37 replies to this topic

[Krzysztof z Bagien]

I keep saying that I can't see a way Elo matchmaking system can work in 8v8 game. It can be used effectively (though there are still some concerns) in 1v1, or even Free For All style game (as it is 1v1v1... etc). It could also be effectively used in team games, but only to rate whole teams, like for example in football* (where it is actually used I belive), but not to rate indivitual players!

Consider this: team wins a match against better opponent, than wins against equal opponent, and than loses to better opponent. How should goalkeeper's rating change?

___
* I mean real football, not what you crazy Americans call "football", but Elo system is uset there too AFAIK.


Edit: Deamonition has good point in what he says. There are several systems used in multiplayer team games to rank inividual players so it can be done, but they take more data than simply whether team won or lost into account.

Edited by Krzysztof z Bagien, 12 April 2013 - 02:55 PM.

[Deamonition]

ohtochooseaname, on 12 April 2013 - 02:45 PM, said:

It's not being better than who you are teamed with, it's being better than the bracket you were dropped into: that, according to what they've said in their posts, is how it works. As far as I can tell, the game creates a game of a bracket, say 1500, then starts pulling in players. The more time it takes to pull in players (based on rating and weight), the looser it is in pulling in player. Therefore, if you are above 1300, you'll tend to get pulled into lower rating games (there are more games below you to be pulled into) than your rating and win more (because your opponents and team mates will still on average, average 1300), whereas, if you are below 1300, you'll tend to get pulled into higher rating games than your rating and lose more. I don't believe it re-balances after it makes the two teams, or my W/L ratio would be much smaller. If it did re-balance then it could possibly set you with a bunch of noobs to offset your score, in which case, you'd tend to lose of that offset were very high because it would assume that you could actually carry a team. I don't believe it does this.

The fact that you are on your team reduces the likelihood that a new player will be on your team, thus, your team should win more frequently. You don't have to carry your team or anything like that: a player on the enemy team is on average as skilled as a player on your team (neglecting people, who just started), so your impact doesn't have to be much to have an effect on your win ratio over a large number of games.

Ok, I understand a bit more what you meant. But, and again this is purely based on my experience playing couple of hundreds of games of MWO... I don't think the brackets work at all. I have honestly often been with or against guys that were not at all of the same level as some ppl in the 16 people were. I mean, when you EASILY stomp an 8men team without any of your team being scratched... how can we assume that your bracket concept is properly applied in MWO? And while you could say that sometimes, it could happen, you must actually admit that it happens quite often. At least 1 out of 3 or 4 of my games are complete stomps.

Krzysztof z Bagien, on 12 April 2013 - 02:50 PM, said:

I keep saying that I can't see a way Elo matchmaking system can work in 8v8 game. It can be used effectively (though there are still some concerns) in 1v1, or even Free For All style game (as it is 1v1v1... etc). It could be effectively also used in team games, but only to rate whole teams, like for example in football* (where it is actually used I belive), but not to rate indivitual players!

Consider this: team wins a match against better opponent, than wins against equal opponent, and than loses to better opponent. How should goalkeeper's rating change?

___
* I mean real football, not what you crazy Americans call "football", but Elo system is uset there too.

THANK GOD! The first guy that seems to agree with me! Or at least that says it!

[Krzysztof z Bagien]

Looks like we are only people that are sane here

But, jokes aside...
Elo system is not used in any multiplayer game I know but MWO and LOL and I think there's the reason for it.
Just think about it: you only have such data to work on: player's Elo rating as well as his teammates and opponents ratings and match outcome. Using that you determine if and how to change player's rating. Now - if your performance doesn't have big impact on the game outcome how can game outcome be used to rate your performance? And if your performance is so significant that it basically means you win a match by yourself - than your teammates performance is irrelevant, so how can this match's outcome be used to rate them?

Edit: I also want to add that I belive that supposed experience change brought by Elo introduction (whether it is for good or for bad) is simply a placebo effect - people belive that system is broken so they see that every match is unbalanced and vice versa. I also belive that there's no significant difference between what we have now and random matchmaking.

Edited by Krzysztof z Bagien, 12 April 2013 - 03:56 PM.

[Deamonition]

Krzysztof z Bagien, on 12 April 2013 - 03:09 PM, said:

Looks like we are only people that are sane here

But, jokes aside...
Elo system is not used in any multiplayer game I know but MWO and LOL and I think there's the reason for it.
Just think about it: you only have such data to work on: player's Elo rating as well as his teammates and opponents ratings and match outcome. Using that you determine if and how to change player's rating. Now - if your performance doesn't have big impact on the game outcome how can game outcome be used to rate your performance? And if your performance is so significant that it basically means you win a match by yourself - than your teammates performance is irrelevant, so how can this match's outcome be used to rate them?

As you pointed out, it's really for 1 v 1. Therefore, the bigger the team, the worse the system works. If it would be 2 v 2, that would not be the best system, but that could work. It doesn't work well for 5 v 5 in League of Legends, and it's worse for 8 v 8 in MWO.

[Krzysztof z Bagien]

And we have 12v12 comming soon.

