277 replies to this topic

[Void2258]

Alpha087, on 20 April 2013 - 01:38 PM, said:

Weight balance might be good, but skill balance is far from it from what I've seen... Out of the last dozen or so matches that I've played, more than half of them ended with 0 - 2 kills on one side, and 8 kills on the other.

I'm currently at 273 matches played, so I'm not certain what effect that has on this invisible ELO system, but whatever was changed around has made for a much worse play experience, at least for me. I'm finding myself rage-quitting after multiple bad match-ups more and more now.. As opposed to pre-patch where I could just keep playing for hours on end.

Are you kidding? I have barely seen a single match that is weight balanced, never mind skill balanced. Almost every match is decided the second one team has 3 lights and the other is all heavies and assaults that can't do anything against them flanking. Skill only comes in in the rare match that isn't horribly imbalanced in terms of team compositions.

[AP514]

WE all AGREE...MATCHING is not working as is.....

PLZ FIX IT......

[Znail]

It's a bit amazing that they accidentaly managed to make a really good system like stage 3. It must have been by accident as they else so would they have realised that stage 4 was a bad idea. The stage 4 graph looks like pure static and no relation to actual player skills. One would expect experienced players, maxed out mechs and modules to win more games then total noobs in trial mechs and thus gravitate up in Elo with more games played, like in the stage 3 graph. I have to wonder what they did that so totally broke the Elo system free from player skills? Most Elo systems, even bad ones, would at least show some tendancy for players to get better the more they play, but this one looks totally random static. I guess it takes skill of some kind to fail this badly.

From past experience so would I expect this to take no more then 6-12 months to get fixed. It seems it takes quite a while for them to fix obvious issues, simply because they get into denial as soon as the players complain about something. It simply can't be that players realise right away when something is not working while they failed to do so and thus they stick their heads into the sand.

[Hauser]

Znail, on 08 July 2013 - 08:59 AM, said:

One would expect experienced players, maxed out mechs and modules to win more games then total noobs in trial mechs and thus gravitate up in Elo with more games played, like in the stage 3 graph.

One of the properties of an Elo system is that it determines your skill relative to the population of active players. The light blue section covers players that haven't played much at all. They haven't played enough to be compared to the population at large so their distribution is pretty much a normal distribution.

When you look at the dark blue curve you see that the majority of players that have played more and have been compared against each other. That curve shows a slight progression where the half the players are below average with a slightly longer tail on the right where the good players are.

[PhoenixFire55]

Hauser, on 09 July 2013 - 01:14 AM, said:

When you look at the dark blue curve you see that the majority of players that have played more and have been compared against each other. That curve shows a slight progression where the half the players are below average with a slightly longer tail on the right where the good players are.

Back when I was at school my maths teacher told me that if you make a graph and don't place values on it you are just making a fool of yourself. Those graphs mean absolutely nothing. Light blue and dark blue distributions can be exact same normal distribution in different scales for all we know.

[LoveLost85]

need to get to matchmaking phase 10 or some thing. pretty colored graphs really mean squat. how is it I have hundreds of hours of play time on multiple mastered mechs and I still get grouped with people that say "yayz, I hopez to get mah first kill now" or "this iz mah 3rd match, this game so neat", coupled with the fact that 99% of every single game is a stomp one way or the other. seriously. goes for pugging and 4 man.

[Znail]

Hauser, on 09 July 2013 - 01:14 AM, said:

One of the properties of an Elo system is that it determines your skill relative to the population of active players. The light blue section covers players that haven't played much at all. They haven't played enough to be compared to the population at large so their distribution is pretty much a normal distribution.

When you look at the dark blue curve you see that the majority of players that have played more and have been compared against each other. That curve shows a slight progression where the half the players are below average with a slightly longer tail on the right where the good players are.

Forum ate my post

The short story is that the expected Elo curve looks like the red graph and the blue one does a poor job of trying to look like the red one and thus a poor job of showing the player skill.

