Changing the Rating System

[Ithilwen.1529]

@"Ben Phongluangtham.1065"

I have been reading the C# implementation for Glicko 2 available on GitHub. It doesn't appear to actually follow the glicko 2 algorithm as published, though I admit.. I'm not skilled in Calculus.

Far more important is what's found here: https://www.guildwars2.com/en/news/finding-the-perfect-match/ .

If you look at the chart, found mid page, you will find that a mismatch loss is penalized MORE than a close loss. This is essentially the same as the s1-s2 win streaking and heavily penalizing losses. In my opinion the thinking is fundamentally flawed.

A fair system would do the opposite. In a heavy mismatch, the losing side would suffer lower penalties.

[Dreddo.9865]

@Ithilwen.1529 said:A fair system would do the opposite. In a heavy mismatch, the losing side would suffer lower penalties.Allow me to add to your comment. I just had this game where one of our teammates disconnects early on so we have to go 4v5 almost the full game. And we win. Guess what. The game awarded us a +12 like we had a normal win. I mean it was entertaining to win a 4v5 but would be more fair to be awarded for the extra effort. This is one of the small things the current system doesn't take into consideration.

[SlippyCheeze.5483]

@Ithilwen.1529 said:@"Ben Phongluangtham.1065"

I have been reading the C# implementation for Glicko 2 available on GitHub. It doesn't appear to actually follow the glicko 2 algorithm as published, though I admit.. I'm not skilled in Calculus.

https://en.wikipedia.org/wiki/Glicko_rating_system gives you the math, which is literally all it is. I suspect your comment WRT calculus is because you can't translate the math into meaning, yes?

Far more important is what's found here: https://www.guildwars2.com/en/news/finding-the-perfect-match/ .

https://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm gives you significant details, including the constants in use in GW2 GLICKO-2. It is linked from the page, and I'm certain you have seen it, but others may not follow all the way through.

If you look at the chart, found mid page, you will find that a mismatch loss is penalized MORE than a close loss. This is essentially the same as the s1-s2 win streaking and heavily penalizing losses. In my opinion the thinking is fundamentally flawed.

Are you talking about the "ladder ranking" chart? If so, that has been deprecated, and is no longer the active mechanism for calculating ... well, anything, really. Previously it was about the awarding of ladder points, which are different from MMR, and were not used as part of the GLICKO-2 system. They are not, in fact, even a concept within it.

To see the current mechanisms, apply the constants documented in the wiki to glicko, and/or review the matchmaking pseudocode to identify both how teams are constructed, and how wins are predicted, as they use the same data. (The win prediction being based on the difference between the scores of the teams assembled.)

[SlippyCheeze.5483]

@Dreddo.9865 said:

@Ithilwen.1529 said:

A fair system would do the opposite. In a heavy mismatch, the losing side would suffer lower penalties.

Allow me to add to your comment. I just had this game where one of our teammates disconnects early on so we have to go 4v5 almost the full game. And we win. Guess what. The game awarded us a +12 like we had a normal win. I mean it was entertaining to win a 4v5 but would be more fair to be awarded for the extra effort. This is one of the small things the current system doesn't take into consideration.

I agree with you. It definitely seems worth increasing the value of a match where you are one person down, and still win, given that you had a significantly lowered chance of winning when you lost that person.

[Sampson.2403]

For those at mid/low rank:

Even if they changed the rating system do you honestly think that the leader boards are going to be any different than they are now? The players who consistently figure out how to maximize their team impact each game will be at the top just like they are now. Any benefits that you feel will help you specifically in your rating/mmr from whatever proposed change will also help everyone else, so the end result will be if your rank X now, you will still be rank X afterwards.

For those at high rank:

Yes, being at the top and losing much more than what you win is annoying, but it's not a broken mechanic or a bug, we all agree that it's due to population - the system is working as is. Some things you can't just "fix" without impairing the integrity of the system as a whole.