But I don't think it will break the game in any way, as I said, matchmaking will still be random, but of a different kind. It just seems to be waste of time and resources to work on it.

[Deamonition]

Krzysztof z Bagien, on 12 April 2013 - 03:22 PM, said:

And we have 12v12 comming soon.

But I don't think it will break the game in any way, as I said, matchmaking will still be random, but of a different kind. It just seems to be waste of time and resources to work on it.

It is a waste of time as long as they keep going with the ELO concept. It would be wonderful (but that will never happen) if they would realize that they should create a completely new ranking concept.

And to be honest, I don't think it would even be complicated to "create" it. Perhaps a bit more work to implement it, but not complicated to have a brainstorm session and create the ******* thing.

[ohtochooseaname]

I'll certainly agree that ELO isn't a great system for MWO. All other systems, however, are vastly more complicated. Do we really want to have PGI try to implement something that complicated when they can't even get the ELO algorithm right??

What happens in an 8 player game is that the system is effectively underdamped. You will tend to oscillate a lot because you don't have enough control over the game, so it'll tend to have a random walk pattern trending about your actual skill rating. What they need is to damp down the oscillations by having the maximum possible ELO change reduce as you get more games. This will allow for rapid centering about your approximate skill location, and then allow the system to trim it down and reduce the number of games greatly above or below your actual skill because you will no longer be jumping all over the place. this is a fairly simple fix compared to other methods, which rate performance like, for example the current match score system.

Truly, the only way to have a zero bias rating system (system goal is to group players together based on their impact in game victory) in this game is to make it based on wins and losses, which is necessarily not a great system because you are on a team and thus don't individually have that much impact. There's no way around that other than to score individual performance, and those scoring methods are all easily gamed and quickly become ridiculously complicated. PGI went the simple route, and given the problems going even this route, I don't think they're ready for a more complicated system.

Edited by ohtochooseaname, 12 April 2013 - 03:58 PM.

[ohtochooseaname]

Deamonition, on 12 April 2013 - 03:24 PM, said:

It is a waste of time as long as they keep going with the ELO concept. It would be wonderful (but that will never happen) if they would realize that they should create a completely new ranking concept.

And to be honest, I don't think it would even be complicated to "create" it. Perhaps a bit more work to implement it, but not complicated to have a brainstorm session and create the ******* thing.

I'm waiting to pass judgement until after the fixed scores are up and running. I remember the days before ELO, and I think things are greatly improved. I especially like the fact that, just because some team runs a bunch of 3L's, they're not going to be facing a bunch of lights, and in fact, will mostly be up against mediums if the lights aren't available. I'm hoping their switching to 100% class to class matchmaking fails miserably, and they go for some sort of average tonnage arrangement.

[Deamonition]

ohtochooseaname, on 12 April 2013 - 03:57 PM, said:

I'll certainly agree that ELO isn't a great system for MWO. All other systems, however, are vastly more complicated. Do we really want to have PGI try to implement something that complicated when they can't even get the ELO algorithm right??

What happens in an 8 player game is that the system is effectively underdamped. You will tend to oscillate a lot because you don't have enough control over the game, so it'll tend to have a random walk pattern trending about your actual skill rating. What they need is to damp down the oscillations by having the maximum possible ELO change reduce as you get more games. This will allow for rapid centering about your approximate skill location, and then allow the system to trim it down and reduce the number of games greatly above or below your actual skill because you will no longer be jumping all over the place. this is a fairly simple fix compared to other methods, which rate performance like, for example the current match score system.

Truly, the only way to have a zero bias rating system (system goal is to group players together based on their impact in game victory) in this game is to make it based on wins and losses, which is necessarily not a great system because you are on a team and thus don't individually have that much impact. There's no way around that other than to score individual performance, and those scoring methods are all easily gamed and quickly become ridiculously complicated. PGI went the simple route, and given the problems going even this route, I don't think they're ready for a more complicated system.

Well, unless you tell me you are an actual expert in coding and implementing ranking methods in video games, allow me to not completely agree regarding the difficulty to change such a system.

I don't quite get how it would be that complicated to either borrow or developp a new ranking method. As you say, it needs to be based on the player impact in a victory. In MWO, it could be either damage done (for bigger mechs), damage taken (sometimes you end up being the only guy being shot at...), points captured, assists, kills. The most difficult part would be to get a scoring system for "helping mechs", such as a Raven with ECM staying with his group to help them not getting shot by missiles.

Now, when you have all that figure out. You get ranks as you get points. For example, say that you go in with an Atlas, end up doing the most damage of your group, taking the most damage, kill 3 and assisting in 4 other kills. That would give you quite an awesome score and you would probably go up in rank, because to be able to do all that, it would mean you are better than the rest of the people that were there.

Ok, that needs mathematics, but I am not going to developp all that for free! So yeah, it's not necessarily simple, but I do not think it is THAT MUCH difficult.

Not to mention that ELO is not a super simple method either. ELO calculation can become quite hard depending what you consider in the calculations.

[Deamonition]

This is such a pain. My win/loss ratio used to be close to 2.0 now it's 1.3

K/D used to be 1.9 now it's 1.74.....