The less then 50 matches player group is still a large group of players and they have played enough matches to spread out some in Elo, and should thus also have played enough matches to impact the veteran players Elo. The way the curve looks is not natural and some heavy manipulation of the data most have been done to go from the red graph to the the blue on.

The big question is what they did? Did they introduce a decay over time/matches played? Only counting the last 100 games? Or did they simply multiple everyones Elo with whatever constant needed to get the graph they wanted? If it's a one time thing like the last, then I would expect the graph to recover in time.

Edited by Znail, 09 July 2013 - 11:55 AM.

[Hauser]

Znail, on 09 July 2013 - 11:33 AM, said:

The big question is what they did? Did they introduce a decay over time/matches played? Only counting the last 100 games? Or did they simply multiple everyones Elo with whatever constant needed to get the graph they wanted? If it's a one time thing like the last, then I would expect the graph to recover in time.

What they did was this:

1. They discovered that the way they were assigning Elo to players was wrong. A strong player defeating a weak player should get a tiny Elo boost. He was getting a big Elo boost. A weak player wining from a strong player should get big boost. He was getting a tiny one. This results in the red graph where a significant amount of the players are humping the maximum Elo barrier. In a well calibrated system this shouldn't happen so easily.

The mistake they made was a programming mistake. A very understandable one too. I made my own implementation of an Elo ranking system to understand how it works and I made exactly the same mistake. The explanation is rather technical though so I'll skip that.

2. After they fixed the mistake they took the match history of all players and fed that through the algorithm. They basically recalculated Elo for every match that was played. That's how they arrived at the blue graph. The blue graph will not ever look like the red graph again. If it does that means something is broken.
.
Rather it should start looking like this graph: http://lolmatches.com/charts

Big buldge around 1300 (most players are average!), then a long tail into 2600. The game is only beginning to show that just now.

Znail, on 09 July 2013 - 11:33 AM, said:

The less then 50 matches player group is still a large group of players and they have played enough matches to spread out some in Elo, and should thus also have played enough matches to impact the veteran players Elo. The way the curve looks is not natural and some heavy manipulation of the data most have been done to go from the red graph to the the blue on.

At the time new players started with 1300 Elo. The peak of the light blue lies at and 1250 the rest of the bulk is even lower. As you can lose 25 points for losing an equal game, that means a good number of new players players have played and lost at least two matches or more. This is to be expected really, they're new players.

Edited by Hauser, 09 July 2013 - 05:13 PM.

[Hellcat420]

matchmaker is still complete garbage. lobbies please.

[Asmosis]

Wispsy, on 22 April 2013 - 01:34 PM, said:

Well you can. Then you increase in Elo score until you cannot, leaving you around the right kind of area. Also you make it out as if one team is always much much better then the other team. This is not really the case in my experience, even if it is 8-0 then it can still be a close game with a number of heavily damaged people getting that first couple of kills then snowballing the numbers advantage. That or one team got the positional advantage and that means an awful lot and can make it seem like your team is worse then they are.

Sometimes you'll find part of that 8-0 winning team have no arms/weapons left, or are 3km away from the fight missing a leg. Like Wispsy says the killcount doesnt always represent, had a LOT of 8-2 wins/losses that were in effect very close. 8-0 normally results from superior tactics, a bit of luck and skill.

Sometimes overwhelming force (200+tonnage advantage) but thats pretty rare compared to the above mentioned reasons.

ELO also forced most of the pug-stomping 4man teams into the upper ELO brackets where the individual pug players are good enough to challenge them, so most new/average or below players shouldnt encounter premades much at all.

Edited by Asmosis, 09 July 2013 - 08:15 PM.

[Xandre Blackheart]

All I can speak from is anecdotal evidence, but I still see an awful lot of 8:0/0:8 games. An awful lot of them. and even more 6:2/2:6 games.

I wonder if the new matchmaking has gone into effect yet?

[PhoenixFire55]

I really like the fact that we have a lot of feedback in this topic. If you can guys, ask your friends to leave a comment here, be it positive or negative. Its a shame we can't see our elo scores, then it would make much more sense f.e. if all positive comments will be one zone of elo and all negative the other zone.

