Guild Wars 2

“… we have also decided to make a change to the scoring matrix that will now cause players to lose ladder points in match-ups where they were favored to win and did not.”

So how does this work? Is this based on my team’s MMR counted together, or simply on my own?

teams …….sdfsdfsdfsd

I would guess it would be your odds of victory. Since all the phrasing coming from Anet does not use the word team

oh hey evan… question: patch notes or whatever said you guys changed the scoring matrix a little, but ive never seen… say… an updated version of the table that was in the blog. how was it updated?

Evan Lesh.3295:

The wiki shows how we calculate odds of victory:
https://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm#Match_Prediction

A couple of matches ago I was against a team I have to assume was premade. We lost 500 to 100~ (my memory fails me) and I lost a point. I have a really difficult time imagining how I was favored to win a match where we got absolutely killed. This happened again a few minutes ago, and I’m sure I’ll lose a point for it, too, and yet I do not believe I could be favored to win either of those matches.

Buster.6137:

Evan Lesh.3295:

The wiki shows how we calculate odds of victory:
https://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm#Match_Prediction

A couple of matches ago I was against a team I have to assume was premade. We lost 500 to 100~ (my memory fails me) and I lost a point. I have a really difficult time imagining how I was favored to win a match where we got absolutely killed. This happened again a few minutes ago, and I’m sure I’ll lose a point for it, too, and yet I do not believe I could be favored to win either of those matches.

I don’t know if it counts ranking for every single class, but if it does over all you might get someone who is playing one class over and over. Some people also learn quicker, have experienced players who told them what to do etc. You win some you lose some. In PvP is always good to be ready to die. It just happens.

Buster.6137:

Evan Lesh.3295:

The wiki shows how we calculate odds of victory:
https://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm#Match_Prediction

A couple of matches ago I was against a team I have to assume was premade. We lost 500 to 100~ (my memory fails me) and I lost a point. I have a really difficult time imagining how I was favored to win a match where we got absolutely killed. This happened again a few minutes ago, and I’m sure I’ll lose a point for it, too, and yet I do not believe I could be favored to win either of those matches.

Based on the wiki, you need to score at least 200 points (with 0% to 20% winning chance) or 300 points (with 20% to 40% winning chance) to achieve no ladder point loss even if you lose the match.

Also based on the wiki, your winning favor is calculated based on average rank (weighted 1), average effective rating (weighted 5) and highest roster size (weighted 2). To my knowledge, rank would mean your current leaderboard standing, rating would be your win/loss rate minus deviation and roster size is weighted to counter premade vs. pug scenario – in favor of solo Q-ers. In my opinion, the weight is too low on rank and too high on rating. It would be the cause for the (mis)matching of high MMR players with novices in a same team to balance against premades and their punishing loss in comparison to high ranking, low MMR players.

Iris Ng.9845:

Buster.6137:

Evan Lesh.3295:

The wiki shows how we calculate odds of victory:
https://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm#Match_Prediction

A couple of matches ago I was against a team I have to assume was premade. We lost 500 to 100~ (my memory fails me) and I lost a point. I have a really difficult time imagining how I was favored to win a match where we got absolutely killed. This happened again a few minutes ago, and I’m sure I’ll lose a point for it, too, and yet I do not believe I could be favored to win either of those matches.

Based on the wiki, you need to score at least 200 points (with 0% to 20% winning chance) or 300 points (with 20% to 40% winning chance) to achieve no ladder point loss even if you lose the match.

Also based on the wiki, your winning favor is calculated based on average rank (weighted 1), average effective rating (weighted 5) and highest roster size (weighted 2). To my knowledge, rank would mean your current leaderboard standing, rating would be your win/loss rate minus deviation and roster size is weighted to counter premade vs. pug scenario – in favor of solo Q-ers. In my opinion, the weight is too low on rank and too high on rating. It would be the cause for the (mis)matching of high MMR players with novices in a same team to balance against premades and their punishing loss in comparison to high ranking, low MMR players.

no, rank is your pvp rank there. people constantly complain theyre matched against dragons as rabbits, and theyre mostly unjustified, but if matchmaking has the opportunity to pair you with a person closer to your rank then it will at a low priority.

edit: your sig link……………

Justin ODell.9517:

Evan Lesh.3295:

insanemaniac.2456:

oh hey evan… question: patch notes or whatever said you guys changed the scoring matrix a little, but ive never seen… say… an updated version of the table that was in the blog. how was it updated?

