Guild Wars 2

I unfortunately missed the reddit AMA the other day because I was at work, but I had a question that I had been hoping to ask the PVP team.

Does anything go into calculating a players MMR other than the games they win/lose and the MMR of the other team that they win/lost against?

Does the matchmaking algorithm analyze thing like how many points you decapped/captured, or the outcome of solo battles or team fights? What about things like how often a player is fighting on/off point vs. skirmishing?

I strongly believe that basing a players “skill” goes beyond wins and losses, if tournament PVP was based on teams of 1 and 1 then of course it would be a perfect representation, but with 5 players on a team i think that expecting the “good” players to carry 4 others is far fetched.

Are there any plans to expand the MMR algorithm to incorporate new factors related to individual accomplishments within ranked play? I think this would have a HUGE impact on the quality of matches, and would really show players how good they truly are.

I would like to see a system where the overriding factor in MMR adjustments is still wins and loses BUT in the case of a loss, to have each player’s individual accomplishments taken into account so that good players aren’t punished because of less-skilled teammates.

I’m not the PvP team, but I can say with certainty that ‘personal score’ isn’t a factor in a player’s MMR. It doesn’t matter how many points you decap, where you fight or how you act within a game. It’s impossible for the system to track what a player is doing during the game, and more importantly whether what they’re doing is actually good or not. You earn personal score for channeling Ferocity in Temple of the Silent Storm, but that doesn’t really help your team win.

Now, with that said, I’m pretty sure MMR does take into account how close you come to winning. If you play against an enemy team that’s rated much higher than you, but come close to beating them, your MMR should improve. It’s like how the system last league gave out pips for coming close when against stronger foes. You won’t get the loser pips anymore, but you can still bolster your MMR in a losing cause by putting up a good fight.. whereas going AFK after losing the first teamfight will cause a blowout score and drop your MMR farther than it would drop from a normal loss.

Xiahou Mao.9701:

I’m not the PvP team, but I can say with certainty that ‘personal score’ isn’t a factor in a player’s MMR. It doesn’t matter how many points you decap, where you fight or how you act within a game. It’s impossible for the system to track what a player is doing during the game, and more importantly whether what they’re doing is actually good or not. You earn personal score for channeling Ferocity in Temple of the Silent Storm, but that doesn’t really help your team win.

Now, with that said, I’m pretty sure MMR does take into account how close you come to winning. If you play against an enemy team that’s rated much higher than you, but come close to beating them, your MMR should improve. It’s like how the system last league gave out pips for coming close when against stronger foes. You won’t get the loser pips anymore, but you can still bolster your MMR in a losing cause by putting up a good fight.. whereas going AFK after losing the first teamfight will cause a blowout score and drop your MMR farther than it would drop from a normal loss.

As a full time modelling engineer I can tell you 100% that it is possible to measure an individual players “skill” with stats derived from what they do in a match.

foogison.5067:

Xiahou Mao.9701:

I’m not the PvP team, but I can say with certainty that ‘personal score’ isn’t a factor in a player’s MMR. It doesn’t matter how many points you decap, where you fight or how you act within a game. It’s impossible for the system to track what a player is doing during the game, and more importantly whether what they’re doing is actually good or not. You earn personal score for channeling Ferocity in Temple of the Silent Storm, but that doesn’t really help your team win.

Now, with that said, I’m pretty sure MMR does take into account how close you come to winning. If you play against an enemy team that’s rated much higher than you, but come close to beating them, your MMR should improve. It’s like how the system last league gave out pips for coming close when against stronger foes. You won’t get the loser pips anymore, but you can still bolster your MMR in a losing cause by putting up a good fight.. whereas going AFK after losing the first teamfight will cause a blowout score and drop your MMR farther than it would drop from a normal loss.

As a full time modelling engineer I can tell you 100% that it is possible to measure an individual players “skill” with stats derived from what they do in a match.

