Idea for a better (extended) match making algorithm

[yujo.3821]

Hi,

after venting a bit off and online about the bizarre forms that match making seems to take from time to time, I thought I should start something constructive.Maybe I am naive and maybe it has all been discussed already but I thought I'd just share my idea and maybe some dev will take it up and try it out or tell me why it won't work.

So what is match making and what is a fair match making in a pvp game?For any algorithm to work on a broad scale you need info about the players and their skill level to match them up evenly.

I'd really love to know the criteria of how gw2's match making works and in which order it prioritizes. As far as I know they use glicko2 and I assume some additions to it that make class stacking more unlikely.

My suggestion is:Could you just widen the base of information that your algorithm uses? Make the skill rating more personal! A win is still important but sometimes even the best players cannnot carry 4 others to a victory. But right now the system counts having four bags of sand as team mates as a skill loss.

You could use the info you already have in the game. It's at the end of each game. It's an information about what really happened in that game. Top stats.Top stats involve:

• Damage: you were alive and you played a build that could do the most damage on your team -> this measures player skill!
• Healing: you were alive and you played a build that could heal yourself and others the most on your team-> see above
• Offfense: you were alive and you played a build that enabled you to contest or capture enemy points -> see above
• Defense: you were alive and you played a build that enabled you to contest or defend capped points -> see above
• Kills: you were alive and you played a build that enabled you to take part in killing the most enemies -> see above
• Revives: you were alive and you played a build that enabled you to revive downed players; it measures your awareness and teamplay as well-> see above
• Deaths: you were killed; this could substract points in the new algorithm: see above

To finish this with a real life example where this is already being done. Hockey (NHL): +- stats -> the player is awarded plus and minus points for when a goal happens while he is on the field.If the opponent scores: -If your own team scores: +

Could some Anet dev please comment on this? Are you even able to create your own algorithm or do you have to use a given one like glicko? Can you add to glicko?There are some possibilities: One is to improve the overall match making. The other is to display the player skill rating more accurately. No matter the outcome of the match.

[Sigmoid.7082]

Top stats and not indicative of player skill and giving them any sort of merit would serve to be aweful due to players chasing them instead of doing what is actually needed to win a match.

[Huskyboy.1053]

This would be feasible if they actually assigned specific roles to people, maybe based on their rune/amulet combination. From there you should expect someone to hit certain benchmarks, i.e. my support scrapper should literally have 0 deaths, whereas my d/p thief can expect to have at least one death and should probably get top kills. Top damage should go to a scourge on Scavenger/Deadshot, and so on. As it is, without some idea of what you're supposed to be doing there's no way to validate whether you actually performed as an individual.

And if you expect Anet to set aside a boatload of time to do what I just suggested, I think you are sorely mistaken.

[SlippyCheeze.5483]

@"yujo.3821" said:For any algorithm to work on a broad scale you need info about the players and their skill level to match them up evenly.

Correct.

I'd really love to know the criteria of how gw2's match making works and in which order it prioritizes. As far as I know they use glicko2 and I assume some additions to it that make class stacking more unlikely.

https://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm tells you more or less everything you need to know.

Could you just widen the base of information that your algorithm uses? Make the skill rating more personal! A win is still important but sometimes even the best players cannnot carry 4 others to a victory. But right now the system counts having four bags of sand as team mates as a skill loss.

You mistakenly believe that this distorts rankings; if four "bags of sand" on your team is common, then it is even more common to face four (or five) of them on the other team, leading you to more victories than your skill level should deliver, not less.

Hope that helps.

[yujo.3821]

@Sigmoid.7082 said:Top stats and not indicative of player skill and giving them any sort of merit would serve to be aweful due to players chasing them instead of doing what is actually needed to win a match.

That's your opinion. I think they are indicative of a player's skill. But I agree with you that it could lead to misuse by players trying to only achieve them.Then again, I never argued that this should be the sole rating system for a player. How about a combination? Winning is still paramount to gaining ranking. But the large loss could be softened by personal stats. I assume you know this experience when you carried a game, got all the top stats and still lost because your team mates clearly weren't good enough for the level of the match. The combination of both ratings could mean that you wouldn't receive e.g. -13 points but instead got a nice 0 or maybe even a +1.Wouldn't that be awesome?