[Chuck.2864]

@Frostmane.9734 said:

@pah.4931 said:

I think what would be better is if ranked was team only, and unranked was solo- or group-q ... and both had pips. One for people who want to compete and get prestige (which should always be based on premade teams) ... the other for casual farming with friends and trying fun builds, etc.

The only problem with that is players who don't have an organized team, or don't want one, can't compete in ranked play. I think both systems need to exist.

I would like to see individual player ratings within a match so that people who play well aren't losing rank when the matchmaking system fails.

Both systems do exist, AT's are where you go to get your competitive organized 5-man team PvP. People who think that a team queue is actually going to be a success with the game's current PvP population are kidding themselves. It will be a dead mode. You can already see how few PUGs are formed for AT's, the LFG tool for that is dead, and the incentive to play the AT is much bigger than a random team queue would provide. So you're looking at mostly static teams, and how many of these exist?

[Chuck.2864]

Accidental double post.

[Dreddo.9865]

@Sampson.2403Allow me to disagree with you.

The fact that every season we have complaints about the MM engine must mean something, right? Even the best player in the world teamed with a bunch of bad players can't win alone. The idea of 'carry people to climb ranks' is flawed and can't represent a fair MM and ranking system. Why? Simply because most of the times this is impossible and others it is so tedious that contradicts with the basic principle of playing a game.

What is the point of playing with people of far lesser skill, people with ignorant builds, people that rotate around the map without a clue, people dying in 500ms and so on? Where is the fun in this? Would this help me become a better player and reach my skill potential? Just a few rhetorical questions. It's not about getting a benefit for yourself, it's about getting better quality and challenging PvP matches. That's the idea.

I understand that a big issue with the current situation is the lack of players. But on the other hand we can't just pretend everything is fine and just move on. Because it isn't.

[Sampson.2403]

@Dreddo.9865 said:@Sampson.2403Allow me to disagree with you.

The fact that every season we have complaints about the MM engine must mean something, right? Even the best player in the world teamed with a bunch of bad players can't win alone. The idea of 'carry people to climb ranks' is flawed and can't represent a fair MM and ranking system. Why? Simply because most of the times this is impossible and others it is so tedious that contradicts with the basic principle of playing a game.

What is the point of playing with people of far lesser skill, people with ignorant builds, people that rotate around the map without a clue, people dying in 500ms and so on? Where is the fun in this? Would this help me become a better player and reach my skill potential? Just a few rhetorical questions. It's not about getting a benefit for yourself, it's about getting better quality and challenging PvP matches. That's the idea.

I understand that a big issue with the current situation is the lack of players. But on the other hand we can't just pretend everything is fine and just move on. Because it isn't.

While I can understand your concerns, what you're really complaining about isn't the rating system, it's the solo Q system.

Also, why are you assuming that the matchmaking is placing you with people far below your skill level when in fact it's designed to do the exact opposite.

[Dreddo.9865]

@Sampson.2403 said:While I can understand your concerns, what you're really complaining about isn't the rating system, it's the solo Q system.

Also, why are you assuming that the matchmaking is placing you with people far below your skill level when in fact it's designed to do the exact opposite.

I never spoke of solo Q. GW2 isn't the only game with a ranking and group matchmaking system.When I am joining a gold 3 match with a firebrand who chases to kill a bunker druid on the road, or a DH that fights a 2v1 10 meters away (from me standing on the point) and dies, and many more such examples, I can safely 'assume' I am being paired with people below my skill level. I can't see the 'exact opposite' scope in a MM design where (even in extreme cases) 200+ rating is allowed for a team to be formed.

[Sampson.2403]

@Dreddo.9865 said:

@"Sampson.2403" said:While I can understand your concerns, what you're really complaining about isn't the rating system, it's the solo Q system.

Also, why are you assuming that the matchmaking is placing you with people far below your skill level when in fact it's designed to do the exact opposite.

