Guild Wars 2

Edit for Anet : please describe in greater details as to the maths of the new variance matchup

" The new values allow for adjustments as much as (devation * 0.45 + 10), capped at 100. In both cases, the roll is plus-or-minus these values multiplied by a random real value in the range (-1,1)"

So if i get this right if a server drops -100 points then

(-100*0.45 + 10) = -35

And then we multiply the random real value, say -1 happened

Therefore

(-1)*(-35) = + 35 since negative times negative = positive?

Anyone knows whats the real maths

also, to answer the OP more directly, here’s how the math works. the basis of the matchups are the “rating” and “deviation” values from the leaderboard (https://leaderboards.guildwars2.com/en/na/wvwmain).

under the original matchup system, you simply sorted the servers by rating and grouped them into threes.

with the first ‘randomized’ matchups, the ‘matchup rating’ can shift up or down by up to (deviation + 40) (capped at 200). once you’ve randomly shifted each rating, the servers are then sorted by matchup rating and grouped into threes, same as before.

now, the amount of random shift has been reduced from (deviation + 40) to (0.45*deviation + 10) and the cap has been reduced to 100. everything else is the same — once the new smaller random shift is applied, the servers are sorted by matchup rating and grouped into threes for matches.

the ‘matchup rating’ is used only for determining the matchups. actual server rating stays as it was, and the rating adjustments at the end of the match continue to be done the way they always have, based on Glicko-2 math.

you don’t actually need to know any Glicko-2 math to calculate the likelihood of getting opponents in matchups, as long as you already know the ratings. but if you don’t know the ratings (for example because the current match isn’t over yet) then Glicko-2 is how you can calculate them (although calculating new ratings requires you to guess what the score outcome of a match will be).

-ken

The answer is clearly 42.

Hmmm i put that formula in excel and got that result. so how did u get 42.

I think it’s because 42 is the answer to everything.

Anyways, I’m pretty sure the deviation it’s talking about it the Ratings Deviation attached to each server with Glicko2. It’s not how much the server moves, so much as how unsure the ratings are that the ratings are right. The longer WvW goes on, the smaller that value should be.

Key word here is “should”….

Takerukun.8924:

I think it’s because 42 is the answer to everything.

Anyways, I’m pretty sure the deviation it’s talking about it the Ratings Deviation attached to each server with Glicko2. It’s not how much the server moves, so much as how unsure the ratings are that the ratings are right. The longer WvW goes on, the smaller that value should be.

Key word here is “should”….

LOL on the link. I did not understand what the professor was saying on glicko 2 system.

Btw can someone please explain the maths to us non-math based PHD holders.

Fuzzion.2504:

LOL on the link. I did not understand what the professor was saying on glicko 2 system.

Btw can someone please explain the maths to us non-math based PHD holders.

Umm…to try and make a long story short:

Each server has the ratings that you see on mos.millenium. Glicko2 looks at the ratings of each server in a match and predicts what the score should be base off the rankings. Based off how the score compares to the predicted scores, it shifts the rankings values to more closely match with what would predict what actually happened. The larger the uncertainty in a server’s rankings (the Ratings Deviation), the more drastic the shift possible. Whenever a score shifts (every week for WvW), the deviation should get smaller.

That’s not 100% accurate, but it’s close enough to get the concept across. Someone feel free to chime in if I messed up horribly somewhere.

Takerukun.8924:

Fuzzion.2504:

LOL on the link. I did not understand what the professor was saying on glicko 2 system.

Btw can someone please explain the maths to us non-math based PHD holders.

Umm…to try and make a long story short:

Each server has the ratings that you see on mos.millenium. Glicko2 looks at the ratings of each server in a match and predicts what the score should be base off the rankings. Based off how the score compares to the predicted scores, it shifts the rankings values to more closely match with what would predict what actually happened. The larger the uncertainty in a server’s rankings (the Ratings Deviation), the more drastic the shift possible. Whenever a score shifts (every week for WvW), the deviation should get smaller.

That’s not 100% accurate, but it’s close enough to get the concept across. Someone feel free to chime in if I messed up horribly somewhere.

I understood that part. But could you put it down into a mathematical formula

Oh. My bad.

The math you used in the first post is essentially correct. It’s just that the deviation is the Ratings Deviation assigned to each server. So, say JQ has a RD of 120 (No idea what the actual number is at the moment, but it’s always positive), the new math would allow for JQ’s Glicko2 score to adjust by:

120 * 0.45 + 10 = 64

It would then roll a random number from -1 to 1, multiply that by 64, and add that to JQ’s Glicko2 score while deciding matchups. So for matchmaking purposes, JQ would be 2,131.477 +- 64 = 2067.477 to 2195.477

However, if the RD was greater than 200, it would cap the +- at 100.

