Elo Threshold Adjustment - Poll
#61
Posted 24 January 2014 - 07:13 AM
Is there an actual point to this poll? This reminds me of all the "Should 3PV be in the game" polls that had the community's answer as a resounding "NO!", yet here we are... with 3rd-person view...
#62
Posted 24 January 2014 - 07:30 AM
I'll update whenever I can gather a 12 man.
#63
Posted 24 January 2014 - 07:42 AM
#64
Posted 24 January 2014 - 07:42 AM
Skyfaller, on 23 January 2014 - 02:19 PM, said:
Its USELESS to match people on win:loss ratio. As a scout I can tag/spot/uav every enemy mech and keep them lit on the map plus do 400 damage and kill 3 of them so I'm doing my job skilfully. Yet if the rest of my team is full of lemmings and they all die without scoring a single kill then tell me, WHY is that hurting *MY* matchmaker rating?
Does anyone know how PGI does their Elo calculation? Is if just win loss? Or is it something that is a more nuanced calculation that takes into account things like average Match Score, and expectation of what is being accomplished by a general class of mech? I have not been able to find it myself and would appreciate it if you linked it to me. Thanks.
#65
Posted 24 January 2014 - 07:43 AM
I can already tell from the responses that no one here is actually busting out a stop watch to check.
#67
Posted 24 January 2014 - 07:45 AM
#68
Posted 24 January 2014 - 07:45 AM
#69
Posted 24 January 2014 - 07:55 AM
AdamBaines, on 24 January 2014 - 07:42 AM, said:
Does anyone know how PGI does their Elo calculation? Is if just win loss? Or is it something that is a more nuanced calculation that takes into account things like average Match Score, and expectation of what is being accomplished by a general class of mech? I have not been able to find it myself and would appreciate it if you linked it to me. Thanks.
http://mwomercs.com/...65#entry1626065
#70
Posted 24 January 2014 - 08:50 AM
Then you have a matchmaker which can say: Ok, here is player x with elo 2000 he is actually playing mech xy and with his skill, he is able to get this mech fully upgraded in 100 games with no need to spend cash. Then the devs say. NO thats not in our interest he will get in 8 out of 10 matches in groups where 10 of them will be beginners and in 5 out of 10 matches hes enemy team will have also and tonnage advantage of serveral 100tons. Then he (the player) will get frustrated and start to spend money cause even his skill doesnt help him to make bigger progression in grinding cause no one can compare a hole team alone. Their is NO F2P game which doesnt do exactly the same foolment to their players. WoT, Warthunder and so MWO all work with the same mechanics. And these mechanics are NOT here to give the players maximum fun to play the game its only here to maximize the profit for the devs. Elo and all this sort of "classifie a player with his skill" works fine in games where its 1vs1. In games with more then one vs one it actually works against the player(s) and for the devs.
#71
Posted 24 January 2014 - 09:34 AM
#72
Posted 24 January 2014 - 09:43 AM
I mean, seriously. For the last few days (actually, weeks, but who's counting), it has been SERIOUSLY out of whack.
You may or may not have read about the matches where one team has nothing but 4 heavies and the rest are lights...and the other team has 6 D-DCs, etc....but they DO happen. With frightening regularity.
I don't know what magic 8-ball or how many 20 sided dice get rolled to determine someone's "ELO" rating (ELO-Extremely Laughable Outcome?) but one thing's for certain...you're not using the simplest piece of data you already have at your fingertips. Total GXP earned. I know it's possible but it's highly unlikely that ANYONE with 1,000,000 GXP earned is going to be playing at a 'beginners level' anymore.
Think about this real quick...the Devs have already said outright that their primary focus on the 'new player' to MWO and not the established community. Cool. Whatever.
Do you REALLY want these new players to get stuck with us crotchety old guys that have been playing the MW title forever?
New Player: "Um...I'm new and don't really understand how to play. Can someone tell me how I move forward?"
Scenario 1- Old Player frustrated about being stuck with new players: "sure, the default is Alt+F4. After that, uninstall the game and go back to playing My Little Pony. Noob."
Scenario 2- Another New Player: "The default is W but you can set it to whatever you need. You really should do the movement tutorial before you play live. Good luck."
#73
Posted 24 January 2014 - 09:51 AM
It's fully reasonable to expect an assault to have a higher probability to win over a light/medium/heavy mech. Especially with the prevelance of Misery/733C/733P/DDC/5M/5S/DS/VTR9B etc. Because of this, the average ELO of all player's assault weight class is slightly biased to be higher than the other weight classes. I can provide a zero-hypothesis proof if need be under the above assumption, but this is true.
The end result is that good assault players will be thrown into 900ish ton drops, and good medium/light pilots will be thrown in with less good assault pilots.