I never spoke of solo Q. GW2 isn't the only game with a ranking and group matchmaking system.When I am joining a gold 3 match with a firebrand who chases to kill a bunker druid on the road, or a DH that fights a 2v1 10 meters away (from me standing on the point) and dies, and many more such examples, I can safely 'assume' I am being paired with people below my skill level. I can't see the 'exact opposite' scope in a MM design where (even in extreme cases) 200+ rating is allowed for a team to be formed.

What do you mean you're not talking about solo q. Everything that you're complaining about is specifically linked to queing solo and getting "random" teammates lol.

If you could que up in 5 man teams in ranked then everything would be ok for you?

[Faux Play.6104]

@Ithilwen.1529 said:@"Ben Phongluangtham.1065"

I have been reading the C# implementation for Glicko 2 available on GitHub. It doesn't appear to actually follow the glicko 2 algorithm as published, though I admit.. I'm not skilled in Calculus.

Far more important is what's found here: https://www.guildwars2.com/en/news/finding-the-perfect-match/ .

If you look at the chart, found mid page, you will find that a mismatch loss is penalized MORE than a close loss. This is essentially the same as the s1-s2 win streaking and heavily penalizing losses. In my opinion the thinking is fundamentally flawed.

A fair system would do the opposite. In a heavy mismatch, the losing side would suffer lower penalties.

You are misreading how it works. if a team with an average rank of 1600 plays a team with an average rank of 1400, the 1600 team could win 8 pts but lose 18. The 1400 team could win 18 but will only lose 8. That is a very rough estimate assuming a rating that has settled.

Where the system has trouble matchmaking. If people on a team are a wide spread of skill levels, the average rating for a team becomes less accurate.

[SlippyCheeze.5483]

@Faux Play.6104 said:

@Ithilwen.1529 said:@"Ben Phongluangtham.1065"

I have been reading the C# implementation for Glicko 2 available on GitHub.

It doesn't appear to actually follow the glicko 2 algorithm as published,

though I admit.. I'm not skilled in Calculus.

Far more important is what's found here:

https://www.guildwars2.com/en/news/finding-the-perfect-match/

.

If you look at the chart, found mid page, you will find that a mismatch loss is penalized MORE than a close loss.

This is essentially the same as the s1-s2 win streaking and heavily penalizing losses. In my opinion the thinking is fundamentally flawed.

A fair system would do the opposite. In a heavy mismatch, the losing side would suffer lower penalties.

You are misreading how it works.

As mentioned above, this is also inaccurate at this point in time: the system that chart references was deprecated and is no longer part of PvP.

[Pinkunicorn of Dethecus.32]

@Brother.1504 said:In the end the rating system needs to convey a general feeling of satisfaction or qualifiable fairness with an out come. Not I win yay 3 matches in a row + 5, +4,+5. I lose 1 match, -24, kitten!! If anet wants to retain players in this mode the rating system needs to feel good.

This is why ranking doesn't matter at all. It's luck of the draw whether you can maintain a 2/3s win-rate with random pugs. I get a few bad matches and I'm ~-60 pts down, meaning I need to win at least 10 more, in a row, just to get back to where I started. There's no point in caring about rating. At all. I only queue on my main in primetime, and I can literally freefall from 1730 to 1600 with people making the most basic of mistakes. It's a farce.

[Faux Play.6104]

@SlippyCheeze.5483 said:

@Faux Play.6104 said:

@Ithilwen.1529 said:@"Ben Phongluangtham.1065"

I have been reading the C# implementation for Glicko 2 available on GitHub.

It doesn't appear to actually follow the glicko 2 algorithm as published,

though I admit.. I'm not skilled in Calculus.

Far more important is what's found here:

https://www.guildwars2.com/en/news/finding-the-perfect-match/

.

If you look at the chart, found mid page, you will find that a mismatch loss is penalized MORE than a close loss.

This is essentially the same as the s1-s2 win streaking and heavily penalizing losses. In my opinion the thinking is fundamentally flawed.

A fair system would do the opposite. In a heavy mismatch, the losing side would suffer lower penalties.

You are misreading how it works.

As mentioned above, this is also inaccurate at this point in time: the system that chart references was deprecated and is no longer part of PvP.

