Suggestion to make people play more games

[Megametzler.5729]

Okay, we all know that a lot of people camp spots on the leaderboard and refuse to play (significantly) more games than 120. This is mostly due to the volatility of the ranking, is it not?

 

So why don't we extend the system that's being used for placement matches and those few matches afterwards? The more games you play, the lower the impact of ranking is. That should be easily implemented by a dampening function:

point_difference = point_difference_at_game_zero / number_of_games (or similarish).

It would not be a complex fix of the ranking system, but it might be an easy and quick one to let people play more.

 

An example: You are at1700. You lose one game. If it was you 11th game, you lose 20 points. If it was your 100th game, you only los 10 points. If it was you 200th you only lose like 5 points. It would not solve the impact of losses in higher tiers, the same would apply for wins. But it would encourage you to keep playing when you have you 120 games and have your high rating.

 

Thoughts? Critics? Improvements?

[Abazigal.3679]

At some point, people spamming games are going to end up either gaining +15, either losing -5. This implies :

- Some players are going to climb higher than they should ( legendary should be no problem..)

- Platinum games are going to look same as silver/gold

- People afking or abandonning games at 300-150, since they only lose -5 anyway

 

Your suggestion " could " be good, but won't really fix the real problem which is match quality and low population. People do not play much at 1500 + because there aren't many players at that rank, and you get frustrating loses quite often.

[DragonFury.6243]

> @"Megametzler.5729" said:

> Okay, we all know that a lot of people camp spots on the leaderboard and refuse to play (significantly) more games than 120. This is mostly due to the volatility of the ranking, is it not?

>

> So why don't we extend the system that's being used for placement matches and those few matches afterwards? The more games you play, the lower the impact of ranking is. That should be easily implemented by a dampening function:

> point_difference = point_difference_at_game_zero / number_of_games (or similarish).

> It would not be a complex fix of the ranking system, but it might be an easy and quick one to let people play more.

>

> An example: You are at1700. You lose one game. If it was you 11th game, you lose 20 points. If it was your 100th game, you only los 10 points. If it was you 200th you only lose like 5 points. It would not solve the impact of losses in higher tiers, the same would apply for wins. But it would encourage you to keep playing when you have you 120 games and have your high rating.

>

> Thoughts? Critics? Improvements?

 

As long As their is 250 player in the leaderboard i dont think ANET will care to do all that

[ReaverKane.7598]

It less about volatility but compression...

Seasons are compressed into a tiny window, which leads to a grind mentality, and less chance to make up for slip-ups.

That coupled with the uncertainty of who you'll be placed with leads to the mentality.

Basically the whole league system is broken and poorly managed, which leads to a bad experience, and toxic behaviour.

If you want improvements, seasons need to be longer, teams need to be brought back, and wins/rating need to impact your rewards.

[Megametzler.5729]

> @"Abazigal.3679" said:

> At some point, people spamming games are going to end up either gaining +15, either losing -5. This implies :

> - Some players are going to climb higher than they should ( legendary should be no problem..)

> - Platinum games are going to look same as silver/gold

> - People afking or abandonning games at 300-150, since they only lose -5 anyway

>

> Your suggestion " could " be good, but won't really fix the real problem which is match quality and low population. People do not play much at 1500 + because there aren't many players at that rank, and you get frustrating loses quite often.

 

No no, the same would apply for wins, of course. It would just reduce volatility for more games played.

[pah.4931]

How To Make People Play More Games, Game Development 101:

 

1) Make the game as fun as possible.

 

2) Remove systems that actually encourage inactivity (bogus, terrible ranking system)

 

 

Thank you.

[Abazigal.3679]

> @"ReaverKane.7598" said:

> It less about volatility but compression...

> Seasons are compressed into a tiny window, which leads to a grind mentality, and less chance to make up for slip-ups.

> That coupled with the uncertainty of who you'll be placed with leads to the mentality.

> Basically the whole league system is broken and poorly managed, which leads to a bad experience, and toxic behaviour.

> If you want improvements, seasons need to be longer, teams need to be brought back, and wins/rating need to impact your rewards.

 

This x1000. It's however going to take a while until people consider taking off the season system, since it happens in almost every single game.

[SlippyCheeze.5483]

> @"Megametzler.5729" said:

> Okay, we all know that a lot of people camp spots on the leaderboard and refuse to play (significantly) more games than 120. This is mostly due to the volatility of the ranking, is it not?

>

> So why don't we extend the system that's being used for placement matches and those few matches afterwards? The more games you play, the lower the impact of ranking is. That should be easily implemented by a dampening function:

> point_difference = point_difference_at_game_zero / number_of_games (or similarish).

> It would not be a complex fix of the ranking system, but it might be an easy and quick one to let people play more.

 

I always find suggestions like this really fun. Your idea is great. Wonderful. Amazing.

 

Your idea is also the exact change that differentiates GLICKO-2 (used by ANet) from ELO, which is that it reduces the changes in ranking over time as the "confidence" value increases. That is changed by how accurately the estimate of your skill -- the MMR -- matches the observed reality.

 

> Thoughts? Critics? Improvements?

 

You should be able to find the PvP algorithm documentation on the gw2 wiki, and use that in combination with the documentation about the ELO, GLICKO-2, and TrueSkill systems, to better understand how the MMR and matchmaking system in GW2 work -- as well as what advances (eg: TrueSkill) have been attempted over the status-quo.

[ReaverKane.7598]

> @"Abazigal.3679" said:

> > @"ReaverKane.7598" said:

> > It less about volatility but compression...

> > Seasons are compressed into a tiny window, which leads to a grind mentality, and less chance to make up for slip-ups.

> > That coupled with the uncertainty of who you'll be placed with leads to the mentality.

> > Basically the whole league system is broken and poorly managed, which leads to a bad experience, and toxic behaviour.

> > If you want improvements, seasons need to be longer, teams need to be brought back, and wins/rating need to impact your rewards.

>

> This x1000. It's however going to take a while until people consider taking off the season system, since it happens in almost every single game.

 

I'm not arguing taking it off, but making it decent (as in longer, and less conducive to toxic behaviour).

The only other game i know of that has 2 months seasons is Overwatch. But still the ranked rewards are awarded at the end of the season.

Heroes of the Storm has 4 months long seasons.

Dota 2 season's are now 6 months.

League of Legends has a almost year long season, and you can still play ranked in the off-season to improve your MMR and get better placements.

Smite, looking at their wiki has 11 months long seasons.

 

The problem with GW2 is that Ranked PvP is built for players to farm rewards not as a competition. So as in any other grindy content players will find ways to optimize the farm and minimize their effort. So it leads to people win-trading, people afking, people throwing, because the reward is at the end of a match, regardless of you win or lose. So losing on purpose maximizes the rewards over time, because it makes games shorter. And because seasons are short, and you want to make the most games possible, a lot of people are willing to lose on purpose.

 

What happens is that, because you can't have premade teams, not even duo premades at the highest tiers. Coupled with the real possibility that you'll be matched with much lower ranked players (which are more likely to be there just for the reward farms), this increases the uncertainty of having a fair match, and it's almost certain you'll be playing below your rank's difficulty tier, because not enough people to matchmake. So those that do care about their rank, will end up playing less, to minimize the risk of getting a bad team mate, than people at lower ranks that just want their Byzantium chest on repeat.

 

So what i'm saying is, increase the length of seasons, which allows for players to feel less pressured to play several matches daily. This will work towards increasing the number of players. Just like Living World Season One turned a lot of people of the game because of the forced pace, this is having the same effect on a lot of people.

 

After that, remove the pips/chests reward system.

Give players the "regular" Unranked per match reward in ranked.

Add a repeatable achievement that awards Shards, gold and whatever goodies seems plausible, for every 5-10 ranked wins.

Add a per-season reward for each time you climb a rank.

Add a end-season reward per rank, instead of just for the top players.

 

This will "force" players to care about their outcome, decreasing the likelihood that a player will throw or AFK the match. Because the "real" rewards are now gated behind actual performance.

And, sure this will remove a lot of people from PvP, those that are just there for the rewards, but it's like excising a cancer. If you want the pvp community to have some semblance of health, those people have to go.

 

When you do that, improving the reporting system, the automated detection of toxic behaviours (for example, flagging players with really bad KDA for potential feeding).

Enforce stricter punishment... I mean 6 months ban for using **hacks**? That's just weak. I mean it's **hacks**!! Some companies **sue** people for hacking... Arena Net gives 6 months bans. They're so afraid of reducing the PvP community that they won't do what's needed to prevent it from continuing to decay.

Get rid of the bad seeds, sure it'll reduce it further, but then that'll allow it to grow, or at least die sooner rather than later.