Solution:
Assuming the central limit theorem applies to the ELO distribution, we should rescale ELO values in each weight class to be N(0,1) distributed (N(m,s) is the Normal (or Gaussian) distribution with mean 'm' and standard deviation 's'). This way a good assault pilot is equally likely to be matched with a good pilot of any other weight class.
Assuming each weight class can be considered to be normally distributed (thanks to central limit theorem) as N_x(m_x, s_x) where subscript 'x' is weight class. Let each player's ELO for the weight class they are currently dropping in be: E_i, where subscript 'i' is the player. Calculate a matchmaking ELO, M_i, by normalizing E_i as this:
M_i = (E_i - m_ix) / s_ix
where subscript 'ix' is used to denote the weight class 'x' that pilot 'i' is dropping with. Then M_i will have a normal distribution, N(0,1).
Using the matchmaker ELO (M_i) instead of the player ELO (E_i) should help the matchmaker make better balanced teams on all ELO ranges with better weight class variety in the drops without introducing tonnage limits (yet). It is also a very simple change to make. I'm sure you already have the statistics on each ELO weight class, if not, they are easy to calculate (wikipedia will tell you how, you should use the population mean/standard deviation, not sample mean/standard deviation as you have the full population).
/Li
Edited by Li Song, 24 January 2014 - 09:55 AM.
#74
Posted 24 January 2014 - 09:57 AM
I was in a decent place before the early Jan patch, where my random pug teams generally consisted of similar players, with a light sprinkling of noobs but after the patch that loosened it up, my teams turned to half trial-mechs overnight and match quality plummeted sharply. Sure wait times were under a minute (often less than 30sec) but the matches is was making were very poor, and games not much fun.
Whatever the adjustments that were made on the 23rd were, seemed a much more reasonable selection criteria, at least for me. There were still plenty of wafflestomps, but even those seemed so much better fought. Losing under those conditions is much less frusty, and winning that much sweeter.
#75
Posted 24 January 2014 - 09:59 AM
#76
Posted 24 January 2014 - 10:28 AM
Denolven, on 24 January 2014 - 05:44 AM, said:
Wouldn't it make more sense to make the spread depend on the distance from the average, instead of just using the same range everywhere? Actually, the best would be to make it depend on the amount of people in the range, but depending on math skill it might be harder to implement. The "distance from average" attempt is easier and probably a good enough estimation, as long as the standard deviation of the curve doesn't change heavily.
In the middle of the gauss curve we can have a small range, because there are still enough people in this range to always find enough matches. Then as we get more to the outsides and the amount of people there becomes lower, the range would become larger to compensate.
The classical "one size does NOT fit all"...
EDIT: In case I missed something and the range is dynamic already (growing when matchmaking waits too long), then all that needs to be changed is the function that handles the range increment per time. We could have a really really small range as start, as long as the way the range gets enlarged is handled appropriately. It works pretty much like a logarithmic zoom function (used in zoomable user interfaces, there are scientific papers about this)
I like this idea, with one additional change: bias the range away from the center of the X-axis. So, for players with above average scores, the system will prefer to match against players with even higher ELO, while for those with below average scores (me) the system will prefer opponents with even lower ELO. This would have a flattening effect on the curve over time, driving players away from the edges and center towards the middle-low and middle-high sections.
Edited by lowe0, 24 January 2014 - 10:30 AM.
#77
Posted 24 January 2014 - 10:29 AM
#78
Posted 24 January 2014 - 11:56 AM
Can I just have mine reset so I can play the game solo again?
#79
Posted 24 January 2014 - 12:04 PM
Can you look at matching to a range instead of high/low to target? By matching high/low to target you effectively hang rookies off of every veteran to carry in many matches. If the vet isn't in his top mech, dropping with his friends on VOIP he already isn't carrying his max weight in Elo to begin with.
A better match experience might be just matching to a range. This may also give more wiggle room to weight-matching, at least for mid-range matches. For a game at ~2000 Elo for example just weight-match players as close to 2,000 on both teams as possible. The critical piece is that you leave one or two slots on each team open for those outlier high Elo players who don't have full weight balanced matches available at their range. Try to match one or two on each team but still, one high-Elo player has less impact on the results of a 12v12 match than 3 low Elo players.
Even if the variance between total Elo on both teams is 200, 300, even 500 points so long as it's just a byproduct of 20 or 50 points variance between 10 players it's not as significant as having 2 or 3 or 4 players with scores 100, 200, 300 points below the other teams average.
It might give a better experience and result in more balanced actual matches.
#80
Posted 24 January 2014 - 12:07 PM
Does PGI not track this data internally?
Is this why when we ask questions relating to weapon use statistics and the like, it is never produced?
3 user(s) are reading this topic
0 members, 3 guests, 0 anonymous users