After doing that to every server, it would take the “new” ratings, and order them into matches like the way tiers originally worked.

As to the actual RD’s of each server, I have no idea where you would find them. There was an Anet post a couple months back that had them, but they’ve changed a lot since then. And I don’t particularily want to go through and try to calculate them all from there by hand. There’s also trial and error to try and get the same spreadsheet Anet just posted in the blog, but that would take more time than it’s worth.

If you’re desperate to know them, you could PM Weris from mos.millenium to ask for the current values, but I’d rather not bother him.

Not sure if that helps any. I’ll check tomorrow to see if I failed horribly at understanding your question again. >__<

So what we have here then is that a server may have +64 and may then have that work stripped down to -64 so say if JQ was 2100, and was meant to be 2164 it would now boe 2100 minus 64 to 2036? If this is so then winning is no longer a big deal ? I must be wrong.

The random-roll modification is only a temporary thing to have some variation in the match-making order. But your permanent rating is only modified by the outcome of matches.

Deviation you can see at https://leaderboards.guildwars2.com/de/na/wvw
it’s between 150-170 so the current maximal modification of your rank by the random roll is approx. 86

So TC with a predicted rank of 1942 after this weeks match (http://mos.millenium.org/na/matchups) may get a modified match rank of
2028 (roll a +1) till 1856 (roll a -1)
2028 (roll a +1) is behind the unmodified SoR with 2051, but SoR rolls a dice as well.
if it is between -0.26 and +1 SoR stays 3rd and T1 match is BG – JQ – SoR
if it is -1 till -0.27 SoR ends on 4th behind the +1 rolling TC and T1 match would be BG-JQ-TC

The roll also influence the color you get in a match. Currently JQ and BG will always be in T1, but if JQ rolls -1 and SoR rolls +1 then JQ is red and SoR is blue (or green).
If JQ rolls +1 and BG rolls -1 then JQ is green and BG blue (or red).

*So TC with a predicted rank of 1942 after this weeks match may get a modified match rank of
2028 (roll a +1) till 1856 (roll a -1) *

so you are saying that the random roll of the dice has a very significant effect on rank scores?

Because to go from 2028 to 1856 is range of 172. Is this correct?

No, RNG does not affect rank points, but matchups are calculated, based on rank points modified by random roll, instead of plain ranking

I appreciate your answer.

Anet, could you please clarify the maths for us.

-Fuzz

Ken is our best bet.

Anet’s spreadsheet they posted can’t be right. The percentages are all messed up.

Johje Holan.4607:

Ken is our best bet.

Anet’s spreadsheet they posted can’t be right. The percentages are all messed up.

It probably was correct once, but it is ages old as you can see on the order of servers.
If you look at EU-order and compare it to history I would say, it shows the probabilities of the ranking valid at the 15 November

As server rankings changed in the meantime probability did as well.

So another question the needs to be answered by anet is what week did they they use for tha ranking ontop of a explanation of the maths. And a bonus question is hen will season 2 start

Johje Holan.4607:

Ken is our best bet.

Anet’s spreadsheet they posted can’t be right. The percentages are all messed up.

up-to-date numbers are posted here:
https://forum-en.gw2archive.eu/forum/wuv/matchups/Who-would-you-like-to-face-next-week/3367752

assuming my code is right, of course. there’s always the possibility that it’s wrong but at this point it’s been reviewed pretty carefully (and not just by me, since the source is posted in the API forum other people have been able to review it as well).

the percentages posted are probably correct (even though they look a bit odd) but they are almost certainly based on old ratings. sometimes the numbers come out oddly because your ideal opponent gets dragged into another matchup. this kind of thing even happened from time to time under the old system (although not as often).

-ken

Something is wrong with new system, looking at new table (https://d3b4yo2b5lbfy.cloudfront.net/wp-content/uploads/2013/12/6f68004.png) Gunnar’s Hold shouldn’t be matched up against Baruch Bay and Augury Rock, yet GH got exactly them this reset.
Am i wrong?

Sokre.5714:

Something is wrong with new system, looking at new table (https://d3b4yo2b5lbfy.cloudfront.net/wp-content/uploads/2013/12/6f68004.png) Gunnar’s Hold shouldn’t be matched up against Baruch Bay and Augury Rock, yet GH got exactly them this reset.
Am i wrong?