 

Finally, if they really care about the competitive nature, bring back ranked teams, and maybe add Draft picks and bans on the pre-game. These two additions will greatly reduce the impact of their inability to properly balance the classes (draft picks and bans allows players to pick counters and ban broken builds, and team queues get rid of the problem by simply playing around it with proper strategy).

Not only that but with pre-made teams you're getting rid of the team uncertainty that prevents top players from playing as much after they've reached their desired rank.

 

I'm not going to ask for better balance, because on one hand that's a no-brainer and needn't be mentioned, and on the other hand, i'm completely sceptical of Arena Net's ability to ever balance the game properly.

 

Right now, GW2's sPvP is a decaying mess, and the fault lies entirely on Arena Net and their strict adherence to flawed philosophies. It's their adherence to a poorly thought up, and half-implemented vote (a lot of people voted in favour of a test-season, not a permanent change) to remove teams, that limits player's options to deal with toxic behaviour.

It's their strict adherence to a nonsense quarterly balance schedule that allows for imbalanced builds and toxic metas to fester for months and years (which means for **several** of their seasons).

It's their strict adherence to tiny short seasons, trying to play on people's completionist nature to rush the content while masking it's issues that drives a lot of players away from the content.

It's their reliance on PvE reward farms to instigate players to play PvP (to the point where it's easier to farm pve content in PvP, like dungeon rewards, than in PvE) coupled with the short seasons that creates the farm mentality.

And sure the short seasons and the farms worked in the short term, but as the game grows older, and people already farmed their stuff, and because they pretty much removed the reasons why actual competitive players, for the most part, would care about the game, the population will dwindle to nothing.

 

 

 

 

 

[Megametzler.5729]

> @"SlippyCheeze.5483" said:

> > @"Megametzler.5729" said:

> > Okay, we all know that a lot of people camp spots on the leaderboard and refuse to play (significantly) more games than 120. This is mostly due to the volatility of the ranking, is it not?

> >

> > So why don't we extend the system that's being used for placement matches and those few matches afterwards? The more games you play, the lower the impact of ranking is. That should be easily implemented by a dampening function:

> > point_difference = point_difference_at_game_zero / number_of_games (or similarish).

> > It would not be a complex fix of the ranking system, but it might be an easy and quick one to let people play more.

>

> I always find suggestions like this really fun. Your idea is great. Wonderful. Amazing.

>

> Your idea is also the exact change that differentiates GLICKO-2 (used by ANet) from ELO, which is that it reduces the changes in ranking over time as the "confidence" value increases. That is changed by how accurately the estimate of your skill -- the MMR -- matches the observed reality.

>

> > Thoughts? Critics? Improvements?

>

> You should be able to find the PvP algorithm documentation on the gw2 wiki, and use that in combination with the documentation about the ELO, GLICKO-2, and TrueSkill systems, to better understand how the MMR and matchmaking system in GW2 work -- as well as what advances (eg: TrueSkill) have been attempted over the status-quo.

 

But it does not include a dampening in the number of games played, as far as I know and as far as I know the system. I am with you that GLICKO is a good system which determines the actual rating quite well. In the end, it always does.

 

The problem is, it is really volatile. I usually play exactly 120 matches in a season, maybe a couple more. Why so? Because if I lose my 120th game, I lose like 20-25 points. Yes, I can play 1 or 2 matches more to (usually) regain my ranking. But it makes me really nervous about those very last matches, while I do not care so much about some matches in between. So if the impact of late matches is being reduced, people might want to play without caring so much. You rarely see players with more than 150 games at the top (yes, that _might_ also be because they don't have the time).

 

As you said, the absolute majority of players are at their correct ranking after a couple of tens games played. So it would hardly have a big impact on the actual ladder. But maybe it would even reduce toxicity in the end, because of less impact.

 

By the way, as far as I understood it, Trueskill might somehow use this. More games played lead to a lower level of uncertainty lead to less impact of last games.

[SlippyCheeze.5483]

> @"Megametzler.5729" said:

> > @"SlippyCheeze.5483" said:

> > > @"Megametzler.5729" said:

> > > Okay, we all know that a lot of people camp spots on the leaderboard and refuse to play (significantly) more games than 120. This is mostly due to the volatility of the ranking, is it not?

> > >

> > > So why don't we extend the system that's being used for placement matches and those few matches afterwards? The more games you play, the lower the impact of ranking is. That should be easily implemented by a dampening function:

> > > point_difference = point_difference_at_game_zero / number_of_games (or similarish).

> > > It would not be a complex fix of the ranking system, but it might be an easy and quick one to let people play more.

> >

> > I always find suggestions like this really fun. Your idea is great. Wonderful. Amazing.

> >

> > Your idea is also the exact change that differentiates GLICKO-2 (used by ANet) from ELO, which is that it reduces the changes in ranking over time as the "confidence" value increases. That is changed by how accurately the estimate of your skill -- the MMR -- matches the observed reality.

> >

> > > Thoughts? Critics? Improvements?

> >

> > You should be able to find the PvP algorithm documentation on the gw2 wiki, and use that in combination with the documentation about the ELO, GLICKO-2, and TrueSkill systems, to better understand how the MMR and matchmaking system in GW2 work -- as well as what advances (eg: TrueSkill) have been attempted over the status-quo.

>

> But it does not include a dampening in the number of games played, as far as I know and as far as I know the system. I am with you that GLICKO is a good system which determines the actual rating quite well. In the end, it always does.

 

The "confidence" system is exactly a damping mechanism, reducing the size of changes. As "confidence" increases, the size of changes decreases. It is not a one-way ratchet, in the sense that you can decrease confidence again by repeatedly violating expectations, but that should be extremely rare. It also doesn't last, because as soon as you will rapidly return to the origin and confidence based on actual performance.

 

> The problem is, it is really volatile. I usually play exactly 120 matches in a season, maybe a couple more. Why so? Because if I lose my 120th game, I lose like 20-25 points. Yes, I can play 1 or 2 matches more to (usually) regain my ranking.

 

You are describing an unexpected loss, followed by two expected wins. That is the other factor that adjusts the magnitude of change, along with confidence. If your confidence were also low, you would see a much bigger change in either direction in those matches.

 

> By the way, as far as I understood it, Trueskill might somehow use this. More games played lead to a lower level of uncertainty lead to less impact of last games.

 

No, it just does more -- though not well documented -- stuff to try and better assign performance values to players based on more than a boolean "did the team win or not" result. It uses the same exact system, including confidence, to manage the actual rating.

 

So, while it may be able to better recognize individual player performance in any given match, it doesn't change the fundamental structure of the process.

[Megametzler.5729]

> @"SlippyCheeze.5483" said:

> > @"Megametzler.5729" said:

> > > @"SlippyCheeze.5483" said:

> > > > @"Megametzler.5729" said:

> > > > Okay, we all know that a lot of people camp spots on the leaderboard and refuse to play (significantly) more games than 120. This is mostly due to the volatility of the ranking, is it not?

> > > >

> > > > So why don't we extend the system that's being used for placement matches and those few matches afterwards? The more games you play, the lower the impact of ranking is. That should be easily implemented by a dampening function:

> > > > point_difference = point_difference_at_game_zero / number_of_games (or similarish).

> > > > It would not be a complex fix of the ranking system, but it might be an easy and quick one to let people play more.

> > >

> > > I always find suggestions like this really fun. Your idea is great. Wonderful. Amazing.

> > >

> > > Your idea is also the exact change that differentiates GLICKO-2 (used by ANet) from ELO, which is that it reduces the changes in ranking over time as the "confidence" value increases. That is changed by how accurately the estimate of your skill -- the MMR -- matches the observed reality.

> > >

> > > > Thoughts? Critics? Improvements?

> > >

> > > You should be able to find the PvP algorithm documentation on the gw2 wiki, and use that in combination with the documentation about the ELO, GLICKO-2, and TrueSkill systems, to better understand how the MMR and matchmaking system in GW2 work -- as well as what advances (eg: TrueSkill) have been attempted over the status-quo.

> >

> > But it does not include a dampening in the number of games played, as far as I know and as far as I know the system. I am with you that GLICKO is a good system which determines the actual rating quite well. In the end, it always does.

>

> The "confidence" system is exactly a damping mechanism, reducing the size of changes. As "confidence" increases, the size of changes decreases. It is not a one-way ratchet, in the sense that you can decrease confidence again by repeatedly violating expectations, but that should be extremely rare. It also doesn't last, because as soon as you will rapidly return to the origin and confidence based on actual performance.