Still need to update the wiki.

Updated.

cool thanks!

edit: i drewed a pitcher

Buster.6137:

I have a really difficult time imagining how I was favored to win a match where we got absolutely killed. This happened again a few minutes ago, and I’m sure I’ll lose a point for it, too, and yet I do not believe I could be favored to win either of those matches.

You probably weren’t actually favored to win those matches. Just because their algorithm predicts that you would be favored does not mean you would actually be favored – it has very, very clear flaws when handling premades vs pugs, for instance.

Justin, I can’t help but ask: Why do you insist on allowing for an organized team (aka: “premades”) to fight PUG teams?

By now you probably must have realized what any good competitive player could have already told you: it’s a bad idea. In the realm of the terribad ideas.
You must know this.

Why do you allow for this?
Oh, I can guess the reason – not enough player population. But in the long run, if you keep allowing for this travesti of a competition/balance, you’ll lose more players than if you split for “Random Arenas” and “Team Arenas”.

It’s as if you do not believe yourself that sPvP can be improved…

Justin ODell.9517:

Ensign.2189:

Just because their algorithm predicts that you would be favored does not mean you would actually be favored – it has very, very clear flaws when handling premades vs pugs, for instance.

At least in aggregate, we have a very strong, linear prediction gradient for accuracy. Which means our match prediction is pretty kitten good, but we could always make improvements.

Roster size (i.e. premade count) accounts for 25% of a match’s prediction. I do wonder if it should be more.

The matchmaking is really good so ty for that. I just wish we could get mmr leaderboards

Bio Flame.4276:

Justin, I can’t help but ask: Why do you insist on allowing for an organized team (aka: “premades”) to fight PUG teams?

By now you probably must have realized what any good competitive player could have already told you: it’s a bad idea. In the realm of the terribad ideas.
You must know this.

Why do you allow for this?
Oh, I can guess the reason – not enough player population. But in the long run, if you keep allowing for this travesti of a competition/balance, you’ll lose more players than if you split for “Random Arenas” and “Team Arenas”.

It’s as if you do not believe yourself that sPvP can be improved…

This is over stated. Premades vs pugs. You can win most premades with solo queue and in fact most premades dont use ts or tryharc anymore. Mostly people qq about it to give them an excuse for losing because they arent as good as they think they are

How much does the system take into account the skill of a premade group when working out how much to up their MMR? If I get into a group of 5, I’m not going to be much more effective than I am solo.

Lordrosicky.5813:

Bio Flame.4276:

Justin, I can’t help but ask: Why do you insist on allowing for an organized team (aka: “premades”) to fight PUG teams?

By now you probably must have realized what any good competitive player could have already told you: it’s a bad idea. In the realm of the terribad ideas.
You must know this.

Why do you allow for this?
Oh, I can guess the reason – not enough player population. But in the long run, if you keep allowing for this travesti of a competition/balance, you’ll lose more players than if you split for “Random Arenas” and “Team Arenas”.

It’s as if you do not believe yourself that sPvP can be improved…

This is over stated. Premades vs pugs. You can win most premades with solo queue and in fact most premades dont use ts or tryharc anymore. Mostly people qq about it to give them an excuse for losing because they arent as good as they think they are

You can win if they are bad or your team is very good, true.
But what does that prove? That given a large skill-gap, a good team can beat a bad team?

That is not the point. The point is that an organized team has chosen their classes, their builds and skills in order to become more than the sum of all their parts. An organized team has TS, is able to adjust to the fight faster than their PUG counterparts, courtesy of voice coms.
That is a very different playing field when compared to 5 random strangers, with builds which are probably not optimized with their team mates and have no quick voice coms.

We’re in 2015. PUG team should NOT face organized teams. Period.