Not sure what chart you are talking about. The system uses Glicko2. The numbers I put in were rough estimates.

https://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm

[Brother.1504]

Is anyone satisfied with gw2 spvp matchmaking?

[GoQ.1048]

frustration snowballed and later on manifested as an excaptable sideeffect - it's to late to be.

behind our screens we all humans. a system and society based on good & bad judgement doesnt acknowledge that. u take on a role in pvp and by that the human is placed in a very small frame of what he actually is. and when the match is over u get judged. you are either good or bad / usefull or useless. anybodies ego will see through and prevent that. you either turn your back on such a mentaly ill environment or you get sarcastic first and cynic later on. we all know instantaniously that a small (good or bad fullfilled) role doesnt define us and seperate ourselfs from that system the one way or the other. those who continue to play it are basicly without empathy and sympathy for anybody but themselfs.

a judgement of a whole person - and any gamer is that - in good or bad - will lead to frustration. and that frustration snowballed for so long now it became part of the system and manifested into subconscious social conformaty. you except it or you can not take part.

right now there is no reward for those matches you are forced to carry but at the same time you get bombarded by the sheer amount of those matches and therefore driven even faster of seperating urself from the judgement at the end of the match.

i was an high lvl athlet back than and i couldnt live without a competetive field. but after a competition not only your body but also your mind needed a rest. ingame you can just have competition after competition and your mind gets drained by that and the chip on your shoulder gets louder and louder.competition - with exception of very high skilled ppl - negates playfullness. you try to be on peak and act within your frame at the best manner possible. and we all grew up by knowing that games are a playfull learning experience. if you are far ahead of any competition you can get that playfullness back to a certain extend but all in all competition negates anything that is gaming.

and ontop of all that frustration prevents you from getting solutions. anger hormons block the part of our brain that is used for creativity - a needed tool to find solutions to problems cause else u only act upon instinct. it's known as frontal lope shutdown.

maybe a cap of 2 ranked games per day would prevent our animals to come out. this + a better reward for carry games.

[SlippyCheeze.5483]

@Faux Play.6104 said:

@SlippyCheeze.5483 said:

@Faux Play.6104 said:

@Ithilwen.1529 said:@"Ben Phongluangtham.1065"

I have been reading the C# implementation for Glicko 2 available on GitHub.

It doesn't appear to actually follow the glicko 2 algorithm as published,

though I admit.. I'm not skilled in Calculus.

Far more important is what's found here:

https://www.guildwars2.com/en/news/finding-the-perfect-match/

.

If you look at the chart, found mid page, you will find that a mismatch loss is penalized MORE than a close loss.

This is essentially the same as the s1-s2 win streaking and heavily penalizing losses. In my opinion the thinking is fundamentally flawed.

A fair system would do the opposite. In a heavy mismatch, the losing side would suffer lower penalties.

You are misreading how it works.

As mentioned above, this is also inaccurate at this point in time: the system that chart references was deprecated and is no longer part of PvP.

Not sure what chart you are talking about. The system uses Glicko2. The numbers I put in were rough estimates.

https://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm

Oh. The comment about "look at the chart, found mid page", which references the older match prediction / ladder systems, and is not relevant. The underlying algorithm is indeed glicko2, as documented in https://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm along with the constants they use.

[Sampson.2403]

@Brother.1504 said:Is anyone satisfied with gw2 spvp matchmaking?

What alternative is there? For the MM system to only create games where all 5 players are super close to each other in MMR? That would create insanely long que times. This would be like Lebron James refusing to play another basketball game unless everyone else on his team is as good as he is. It's just simply unrealistic.

I do think that the points won / lost could be revisited though. No matter how you slice it, having no choice but to que up solo into random team games where you can potentially lose 5 times more points than you can win isn't very healthy.

[milego.4830]

As I said before, this rating system needs to go. It's disheartening and actually rewards people that play less games.