>

> > The problem is, it is really volatile. I usually play exactly 120 matches in a season, maybe a couple more. Why so? Because if I lose my 120th game, I lose like 20-25 points. Yes, I can play 1 or 2 matches more to (usually) regain my ranking.

>

> You are describing an unexpected loss, followed by two expected wins. That is the other factor that adjusts the magnitude of change, along with confidence. If your confidence were also low, you would see a much bigger change in either direction in those matches.

>

> > By the way, as far as I understood it, Trueskill might somehow use this. More games played lead to a lower level of uncertainty lead to less impact of last games.

>

> No, it just does more -- though not well documented -- stuff to try and better assign performance values to players based on more than a boolean "did the team win or not" result. It uses the same exact system, including confidence, to manage the actual rating.

>

> So, while it may be able to better recognize individual player performance in any given match, it doesn't change the fundamental structure of the process.

 

That is not what I am experiencing. Even with my 120th or 150th game, I can still have -25 just like in my first game after the placement matches (and the corresponding few games with high rating difference).

 

€: Or +12 of course, losing numbers are just higher usually, so a better example. ;)

[paShadoWn.5723]

They would not play more even with your proposal in place. Because for them it is play for rating, not rating for play.

[maxwelgm.4315]

> @"ReaverKane.7598" said:

> > @"Abazigal.3679" said:

> > > @"ReaverKane.7598" said:

> > > It less about volatility but compression...

> > > Seasons are compressed into a tiny window, which leads to a grind mentality, and less chance to make up for slip-ups.

> > > That coupled with the uncertainty of who you'll be placed with leads to the mentality.

> > > Basically the whole league system is broken and poorly managed, which leads to a bad experience, and toxic behaviour.

> > > If you want improvements, seasons need to be longer, teams need to be brought back, and wins/rating need to impact your rewards.

> >

> > This x1000. It's however going to take a while until people consider taking off the season system, since it happens in almost every single game.

>

> I'm not arguing taking it off, but making it decent (as in longer, and less conducive to toxic behaviour).

> The only other game i know of that has 2 months seasons is Overwatch. But still the ranked rewards are awarded at the end of the season.

> Heroes of the Storm has 4 months long seasons.

> Dota 2 season's are now 6 months.

> League of Legends has a almost year long season, and you can still play ranked in the off-season to improve your MMR and get better placements.

> Smite, looking at their wiki has 11 months long seasons.

>

> The problem with GW2 is that Ranked PvP is built for players to farm rewards not as a competition. So as in any other grindy content players will find ways to optimize the farm and minimize their effort. So it leads to people win-trading, people afking, people throwing, because the reward is at the end of a match, regardless of you win or lose. So losing on purpose maximizes the rewards over time, because it makes games shorter. And because seasons are short, and you want to make the most games possible, a lot of people are willing to lose on purpose.

>

> What happens is that, because you can't have premade teams, not even duo premades at the highest tiers. Coupled with the real possibility that you'll be matched with much lower ranked players (which are more likely to be there just for the reward farms), this increases the uncertainty of having a fair match, and it's almost certain you'll be playing below your rank's difficulty tier, because not enough people to matchmake. So those that do care about their rank, will end up playing less, to minimize the risk of getting a bad team mate, than people at lower ranks that just want their Byzantium chest on repeat.

>

> So what i'm saying is, increase the length of seasons, which allows for players to feel less pressured to play several matches daily. This will work towards increasing the number of players. Just like Living World Season One turned a lot of people of the game because of the forced pace, this is having the same effect on a lot of people.

>

> After that, remove the pips/chests reward system.

> Give players the "regular" Unranked per match reward in ranked.

> Add a repeatable achievement that awards Shards, gold and whatever goodies seems plausible, for every 5-10 ranked wins.

> Add a per-season reward for each time you climb a rank.

> Add a end-season reward per rank, instead of just for the top players.

>

> This will "force" players to care about their outcome, decreasing the likelihood that a player will throw or AFK the match. Because the "real" rewards are now gated behind actual performance.

> And, sure this will remove a lot of people from PvP, those that are just there for the rewards, but it's like excising a cancer. If you want the pvp community to have some semblance of health, those people have to go.

>

> When you do that, improving the reporting system, the automated detection of toxic behaviours (for example, flagging players with really bad KDA for potential feeding).

> Enforce stricter punishment... I mean 6 months ban for using **hacks**? That's just weak. I mean it's **hacks**!! Some companies **sue** people for hacking... Arena Net gives 6 months bans. They're so afraid of reducing the PvP community that they won't do what's needed to prevent it from continuing to decay.

> Get rid of the bad seeds, sure it'll reduce it further, but then that'll allow it to grow, or at least die sooner rather than later.

>

> Finally, if they really care about the competitive nature, bring back ranked teams, and maybe add Draft picks and bans on the pre-game. These two additions will greatly reduce the impact of their inability to properly balance the classes (draft picks and bans allows players to pick counters and ban broken builds, and team queues get rid of the problem by simply playing around it with proper strategy).

> Not only that but with pre-made teams you're getting rid of the team uncertainty that prevents top players from playing as much after they've reached their desired rank.

>

> I'm not going to ask for better balance, because on one hand that's a no-brainer and needn't be mentioned, and on the other hand, i'm completely sceptical of Arena Net's ability to ever balance the game properly.

>

> Right now, GW2's sPvP is a decaying mess, and the fault lies entirely on Arena Net and their strict adherence to flawed philosophies. It's their adherence to a poorly thought up, and half-implemented vote (a lot of people voted in favour of a test-season, not a permanent change) to remove teams, that limits player's options to deal with toxic behaviour.

> It's their strict adherence to a nonsense quarterly balance schedule that allows for imbalanced builds and toxic metas to fester for months and years (which means for **several** of their seasons).

> It's their strict adherence to tiny short seasons, trying to play on people's completionist nature to rush the content while masking it's issues that drives a lot of players away from the content.

> It's their reliance on PvE reward farms to instigate players to play PvP (to the point where it's easier to farm pve content in PvP, like dungeon rewards, than in PvE) coupled with the short seasons that creates the farm mentality.

> And sure the short seasons and the farms worked in the short term, but as the game grows older, and people already farmed their stuff, and because they pretty much removed the reasons why actual competitive players, for the most part, would care about the game, the population will dwindle to nothing.

>

>

>

>

>

 

These are all good suggestions but your post seems to have two conflicting ideas. Removing rewards from "participation" in PvP will grossly diminish the population that is already (as you put it yourself) low enough for people to get matched too low from their ranks or also too high up (in my case, I find it very annoying that I get matched to top 250 players even myself being in gold III far too often). That would probably lead to the hour long queues players used to experience in Legendary during some of the first seasons. Regardless of whether or not me or you would actually prefer longer queues to having matches with farming players, Anet certainly seems to have enjoyed their decision of increasing population by luring PvE players with free rewards. Perhaps they don't even really mean this to be a competitive thing anymore, and that is why they are currently working on 2v2 maps and such.

[ReaverKane.7598]

> @"maxwelgm.4315" said:

> > @"ReaverKane.7598" said:

> > > @"Abazigal.3679" said:

> > > > @"ReaverKane.7598" said:

> > > > It less about volatility but compression...

> > > > Seasons are compressed into a tiny window, which leads to a grind mentality, and less chance to make up for slip-ups.

> > > > That coupled with the uncertainty of who you'll be placed with leads to the mentality.

> > > > Basically the whole league system is broken and poorly managed, which leads to a bad experience, and toxic behaviour.

> > > > If you want improvements, seasons need to be longer, teams need to be brought back, and wins/rating need to impact your rewards.

> > >

> > > This x1000. It's however going to take a while until people consider taking off the season system, since it happens in almost every single game.

> >

> > I'm not arguing taking it off, but making it decent (as in longer, and less conducive to toxic behaviour).

> > The only other game i know of that has 2 months seasons is Overwatch. But still the ranked rewards are awarded at the end of the season.

> > Heroes of the Storm has 4 months long seasons.

> > Dota 2 season's are now 6 months.

> > League of Legends has a almost year long season, and you can still play ranked in the off-season to improve your MMR and get better placements.

> > Smite, looking at their wiki has 11 months long seasons.

> >

> > The problem with GW2 is that Ranked PvP is built for players to farm rewards not as a competition. So as in any other grindy content players will find ways to optimize the farm and minimize their effort. So it leads to people win-trading, people afking, people throwing, because the reward is at the end of a match, regardless of you win or lose. So losing on purpose maximizes the rewards over time, because it makes games shorter. And because seasons are short, and you want to make the most games possible, a lot of people are willing to lose on purpose.