*Disclaimer – this is only valid is one is looking for competition and a competitive game. *

Justin ODell.9517:

Ensign.2189:

Just because their algorithm predicts that you would be favored does not mean you would actually be favored – it has very, very clear flaws when handling premades vs pugs, for instance.

At least in aggregate, we have a very strong, linear prediction gradient for accuracy. Which means our match prediction is pretty kitten good, but we could always make improvements.

Roster size (i.e. premade count) accounts for 25% of a match’s prediction. I do wonder if it should be more.

The First stage to recovery is admitting that you have F’d up (cough -since the start of this game- cough)

would it be possible to get that admission? i cant tell you have much we all just wanna hear it not this jumping around the issue we want to see you get on the right path as it were

official.7362:

Justin ODell.9517:

Ensign.2189:

Just because their algorithm predicts that you would be favored does not mean you would actually be favored – it has very, very clear flaws when handling premades vs pugs, for instance.

At least in aggregate, we have a very strong, linear prediction gradient for accuracy. Which means our match prediction is pretty kitten good, but we could always make improvements.

Roster size (i.e. premade count) accounts for 25% of a match’s prediction. I do wonder if it should be more.

The First stage to recovery is admitting that you have F’d up (cough -since the start of this game- cough)

would it be possible to get that admission? i cant tell you have much we all just wanna hear it not this jumping around the issue we want to see you get on the right path as it were

the best tip i can give u —-
SNEAK INTO THE WORLD OF WARCRAFT OFFICES WHEN NO ONES AROUND BUT THE JANITORS, AT THAT TIME YOU BRING IN THE NEIGHBORHOOD HASSY AT HUSSY HASSY AND HAIRY’S BOOTyBAKERY. SHE WILL DISTRACT THE JANITOR JUST LONG ENOUGH FOR YOU TO GET IN AND GET OUT SAFELY WITH DOCUMENTS TO LEARN HOW WOW’s ALGORITHM WORKS TO ONCE AND FOR ALL SAVE PVP IN GW2 hurrayyyyyyyyy!~!!!!#!#!#! zomg cruuk u saved us ALL HAIL CRUUK

ps ill take your deleting this as a sign that you have undergone this brave but necessary quest

Hey Evan and Justin.

Bring back solo que please

Much love in advance. Kthanxbye

Justin ODell.9517:

Which means our match prediction is pretty kitten good, but we could always make improvements.

You’re using an entirely linear model, instead of a normal (Gaussian) model (or, if you wanted to be a bit more sophisticated, a model with an empirically measured underlying performance distribution). When everything is well behaved (teams of roughly equal rating, same premade count) there shouldn’t be a meaningful difference between these options, but I am extraordinarily skeptical that your algorithm doesn’t take a huge dump when handling edge cases.

Double that skepticism when handling premades – not only is the impact of forming a premade non-linear on match outcomes, but its impact varies with rating (that is, very good players forming a premade benefit more from a premade than weaker players) in every other game I am aware of.

Justin ODell.9517:

Roster size (i.e. premade count) accounts for 25% of a match’s prediction. I do wonder if it should be more.

I assume you are referring to the roster ‘weight’ variable? That’s not how that thing works – it’s actually a variable that affects the impact of rating and rank differences on outcomes, in addition to premade status. For example, when the ‘weight’ is set equal to 2, a full premade with average rating 160 points higher than the other team would be expected to win 75% of the time; if the premade ‘weight’ is set to 4, that same premade vs premade match would only expect the former to win 70% of the time.

Every time you tweak that premade variable, you are actually shifting what individual ratings mean. For example, with the numbers listed on the wiki it takes a 320 point difference in rating before you predict one team will win 100% of the time, but with premade weight set to 4 that number increases to 400.

The value you actually would want to tweak to independently alter the impact on premades on predicted outcome would be the premade ‘spread’ variable.

As for how much that ‘spread’ should actually be, that’s not difficult to test, if a bit complicated – you’d want to bootstrap samples of solo queue and full team queue games, and find the value of ‘spread’ such that both sets converge on the same ratings.