[yujo.3821]

@SlippyCheeze.5483 said:

@"yujo.3821" said:For any algorithm to work on a broad scale you need info about the players and their skill level to match them up evenly.

Correct.

I'd really love to know the criteria of how gw2's match making works and in which order it prioritizes. As far as I know they use glicko2 and I assume some additions to it that make class stacking more unlikely.

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

tells you more or less everything you need to know.

Could you just widen the base of information that your algorithm uses? Make the skill rating more personal! A win is still important but sometimes even the best players cannnot carry 4 others to a victory. But right now the system counts having four bags of sand as team mates as a skill loss.

You mistakenly believe that this distorts rankings; if four "bags of sand" on your team is common, then it is even more common to face four (or five) of them on the other team, leading you to

more

victories than your skill level should deliver, not less.

Hope that helps.

Your logic is incorrect. Why would someone other have it harder than me? I don't think so. I am not talking about just winning and that I feel at a disadvantage. I am talking about match quality. I don't assume I am the only one that experiences this. I very much agree that this is the reality for many other players. In fact I do know so by watching twitch streams of good players having the same experience albeit on a different skill level. I am talking about people not being matched correctly and this happens stochastically. It's not a simple: Team A is mismatched and also Team B is mismatched.It's more like Team A is mismatched in itself by the skill level of the players on its own team. While also being matched up against a team that may or may not be on an equal level. The calculation of the skill level is the fault here. You can very easily corrupt the outcome of an algorithm by giving it faulty input. And that faulty ingredient is the pseudo skill level of the player the system assumes.

[SlippyCheeze.5483]

@yujo.3821 said:

@SlippyCheeze.5483 said:

@yujo.3821 said:For any algorithm to work on a broad scale you need info about the players and their skill level to match them up evenly.

Correct.

I'd really love to know the criteria of how gw2's match making works and in which order it prioritizes. As far as I know they use glicko2 and I assume some additions to it that make class stacking more unlikely.

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

tells you more or less everything you need to know.

Could you just widen the base of information that your algorithm uses? Make the skill rating more personal! A win is still important but sometimes even the best players cannnot carry 4 others to a victory. But right now the system counts having four bags of sand as team mates as a skill loss.

You mistakenly believe that this distorts rankings; if four "bags of sand" on your team is common, then it is even more common to face four (or five) of them on the other team, leading you to

more

victories than your skill level should deliver, not less.

Hope that helps.

Your logic is incorrect. Why would someone other have it harder than me?

Simple: there are ten slots allocated in a match, five to a side, right? On your side, one slot is always filled with a "not a bag of sand" player, you. That means when a "bag of sand" is added the odds are 5/9 that they place on the other team, and 4/9 they place on your team.

That is how you are more likely to face a team with "bags of sand" than to be part of a team with them: as long as the invariant "you are not a bag of sand" remains true, statistically you are more likely to face them than fight with them.

I am not talking about just winning and that I feel at a disadvantage. I am talking about match quality.

Then you are not talking about the MMR at all, as it does not even remotely attempt to model "quality", only the probability of victory or loss. That is a perfectly reasonable thing to talk about, and who knows, a matchmaking system that was focused on quality would be extremely interesting, provided you have a suitable definition of quality other than the existing MMR systems "a balanced match is a quality match".

It's more like Team A is mismatched in itself by the skill level of the players on its own team. While also being matched up against a team that may or may not be on an equal level. The calculation of the skill level is the fault here. You can very easily corrupt the outcome of an algorithm by giving it faulty input. And that faulty ingredient is the pseudo skill level of the player the system assumes.

OK. How is the concrete system outlined in the wiki using "faulty input", then? Specifics, not just hand-waving, please.