> >

> > What happens is that, because you can't have premade teams, not even duo premades at the highest tiers. Coupled with the real possibility that you'll be matched with much lower ranked players (which are more likely to be there just for the reward farms), this increases the uncertainty of having a fair match, and it's almost certain you'll be playing below your rank's difficulty tier, because not enough people to matchmake. So those that do care about their rank, will end up playing less, to minimize the risk of getting a bad team mate, than people at lower ranks that just want their Byzantium chest on repeat.

> >

> > So what i'm saying is, increase the length of seasons, which allows for players to feel less pressured to play several matches daily. This will work towards increasing the number of players. Just like Living World Season One turned a lot of people of the game because of the forced pace, this is having the same effect on a lot of people.

> >

> > After that, remove the pips/chests reward system.

> > Give players the "regular" Unranked per match reward in ranked.

> > Add a repeatable achievement that awards Shards, gold and whatever goodies seems plausible, for every 5-10 ranked wins.

> > Add a per-season reward for each time you climb a rank.

> > Add a end-season reward per rank, instead of just for the top players.

> >

> > This will "force" players to care about their outcome, decreasing the likelihood that a player will throw or AFK the match. Because the "real" rewards are now gated behind actual performance.

> > And, sure this will remove a lot of people from PvP, those that are just there for the rewards, but it's like excising a cancer. If you want the pvp community to have some semblance of health, those people have to go.

> >

> > When you do that, improving the reporting system, the automated detection of toxic behaviours (for example, flagging players with really bad KDA for potential feeding).

> > Enforce stricter punishment... I mean 6 months ban for using **hacks**? That's just weak. I mean it's **hacks**!! Some companies **sue** people for hacking... Arena Net gives 6 months bans. They're so afraid of reducing the PvP community that they won't do what's needed to prevent it from continuing to decay.

> > Get rid of the bad seeds, sure it'll reduce it further, but then that'll allow it to grow, or at least die sooner rather than later.

> >

> > Finally, if they really care about the competitive nature, bring back ranked teams, and maybe add Draft picks and bans on the pre-game. These two additions will greatly reduce the impact of their inability to properly balance the classes (draft picks and bans allows players to pick counters and ban broken builds, and team queues get rid of the problem by simply playing around it with proper strategy).

> > Not only that but with pre-made teams you're getting rid of the team uncertainty that prevents top players from playing as much after they've reached their desired rank.

> >

> > I'm not going to ask for better balance, because on one hand that's a no-brainer and needn't be mentioned, and on the other hand, i'm completely sceptical of Arena Net's ability to ever balance the game properly.

> >

> > Right now, GW2's sPvP is a decaying mess, and the fault lies entirely on Arena Net and their strict adherence to flawed philosophies. It's their adherence to a poorly thought up, and half-implemented vote (a lot of people voted in favour of a test-season, not a permanent change) to remove teams, that limits player's options to deal with toxic behaviour.

> > It's their strict adherence to a nonsense quarterly balance schedule that allows for imbalanced builds and toxic metas to fester for months and years (which means for **several** of their seasons).

> > It's their strict adherence to tiny short seasons, trying to play on people's completionist nature to rush the content while masking it's issues that drives a lot of players away from the content.

> > It's their reliance on PvE reward farms to instigate players to play PvP (to the point where it's easier to farm pve content in PvP, like dungeon rewards, than in PvE) coupled with the short seasons that creates the farm mentality.

> > And sure the short seasons and the farms worked in the short term, but as the game grows older, and people already farmed their stuff, and because they pretty much removed the reasons why actual competitive players, for the most part, would care about the game, the population will dwindle to nothing.

> >

> >

> >

> >

> >

>

> These are all good suggestions but your post seems to have two conflicting ideas. Removing rewards from "participation" in PvP will grossly diminish the population that is already (as you put it yourself) low enough for people to get matched too low from their ranks or also too high up (in my case, I find it very annoying that I get matched to top 250 players even myself being in gold III far too often). That would probably lead to the hour long queues players used to experience in Legendary during some of the first seasons. Regardless of whether or not me or you would actually prefer longer queues to having matches with farming players, Anet certainly seems to have enjoyed their decision of increasing population by luring PvE players with free rewards. Perhaps they don't even really mean this to be a competitive thing anymore, and that is why they are currently working on 2v2 maps and such.

 

It is not conflicting... I explained it several times. For the community to grow healthy and competitive, those players that are actively eroding it need to be disincentivized to play. Hence moving the rewards into a competitive spot, adding teams back (i mean the majority of people in PvP aren't there for the PvP, only a small minority of GW2 players have rank 80 in pvp - less than the people playing Raids, which isn't that big of a feat, and still the removal of teams was made an open vote to the whole community) all this would move GW2 back to a more e-sport like point.

The result would be that competitive minded players would have an incentive to come back, and the people interested in the farm would either continue, but be less toxic (the rewards would still be there, they just need to care about their match outcomes more), or leave, which would reduce the population in the short term, that's true.

 

But in the long term, making PvP better will attract people back to it. I don't have the data, but i almost bet the peak of PvP population was during the content drought before HoT came out. Back then you didn't have the farmable rewards, you didn't have leagues, and people were still playing the heck out of PvP, because it was fun, it was reasonably balanced and you could play with your friends in all modes, and they were constantly advertising World Championships, having PvP segments on Arena Net streams, etc, so people wanted to get back at it more.. So rewards aren't that important if you think about it, it's the quality of the content, and the perception of that quality.

Arena Net wanted to inflate their numbers through the easy pickings, and ended up with a skewed pvp population. And they started balancing and managing the content for the "i'm just here for the loot" population, instead of the competitive minded players. Obvious results ensue.

[SlippyCheeze.5483]

> @"Megametzler.5729" said:

> > @"SlippyCheeze.5483" said:

> > > @"Megametzler.5729" said:

> > > > @"SlippyCheeze.5483" said:

> > > > > @"Megametzler.5729" said:

> > > > > Okay, we all know that a lot of people camp spots on the leaderboard and refuse to play (significantly) more games than 120. This is mostly due to the volatility of the ranking, is it not?

> > > > >

> > > > > So why don't we extend the system that's being used for placement matches and those few matches afterwards? The more games you play, the lower the impact of ranking is. That should be easily implemented by a dampening function:

> > > > > point_difference = point_difference_at_game_zero / number_of_games (or similarish).

> > > > > It would not be a complex fix of the ranking system, but it might be an easy and quick one to let people play more.

> > > >

> > > > I always find suggestions like this really fun. Your idea is great. Wonderful. Amazing.

> > > >

> > > > Your idea is also the exact change that differentiates GLICKO-2 (used by ANet) from ELO, which is that it reduces the changes in ranking over time as the "confidence" value increases. That is changed by how accurately the estimate of your skill -- the MMR -- matches the observed reality.

> > > >

> > > > > Thoughts? Critics? Improvements?

> > > >

> > > > You should be able to find the PvP algorithm documentation on the gw2 wiki, and use that in combination with the documentation about the ELO, GLICKO-2, and TrueSkill systems, to better understand how the MMR and matchmaking system in GW2 work -- as well as what advances (eg: TrueSkill) have been attempted over the status-quo.

> > >

> > > But it does not include a dampening in the number of games played, as far as I know and as far as I know the system. I am with you that GLICKO is a good system which determines the actual rating quite well. In the end, it always does.

> >

> > The "confidence" system is exactly a damping mechanism, reducing the size of changes. As "confidence" increases, the size of changes decreases. It is not a one-way ratchet, in the sense that you can decrease confidence again by repeatedly violating expectations, but that should be extremely rare. It also doesn't last, because as soon as you will rapidly return to the origin and confidence based on actual performance.

> >

> > > The problem is, it is really volatile. I usually play exactly 120 matches in a season, maybe a couple more. Why so? Because if I lose my 120th game, I lose like 20-25 points. Yes, I can play 1 or 2 matches more to (usually) regain my ranking.

> >

> > You are describing an unexpected loss, followed by two expected wins. That is the other factor that adjusts the magnitude of change, along with confidence. If your confidence were also low, you would see a much bigger change in either direction in those matches.

> >

> > > By the way, as far as I understood it, Trueskill might somehow use this. More games played lead to a lower level of uncertainty lead to less impact of last games.

> >

> > No, it just does more -- though not well documented -- stuff to try and better assign performance values to players based on more than a boolean "did the team win or not" result. It uses the same exact system, including confidence, to manage the actual rating.

> >

> > So, while it may be able to better recognize individual player performance in any given match, it doesn't change the fundamental structure of the process.

>

> That is not what I am experiencing. Even with my 120th or 150th game, I can still have -25 just like in my first game after the placement matches (and the corresponding few games with high rating difference).

