Guild Wars 2

I’m not sure if this should be reported as a bug but there is an issue with the current modeling in the WvW ratings system. I assume that this link is still a valid description of the strength of victory modeling. https://forum-en.gw2archive.eu/forum/game/wuv/The-math-behind-WvW-ratings/first#post2034699

The problem is that the Sine function model that is currently used to award the actual score in the glicko rating deviates considerably from the actual real distribution of results. I retrieved a history of scores from http://mos.millenium.org/na/matchups and parsed the results and converted the scores into the portions used by the Anet Glicko system. I then converted that distribution into a probability distribution that the rating system uses to assign the “actual” score that is use to correct the ratings. I’ve attached an image of the plots that I did for this explanation.

The black curve is the real distribution of scores. It was normally distributed about 0.5 with a SD of 0.15.

The Green curve is the Sine function curve that is currently used. This curve currently over rewards lower performing servers and under rewards higher performing servers. The worst error occurs at 0.7 and 0.3 on the x axis. In these locations the error is about 0.11. this means that higher performing servers (0.7) performing in this range are getting about a 10% penalty in score and the lower performing (0.3) server is getting a 10% bonus. It doesn’t matter what your server rating is, if you perform in the error range you being rewarded improperly.

The real problem with this is when a server is expected to have a really high score not to lose rating points. The best example of this right now is Dragonbrand (1839.9965). They are currently competing against Stormbluff Isle (1667.1971) and Maguuma (1534.7854).

According the current system, Dragonbrand would have an expected score of 0.71 against Stormbluff Isle. According to the sine function currently used, they would Dragonbrand would need to have a score portion of 0.64. But according to the real data all that should be required to achieve that score is 0.59. That 0.05 difference is significant over the course of the week where the overall proportion shrinks as the scores go up. Against Maguuma the situation is worse. The Sine function requires a portion of 0.73 when only a 0.65 is justified by the historical data. The conclusion is that Dragonbrand (and any top ranked server in a teir) must work harder than necessary not to be pulled back down. And the lower ranked servers will have an easier time pull the top down.

If the servers were pooled and randomly competed this wouldn’t be a problem. It would simply result in a virtual floor and ceiling for the ratings. Downward pressure on the top and upward pressure on the bottom. Everything pushed to the middle. But when you have tiers, this makes Tier mobility difficult. Servers pushing up have to put forth a monumental effort to break in to the next tier and the servers on the bottom of the a Tier get protected by the system even after they stop performing at a level worthy of their tier. To be more blunt, it screws servers like Dragonbrand and Tarnished Coast and benefits servers like Blackgate and Sea of Sorrows who have had long periods of poor performance and still refuse to drop down to the next tier.

In a Strength of Victory system like the one Anet is using, the score awarded should be consistent with the distribution of the historical scoring data. I understand that when a system is initiated and there is no data to reference you must guess. But now there is plenty of data to use for reference.

I’m not sure where the Sine function came from but modeling normally distributed data can be done with a logistic curve much more easily. I believe this is why Elo used it. I have modeled a logistic curve to fit the historical data and it shown in the chart as the red line. The sigmoid midpoint is 0.5, and the k value is 10.91. This k value gives the lowest sum of squares of the difference between the historical data and the logistic model.

You should consider fixing this to reward hard working servers that are trying to push up the ranks and maybe more importantly to properly punish servers that are coasting at the bottom of upper tiers.

Just as an aside, reducing the range of the MMR variability by capping it at 100 has also aggravated the problem because it forces servers like Dragonbrand and Tarnished Coast to compete against lower ranked servers more often which keeps them weighed down.

I’ve also attached a csv of some of the derived data. I’ve included column headers but it is still a bit messy.
Columns
Proportion: a score proportion
Count: The number of times the proportion block (by 0.01) appeared.
Cumulative: Cumulative count.
%cumulative: total Probability distribution of real proportions data.
Sine: the result of the proportion as a function of the Sine function currently used.
Corrected: The result of the proportion as a function of the logistic function.
diff^2: squared diff of %cumulative and Corrected.

EDIT: Changed the x axis on the graph to be more consistent with my comments.

Interesting numbers, it makes a lot of sense actually. I remember how over the course of two years my server slowly crawled its way upwards, but would end up bogged down with the same matchups over and over at every step of the way. That caused a strong erosion of player interest, and people transfered out even though we were winning. We would have to overperform on a systematical basis in order not to loose points and stagnate even more.

