your damage/kills don't mean jack squat if it was in the wrong place at the wrong time, and didn't help your team win. You can blame your team all you want, but the only common factor in all your games is YOU. Also, http://en.wikipedia....93Kruger_effect

I missed the part where I blamed anything on my team mates...

You seem to have missed the part where I pointed out how the current match making system causes near 50% percentile players to have sub 50% win rates, by virtue of the fact they will be pitted against pre-made teams 66% of the time (when a single pre-made is present, which is frequently).

As Mackman pointed out, your total win/loss rate can be miss leading. I have played a large number of matches in intentionally sub-par or gimped builds as part of research for a mech review guide I am writing. But I also collect lots of stats to analyze too, so I know that when I am "trying" my win rate is still in the mid 40% for assault matches, and the high 40% for conquest.

Your opinion that win rate directly reflects a person ability to help their team win is flawed in this situation, since a single player can do little to counter the overwhelming advantage a 4 man coordinated team has.


Whether I suck or not has no bearing on the fact that the current matchmaking system does disconnect win rate from skill because of the pre-made dynamic. I don't really give a flip if I am top 70% or bottom 30%, I look forward to the Elo putting me where I belong where I can play in fair matches.

Edited by 80Bit, 18 February 2013 - 09:27 PM.

[JPsi]

Possibly another thing to note. The data collected over the last few weeks may not be enough to immediately get it all right. Players may still run into very inaccurate match ups early on.

So with that in mind, don't expect an immediate improvement (hopefully there will be one anyway). Complaining if it hasn't worked in the first day of implementation is probably not constructive.

[Grits N Gravy]

Mackman, on 18 February 2013 - 08:26 PM, said:

How the hell could it make things worse than a purely random distribution? How? Even if it literally just divides the player base in half, it would give us better matches than we have right now.

It has the possibility of making matchmaking worse than random, because of Elo score distortion as a result of the effect of grouping. Not only those who group but also those who do not. Your not going to be seeing matches of people with the same Elo scores. You're going to get matches within 2 standard deviations of your score. The effect of grouping destroys the validity of the standard deviation and hence destroys the accuracy of the predictive properties of Elo scores. When you rely on standard probability for matchmaking, over time the skill distribution of your and opponents team's will normalize. In an Elo system you can't make that prediction give population size and a distorted standard deviation.

JPsi, on 18 February 2013 - 08:23 PM, said:

Just going to respond to that part.
A) The Elo hell thing has been debated back and forth all over those forums, some believe it some don't. It still falls under the 'Myth" category.
B )Their reasons for moving from Elo as stated, was that the Elo score felt near meaningless to the individual player. Moving up a few ranks when you are ranked against 20000+ other players didn't feel meaningful.

You've asserted that they moved away from Elo due to the Elo Hell scenario. I've yet to see where thats been said by any Riot employee. Also just another little note, they currently still have Elo, even with the leagues, its still used for their matchmaking, just not directly for their rankings. It is now hidden.
It is also used indirectly for rankings, the reduced LP earnings at about the 90 point range are influenced in severity and duration by the players Elo Score. Can go find the relevant dev posts if you'd like, just too lazy to do so now.

And they've had to adjust the LP earnings, because people were in effect in Elo hell. Players not getting enough LP from a win and losses costing a disproportional amount of LP relative to a win.
http://na.leagueofle...31#post34184931

80Bit, on 18 February 2013 - 08:07 PM, said:

Since a lot of what you say is based off of our "low population", can you let us know what that low population number is, and where you are getting it?

It was around 5k before they took the info out of the game. You need at least 20,000k players to have a sample size of 2000 for a decent que. 2000 is considered the minimal pool size to have an acceptable margin of error in polling. I doubt it is at 20,000 now. In reality neither of has exact figures as to the population size. Though the fact of the matter is Elo systems have minimal population thresholds necessary for them to work. From anecdotal experience I don't think MWO has the population to support a Elo system that will work well.

80Bit, on 18 February 2013 - 08:07 PM, said:

-snip- for brevity
solo player has a 4 in 12 (33.3%) chance to be placed on the side with the pre-made group, and an 8 in 12 (66%) chance to be placed on the side opposing the pre-made group. Pre-made groups win the vast majority of matches they play, and thus, solo players of almost all skill levels are losing more than 50% of their matches. The current matchmaking fails to deliver fair matches to all players.


So to summarize the current state of MWO match making is not matching the laws of probability you are talking about. It is my belief that the Elo structure will indeed improve match quality by helping correct the pre-made placement problem in a round about way.