Justin ODell.9517:

Ensign.2189:

Just because their algorithm predicts that you would be favored does not mean you would actually be favored – it has very, very clear flaws when handling premades vs pugs, for instance.

At least in aggregate, we have a very strong, linear prediction gradient for accuracy. Which means our match prediction is pretty kitten good, but we could always make improvements.

Roster size (i.e. premade count) accounts for 25% of a match’s prediction. I do wonder if it should be more.

Define strong.

wwwes.1398:

Justin ODell.9517:

Which means our match prediction is pretty kitten good

Define strong.

he actually did…

Out of curiosity, why use Your Team’s Score to assign points instead of margin of victory/loss? This system doesn’t reward your team for winning by a lot, just winning.

Justin ODell.9517:

Ensign.2189:

Just because their algorithm predicts that you would be favored does not mean you would actually be favored – it has very, very clear flaws when handling premades vs pugs, for instance.

At least in aggregate, we have a very strong, linear prediction gradient for accuracy. Which means our match prediction is pretty kitten good, but we could always make improvements.

Roster size (i.e. premade count) accounts for 25% of a match’s prediction. I do wonder if it should be more.

I think it should be more. Going up against a premade team is sometimes a death sentence before you even begin. Up that percentage

well from my records of last 2 nights the prediction is about 50% which is same as coin toss.
April 1, 7 – 0 Gain 7 pts
April 2, 1-9 Gain 1 lost 6
So the prediction was wrong 6 of 10 matches last night and most likely wrong 3-4 out of 7 on the first. Avg success rate around 50%…

insanemaniac.2456:

wwwes.1398:

Justin ODell.9517:

Which means our match prediction is pretty kitten good

Define strong.

he actually did…

Well, no. Is strong guessing correctly 95% of the time? 80% of the time?

I have complained about the favour to win algorithm myself but I don’t think it is as kittenome people are making it out to be. I have no problem facing off against premades or partly premades as a solo as long as I don’t end up losing points when my team is highly outmatched.

I think the 25% mentioned by Justin above should maybe be increased to something more like 35 or 40. The biggest advantage a premade has is being able to bring a decent team comp. I realise most of the premades you face do not use voice comms but even having a triple queue on your team that have builds that synergise is a huge advantage over solo queuers who get 5 random classes with random builds teamed up.

I would also remove the pvp rank completely as it is probably the worst judge of how good someone is at pvp. In fact if you removed the pvp rank and added the weight to premade you would probably have a decent estimate.

haha bad @s some is changed to kittenome

Basket of Boxes.1976:

I have complained about the favour to win algorithm myself but I don’t think it is as kittenome people are making it out to be.

Their rating algorithm is based on Glicko2, but their favored to win algorithm is inconsistent with Glicko2. You don’t need any rigorous statistical analysis to see the problem.

Ensign.2189:

Basket of Boxes.1976:

I have complained about the favour to win algorithm myself but I don’t think it is as kittenome people are making it out to be.

Their rating algorithm is based on Glicko2, but their favored to win algorithm is inconsistent with Glicko2. You don’t need any rigorous statistical analysis to see the problem.

No, you just need to understand that Glicko2 is meant to rate 1v1 scenarios and some minor adjustments are required for team situations.

Basket of Boxes.1976:

I have complained about the favour to win algorithm myself but I don’t think it is as kittenome people are making it out to be. I have no problem facing off against premades or partly premades as a solo as long as I don’t end up losing points when my team is highly outmatched.

I think the 25% mentioned by Justin above should maybe be increased to something more like 35 or 40. The biggest advantage a premade has is being able to bring a decent team comp. I realise most of the premades you face do not use voice comms but even having a triple queue on your team that have builds that synergise is a huge advantage over solo queuers who get 5 random classes with random builds teamed up.

I would also remove the pvp rank completely as it is probably the worst judge of how good someone is at pvp. In fact if you removed the pvp rank and added the weight to premade you would probably have a decent estimate.

I’ll second this. If I could lose my rank and go back down to Rabbit to get outmatched less, I’d do it in a heartbeat. When you go 5-20 in a single weekend, getting a rank up chest mostly just feels like sarcasm.