I agree that MU fluidity should be increased. Right now the mid and lower tiers are bleeding out of players anyway, as the recent changes have failed to cause significant enthusiasm (euphemism, much?), so enabling faster server movements could be an interesting experiment.

Very interesting. I wonder what Habib Loew would say about this if he still works at Anet. His explanations and calculus formulas in the original thread were also very in depth.

Habib Loew.6239:

wAGlickoScore = (sin((wAPercent – 0.5) * Pi) + 1) * 0.5

?=). ¿¿¿¿

Always wondered why it seemed so off from what actually takes place. Should have figured it was an error in the code, or bad math whatever.

Makes me think the KISS method would have been best. Like winner up -loser down. Or Johje’s suggestion of a modified glicko-based swiss.

displayname.8315:

Makes me think the KISS method would have been best. Like winner up -loser down.

But what would you do with the third server? If its +1/-1/0 on a server ladder you would end up with servers sharing the same rank. If T1 contain servers A, B, C at ranks 1, 2, 3 respectivly and B wins while C looses… You would have A at 1, B also at 1 and C at 4, with no server capable of being 2 lol.

Dawdler.8521:

displayname.8315:

Makes me think the KISS method would have been best. Like winner up -loser down.

But what would you do with the third server? If its +1/-1/0 on a server ladder you would end up with servers sharing the same rank. If T1 contain servers A, B, C at ranks 1, 2, 3 respectivly and B wins while C looses… You would have A at 1, B also at 1 and C at 4, with no server capable of being 2 lol.

With winner up – loser down its as simple as win your matchup move up a tier, lose and move down a tier, place second and remain in your current tier. No need for +1 points or sine waves.

This would result in movement only between adjacent tiers, while still allowing stacked servers to move in the correct upwards direction each week.

Either way maybe they’ll “fix the error” and we’ll get more accurate, expedient movement.

Just curious, but did you use your model or analyze the EU data to compare?

In light of this data, or in conjunction with it, visually comparing the NA v EU evolution ladders begs several questions.

1) In looking at the left edge of the data set, it is clear through initial variability in glicko
that, EU’s glicko fluctuated greatly for nearly 6 months, while NA’s seemed to settle in at 3 months, and then shake up around the 6 month mark too. So what happened roughly 26 weeks in?

I believe that in NA, free transfers (and cheap alt accounts) played the largest role as we began stacking servers around 3 months into it. Even then, differences in playstyles (GvG, PPT, fight, etc.) were becoming an issue for NA servers, and people began to choose sides through free transfers. Don’t know if EU had the same issues?

2) In looking at the right side of the data sets, it appears that NA is seeing variability really only in the lower tiers (lower right hand quadrant) while EU sees it’s highest variability occurring in the upper right hand quadrant. This seems to indicate that EU is stacking into the top HALF of its servers while NA has chosen to stack into it’s top 3-4 servers. It also does appear that even EU is having to transfer and stack over time, but I’ll let an EU person comment on that as they will know better what is happening.

To drive that point home, since 9 months into the game, NA has only had 4 servers in 4th place…FOUR!!! In 2.5 years!!! Now it’s worth mentioning that SF maintained #1 for an impressive (almost) 52 weeks in this timeframe, but they faced 6 opponents in that time, and Kodash was their longest at 12 weeks. NA’s 4th place has been stays of 30ish weeks.

3) Finally, the data would seem to indicate that glicko rewards more balanced servers through more variability. Currently, EU has 15 servers within 300pts of each other (1618 to 1902), NA has 7 (1820-2160). Similarly, NA has chosen to go with 4 “Very High” servers and 2 “High” servers, while EU went for 2 “Very High” and 11 “High”…(11…insane)!!!!!

just some thoughts

~psyco~

Don’t have time to learn how the Glitchy 2.0 Scoring works but they should have done this instead of the Glycol one

1st place moves up on tier
2nd place stays on tier
3rd place moves down on tier

why do complex math if you can make it simple with better results……?

Norbe.7630:

Don’t have time to learn how the Glitchy 2.0 Scoring works but they should have done this instead of the Glycol one

1st place moves up on tier
2nd place stays on tier
3rd place moves down on tier

why do complex math if you can make it simple with better results……?