Don't come and tell me that this system is working well....

[Syllogy]

Deamonition, on 12 April 2013 - 07:41 PM, said:

This is such a pain. My win/loss ratio used to be close to 2.0 now it's 1.3

K/D used to be 1.9 now it's 1.74.....

Don't come and tell me that this system is working well....

ERMAGERD! Fix coming Tuesday

[ohtochooseaname]

Deamonition, on 12 April 2013 - 07:41 PM, said:

This is such a pain. My win/loss ratio used to be close to 2.0 now it's 1.3

K/D used to be 1.9 now it's 1.74.....

Don't come and tell me that this system is working well....

If you are being matched with equal players, your W/L ratio and K/D ratio should trend towards 1, especially if you are solo pugging. The fact that yours are going down means that it is actually working for you and you are finding more difficult opponents on average than you were before the match making, which means your games are getting closer to being evenly matched.

[ohtochooseaname]

Deamonition, on 12 April 2013 - 05:05 PM, said:

Well, unless you tell me you are an actual expert in coding and implementing ranking methods in video games, allow me to not completely agree regarding the difficulty to change such a system.

Ok, that needs mathematics, but I am not going to developp all that for free! So yeah, it's not necessarily simple, but I do not think it is THAT MUCH difficult.

Not to mention that ELO is not a super simple method either. ELO calculation can become quite hard depending what you consider in the calculations.

I deal with coding for numerical and statistical models fairly regularly on a completely different application, so I tend to talk about these things as if the person were the object under measurement, which is basically what the ELO system is.

The fact is that their current ELO system is basically as simple as you can make a matchmaking system since it only includes wins and losses for a given class. It is one simple calculation with one variable being tracked for each player and a binary outcome set. Any additional stats being used in the calculations end up making the system incredibly more complicated because there is an exponential relationship between the number of variables and how "complicated" the system is as they inter-relate. The ELO system they have implemented is also, as I have said before, an unbiased system. All other tracking methods (meaning not Wins/Losses) of competence will be less accurate, can be gamed by the players, but will converge faster (meaning you will reach your final ELO faster). They will also tend to be biased against certain classes, as the current match score system is against lights.

Writing a code to track someone's contribution in a "role warfare" environment is something you could have 10 people work on year round and still not end up with anything good simply because there is just too much to track. PGI went with simple and unbiased, and I can respect that. What I can't respect is the fact that they got it completely wrong for several months.

[Henrylw]

my record team wipe: 14 times continuous. ;;; i dont know whats balance means...i just play for 3 days but i have to facing a pro players..

[Hauser]

Henrylw, on 16 April 2013 - 02:10 AM, said:

my record team wipe: 14 times continuous. ;;; i dont know whats balance means...i just play for 3 days but i have to facing a pro players..

Nah. Nothing so bad really.

You're in a trial mech. They're stock mechs from the table top and while they are serviceable they are simply outclassed by custom build mechs. I reckon you'll want to buy your own mech some time soon. Once you have your own mech you can also unlock piloting skills that make your mech better. Especially unlocking the elites gives you a huge boost.

Since you are new you are going up against average players. Elo needs some time to sort things out and send you to the bottom. You'll come back up once you can handle your mech.

Finally you might want to get on TS. Tell people you're new, ask questions, they'll help you out.

http://mwomercs.com/...mspeak-servers/

edit:

You might also want to unlock your arms and disable throttle decay. They're supposed to make the game easier for new players but I think they are very restrictive and really hamper anybody seriously trying to learn his mech. You can do that in options.

Edited by Hauser, 16 April 2013 - 05:19 AM.

[The Strange]

Deamonition, on 12 April 2013 - 07:41 PM, said:

This is such a pain. My win/loss ratio used to be close to 2.0 now it's 1.3

K/D used to be 1.9 now it's 1.74.....

Don't come and tell me that this system is working well....

The W/L average is normalizing, as is the intended effect of Elo. The K/D ratio in this game is a joke though. Don't get hung up on it. I don't know how many times I have worked a Mech down to nothing just to have someone else run by and take the final shot to get the "kill".

[Hauser]

The Strange, on 16 April 2013 - 06:54 AM, said:

The W/L average is normalizing, as is the intended effect of Elo. The K/D ratio in this game is a joke though. Don't get hung up on it. I don't know how many times I have worked a Mech down to nothing just to have someone else run by and take the final shot to get the "kill".

Stealing kills is easy though. You press R, you see which part of the mech has been damaged. Shoot it.

Edited by Hauser, 16 April 2013 - 08:14 AM.

[ohtochooseaname]

I don't worry about stealing kills, etc. Nor do I worry about other people stealing mine. I figure, it averages out. (though, people with a high single point alpha tend to steal kills a lot more frequently on accident than, say a 5D spider). IMO, you always kill the enemy as quickly as possible to reduce the impact to the rest of your team. So, when it goes down (unless it's a slaughter fest), I always appreciate the help, especially if I'm close to death. In fact, I'll even *gasp* turn my armored back to the enemy to allow my allies to kill them while I avoid being killed. If your K/DR is greater than 1, it's usually better to do this than to risk dying, anyway.