The system is right, the table in ANet Blog is computed from a 1 month old rating.
Look at https://forum-en.gw2archive.eu/forum/wuv/matchups/Who-would-you-like-to-face-next-week/3367752 to see the actual probabilities.
Baruch Bay [SP] vs Gunnar’s Hold = 0.5% i.e. it can happen 1 in 200 games. You rolled very close to +1 and BB and AR both rolled close to -1.

Dayra.7405:

Deviation you can see at https://leaderboards.guildwars2.com/de/na/wvw

How did I miss that…? Don’t mind me as I bookmark and +1 your post.

Thank you!

Thank you Dayra for clarification.

Thanks for the answer, i shall quote you in my weekly score updates.

I just hope we’re not back to the old days of same match-ups over and over and over again. That was super-lame.

I just want to know why SOR and TC aren’t matched against one another as the rank 3 and 4 servers.

Cottontail.8526:

I just want to know why SOR and TC aren’t matched against one another as the rank 3 and 4 servers.

Given the current ratings, these 2 are the only ones that can take the 3rd spot in T1. (SoS and FA cannot “rollup” enough to be ahead a “downrolling” SoR)
But there is only 1 spot free (Neither JQ nor BG are able to roll out of T1) so it’s either SoR or TC in T1 and the other in T2. Therefore they cannot match.

One has to be the punching ball of JQ and BG.

Cottontail.8526:

I just want to know why SOR and TC aren’t matched against one another as the rank 3 and 4 servers.

the way the ratings are right now, JQ and BG are guaranteed to always be in tier 1. the only question is, who will fill the third spot in the T1 match? either SoR or TC has to do it — no other server has a rating high enough now that the randomness has been reduced.

if SoR and TC want to be able to play each other, you need to lose as badly as possible to SoS every time you face them. this will raise SoS’s rating high enough so that sometimes they will play against JQ and BG, and when that happens, that will make it possible for SoR and TC to play each other (unless you get unlucky that week and one or the other of you plays in T3 instead).

the other way for SoR and TC to play each other is for either JQ or BG to implode and drop to rank 3 or lower. that would make it possible for SoR and TC to face each other in the #2 and #3 spots of tier 1.

-ken

Anet, you went too far with the “drastically reduced” RNG. Raise it back up a bit. Why you guys always go too far?

Also, you should seperate the EU and NA RNG. I’m guessing that’s why you went so low. But you probably can’t do that can you.

yeah anet u did go wayy too far. now it feels like t3 with tc/fa matchups for the next 20 weeks all over again. the variety was way more fun. the only thing that was terrible was the leagues as servers like nsp had to fight fa orsbi which we all told u would happen and be a terrible blowout matchup.
now u over react and put us back to the good old boring same matchups for week after week after week.
i think i will just play eotm if this new matchup system stays like this!

selan.8354:

yeah anet u did go wayy too far. now it feels like t3 with tc/fa matchups for the next 20 weeks all over again. the variety was way more fun. the only thing that was terrible was the leagues as servers like nsp had to fight fa orsbi which we all told u would happen and be a terrible blowout matchup.
now u over react and put us back to the good old boring same matchups for week after week after week.
i think i will just play eotm if this new matchup system stays like this!

Anet always over-nerfs.

As for EoTM, you’ll be seeing the same matchups there as in your regular WvW matchup, so how does that help?

I don’t think ANet did over-do in nerfing variation.

The player did over-do in concentration on the top-server.

There is just no one that want to play against top 3, be it NA or EU.

Dayra.7405:

I don’t think ANet did over-do in nerfing variation.

The player did over-do in concentration on the top-server.

There is just no one that want to play against top 3, be it NA or EU.

I think it’s fine for the top three to be matched up for the rest of eternity — those who stacked to T1 deserve it, but considering how all other matchups turned out this week I’d say it’s likely they over-nerfed.

Admittedly this week NA rolled quite tier-aligned, but this is always a possibility and does not mean that it will always do it.
Admittedly NA has many holes in ranking. But if I remember the million posts against the random system everyone argued that these differences in ranking are adequate and matches over these differences will only result in blow-outs.

If I look at Ken’s probability list for NA
https://dviw3bl0enbyw.cloudfront.net/uploads/forum_attachment/file/125866/na1-20131213.txt
then nearly every server outside T1 has 4-5 server with probability over 10% to meet.
3 out of 5 are 8 different matches possible per server.