Sure you can but will it be meaningful. You can score points in game in ways that don’t lead to wins. It is very common in losses for example sometimes quite lopsided losses for half the losing team to have higher scores then most of the people on the winning team. Also, when I group with friends because of different roles we can have widely different scores at the end of the match I don’t assume Im contributing more because my score is higher nor do they.

brannigan.9831:

foogison.5067:

Xiahou Mao.9701:

I’m not the PvP team, but I can say with certainty that ‘personal score’ isn’t a factor in a player’s MMR. It doesn’t matter how many points you decap, where you fight or how you act within a game. It’s impossible for the system to track what a player is doing during the game, and more importantly whether what they’re doing is actually good or not. You earn personal score for channeling Ferocity in Temple of the Silent Storm, but that doesn’t really help your team win.

Now, with that said, I’m pretty sure MMR does take into account how close you come to winning. If you play against an enemy team that’s rated much higher than you, but come close to beating them, your MMR should improve. It’s like how the system last league gave out pips for coming close when against stronger foes. You won’t get the loser pips anymore, but you can still bolster your MMR in a losing cause by putting up a good fight.. whereas going AFK after losing the first teamfight will cause a blowout score and drop your MMR farther than it would drop from a normal loss.

As a full time modelling engineer I can tell you 100% that it is possible to measure an individual players “skill” with stats derived from what they do in a match.

Sure you can but will it be meaningful. You can score points in game in ways that don’t lead to wins. It is very common in losses for example sometimes quite lopsided losses for half the losing team to have higher scores then most of the people on the winning team. Also, when I group with friends because of different roles we can have widely different scores at the end of the match I don’t assume Im contributing more because my score is higher nor do they.

I think you mis-understood what I was saying. I do not mean the score that is awarded through captures/kills/communes.

I was referring to the statistics behind damage dealt, players kills, points capped/decapped, time spent defending, time spent on offense, health given to teammates, revives, etc etc etc.

The game’s engine already tracks all of these things. I am proposing to assign weight to different actions and include these in scoring a players matchmaking rating.

foogison.5067:

I unfortunately missed the reddit AMA the other day because I was at work, but I had a question that I had been hoping to ask the PVP team.

Does anything go into calculating a players MMR other than the games they win/lose and the MMR of the other team that they win/lost against?

Does the matchmaking algorithm analyze thing like how many points you decapped/captured, or the outcome of solo battles or team fights? What about things like how often a player is fighting on/off point vs. skirmishing?

I strongly believe that basing a players “skill” goes beyond wins and losses, if tournament PVP was based on teams of 1 and 1 then of course it would be a perfect representation, but with 5 players on a team i think that expecting the “good” players to carry 4 others is far fetched.

Are there any plans to expand the MMR algorithm to incorporate new factors related to individual accomplishments within ranked play? I think this would have a HUGE impact on the quality of matches, and would really show players how good they truly are.

I would like to see a system where the overriding factor in MMR adjustments is still wins and loses BUT in the case of a loss, to have each player’s individual accomplishments taken into account so that good players aren’t punished because of less-skilled teammates.

Good question. I want to know that too. Since MMR is affecting our random soloq matchup, we need to know how to work our MMR to get it better.

Nothing goes into MMR adjustments besides match outcomes (win / loss) and pre-match MMRs.

You’re going to be hard pressed to find in-game variables that’ll improve player ratings without hurting the integrity of the game. Sure, there might be some general trends that hold true like ‘doing damage is good!’ and ‘dying is bad!’, but including those in the outcome of the match distorts strategic choices – am I going to get more credit for a win if I play a DPS focused class than if I had played a support class? Should I dump damage into the tank to up my participation, or should I keep trying to catch that Mesmer/Thief? Is continuing to fight this 2v1 to the death going to cost me more on contribution than it adds to my chance of winning? Figuring out what happened, and who contributed more or less to winning a match isn’t impossible – but it’s really, really hard, and if you get it wrong you’re introducing perverse incentives.

There are much, much bigger benefits to be had from digging into the data to understand the right way to construct a team rating from individual player ratings (hint: adding them all together is not it), tackling heteroscedasticity across the rating range, refining the shape of your outcomes curve, engineering a dynamic k-value to account for the informational value of a given match for each player. All of these are doable without distorting any incentives and could pay huge dividends on match quality and rating integrity – and are mostly unsolved problems that would apply to a variety of games.