>

> €: Or +12 of course, losing numbers are just higher usually, so a better example. ;)

 

If you have two different magnitudes of change, ±12 and ±25, then you are describing the effects of full confidence, and the effects of "did you deliver the expected result".

 

If you *only* experience +12 and -25, then you are consistently being placed in matches you are expected to win, so when you win (and match expectation) you get a small change, and when you lose (and violate expectation) you get the larger change. (Observe that ±12 would be half of ±25, rounded down.)

 

I'm surprised you would consistently be placed in matches you are expected to win, but ultimately, if that is what is happening you should try and address that specific quirk in the matchmaker, rather than complaining that the algorithm isn't behaving as it "should" when, in fact, it works; the thing consuming the MMR and building teams is what is not working as expected.

 

Personally, I'd bet that you simply remember more instances of this, because it is the annoying version, than instances where you unexpectedly win or expectedly lose; to confirm or refute that you could grab your 10 most recent matches from the API and post the edited subset here with the actual scores. Seeing 10, all expected win, matches would be very convincing that there is something odd going on in matchmaking, at least to me.

[Megametzler.5729]

> @"SlippyCheeze.5483" said:

> > @"Megametzler.5729" said:

> > > @"SlippyCheeze.5483" said:

> > > > @"Megametzler.5729" said:

> > > > > @"SlippyCheeze.5483" said:

> > > > > > @"Megametzler.5729" said:

> > > > > > Okay, we all know that a lot of people camp spots on the leaderboard and refuse to play (significantly) more games than 120. This is mostly due to the volatility of the ranking, is it not?

> > > > > >

> > > > > > So why don't we extend the system that's being used for placement matches and those few matches afterwards? The more games you play, the lower the impact of ranking is. That should be easily implemented by a dampening function:

> > > > > > point_difference = point_difference_at_game_zero / number_of_games (or similarish).

> > > > > > It would not be a complex fix of the ranking system, but it might be an easy and quick one to let people play more.

> > > > >

> > > > > I always find suggestions like this really fun. Your idea is great. Wonderful. Amazing.

> > > > >

> > > > > Your idea is also the exact change that differentiates GLICKO-2 (used by ANet) from ELO, which is that it reduces the changes in ranking over time as the "confidence" value increases. That is changed by how accurately the estimate of your skill -- the MMR -- matches the observed reality.

> > > > >

> > > > > > Thoughts? Critics? Improvements?

> > > > >

> > > > > You should be able to find the PvP algorithm documentation on the gw2 wiki, and use that in combination with the documentation about the ELO, GLICKO-2, and TrueSkill systems, to better understand how the MMR and matchmaking system in GW2 work -- as well as what advances (eg: TrueSkill) have been attempted over the status-quo.

> > > >

> > > > But it does not include a dampening in the number of games played, as far as I know and as far as I know the system. I am with you that GLICKO is a good system which determines the actual rating quite well. In the end, it always does.

> > >

> > > The "confidence" system is exactly a damping mechanism, reducing the size of changes. As "confidence" increases, the size of changes decreases. It is not a one-way ratchet, in the sense that you can decrease confidence again by repeatedly violating expectations, but that should be extremely rare. It also doesn't last, because as soon as you will rapidly return to the origin and confidence based on actual performance.

> > >

> > > > The problem is, it is really volatile. I usually play exactly 120 matches in a season, maybe a couple more. Why so? Because if I lose my 120th game, I lose like 20-25 points. Yes, I can play 1 or 2 matches more to (usually) regain my ranking.

> > >

> > > You are describing an unexpected loss, followed by two expected wins. That is the other factor that adjusts the magnitude of change, along with confidence. If your confidence were also low, you would see a much bigger change in either direction in those matches.

> > >

> > > > By the way, as far as I understood it, Trueskill might somehow use this. More games played lead to a lower level of uncertainty lead to less impact of last games.

> > >

> > > No, it just does more -- though not well documented -- stuff to try and better assign performance values to players based on more than a boolean "did the team win or not" result. It uses the same exact system, including confidence, to manage the actual rating.

> > >

> > > So, while it may be able to better recognize individual player performance in any given match, it doesn't change the fundamental structure of the process.

> >

> > That is not what I am experiencing. Even with my 120th or 150th game, I can still have -25 just like in my first game after the placement matches (and the corresponding few games with high rating difference).

> >

> > €: Or +12 of course, losing numbers are just higher usually, so a better example. ;)

>

> If you have two different magnitudes of change, ±12 and ±25, then you are describing the effects of full confidence, and the effects of "did you deliver the expected result".

>

> If you *only* experience +12 and -25, then you are consistently being placed in matches you are expected to win, so when you win (and match expectation) you get a small change, and when you lose (and violate expectation) you get the larger change. (Observe that ±12 would be half of ±25, rounded down.)

>

> I'm surprised you would consistently be placed in matches you are expected to win, but ultimately, if that is what is happening you should try and address that specific quirk in the matchmaker, rather than complaining that the algorithm isn't behaving as it "should" when, in fact, it works; the thing consuming the MMR and building teams is what is not working as expected.

>

> Personally, I'd bet that you simply remember more instances of this, because it is the annoying version, than instances where you unexpectedly win or expectedly lose; to confirm or refute that you could grab your 10 most recent matches from the API and post the edited subset here with the actual scores. Seeing 10, all expected win, matches would be very convincing that there is something odd going on in matchmaking, at least to me.

 

Ah. I think you got my idea wrong. :p

 

-25 and +12 happens a lot in plat 3. But that is not the main issue. My point is the same for gold and all - my point is not about "playing according to the estimated skill" or not. If you put a dampening function over the actual rating loss/gain, you latter games do not have so much impact anymore and the resulting jumps get reduced - even after 100 games, I can have a bad day and end up in 1650. When I have a good day, I end up in 1750. This could force people to be really careful and get toxic especially for the very last games. If the last games did not have such a big impact - because the MMR should already be close to your actual skill rating -, this would be reduced.

 

Trueskill might have this system implemented, because _a higher number of games reduces your standard deviation_. GLICKO - as far as I know - does not. As I said, the same rating for me, the same enemies, the same result leads to the same loss/gain at the 20th game as in the 150th game. The idea already exists in the placement matches and the few games afterwards, where the effect is increased. I just suggest to extend this specific functionality to decrease the ranking volatility, let people play more games without worrying too much and - maybe - reduce nervousness for the last games.

 

€: Why is the thread being hijacked by the same old critics without proper solutions...? It feels like some guys just copy & paste from other threads. :/

[SlippyCheeze.5483]

> @"Megametzler.5729" said:

> > @"SlippyCheeze.5483" said:

> > > @"Megametzler.5729" said:

> > > > @"SlippyCheeze.5483" said:

> > > > > @"Megametzler.5729" said:

> > > > > > @"SlippyCheeze.5483" said:

> > > > > > > @"Megametzler.5729" said:

> > > > > > > Okay, we all know that a lot of people camp spots on the leaderboard and refuse to play (significantly) more games than 120. This is mostly due to the volatility of the ranking, is it not?

> > > > > > >

> > > > > > > So why don't we extend the system that's being used for placement matches and those few matches afterwards? The more games you play, the lower the impact of ranking is. That should be easily implemented by a dampening function:

> > > > > > > point_difference = point_difference_at_game_zero / number_of_games (or similarish).

> > > > > > > It would not be a complex fix of the ranking system, but it might be an easy and quick one to let people play more.

> > > > > >

> > > > > > I always find suggestions like this really fun. Your idea is great. Wonderful. Amazing.

> > > > > >

> > > > > > Your idea is also the exact change that differentiates GLICKO-2 (used by ANet) from ELO, which is that it reduces the changes in ranking over time as the "confidence" value increases. That is changed by how accurately the estimate of your skill -- the MMR -- matches the observed reality.

> > > > > >

> > > > > > > Thoughts? Critics? Improvements?

> > > > > >

> > > > > > You should be able to find the PvP algorithm documentation on the gw2 wiki, and use that in combination with the documentation about the ELO, GLICKO-2, and TrueSkill systems, to better understand how the MMR and matchmaking system in GW2 work -- as well as what advances (eg: TrueSkill) have been attempted over the status-quo.

> > > > >

> > > > > But it does not include a dampening in the number of games played, as far as I know and as far as I know the system. I am with you that GLICKO is a good system which determines the actual rating quite well. In the end, it always does.

> > > >

> > > > The "confidence" system is exactly a damping mechanism, reducing the size of changes. As "confidence" increases, the size of changes decreases. It is not a one-way ratchet, in the sense that you can decrease confidence again by repeatedly violating expectations, but that should be extremely rare. It also doesn't last, because as soon as you will rapidly return to the origin and confidence based on actual performance.