This is a good idea, except for the special cases of top and bottom tiers where there are no more rooms for the top and bottom servers to move.

Using the top as an example, top and the runner up would have to match up again unless special rules are added like forcing the runner up server to the second tier. But this would means the runner up would play the third place again in the second tier.

Moreover, since the top 6 NA servers have been playing together for a while many guild leaders recongnize each other and often fought each other for fun more than for the scores of the servers, even without an official GvG match. They could prefer to play one server over another as some posts claimed of conspiracy of two server not wanting another in the tier because of different play styles, ppt vs ppk.

Since the active population of these top servers are probably more than 50 percent of the total population of all servers, any drastic changes that could bring uneven matches and probably a lot complains.

WvW has evolved from its original structure due to the long term staled matches and Anet does not seem to understand its evolution and some of the updates caused many displeasure and disappearance of players.

Lord of Rings.5371:

WvW has evolved from its original structure due to the long term staled matches and Anet does seem to understand it evolution and some of the updates caused many displeasure and disappearance of players.

I think you’re giving anet too much credit here. Right up to HOT they have demonstrated very little understanding of the mode at all. The overhaul will tell if they have a better understanding after the HOT failure.

morrolan.9608:

Lord of Rings.5371:

WvW has evolved from its original structure due to the long term staled matches and Anet does seem to understand it evolution and some of the updates caused many displeasure and disappearance of players.

I think you’re giving anet too much credit here. Right up to HOT they have demonstrated very little understanding of the mode at all. The overhaul will tell if they have a better understanding after the HOT failure.

You are right.

I have corrected my post.

Thx.

psyco.4082:

Just curious, but did you use your model or analyze the EU data to compare?

I used NA and EU data. Anything I could get from millenium.

I simply commented on the consequences of the math. There are other behaviors that could cause server movements through the tiers. Population stacking being one of them.

The question I would have is whether the population stacking is a natural phenomenon or one brought on by the need to have a significant advantage to move up to the next tier. Because of the difficulty in moving between tiers, servers may need to recruit to increase their population to make the move.

This essentially results in tiers sorted by population so why have a rating at all.

The tragedy of this is that lower population server with better skill can still get tier locked because of the inability to achieve the ridiculous results that are required.

I’m not an expert, but as far as I can tell the sine is just used to determine the degree of a servers win/loss, this value is then fed into the glicko algorithm to determine the points each server gets for the matchup, so the sine itself does not determine the points each server gets.

Macilien.3078:

I’m not an expert, but as far as I can tell the sine is just used to determine the degree of a servers win/loss, this value is then fed into the glicko algorithm to determine the points each server gets for the matchup, so the sine itself does not determine the points each server gets.

The sine function is used to report the “actual” score that is compared to the “expected” score. In Elo and in Glicko they are used the same (Actual-Expected) x some multiplier.

By not having the calculation of the actual score tied to the historical distribution of results in their strength of victory system, they are not properly rewarding the effort.

Lord of Rings.5371:

Norbe.7630:

Don’t have time to learn how the Glitchy 2.0 Scoring works but they should have done this instead of the Glycol one

1st place moves up on tier
2nd place stays on tier
3rd place moves down on tier

why do complex math if you can make it simple with better results……?

Using the top as an example, top and the runner up would have to match up again unless special rules are added

The #1 and #2 servers always face each other anyways. At least they would have opponents cycling in.

From playing T1 recently they do have a bit of a higher avg of total players than T2. Mostly just NA pugs stacking things up. With a few transfers it would balance quick enough. I know the gem costs to transfer are a little too high but maybe something will change on that too.

Thanks for not letting me down, displayname.

No. KISS is not always the correct path, and 1-up-1-down is not better or more virtuous for being easily understood with basic arithmetic. This was shown, on this forum, years ago. It doesn’t work on this, a tiered ladder with 3-party matches. Thinking otherwise means you haven’t thought or worked through its consequences over time: the basic arithmetic, played out multiple times, for multiple matches, in sequence.

Unfortunate that this forum’s search is forever broken. This was a very longly debated subject back when ANet was still fiddling with the thing, and particularly at the month where they tightened their per-server random seed from 1.0 to 0.5. It was then shown here what a model of 1-up-1-down looks like over time in this, and it wasn’t good. That’s easy to replicate, if you’re interested. Just do it.