[HeadCrowned.6834]

This has been discussed before. Top stats won't work as an indication of how well a player played during a match. Its too highly dependent on professions for example. Some professions can get 4 or 5 top stats in one match, others max 2 or if lucky, 3.

[BikeIsGone.8675]

sry to burst your bubble, but having a matchmaker have to take builds into consideration kills this "suggestion".Having a system accurately evaluate every possible build for a class is physically impossible.

[yujo.3821]

@"BikeIsGone.8675" said:sry to burst your bubble, but having a matchmaker have to take builds into consideration kills this "suggestion".Having a system accurately evaluate every possible build for a class is physically impossible.

Never said that. It doesn't consider the build per se. Read again.

[yujo.3821]

@HeadCrowned.6834 said:This has been discussed before. Top stats won't work as an indication of how well a player played during a match. Its too highly dependent on professions for example. Some professions can get 4 or 5 top stats in one match, others max 2 or if lucky, 3.

I agree that some classes have it harder than others like thief for example. But why are they not an indicator? They sure are. I mostly play a bruiser build. On higher rankings like plat 2 I wont get more than 1 or 2 top stats if at all. On lower rankings I sometimes get 4-6 top stats. My point is still valid. A bad player on a build that carries him will still get demolished by good players. Right now people complain about op scourges and mesmers and many players play the meta build of a specific profession....As i said in my original post wins should still matter the most, I just don't like it that your personal skill level doesn't matter directly for your ranking. It's mostly the luck of the team mates you get.

[MithranArkanere.8957]

What we need isn't better matchmaking, it's a drafting process so we can ban problematic builds at the start of ranked and tournament matches, since balance updates take ages to deal with them, but players can identify them quickly in just a few matches after every update.

But to be able to have a drafting process, people would have to be able to queue with multiple characters, multiple pre-defined builds. Not just the current character.And to be able to have multiple predefined builds, we'd need some sort of template system to store them, and some panel to list the ones we want to queue with.

Once we have templates, a way to queue with pairs of characters and the template linked to them instead only the current character, and a drafting process at the start of the match to get rid of problematic builds, matches will begin to be less one-sided.

There would have to be many other details, like considering builds that are very similar as the same build so people don't just queue with 20 variations of one build.And once the drafting process is over, the current characters and their builds would have to be locked until the end of the match.

But we'd never get any close to balanced matches until part of it is left on the hands of the players. You see it in every player-organized tournament and competition for every game. There's always some banned character, skill weapon or build because players always get better at identifying that kind of issue.It all boils down to "either everyone uses it and make the game less fun, or we just get rid of it and force people to actually develop skills to get fun matches".

Changes in matchmaking will not work better for that than a drafting process.

[yujo.3821]

@SlippyCheeze.5483 said:

@yujo.3821 said:

@SlippyCheeze.5483 said:

@yujo.3821 said:For any algorithm to work on a broad scale you need info about the players and their skill level to match them up evenly.

Correct.

I'd really love to know the criteria of how gw2's match making works and in which order it prioritizes. As far as I know they use glicko2 and I assume some additions to it that make class stacking more unlikely.

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

tells you more or less everything you need to know.

Could you just widen the base of information that your algorithm uses? Make the skill rating more personal! A win is still important but sometimes even the best players cannnot carry 4 others to a victory. But right now the system counts having four bags of sand as team mates as a skill loss.

You mistakenly believe that this distorts rankings; if four "bags of sand" on your team is common, then it is even more common to face four (or five) of them on the other team, leading you to

more

victories than your skill level should deliver, not less.

Hope that helps.

Your logic is incorrect. Why would someone other have it harder than me?

Simple: there are ten slots allocated in a match, five to a side, right? On your side, one slot is always filled with a "not a bag of sand" player, you. That means when a "bag of sand" is added the odds are 5/9 that they place on the other team, and 4/9 they place on your team.

That is how you are more likely to face a team with "bags of sand" than to be part of a team with them: as long as the invariant "you are not a bag of sand" remains true, statistically you are more likely to face them than fight with them.