> > > >

> > > > > The problem is, it is really volatile. I usually play exactly 120 matches in a season, maybe a couple more. Why so? Because if I lose my 120th game, I lose like 20-25 points. Yes, I can play 1 or 2 matches more to (usually) regain my ranking.

> > > >

> > > > You are describing an unexpected loss, followed by two expected wins. That is the other factor that adjusts the magnitude of change, along with confidence. If your confidence were also low, you would see a much bigger change in either direction in those matches.

> > > >

> > > > > By the way, as far as I understood it, Trueskill might somehow use this. More games played lead to a lower level of uncertainty lead to less impact of last games.

> > > >

> > > > No, it just does more -- though not well documented -- stuff to try and better assign performance values to players based on more than a boolean "did the team win or not" result. It uses the same exact system, including confidence, to manage the actual rating.

> > > >

> > > > So, while it may be able to better recognize individual player performance in any given match, it doesn't change the fundamental structure of the process.

> > >

> > > That is not what I am experiencing. Even with my 120th or 150th game, I can still have -25 just like in my first game after the placement matches (and the corresponding few games with high rating difference).

> > >

> > > €: Or +12 of course, losing numbers are just higher usually, so a better example. ;)

> >

> > If you have two different magnitudes of change, ±12 and ±25, then you are describing the effects of full confidence, and the effects of "did you deliver the expected result".

> >

> > If you *only* experience +12 and -25, then you are consistently being placed in matches you are expected to win, so when you win (and match expectation) you get a small change, and when you lose (and violate expectation) you get the larger change. (Observe that ±12 would be half of ±25, rounded down.)

> >

> > I'm surprised you would consistently be placed in matches you are expected to win, but ultimately, if that is what is happening you should try and address that specific quirk in the matchmaker, rather than complaining that the algorithm isn't behaving as it "should" when, in fact, it works; the thing consuming the MMR and building teams is what is not working as expected.

> >

> > Personally, I'd bet that you simply remember more instances of this, because it is the annoying version, than instances where you unexpectedly win or expectedly lose; to confirm or refute that you could grab your 10 most recent matches from the API and post the edited subset here with the actual scores. Seeing 10, all expected win, matches would be very convincing that there is something odd going on in matchmaking, at least to me.

>

> Ah. I think you got my idea wrong. :p

>

> -25 and +12 happens a lot in plat 3. But that is not the main issue. My point is the same for gold and all - my point is not about "playing according to the estimated skill" or not. If you put a dampening function over the actual rating loss/gain, you latter games do not have so much impact anymore and the resulting jumps get reduced - even after 100 games, I can have a bad day and end up in 1650. When I have a good day, I end up in 1750. This could force people to be really careful and get toxic especially for the very last games. If the last games did not have such a big impact - because the MMR should already be close to your actual skill rating -, this would be reduced.

>

> Trueskill might have this system implemented, because _a higher number of games reduces your standard deviation_. GLICKO - as far as I know - does not.

 

I'm pretty sure that "confidence" is exactly what you are talking about here, that being the one thing that differentiates GLICKO and ELO. It reduces (or increases) the magnitude of change based on how accurately the algorithms predictions match reality.

 

> €: Why is the thread being hijacked by the same old critics without proper solutions...? It feels like some guys just copy & paste from other threads. :/

 

I don't copy-paste, but ... this suggestion is exactly what those other thread have said, worded differently. Expecting a different response to the same question would require that something had changed between the two threads. Sadly, it hasn't.

 

Well, or that these comments are made in bad faith and/or ignorance, I guess, along with an inability to keep the wrong idea straight between two different threads.

 

Ultimately, I'm not trying to hijack your thread, I'm trying to point out that you are asking for something that already exists as part of the algorithm, and so, make the point that while your idea is not wrong, it is "not wrong" because that is exactly what makes GLICKO better than ELO. That is, Mark Glickman absolutely agrees with you. ;)

[Megametzler.5729]

> @"SlippyCheeze.5483" said:

> > @"Megametzler.5729" said:

> > > @"SlippyCheeze.5483" said:

> > > > @"Megametzler.5729" said:

> > > > > @"SlippyCheeze.5483" said:

> > > > > > @"Megametzler.5729" said:

> > > > > > > @"SlippyCheeze.5483" said:

> > > > > > > > @"Megametzler.5729" said:

> > > > > > > > Okay, we all know that a lot of people camp spots on the leaderboard and refuse to play (significantly) more games than 120. This is mostly due to the volatility of the ranking, is it not?

> > > > > > > >

> > > > > > > > So why don't we extend the system that's being used for placement matches and those few matches afterwards? The more games you play, the lower the impact of ranking is. That should be easily implemented by a dampening function:

> > > > > > > > point_difference = point_difference_at_game_zero / number_of_games (or similarish).

> > > > > > > > It would not be a complex fix of the ranking system, but it might be an easy and quick one to let people play more.

> > > > > > >

> > > > > > > I always find suggestions like this really fun. Your idea is great. Wonderful. Amazing.

> > > > > > >

> > > > > > > Your idea is also the exact change that differentiates GLICKO-2 (used by ANet) from ELO, which is that it reduces the changes in ranking over time as the "confidence" value increases. That is changed by how accurately the estimate of your skill -- the MMR -- matches the observed reality.

> > > > > > >

> > > > > > > > Thoughts? Critics? Improvements?

> > > > > > >

> > > > > > > You should be able to find the PvP algorithm documentation on the gw2 wiki, and use that in combination with the documentation about the ELO, GLICKO-2, and TrueSkill systems, to better understand how the MMR and matchmaking system in GW2 work -- as well as what advances (eg: TrueSkill) have been attempted over the status-quo.

> > > > > >

> > > > > > But it does not include a dampening in the number of games played, as far as I know and as far as I know the system. I am with you that GLICKO is a good system which determines the actual rating quite well. In the end, it always does.

> > > > >

> > > > > The "confidence" system is exactly a damping mechanism, reducing the size of changes. As "confidence" increases, the size of changes decreases. It is not a one-way ratchet, in the sense that you can decrease confidence again by repeatedly violating expectations, but that should be extremely rare. It also doesn't last, because as soon as you will rapidly return to the origin and confidence based on actual performance.

> > > > >

> > > > > > The problem is, it is really volatile. I usually play exactly 120 matches in a season, maybe a couple more. Why so? Because if I lose my 120th game, I lose like 20-25 points. Yes, I can play 1 or 2 matches more to (usually) regain my ranking.

> > > > >

> > > > > You are describing an unexpected loss, followed by two expected wins. That is the other factor that adjusts the magnitude of change, along with confidence. If your confidence were also low, you would see a much bigger change in either direction in those matches.

> > > > >

> > > > > > By the way, as far as I understood it, Trueskill might somehow use this. More games played lead to a lower level of uncertainty lead to less impact of last games.

> > > > >

> > > > > No, it just does more -- though not well documented -- stuff to try and better assign performance values to players based on more than a boolean "did the team win or not" result. It uses the same exact system, including confidence, to manage the actual rating.

> > > > >

> > > > > So, while it may be able to better recognize individual player performance in any given match, it doesn't change the fundamental structure of the process.

> > > >

> > > > That is not what I am experiencing. Even with my 120th or 150th game, I can still have -25 just like in my first game after the placement matches (and the corresponding few games with high rating difference).

> > > >

> > > > €: Or +12 of course, losing numbers are just higher usually, so a better example. ;)

> > >

> > > If you have two different magnitudes of change, ±12 and ±25, then you are describing the effects of full confidence, and the effects of "did you deliver the expected result".

> > >

> > > If you *only* experience +12 and -25, then you are consistently being placed in matches you are expected to win, so when you win (and match expectation) you get a small change, and when you lose (and violate expectation) you get the larger change. (Observe that ±12 would be half of ±25, rounded down.)

> > >

> > > I'm surprised you would consistently be placed in matches you are expected to win, but ultimately, if that is what is happening you should try and address that specific quirk in the matchmaker, rather than complaining that the algorithm isn't behaving as it "should" when, in fact, it works; the thing consuming the MMR and building teams is what is not working as expected.

> > >

> > > Personally, I'd bet that you simply remember more instances of this, because it is the annoying version, than instances where you unexpectedly win or expectedly lose; to confirm or refute that you could grab your 10 most recent matches from the API and post the edited subset here with the actual scores. Seeing 10, all expected win, matches would be very convincing that there is something odd going on in matchmaking, at least to me.