Virtute.8251:

Thanks for not letting me down, displayname.

No. KISS is not always the correct path, and 1-up-1-down is not better or more virtuous for being easily understood with basic arithmetic. This was shown, on this forum, years ago. It doesn’t work on this, a tiered ladder with 3-party matches. Thinking otherwise means you haven’t thought or worked through its consequences over time: the basic arithmetic, played out multiple times, for multiple matches, in sequence.

Unfortunate that this forum’s search is forever broken. This was a very longly debated subject back when ANet was still fiddling with the thing, and particularly at the month where they tightened their per-server random seed from 1.0 to 0.5. It was then shown here what a model of 1-up-1-down looks like over time in this, and it wasn’t good. That’s easy to replicate, if you’re interested. Just do it.

Oh hey anvil

I don’t see the problem with winners moving upwards instead of downwards, what are the consequences? Are you saying having variety in matchups in a no-no because of a few stacked servers? That is a whole different enchilada and should be considered a problem to be solved not something to be worked around.

Random seeds or whatever is not what 1-up-1-down is about. Typically the most stacked server is the winner and the smallest server is the loser. Very simple, and at the same time rewards winners and removes the need for glicko pushes.

Id just be happy with a simple ranking system by win-place-show and match servers by their rank instead of —-

Habib Loew.6239:

wAGlickoScore = (sin((wAPercent – 0.5) * Pi) + 1) * 0.5

+randomizer formula.. whatever that is

Dawdler.8521:

displayname.8315:

Makes me think the KISS method would have been best. Like winner up -loser down.

But what would you do with the third server? If its +1/-1/0 on a server ladder you would end up with servers sharing the same rank. If T1 contain servers A, B, C at ranks 1, 2, 3 respectivly and B wins while C looses… You would have A at 1, B also at 1 and C at 4, with no server capable of being 2 lol.

After thinking about a rank points system I kinda like the idea of ranking/matching by meritocracy.

I think it might be better if only the winner of a tier gets +1 rank point while the other 2 either stay the same or get replaced by a +1 rank server.

This would make it so a server would have to win a tier 3 times to move up. They would have to win a tier once to be safe from dropping.

In that scenario if B, the rank 2 server wins they would replace the A server as rank 1 and A would become rank 2. If the rank 4 server wins T2 they would replace C the rank 3 server and C would become rank 4 in T2.

Because there are so few servers in the field of competition and because they all must compete every week, a simple ladder would have probably been best for WvW.

The glicko would work great for Structured PvP though using a simple win/lose/draw scoring system.

1-up-1-down would drastically increase the frequency of (even more) imbalanced matches. For each tier, the middle server will tend to get stuck, and the #1 and #3 will tend to bounce constantly between their natural tier and the neighbor tier. I would expect the players to quickly realize that fact, then fall into a pattern of off-weeks and on-weeks where a majority would avoid the match following a 1-up that pushes them into a guaranteed curb-stomp loss. Counterpoint, of course, would be the karma train swap meet for the guaranteed win week. That’s what happens now, just less frequently.

1-up-1-down can work in competitions with 2 opponents per match. The third opponent is why it won’t work here. Furthermore, it gives no real weight to historical performance, and so it is too volatile.

It’s completely okay, for example, that NA Tier 1 servers have tended to stick in that tier, because history has proven that all three of them are nearly guaranteed to win if they match against any of three Tier 2 servers. That history is reflected in the wide separation between those tiers today. The fact that a Tier 2 server can inflate to Tier 1 population and activity was proven last year, by Yak’s Bend, and the fact that a fall from Tier 1 will deflate the server was proven most recently by Tarnished Coast.

The WvW leaderboard was not designed as a short-term competitive ladder. We, the players, turned it into a meta-competition, with websites like mos.millenium.org and our own behavior. That got reinforced by ArenaNet later, with their WvW seasonal tournaments, but they lock the leaderboards into leagues for those things because the leaderboard isn’t actually designed for it. It needs the league meta-tiers to make the tournament meaningful for anyone below tier 2, and it needs forced shuffling of servers during the tournament, because the rankings are intentionally designed to not do that.

ArenaNet’s use of a modified Glicko 2 ranking is purely for matching nearly equal opponents, where more than 3 opponents (servers) was also a systematic requirement of the original design, because they did not have a solid “megaserver” technology—which, by the way, is a fancy way of naming extremely redundant and wide network topologies, a/k/a, high-availability clusters, and “cloud”.