[lLobo.7960]

@Ithilwen.1529 said:If you look at the chart, found mid page, you will find that a mismatch loss is penalized MORE than a close loss. This is essentially the same as the s1-s2 win streaking and heavily penalizing losses. In my opinion the thinking is fundamentally flawed.

A fair system would do the opposite. In a heavy mismatch, the losing side would suffer lower penalties.

There are a series of issues here.One is that people see rating as a reward and that they should be rewarded more for a greater effort.Rating is not a rewardThe idea of having greater losses on blowouts is not punish you, is to lower your rating so you go down to a tier where you don't get blowouts and can have more competitive matches.Same with low rating climbs for close wins. The idea is to keep you in a tier where you are having competitive matches.If you win a blowout you should be awarded a good rating so you don't ruin the matches of people on that tier...But to accept this people need to understand that their rating gains and losses are not rewards or penalties. Its just a system to try to give you competitive matches.

Rewards for close games, come backs and all those things should be given as pips, chests and other things.

[Bigbosos.2063]

@"AngelLovesFredrik.6741" said:

@SlippyCheeze.5483 said:

@Ithilwen.1529 said:

@"AngelLovesFredrik.6741" said:

@Ithilwen.1529 said:The problem isn't the system, it's the configuration.

When I played rated Chess, I wasn't put against Master rated players. It wasn't a fair match.

In the instance that I was matched against a rated Master.. I actually gained rating for losing, because of the mismatch.

ANET doesn't appear to allow gain of points for losing a mismatch, nor losing points for a mismatch win. That's the first problem.

This is an important aspect of the rating system that is not being used.

So, the ratings are distorted from where they would be.

The second issue is that ANET permits too wide a difference in player skill levels in one match. This goes back to the first point.

The third and lesser issue is that the classes

still

aren't genuinely balanced. That's a different discussion.

TLDR: The rating system doesn't appear to be being used as designed. So, the results tend to be distorted.

Gaining points for losing and losing points for winning should never be a part of anything competitive.

I've been reading up on glicko-2 and reading C# code for implementing it. I'm no longer sure I'm right on what I said before.

GLICKO-2 is designed so that you only ever gain points on a win, and lose points on a loss, with the magnitude of change determined by the question "was this result the one we expected?"

Gaining small amounts, and losing large amounts, means that the matchmaking system is consistently putting you in matches you are expected to win -- basically, instead of a "fair" distribution of matches, with a 50/50 expectation that you will win or lose, you are in a biased distribution where you have a 100 percent expectation that you will win. Congratulations, the game tipped the scales

in your favour

.

So, when you match expectations and win you gain a small amount, and when you break expectations and lose you lose a large amount. If you were in a "fair" distribution of matches, you would also see small losses of rating (expected to lose, and did) and at least sometimes large gains (expected to lose, but won), but ... the evidence you supply says that the game actually gives you what it expects to be easy wins all the time.

Congratulations. All y'all playing PvP on easy mode, I guess. (or, IDK, maybe y'all forgetting when you lose small, or win big, because they don't stand out in your memory.)

This is not the full truth.The version of Glicko-2 that Guild Wars is operating, doesn't take team rating into account. This means that if I queue at 1800 rating off-peak and the closest players to my rating are hovering around 1650, I am expected to win every single game. Further-more. To balance out my 1800 rating, people with 1500 rating gets put into my team against the middle 1650 players. This makes it so that the legend player gets put in a seemlingly un-winnable scenario while the match-maker (Glicko-2) sees this as a 100% guaranteed win.What needs to be done instead of this is that Glicko counts the

Average Rating

of

Everyone

in the team and distribute the won / lost rating evenly.

IIRC, don't remember which one of ANet employees once said the matchmaking ensures that the average rating of teams in a match never differs more than 50 points. The issue is that they specifically underlined that standard deviation is still an issue and unaccounted for.

What you said is extremely correct, and your solution makes much more sense from a statitics and logical point of view than just accounting for the average team rating.

[Bigbosos.2063]

@Dreddo.9865 said:IMHO the MM engine should value more total ranked matches played. That should be the decisive factor when picking players to form a team. Then of course personal performance should also count, and maybe they could present this indirectly like a small + on the rank points gained for win or a small - for points lost for a loss.