> >

> > Ah. I think you got my idea wrong. :p

> >

> > -25 and +12 happens a lot in plat 3. But that is not the main issue. My point is the same for gold and all - my point is not about "playing according to the estimated skill" or not. If you put a dampening function over the actual rating loss/gain, you latter games do not have so much impact anymore and the resulting jumps get reduced - even after 100 games, I can have a bad day and end up in 1650. When I have a good day, I end up in 1750. This could force people to be really careful and get toxic especially for the very last games. If the last games did not have such a big impact - because the MMR should already be close to your actual skill rating -, this would be reduced.

> >

> > Trueskill might have this system implemented, because _a higher number of games reduces your standard deviation_. GLICKO - as far as I know - does not.

>

> I'm pretty sure that "confidence" is exactly what you are talking about here, that being the one thing that differentiates GLICKO and ELO. It reduces (or increases) the magnitude of change based on how accurately the algorithms predictions match reality.

>

> > €: Why is the thread being hijacked by the same old critics without proper solutions...? It feels like some guys just copy & paste from other threads. :/

>

> I don't copy-paste, but ... this suggestion is exactly what those other thread have said, worded differently. Expecting a different response to the same question would require that something had changed between the two threads. Sadly, it hasn't.

>

> Well, or that these comments are made in bad faith and/or ignorance, I guess, along with an inability to keep the wrong idea straight between two different threads.

>

> Ultimately, I'm not trying to hijack your thread, I'm trying to point out that you are asking for something that already exists as part of the algorithm, and so, make the point that while your idea is not wrong, it is "not wrong" because that is exactly what makes GLICKO better than ELO. That is, Mark Glickman absolutely agrees with you. ;)

 

I did not mean you, it is good to discuss this. :p Compression for example has absolutely nothing to do with this topic...

 

I didn't read anything about confidence in the wiki and I never experienced something like that ingame... maybe it is implemented but occurs way too late? :hushed:

[SlippyCheeze.5483]

> @"Megametzler.5729" said:

> > @"SlippyCheeze.5483" said:

> > > @"Megametzler.5729" said:

> > > > @"SlippyCheeze.5483" said:

> > > > > @"Megametzler.5729" said:

> > > > > > @"SlippyCheeze.5483" said:

> > > > > > > @"Megametzler.5729" said:

> > > > > > > > @"SlippyCheeze.5483" said:

> > > > > > > > > @"Megametzler.5729" said:

> > > > > > > > > Okay, we all know that a lot of people camp spots on the leaderboard and refuse to play (significantly) more games than 120. This is mostly due to the volatility of the ranking, is it not?

> > > > > > > > >

> > > > > > > > > So why don't we extend the system that's being used for placement matches and those few matches afterwards? The more games you play, the lower the impact of ranking is. That should be easily implemented by a dampening function:

> > > > > > > > > point_difference = point_difference_at_game_zero / number_of_games (or similarish).

> > > > > > > > > It would not be a complex fix of the ranking system, but it might be an easy and quick one to let people play more.

> > > > > > > >

> > > > > > > > I always find suggestions like this really fun. Your idea is great. Wonderful. Amazing.

> > > > > > > >

> > > > > > > > Your idea is also the exact change that differentiates GLICKO-2 (used by ANet) from ELO, which is that it reduces the changes in ranking over time as the "confidence" value increases. That is changed by how accurately the estimate of your skill -- the MMR -- matches the observed reality.

> > > > > > > >

> > > > > > > > > Thoughts? Critics? Improvements?

> > > > > > > >

> > > > > > > > You should be able to find the PvP algorithm documentation on the gw2 wiki, and use that in combination with the documentation about the ELO, GLICKO-2, and TrueSkill systems, to better understand how the MMR and matchmaking system in GW2 work -- as well as what advances (eg: TrueSkill) have been attempted over the status-quo.

> > > > > > >

> > > > > > > But it does not include a dampening in the number of games played, as far as I know and as far as I know the system. I am with you that GLICKO is a good system which determines the actual rating quite well. In the end, it always does.

> > > > > >

> > > > > > The "confidence" system is exactly a damping mechanism, reducing the size of changes. As "confidence" increases, the size of changes decreases. It is not a one-way ratchet, in the sense that you can decrease confidence again by repeatedly violating expectations, but that should be extremely rare. It also doesn't last, because as soon as you will rapidly return to the origin and confidence based on actual performance.

> > > > > >

> > > > > > > The problem is, it is really volatile. I usually play exactly 120 matches in a season, maybe a couple more. Why so? Because if I lose my 120th game, I lose like 20-25 points. Yes, I can play 1 or 2 matches more to (usually) regain my ranking.

> > > > > >

> > > > > > You are describing an unexpected loss, followed by two expected wins. That is the other factor that adjusts the magnitude of change, along with confidence. If your confidence were also low, you would see a much bigger change in either direction in those matches.

> > > > > >

> > > > > > > By the way, as far as I understood it, Trueskill might somehow use this. More games played lead to a lower level of uncertainty lead to less impact of last games.

> > > > > >

> > > > > > No, it just does more -- though not well documented -- stuff to try and better assign performance values to players based on more than a boolean "did the team win or not" result. It uses the same exact system, including confidence, to manage the actual rating.

> > > > > >

> > > > > > So, while it may be able to better recognize individual player performance in any given match, it doesn't change the fundamental structure of the process.

> > > > >

> > > > > That is not what I am experiencing. Even with my 120th or 150th game, I can still have -25 just like in my first game after the placement matches (and the corresponding few games with high rating difference).

> > > > >

> > > > > €: Or +12 of course, losing numbers are just higher usually, so a better example. ;)

> > > >

> > > > If you have two different magnitudes of change, ±12 and ±25, then you are describing the effects of full confidence, and the effects of "did you deliver the expected result".

> > > >

> > > > If you *only* experience +12 and -25, then you are consistently being placed in matches you are expected to win, so when you win (and match expectation) you get a small change, and when you lose (and violate expectation) you get the larger change. (Observe that ±12 would be half of ±25, rounded down.)

> > > >

> > > > I'm surprised you would consistently be placed in matches you are expected to win, but ultimately, if that is what is happening you should try and address that specific quirk in the matchmaker, rather than complaining that the algorithm isn't behaving as it "should" when, in fact, it works; the thing consuming the MMR and building teams is what is not working as expected.

> > > >

> > > > Personally, I'd bet that you simply remember more instances of this, because it is the annoying version, than instances where you unexpectedly win or expectedly lose; to confirm or refute that you could grab your 10 most recent matches from the API and post the edited subset here with the actual scores. Seeing 10, all expected win, matches would be very convincing that there is something odd going on in matchmaking, at least to me.

> > >

> > > Ah. I think you got my idea wrong. :p

> > >

> > > -25 and +12 happens a lot in plat 3. But that is not the main issue. My point is the same for gold and all - my point is not about "playing according to the estimated skill" or not. If you put a dampening function over the actual rating loss/gain, you latter games do not have so much impact anymore and the resulting jumps get reduced - even after 100 games, I can have a bad day and end up in 1650. When I have a good day, I end up in 1750. This could force people to be really careful and get toxic especially for the very last games. If the last games did not have such a big impact - because the MMR should already be close to your actual skill rating -, this would be reduced.

> > >

> > > Trueskill might have this system implemented, because _a higher number of games reduces your standard deviation_. GLICKO - as far as I know - does not.

> >

> > I'm pretty sure that "confidence" is exactly what you are talking about here, that being the one thing that differentiates GLICKO and ELO. It reduces (or increases) the magnitude of change based on how accurately the algorithms predictions match reality.

> >

> > > €: Why is the thread being hijacked by the same old critics without proper solutions...? It feels like some guys just copy & paste from other threads. :/

> >

> > I don't copy-paste, but ... this suggestion is exactly what those other thread have said, worded differently. Expecting a different response to the same question would require that something had changed between the two threads. Sadly, it hasn't.

> >

> > Well, or that these comments are made in bad faith and/or ignorance, I guess, along with an inability to keep the wrong idea straight between two different threads.

> >

> > Ultimately, I'm not trying to hijack your thread, I'm trying to point out that you are asking for something that already exists as part of the algorithm, and so, make the point that while your idea is not wrong, it is "not wrong" because that is exactly what makes GLICKO better than ELO. That is, Mark Glickman absolutely agrees with you. ;)

>

> I did not mean you, it is good to discuss this. :p Compression for example has absolutely nothing to do with this topic...

>

> I didn't read anything about confidence in the wiki and I never experienced something like that ingame... maybe it is implemented but occurs way too late? :hushed:

 

That is entirely possible. Like, I'm not trying to say the current thing is perfect, by any stretch of the imagination. It just isn't as terrible as some people think. Even if it is mean to them and puts them in a lower tier than they think they should be. ;) (and I don't mean you when I say that.)