It was arguably more interesting to do it this way, with the server leaderboard, rather than forever lock a trio of servers into a perpetual battle. That was a novel spin on the Tri-Realm RvR concept, and it worked, for better and worse.

Every time a significantly imbalanced match has occurred, there has been no end to the complaints, up to and including that being one of the leading reasons for the deletion and banning of the old “matchup threads” here.

TorquedSoul has shown a way to improve the ranking method now, with data that didn’t exist 3.5 years ago, but aside from some slight changes like that, the method itself has always done what it is designed to do, in an appropriate time frame, given the fact that the matches each last 1 week. Torqued’s claim is very specific, and it isn’t that ArenaNet Glicko is the wrong thing to do. Rather, it’s that ArenaNet Glicko is doing a part of its job slightly wrong.

I don’t think that a simple ladder would do anything good for the week-to-week experience. The weekly competition of WvW is ideally between the three most-similar opponents; not some make-believe idea of an opponent that doesn’t exist, or one that would be fun to fight for one night until you realize that they have so many people logged in that you can’t possibly fight it.

Virtute.8251:

1-up-1-down would drastically increase the frequency of (even more) imbalanced matches. For each tier, the middle server will tend to get stuck, and the #1 and #3 will tend to bounce constantly between their natural tier and the neighbor tier. I would expect the players to quickly realize that fact, then fall into a pattern of off-weeks and on-weeks where a majority would avoid the match following a 1-up that pushes them into a guaranteed curb-stomp loss. Counterpoint, of course, would be the karma train swap meet for the guaranteed win week. That’s what happens now, just less frequently.

1-up-1-down can work in competitions with 2 opponents per match. The third opponent is why it won’t work here. Furthermore, it gives no real weight to historical performance, and so it is too volatile.

This, right here.

1-up-1-down would be great only if:

Matches were of only two servers
- OR -
The disparity between tiers/servers across all tiers would approach zero.

Otherwise, you have middle-score stagnancy and wildly-swinging weeks as described above. Middle-tier stagnancy is already an issue, but the wild swings due to tier disparity are rare.

The best would probably be some kind of weighted system which measures score/tier growth perhaps with some weighted data such as population changes and coverage hour graph analysis. If a server gets stacked, this will be visible in the data, and rapid tier acceleration/winning streaks will push the server up and up faster rather than needing to wait for glicko to catch up.

The most balanced possible matchups would require a change from the server-exclusive model, though, as heuristic functions with some genetic algorithms using lots of data as a backbone could pair multiple servers together to cover multiple heuristic needs to create the most balanced encounters, while mutation functions keep the server combinations fresh.

Of course, this comes at the cost of keeping things solely server-based, which the community is divided on. Sustainable balanced matches without stagnancy versus server communities; take your pick, as both isn’t possible as far as anyone’s aware.

It’s true that waiting for glicko to catch up is a painful experience. However, the necessary evil in it is the fact that it is the waiting which helps prevent player-intentional volatility. If the players are allowed to induce match changing volatility within one or two matches (again, the off-week/on-week problem) then it becomes possible for them to reliably coordinate a boom/bust cycle. If they only have to wait a week or two between phases of that oscillation, then it becomes inevitable that it will happen.

Waiting for glicko to catch up has been the deterrent from that behavior. That behavior is the reason that Yak’s Bend rose to Tier 1 last year, and it was Tarnished Coast’s refusal to be patient with the glicko rankings that stopped the cycle from happening.

The real reason that the NA ladder has such a different history as compared to the EU ladder, as shown in the narrower gaps between EU tiers, is that NA and EU players have a very different sense of sportsmanship for a game of this type. The NA players have always stacked the top servers to “full”, until there was no one left who could be convinced to do it, and they repeated that until now, when I get whispers from Tier 1 people scrounging desperately for people they can’t find.