I'll say again, there is no reason for PGI to hide our elo scores from each other if elo is working, just as there is no reason not to provide us with values on elo distribution graphs and new graphs of same distributions. I think the only reason they do it is because its broken.

[Hauser]

Interesting quote from another topic:

Blue Footed Booby, on 08 July 2013 - 11:34 AM, said:

Thought experiment:
Can you think of any reasons for those low scores other than those players being bad? How might you check whether one of them is in fact the case?

Quote

The Dunning–Kruger effect is a cognitive bias in which unskilled individuals suffer from illusory superiority, mistakenly rating their ability much higher than average. This bias is attributed to a metacognitive inability of the unskilled to recognize their mistakes

I've seen competent lances get mowed down because they parked themselves at a chokepoint, only to have the other lance scatter to the winds instead of providing fire support. I've also seen matches get lost because no one on the team scouted enough to find that the entire other team was moving in a single blob toward one of the now-separated friendly lances. In both cases the folks who survived to the end of the match actually contributed the least to the fight (and indeed survived because they were farthest away from the heaviest fighting), but sometimes they get the highest score simply because the people trying to play correctly got steamrolled too fast, and then the should-have-been scouts or fire support take pot shots at the weakened enemy. See where I'm going with this?

Edited by Hauser, 10 July 2013 - 04:25 AM.

[Xandre Blackheart]

Update. I'm sorry guys but I have to be brutally honest here. Fu*k your current ELO system whatever it is. An entire evening of nothing but 8:0 7:1 1:7 0:8 games. One "close" game with a 4-8 result. I realize this is anecdotal, but the relevant part is that's my perception of the current state of the game, as well as the 3 other poor ******** I was playing with.

I just had a 3 week hiatus, looks like another even longer hiatus is coming up.

[Victor Morson]

Xandre Blackheart, on 10 July 2013 - 10:37 PM, said:

Update. I'm sorry guys but I have to be brutally honest here. Fu*k your current ELO system whatever it is. An entire evening of nothing but 8:0 7:1 1:7 0:8 games. One "close" game with a 4-8 result. I realize this is anecdotal, but the relevant part is that's my perception of the current state of the game, as well as the 3 other poor ******** I was playing with.

I just had a 3 week hiatus, looks like another even longer hiatus is coming up.

Weight matching is 100 times more important than skill matching, honestly, in that environment. And not just "class for class" because I'm sorry, that Dragon does not have a value of a Cataphract, period.

[PhoenixFire55]

Victor Morson, on 14 July 2013 - 03:15 PM, said:

Weight matching is 100 times more important than skill matching, honestly, in that environment. And not just "class for class" because I'm sorry, that Dragon does not have a value of a Cataphract, period.

Weight matching isn't more important than skill matching. Problem is we don't have any skill matching at this point. Current elo scores have nothing to do with skill.

[Lugh]

PhoenixFire55, on 15 July 2013 - 01:23 AM, said:

Weight matching isn't more important than skill matching. Problem is we don't have any skill matching at this point. Current elo scores have nothing to do with skill.

Correct. I could be the WORST shot ever in MWO and pilot a light and scout and cap for wins. My inability to fight handicaps my team even though my 'ELO score' shows I've got a quite good win /loss ratio...

[No7]

Where can I find information from PGI on how the matchmaker compensates when it doesn't find a perfect match?

[Hauser]

No7, on 18 July 2013 - 01:24 AM, said:

Where can I find information from PGI on how the matchmaker compensates when it doesn't find a perfect match?

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

In short, the match maker tries to find people near a specific elo value and minimize the weight difference for players. If it can't find new people it will slowly accept greater diversions from the target Elo and greater weight difference. After two minute it gives up and just puts stuff together.

[No7]

Hauser, on 18 July 2013 - 08:21 AM, said:

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

Awesome, thanks!

That was the post I was looking for.

7

Edited by No7, 18 July 2013 - 09:33 PM.