 

The biggest thing to remember is that it normally takes around 10 games for MMR to converge to high confidence in GLICKO, which is coincidentally the same number of "placement" games -- games where changes to MMR are hidden. Which means, for the vast majority of players, you never see the huge adjustments, only the relatively small once that polish from "reasonably ok" to "good" estimates.

 

If you really want to see the big changes, tank your rating deliberately and then return to playing normally, and you should see larger adjustments. (Like that one person who posted asking why they were seeing big jumps post-placement, which is almost certainly because their bad luck got them placed low due to randomly getting an exceptionally long string of people tanking their ratings and/or afking in placement.)

[Megametzler.5729]

> @"SlippyCheeze.5483" said:

> > @"Megametzler.5729" said:

> > > @"SlippyCheeze.5483" said:

> > > > @"Megametzler.5729" said:

> > > > > @"SlippyCheeze.5483" said:

> > > > > > @"Megametzler.5729" said:

> > > > > > > @"SlippyCheeze.5483" said:

> > > > > > > > @"Megametzler.5729" said:

> > > > > > > > > @"SlippyCheeze.5483" said:

> > > > > > > > > > @"Megametzler.5729" said:

> > > > > > > > > > Okay, we all know that a lot of people camp spots on the leaderboard and refuse to play (significantly) more games than 120. This is mostly due to the volatility of the ranking, is it not?

> > > > > > > > > >

> > > > > > > > > > So why don't we extend the system that's being used for placement matches and those few matches afterwards? The more games you play, the lower the impact of ranking is. That should be easily implemented by a dampening function:

> > > > > > > > > > point_difference = point_difference_at_game_zero / number_of_games (or similarish).

> > > > > > > > > > It would not be a complex fix of the ranking system, but it might be an easy and quick one to let people play more.

> > > > > > > > >

> > > > > > > > > I always find suggestions like this really fun. Your idea is great. Wonderful. Amazing.

> > > > > > > > >

> > > > > > > > > Your idea is also the exact change that differentiates GLICKO-2 (used by ANet) from ELO, which is that it reduces the changes in ranking over time as the "confidence" value increases. That is changed by how accurately the estimate of your skill -- the MMR -- matches the observed reality.

> > > > > > > > >

> > > > > > > > > > Thoughts? Critics? Improvements?

> > > > > > > > >

> > > > > > > > > You should be able to find the PvP algorithm documentation on the gw2 wiki, and use that in combination with the documentation about the ELO, GLICKO-2, and TrueSkill systems, to better understand how the MMR and matchmaking system in GW2 work -- as well as what advances (eg: TrueSkill) have been attempted over the status-quo.

> > > > > > > >

> > > > > > > > But it does not include a dampening in the number of games played, as far as I know and as far as I know the system. I am with you that GLICKO is a good system which determines the actual rating quite well. In the end, it always does.

> > > > > > >

> > > > > > > The "confidence" system is exactly a damping mechanism, reducing the size of changes. As "confidence" increases, the size of changes decreases. It is not a one-way ratchet, in the sense that you can decrease confidence again by repeatedly violating expectations, but that should be extremely rare. It also doesn't last, because as soon as you will rapidly return to the origin and confidence based on actual performance.

> > > > > > >

> > > > > > > > The problem is, it is really volatile. I usually play exactly 120 matches in a season, maybe a couple more. Why so? Because if I lose my 120th game, I lose like 20-25 points. Yes, I can play 1 or 2 matches more to (usually) regain my ranking.

> > > > > > >

> > > > > > > You are describing an unexpected loss, followed by two expected wins. That is the other factor that adjusts the magnitude of change, along with confidence. If your confidence were also low, you would see a much bigger change in either direction in those matches.

> > > > > > >

> > > > > > > > By the way, as far as I understood it, Trueskill might somehow use this. More games played lead to a lower level of uncertainty lead to less impact of last games.

> > > > > > >

> > > > > > > No, it just does more -- though not well documented -- stuff to try and better assign performance values to players based on more than a boolean "did the team win or not" result. It uses the same exact system, including confidence, to manage the actual rating.

> > > > > > >

> > > > > > > So, while it may be able to better recognize individual player performance in any given match, it doesn't change the fundamental structure of the process.

> > > > > >

> > > > > > That is not what I am experiencing. Even with my 120th or 150th game, I can still have -25 just like in my first game after the placement matches (and the corresponding few games with high rating difference).

> > > > > >

> > > > > > €: Or +12 of course, losing numbers are just higher usually, so a better example. ;)

> > > > >

> > > > > If you have two different magnitudes of change, ±12 and ±25, then you are describing the effects of full confidence, and the effects of "did you deliver the expected result".

> > > > >

> > > > > If you *only* experience +12 and -25, then you are consistently being placed in matches you are expected to win, so when you win (and match expectation) you get a small change, and when you lose (and violate expectation) you get the larger change. (Observe that ±12 would be half of ±25, rounded down.)

> > > > >

> > > > > I'm surprised you would consistently be placed in matches you are expected to win, but ultimately, if that is what is happening you should try and address that specific quirk in the matchmaker, rather than complaining that the algorithm isn't behaving as it "should" when, in fact, it works; the thing consuming the MMR and building teams is what is not working as expected.

> > > > >

> > > > > Personally, I'd bet that you simply remember more instances of this, because it is the annoying version, than instances where you unexpectedly win or expectedly lose; to confirm or refute that you could grab your 10 most recent matches from the API and post the edited subset here with the actual scores. Seeing 10, all expected win, matches would be very convincing that there is something odd going on in matchmaking, at least to me.

> > > >

> > > > Ah. I think you got my idea wrong. :p

> > > >

> > > > -25 and +12 happens a lot in plat 3. But that is not the main issue. My point is the same for gold and all - my point is not about "playing according to the estimated skill" or not. If you put a dampening function over the actual rating loss/gain, you latter games do not have so much impact anymore and the resulting jumps get reduced - even after 100 games, I can have a bad day and end up in 1650. When I have a good day, I end up in 1750. This could force people to be really careful and get toxic especially for the very last games. If the last games did not have such a big impact - because the MMR should already be close to your actual skill rating -, this would be reduced.

> > > >

> > > > Trueskill might have this system implemented, because _a higher number of games reduces your standard deviation_. GLICKO - as far as I know - does not.

> > >

> > > I'm pretty sure that "confidence" is exactly what you are talking about here, that being the one thing that differentiates GLICKO and ELO. It reduces (or increases) the magnitude of change based on how accurately the algorithms predictions match reality.

> > >

> > > > €: Why is the thread being hijacked by the same old critics without proper solutions...? It feels like some guys just copy & paste from other threads. :/

> > >

> > > I don't copy-paste, but ... this suggestion is exactly what those other thread have said, worded differently. Expecting a different response to the same question would require that something had changed between the two threads. Sadly, it hasn't.

> > >

> > > Well, or that these comments are made in bad faith and/or ignorance, I guess, along with an inability to keep the wrong idea straight between two different threads.

> > >

> > > Ultimately, I'm not trying to hijack your thread, I'm trying to point out that you are asking for something that already exists as part of the algorithm, and so, make the point that while your idea is not wrong, it is "not wrong" because that is exactly what makes GLICKO better than ELO. That is, Mark Glickman absolutely agrees with you. ;)

> >

> > I did not mean you, it is good to discuss this. :p Compression for example has absolutely nothing to do with this topic...

> >

> > I didn't read anything about confidence in the wiki and I never experienced something like that ingame... maybe it is implemented but occurs way too late? :hushed:

>

> That is entirely possible. Like, I'm not trying to say the current thing is perfect, by any stretch of the imagination. It just isn't as terrible as some people think. Even if it is mean to them and puts them in a lower tier than they think they should be. ;) (and I don't mean you when I say that.)

>

> The biggest thing to remember is that it normally takes around 10 games for MMR to converge to high confidence in GLICKO, which is coincidentally the same number of "placement" games -- games where changes to MMR are hidden. Which means, for the vast majority of players, you never see the huge adjustments, only the relatively small once that polish from "reasonably ok" to "good" estimates.

>

> If you really want to see the big changes, tank your rating deliberately and then return to playing normally, and you should see larger adjustments. (Like that one person who posted asking why they were seeing big jumps post-placement, which is almost certainly because their bad luck got them placed low due to randomly getting an exceptionally long string of people tanking their ratings and/or afking in placement.)

 

Yeah, you are actually right - the whole thing seems to be implemented in GLICKO. Then there just seems to be a major implementation fault. Mr. Glickman even states the topic of how to derive c - the confidence factor or so - to not make the rating stable, but not too volatile when many games are played. That's probably what went kind of wrong here, because - as I mentioned - I never experienced this reduction of rating difference...