ok heres the comparison between Glicko and KISS

with Glicko:
T1 = you will face your 2 opponents for a long time, very long time till you get bored
T2-T7 = you will face different opponents for long time or short time
T8 = same as T1 but slightly better

with KISS:
T1 = you will face 2 or more opponents within a short time
T2-T7 = you will face 3 or more opponents within a short time
T8 = same as T1

ok lets talk about players and ask ourselves:
-would you like to face different opponents on your server or you prefer to fight the same people for eternity?
-why does one transfer to a higher tier or a lower tier, is it the fights that they cant get on their tier or the boredom of facing the same faces everyday?
-is Glicko system really based on the skills of the world or the population and coverage?
-and so on and so forth………

Glicko system, as is done here, is a reflection of population and coverage, seen through how that influences PPT. The outcome of PPT across an entire week is determined primarily by which server has the most people focused on gaining PPT during the times when their opponents have fewer thusly focused people. The outcome is only secondarily determined by PPT gained through victories against roughly equal opponents.

The thought, “I want to fight new people”, is the primary reason to transfer to a different server.

The thought, “I want to win against these people”, is the primary reason to recruit transferring people into your server. That secondarily leads to eventually fighting against new people, as your server will either ascend the ladder or inspire your opponents to also recruit.

None of this is indicating that either server transfer or glicko ranking of servers are bad things which are wrongly used or implemented.

The use of glicko for a limited size scoring pool is a flawed concept. It was never ment for such a small sample size such as Gw2 wvw. It was ment for unlimited numbers of players who may never actually play against each other. Glicko is a much too complicated formula to be used in this situation.

A 1 up 1 down ladder system, while not a perfect system, would handle this situation with wvw matchups much better than the current glicko system. Yes, there would be people abuse the system, but it wouldn’t have the system ending problems you are suggesting. Now I’m not saying it is the system we need to be using, I’m saying it would be better than stupid over complex system we are currently using. I spent years playing in a 1 up 1 down ladder system, and it would be much more difficult to abuse than you think, mostly because of the complexity of wvw. Not only would you have to get all the guild on a server to agree t o throw matches, you’d also have to get all the pugs and pugmanders to agree, and that would be nearly impossible to do. It’s been done using the current system. Mag just recently did it when they decided to tank so hard they were playing down 2 tiers.

I’m not entirely sure what the solution to this problem is, but it sure as hell isn’t glicko. The current situation of DB being a perfect example. DB got to tier 3 and immediately started out scoring the rest of tier 3, usually by almost double the combined score of the other 2 servers. The real problem with this isn’t DB being stuck in tier 3, it is the fact that the other servers are stuck playing against DB. While DB has been trying to get out of tier 3 they have wrecked 3 different servers. Magumma tanked hard, Crystal Desert had its best guild transfer, and now Bannock has quit. Both SBI and Mag would have been a much better matchup against SoS months ago. Now we have a server who has been out scoring the other 2 servers by almost double the combined scores of the other 2 servers for months loosing glicko instead of gaining glicko. GG anet

Still, some people want 1-up-1-down because it’s very obvious looking and has an instant reward effect, like loot drops that force you to press the loot pickup key all night, and a gold, silver, or bronze medal for every moment you spend near a building or NPC.

How it would not work, and would break down, is more difficult because it is long-term thinking, and might send you into a spreadsheet to illustrate.

The problem with Dragonbrand right now would not be fixed by not using glicko to rank them. That problem is the fact that there are too few servers in the pool, regardless of the ranking method used, to offer every server a good match of two other servers. It is a hard problem, and you can not solve it with a new leaderboard concept. We’ve been over this before.

To ask for 1-up-1-down is effectively the same as asking to have the problem made worse, for more people.

There is no amount of PVF’N or glicko theory crafting that can prove or disprove if a 1 up 1 down concept would or would not work. The only way to know, is to try it, but that would a bit extreme at this point. A simple modification to the formula would be the easiest solution to the current problem. Removing the portion that is weighted against the higher server may fix the problem. It may also benefit to leave some to weighted score toward the lower server to prevent tanking, but the restrictions on the winning server needs to be removed so that they can move up in a natural progression.

WTB phone that doesn’t autocorrect server to service every fn time…….

Glicko is good if your playing PvP as a guy said before
its good if the mode is 1v1 or your a single guy who plays a mode with some random guys as your team, random everytime

now the WvW, its a almost fixed team gameplay till you change your roster
meaning its like a fixed team, like Basketball tounaments or Dota2 tournaments

how do you rank multiple teams on a MOBA tournament, “elimination mode”
how do you rank a single MOBA player, “elo/glicko mode”

now what is WvW, a team or a singularity to be assessed?

imagine Olympics or NBA or MOBA tourneys would use glicko, that would be disasterrific

that my friends is an effective way of saying “wtf”

Bearded.6485:

There is no amount of PVF’N or glicko theory crafting that can prove or disprove if a 1 up 1 down concept would or would not work. The only way to know, is to try it, but that would a bit extreme at this point.