You over state the effect of grouping on win rate of a pure pug dropper. A pug player doesn't only play matches against and with premades. His chances of facing a premade are relative to the number of players in the que and what percentage of those players are in a 4 person premade. Which is probably at most 15 percent. If you noticed in my OP I stated that win rate for an average player should be 48 +-5%. 5% more than accounts for the effect of facing pugs have, as realistically 5% of a que at a given time is a 4 premade.

I pretty much only pug, or run with 1 other person. My win rate is 75% and I don't own an ECM mech, k2 or atlas. The only evidence to support your claim that the match making fails to deliver is as anecdotal as my win rate. Win rates won't always reflect potions of damage boards. As playing for wins is different than playing for damage and kills. A more accurate statement is that Matchmaking does not always deliver fair matches to all players. Going to an Elo system won't make it any better because of population size, and a distorted standard deviation of Elo scores. It's a no win, Elo scores in > 6v6 games have proven a low predictive accuracy.

I argue the time would have been better spent doing a BV system to insure better mech parity, ie reducing play advantage based on mech type, and coming up with a dynamic leader board system where people could stat brag. Because this time would have more effect on player retention and draw than an Elo system which only accurately predicts wins 40% of the time.
http://research.micr...t.aspx?id=67956

Edited by Grits N Gravy, 18 February 2013 - 11:13 PM.

[JPsi]

Grits N Gravy, on 18 February 2013 - 11:04 PM, said:

And they've had to adjust the LP earnings, because people were in effect in Elo hell. Players not getting enough LP from a win and losses costing a disproportional amount of LP relative to a win.
http://na.leagueofle...31#post34184931

That wasn't an Elo hell. That was a direct result of them being overly cautious with their new system.
"This check was tuned too aggressively and was causing the reduced LP gains that many players were seeing."
Not a bug, not an Elo hell. The Elo in the background was still working fine. Merely a new system going through its paces and getting tested.

On your point about grouping, it gets hard to tell. Details on how they are handling it are not exactly common knowledge. It's a tangled mess dealing with grouping, but adjustments can be made for it.

Again the same thing with populations, unfortunately all we have is anecdotal evidence on numbers. It's too hard to say on which side the coin will fall.

My take on it is this: It isn't going to be perfect, not every match will be fair. Overall we will probably see an improvement.

Edited by JPsi, 18 February 2013 - 11:50 PM.

[Grits N Gravy]

Mackman, on 18 February 2013 - 05:02 PM, said:

I'm not sure if the same will hold true for the opposite scenarios: If you win against a team you're "supposed" to win against (if your Elo is already higher than the team you beat), does your Elo still go up and does their Elo still go down? I don't know: In League of Legends, it goes up less if you were supposed to win (and down less if you were supposed to lose), and vise versa.

There's no "check w/l after X amount of matches to see if it's 50%." The 50% is the end result, not the mechanism by which it achieves the result.

In MWO Elo scores go up and down based on the probability of victory as predicted by elo scores with a maxium loss and gain of 50 points per match.

Quote

Old Rating = 1350
Maximum Change Allowed = +50 for a win, -50 for a loss (as seen in Figure 1)
WinFlag = 1
Probability of Winning = 0.41

1350 Players new ranking = 1350 + 50 x (1 – 0.41)
= 1380

1410 Players new ranking = 1410 – 50 x (1 – 0.41)
= 1381

A player’s rating will only go down if they are beaten by a player who has a lower rating than theirs. In this case, if the 1350 player lost, their score would not change since the Match Maker was correct in its prediction.

https://mwomercs.com...79-matchmaking/

Inverse the results and you can see that the win only grants the higher ranked player 20.5 points in that match win. When facing lower ranked players you always risk more than u can gain. Thus for every loss to a lower ranked opponent, you have to win a number of consecutive games that is propitiation to the predicted margin of accuracy of the Elo system. In order to get your Elo score to go back up to it's pre loss level.

It's much easier to drop a score than to raise to it. In order to raise it, u need to either beat a lot of lower ranked players consecutive, a few equally ranked players consecutively or couple of higher ranked players. Depending where you are on the bell curve of skill and population size, it may be hard to see higher ranked players. Thus a low risk strategy of getting your score up (facing higher ranked players, you don't risk a drop in score when facing high ranks) may not be feasible. Thus a bad night of play is not evened out by a good night, this is what I refer to as Elo hell. For every bad night you need 1.5 to 2 nights of good play to come out even. This the direct result of the Elo formula and K factors and is one of the fundamental reasons why Elo system can feel unfair.

Edited by Grits N Gravy, 19 February 2013 - 12:42 AM.

[MasterofBlasters]

I'll never be able to do this again...

[Dragonkindred]

What has the Electric Light Orchestra got to do with MWO???