No because alt accounts.

[SlippyCheeze.5483]

@Bigbosos.2063 said:

@"AngelLovesFredrik.6741" said:

@SlippyCheeze.5483 said:

@Ithilwen.1529 said:

@"AngelLovesFredrik.6741" said:

@Ithilwen.1529 said:The problem isn't the system, it's the configuration.

When I played rated Chess, I wasn't put against Master rated players. It wasn't a fair match.

In the instance that I was matched against a rated Master.. I actually gained rating for losing, because of the mismatch.

ANET doesn't appear to allow gain of points for losing a mismatch, nor losing points for a mismatch win. That's the first problem.

This is an important aspect of the rating system that is not being used.

So, the ratings are distorted from where they would be.

The second issue is that ANET permits too wide a difference in player skill levels in one match. This goes back to the first point.

The third and lesser issue is that the classes

still

aren't genuinely balanced. That's a different discussion.

TLDR: The rating system doesn't appear to be being used as designed. So, the results tend to be distorted.

Gaining points for losing and losing points for winning should never be a part of anything competitive.

I've been reading up on glicko-2 and reading C# code for implementing it. I'm no longer sure I'm right on what I said before.

GLICKO-2 is designed so that you only ever gain points on a win, and lose points on a loss, with the magnitude of change determined by the question "was this result the one we expected?"

Gaining small amounts, and losing large amounts, means that the matchmaking system is consistently putting you in matches you are expected to win -- basically, instead of a "fair" distribution of matches, with a 50/50 expectation that you will win or lose, you are in a biased distribution where you have a 100 percent expectation that you will win. Congratulations, the game tipped the scales

in your favour

.

So, when you match expectations and win you gain a small amount, and when you break expectations and lose you lose a large amount. If you were in a "fair" distribution of matches, you would also see small losses of rating (expected to lose, and did) and at least sometimes large gains (expected to lose, but won), but ... the evidence you supply says that the game actually gives you what it expects to be easy wins all the time.

Congratulations. All y'all playing PvP on easy mode, I guess. (or, IDK, maybe y'all forgetting when you lose small, or win big, because they don't stand out in your memory.)

This is not the full truth.The version of Glicko-2 that Guild Wars is operating, doesn't take team rating into account. This means that if I queue at 1800 rating off-peak and the closest players to my rating are hovering around 1650, I am expected to win every single game. Further-more. To balance out my 1800 rating, people with 1500 rating gets put into my team against the middle 1650 players. This makes it so that the legend player gets put in a seemlingly un-winnable scenario while the match-maker (Glicko-2) sees this as a 100% guaranteed win.What needs to be done instead of this is that Glicko counts the

Average Rating

of

Everyone

in the team and distribute the won / lost rating evenly.

IIRC, don't remember which one of ANet employees once said the matchmaking ensures that the average rating of teams in a match never differs more than 50 points. The issue is that they specifically underlined that standard deviation is still an issue and unaccounted for.

You can check out the details, including the constants, over on the wiki if you want. It's all publicly documented, and that page is maintained by the ANet PvP developers, not by player guesswork or anything.

[Faux Play.6104]

@Sampson.2403 said:

@Brother.1504 said:Is anyone satisfied with gw2 spvp matchmaking?

What alternative is there? For the MM system to only create games where all 5 players are super close to each other in MMR? That would create insanely long que times. This would be like Lebron James refusing to play another basketball game unless everyone else on his team is as good as he is. It's just simply unrealistic.

I do think that the points won / lost could be revisited though. No matter how you slice it, having no choice but to que up solo into random team games where you can potentially lose 5 times more points than you can win isn't very healthy.

You can have top 10 people on the same team as gold 1. That is like James waiting 2 minutes to get queued with a random 7th grader vs waiting 10 minutes to queue with someone at D1 college level. The reason top people lose so many pts for a loss is the average team ratings are so low.