Try it in a spreadsheet, like we did the last time a full length novel thread on the subject wrote itself. It’s a very simple thought experiment, but if you need the numbers to slap you in the face, then that is easy to do.

Norbe.7630:

imagine Olympics or NBA or MOBA tourneys would use glicko, that would be disasterrific

that my friends is an effective way of saying “wtf”

You’re confusing WvW servers with teams in a regulated team sport. WvW is not regulated. The teams are of unequal size. The rules are too chaotic for a sport. Of course your analogy doesn’t make sense. The problem is the analogy, and not either WvW or ANet glicko.

WvW leaderboard is not supposed to change often. That’s the entire point of the thing.

This is actually very interesting data that anet should really look into rather than sugar coat xyz or sweep it all under the rug until it stinks so bad they can’t ignore it.

Population for WvW as a game mode is declining at a rapid rate generally as said above “on/off week behavior”. This is especially true in Sea of Sorrows case, Sometimes we compete well in T2 other times things seem up in the air. This is also seen in FA’s case at times.

Take last session where SoS fought SBI & CD. I imagine many thought it’d be better for the T3 servers & SoS’s fresh matchup however SoS lazily flattened the T3 servers just as bad as DB does resulting in essentially the same old match as always for the lower servers. [Insert T2 Server Name Here] curbstomping resolving nothing.

DB does deserve a few straight matches in T2 to allow the trio to get a feel for eachother & see whether DB can be consistent week to week. At this point i’d say server transfers are a bad idea unless you’re just moving a few guildies to the guilds homeworld. I can only see a change of pace & fresh faces for any server may be the only good thing to come out of all this but we shall see how anet fluffs around with this one.

There needs to be a solid system that players have less of an affect on matchups as in part the current state is in essence an effect from.

Needless to say reading above the current glicko system seems to be working as intended but not functioning as it should? Am i reading that right?

Eternalight.4708:

Needless to say reading above the current glicko system seems to be working as intended but not functioning as it should? Am i reading that right?

Yep.

The way to understand it can be something like this: Look at the graph that TorquedSoul attached to OP, and understand that the curving is meant to enforce a diminishing return on both the upward and downward movement of a server’s ranking.

The effect is a form of loss/gain prevention. It intentionally prevents a server from gaining or losing “too much” ranking from a great or terrible performance in a single match. It forces the good or bad performance to be repeated, as a display of consistency on behalf of the server.

That is why ANet Glicko has been good for its purpose, and 1up would be terrible for the same purpose. 1up does an entirely different thing, for an entirely different purpose. ANet Glicko produces a good enough stability, and that’s what it’s meant to do.

The WvW leaderboard is actually just a table sort. The sort condition is “how do these servers perform against their peers, over time”. It is not a sports competition ladder.

What TorquedSoul has shown is that ANet should probably change the type of curve being applied, because the curve they use is too aggressive in punishing the winner and propping up the loser, in individual matches. He’s also shown that in spite of that fact, the servers still eventually get to the ranking they deserve, but they have to work too hard, or be beaten too badly, to get there.

If ANet Glicko didn’t work well enough to eventually rank a server properly, then TorquedSoul would have no data to prove his point with.

I’m not sure how much time (in other words, how many matches) would be trimmed from a situation like Dragonbrand vs Sea of Sorrows right now, if Torque’s change was made. What is certain, however, is that there would still be pain involved for the losing parties, and that is a part of this game type. That’s what you bought and wanted. 1up isn’t going to change that for the better, where loser’s or insufficient winner’s feels are concerned.

Saying that there use to be a full length novel thread about this subject, and telling every one how easy it is to make a spreadsheet proving your point does nothing to actually prove your point. I can say my crap smells like roses, but it doesn’t make it true.

Bearded.6485:

Saying that there use to be a full length novel thread about this subject

https://www.google.com/#q=site:forum-en.guildwars2.com+glicko+one+up+one+down

Bearded.6485:

and telling every one how easy it is to make a spreadsheet proving your point does nothing to actually prove your point.