I am not talking about just winning and that I feel at a disadvantage. I am talking about match

quality

.

Then you are not talking about the MMR at all, as it does not even remotely attempt to model "quality", only the probability of victory or loss. That is a perfectly reasonable thing to talk about, and who knows, a matchmaking system that

was

focused on quality would be extremely interesting, provided you have a suitable definition of quality other than the existing MMR systems "a balanced match is a quality match".

It's more like Team A is mismatched in itself by the skill level of the players on its own team. While also being matched up against a team that may or may not be on an equal level. The calculation of the skill level is the fault here. You can very easily corrupt the outcome of an algorithm by giving it faulty input. And that faulty ingredient is the pseudo skill level of the player the system assumes.

OK. How is the concrete system outlined in the wiki using "faulty input", then? Specifics, not just hand-waving, please.

Your premise is still wrong: You assume that there are ONLY bags of sand and THE player. :-) The problem of a low player base and an algorithm that tries to match players on a team and against a team in a given time indicates that it works on wrong numbers. Complaints about skill gaps of team mates and match results with 500-200 show that this algorithm doesn't create equality or a 50/50 outcome.I agree that quality is not always equality. But the mmr tries to find balance but fails if the data pool is too shallow or the input doesn't match the output.I can see your point that a strictly balanced game that considers the mean skill level should be equal. I just think that by strictly and only drawing your input from the wins and losses doesn't paint the whole picture.

I don't understand your last question. Didn't I already say it? My point is that it uses the input skill level which is only defined by wins. This skill level as an input ignores the simple fact that a player can be dragged down into mmr oblivion if he plays a game with players on his team that are way below his level and loses. His team loses -> God of PvP is a bad player. Skill goes down. MMR! <- faulty

If you added some form of personal stats into the equation then the outcome would represent the reality of the game. While a good player would still remain a good player. A bad player would become more obvious to the system. Not just by collecting death penalties or not doing damage or killing (because he is insta down anyway) but also because the gap in personal points that are collected through playing well (Top stats or sth similar) will widen the gap between bad and good players. Right now the mmr dictates match ups that are only based on wins. It totally ignores players that were carried by better players. A win is a win. For the bad and for the good alike.

I don't know about you but I notice a distinct increase in skill level and match quality in plat 2 compared to the lower divisions. It's more fun to play for me. Closer matches as well.

[HeadCrowned.6834]

@yujo.3821 said:

@HeadCrowned.6834 said:This has been discussed before. Top stats won't work as an indication of how well a player played during a match. Its too highly dependent on professions for example. Some professions can get 4 or 5 top stats in one match, others max 2 or if lucky, 3.

I agree that some classes have it harder than others like thief for example. But why are they not an indicator? They sure are. I mostly play a bruiser build. On higher rankings like plat 2 I wont get more than 1 or 2 top stats if at all. On lower rankings I sometimes get 4-6 top stats. My point is still valid. A bad player on a build that carries him will still get demolished by good players. Right now people complain about op scourges and mesmers and many players play the meta build of a specific profession....As i said in my original post wins should still matter the most, I just don't like it that your personal skill level doesn't matter directly for your ranking. It's mostly the luck of the team mates you get.

A lot of good thieves will get none or maybe 1 top stat. Why would it be fair to reward them less than for example Druid, which can get 4 top stats? Different classes have different roles, and Top stats do not cover all roles. Ergo, Top stats aren't a good indicator of players performance. The only matches in which it COULD be a clear performance indicator are those matches that have 5 top stats awarded to 1 player or a lot of different non-synergetic top stats (such as most healing and most damage) to 1 player.

[SlippyCheeze.5483]

@yujo.3821 said:

@SlippyCheeze.5483 said:

@yujo.3821 said:

@SlippyCheeze.5483 said:

@yujo.3821 said:For any algorithm to work on a broad scale you need info about the players and their skill level to match them up evenly.

Correct.

I'd really love to know the criteria of how gw2's match making works and in which order it prioritizes. As far as I know they use glicko2 and I assume some additions to it that make class stacking more unlikely.