Do you really need help with making a table that would plot 1up/down across multiple matches? I’m only putting the burden of proof on you, the claimant. I’m prepared to be proven wrong.

Bearded.6485:

I can say my crap smells like roses, but it doesn’t make it true.

I hope you get away with that, ‘cuz that’s the Bearded I got along with in TS.

Got along with….not so much, more like tolerated.

To be completely honest, if you were paying attention you and I were saying the exact same thing, but I threw in some random 1 up 1 down stuff, and other nonsense to keep you posting your walls of text that basically repeated yourself over and over again just in a different way and more words.

anyway…… Im bored now, but GG ….. your pvf has gotten weak.

Virtute.8251:

What TorquedSoul has shown is that ANet should probably change the type of curve being applied, because the curve they use is too aggressive in punishing the winner and propping up the loser.

Yep this curve that punishes the winner and props up the loser should go, or be adjusted.

IMO glicko is just bad because it forces people to PPT push, instead of simply winning.

I still like idea of winner gets +1 rank (2nd and 3rd get +0). So matches would not flip flop ever week, but after a few wins the rank up would be enough to move a tier.

Or maybe the curve change.. since a total blowout win could move you were you belong.

Virtute.8251:

You’re confusing WvW servers with teams in a regulated team sport. WvW is not regulated. The teams are of unequal size. The rules are too chaotic for a sport. Of course your analogy doesn’t make sense. The problem is the analogy, and not either WvW or ANet glicko.

WvW leaderboard is not supposed to change often. That’s the entire point of the thing.

as you analogy have done the same to yours……..
WvW defines as Worlds versus Worlds (my analogy)
WvW defines as Population Point Tick (your analogy)

if what i think is correct that Anet’s glicko only defines population then it should not be called “WvW rankings” but “WvW populace” instead

my POV of rankings is of higher standard than yours or theirs
quality over quantity

i shall then leave you some videos of what different servers looks like in all tiers, i have a feeling that you yourself have not visited some tiers to widen your knowledge about an argument on WvW

Upper Tiers are like this https://youtu.be/9r9K7I26JHU?t=32
Lower Tier are like this https://www.youtube.com/watch?v=HdNn5TZu6R8
Mid Tier are like this https://www.youtube.com/watch?v=FmNgz1uwnY8

that is the situation your on if your like on that tier….

Hehehehehehehehehe
The DB salt is amazing on fries.

Eternalight.4708:

Needless to say reading above the current glicko system seems to be working as intended but not functioning as it should? Am i reading that right?

heres the Glicthy 2 system looks like on real world

https://youtu.be/eVs08g68lyU

displayname.8315:

Virtute.8251:

What TorquedSoul has shown is that ANet should probably change the type of curve being applied, because the curve they use is too aggressive in punishing the winner and propping up the loser.

Yep this curve that punishes the winner and props up the loser should go, or be adjusted.

IMO glicko is just bad because it forces people to PPT push, instead of simply winning.

I still like idea of winner gets +1 rank (2nd and 3rd get +0). So matches would not flip flop ever week, but after a few wins the rank up would be enough to move a tier.

Or maybe the curve change.. since a total blowout win could move you were you belong.

Its the strength of victory (SOV) component that Anet added to the glicko 2 system that results in the PPT push. Glicko 2 can be used with the standard Elo approach of win/lose/draw = 1/0/0.5 where the final score doesn’t matter.

SOV works fine so long as the scoring mechanism is based on the historical scoring data and no one is competing against others who have dramatically different ratings that would require a blowout win to avoid losing points. The entire reason for ratings systems is to ensure competitive matches so those situations shouldn’t exist anyway. The PPT grind happens when the match is assured of victory but the players must keep grinding to stay even in rating. That is the result of either poor pairing or poor modeling.

The benefit of an SOV system is that it would naturally reduce oscillations in the rating by dampening the changes based on score provided performance was consistent. Glicko 2 is also suppose to reduce oscillations but it buffers based on volatility in ratings and not an individual match score.

It’s poor pairing here, and that is inevitable. Elo would do the same thing.

In this case, I would say the poor pairing is the result of poor modeling.

The poor pairing is the result of the population flux and server stacking and the flux and stacking is driven by the poor model that makes it difficult to push up into the next tier.