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

tells you more or less everything you need to know.

Could you just widen the base of information that your algorithm uses? Make the skill rating more personal! A win is still important but sometimes even the best players cannnot carry 4 others to a victory. But right now the system counts having four bags of sand as team mates as a skill loss.

You mistakenly believe that this distorts rankings; if four "bags of sand" on your team is common, then it is even more common to face four (or five) of them on the other team, leading you to

more

victories than your skill level should deliver, not less.

Hope that helps.

Your logic is incorrect. Why would someone other have it harder than me?

Simple: there are ten slots allocated in a match, five to a side, right? On your side, one slot is always filled with a "not a bag of sand" player, you. That means when a "bag of sand" is added the odds are 5/9 that they place on the other team, and 4/9 they place on your team.

That is how you are more likely to face a team with "bags of sand" than to be part of a team with them: as long as the invariant "you are not a bag of sand" remains true, statistically you are more likely to face them than fight with them.

I am not talking about just winning and that I feel at a disadvantage. I am talking about match

quality

.

Then you are not talking about the MMR at all, as it does not even remotely attempt to model "quality", only the probability of victory or loss. That is a perfectly reasonable thing to talk about, and who knows, a matchmaking system that

was

focused on quality would be extremely interesting, provided you have a suitable definition of quality other than the existing MMR systems "a balanced match is a quality match".

It's more like Team A is mismatched in itself by the skill level of the players on its own team. While also being matched up against a team that may or may not be on an equal level. The calculation of the skill level is the fault here. You can very easily corrupt the outcome of an algorithm by giving it faulty input. And that faulty ingredient is the pseudo skill level of the player the system assumes.

OK. How is the concrete system outlined in the wiki using "faulty input", then? Specifics, not just hand-waving, please.

Your premise is still wrong: You assume that there are ONLY bags of sand and THE player. :-)

Nope: If you have an even number of good players, the most likely outcome of random distribution is that both sides have the same number of good players. If you have an odd number of good players, there is a 50:50 chance that you get the advantage (+1 good player) or the disadvantage (-1 good player) in the match.

There is one exception to that rule: if there is only one good player, you, in which case you have the advantage automatically.

So, over the course of multiple matches, regardless of the number of good players in each, you will statistically prefer balanced matches ... except if you are the only good player, in which case you tilt the odds in your own favor. That is a very minor overall effect, but there are zero cases where, on average, you have more "bad players meaning you lose" than "good players meaning you win" cases.

The problem of a low player base and an algorithm that tries to match players on a team and against a team in a given time indicates that it works on wrong numbers. Complaints about skill gaps of team mates and match results with 500-200 show that this algorithm doesn't create equality or a 50/50 outcome.

The exact same logic applies to "outlier MMR" players as "bad" players: they may be bad only in the context of this match, and only because they are outskilled by the opposition, or vice-versa, they may "only" be "good" players because they outskill everyone else...

I don't understand your last question. Didn't I already say it? My point is that it uses the input skill level which is only defined by wins. This skill level as an input ignores the simple fact that a player can be dragged down into mmr oblivion if he plays a game with players on his team that are way below his level and loses. His team loses -> God of PvP is a bad player. Skill goes down. MMR! <- faulty

The system anticipates that, and uses two inputs to determine MMR changes: the predicted outcome, and the actual outcome. The actual outcome determines the direction of the change to MMR (eg: win, MMR rises, lose, MMR drops), but the predicted outcome determines the magnitude of the change (eg: prediction is correct, small change, prediction is wrong, large change.)

I don't know about you but I notice a distinct increase in skill level and match quality in plat 2 compared to the lower divisions. It's more fun to play for me. Closer matches as well.

There should be: as skill rises, quality will definitely increase. I suspect the distinct increase is more a case of "large enough change to be noticed", but who knows, maybe there is a large cluster of players who get that far, and then get "stuck" there because that reflects their level of skill compared to the people in that bracket and above.

Good enough to be good, not good enough to be better than the rest of the good players. :)