Guild Wars 2

Here are the estimated matchup probabilities for the upcoming NA matchups. I am assuming here that CD, SF and DH will each get +200 applied to their ratings for matchmaking purposes. The last post I’ve seen from ArenaNet on this topic indicates that CD would get a +200 adjustment this week but said nothing about SF and DH so I am assuming they will get the same adjustment as last week. If these assumptions are wrong then of course the probabilities will be wrong as well.

Here are the estimated matchup probabilities for the upcoming matchups. I am assuming here that CD, SF and DH will each get +200 applied to their ratings for NA matchmaking purposes. It’s unclear whether ArenaNet will actually apply these adjustments or not. It may be best to not trust the NA numbers at all. EU numbers should be fine.

They didn’t say they were adding +200 again, I was assuming they weren’t since they hadn’t said they were adding any [points this reset.

They weren’t. It was a one time glicko bump and they only said they would remove it at a later time.

The problem is this post here:
https://forum-en.gw2archive.eu/forum/game/wuv/Glicko-Rating-Manual-Adjustments-7-29/6266029

It says that CD was going to get a +300 bump the first week, then a +200 bump the following week. But it didn’t say if the following week would be the last week of temporary changes, or just the first week of +200 bumps that would continue a little longer.

Basically they haven’t said how long “temporary” is. Clearly it’s at least 2 weeks, but we don’t know if it’s more, or if it is how much more.

-ken

Well, here are the estimated matchup probabilities for the upcoming matchups. It’s unclear at this point whether ArenaNet is still adjusting matchmaking ratings for the bottom 3 NA servers or not, so I’m posting numbers both with adjustments (“adj” in the filename) and without.

here are the estimated matchup probabilities for the upcoming matchups. It’s still unclear whether ArenaNet is still adjusting matchmaking ratings or not, but at this point unless we see an “impossible” outcome I’m going to assume we are back to normal unadjusted matchups.

Server relinks are supposed to happen tomorrow which “normally” includes a reset of glicko volatility and deviation.

Also, we don’t know who will be host or guest. I wouldn’t expect DH/SF/CD to be hosts again, but who knows… If they intent to do more multi server links (like the quadserver fiasco) they will have to use low pop servers as hosts.

not sure what happened to the supposed resets for volatility and deviation, because the leaderboard (https://leaderboards.guildwars2.com/en/na/wvwmain) doesn’t show any change.

anyway, here are the probabilities for the new matchups (with new linkings applied).

I was wondering that too. Thought volatility and deviation were going to be reset every time servers were relinked.

here are the estimated matchup probababilities for the upcoming matchups. “na1” and “eu1” files show the probabilities of getting any single opponent, “na2” and “eu2” files show the probabilities of getting a particular pair of opponents. the last two files are the estimated rating outcomes for the current matchups, assuming that final scores don’t meaningfully differ from today’s scores.

here are the estimated matchup probababilities for the upcoming matchups. “na1” and “eu1” files show the probabilities of getting any single opponent, “na2” and “eu2” files show the probabilities of getting a particular pair of opponents. the last two files are the estimated rating outcomes for the current matchups, assuming that final scores don’t meaningfully differ from today’s scores.

So FSP vs Piken vs Jade incoming… But I must say I lost faith in calculations after we got a matchup with like 0.0001% chance a while ago, lol.

yeah, humans have a hard time dealing with low probabilities. our brains aren’t wired to correctly process that kind of information.

it doesn’t help that the actual end-of-match scores are sometimes different enough from the estimated scores that the actual ratings end up different from the estimated ratings, which throws off the rest of the calculations. this is the main reason I post the estimated rating files (the last 2); if the estimated ratings turn out to be wrong then the probabilities will be wrong as well.

posting on Thursdays is a balance between estimating matchup outcomes too early to be accurate, versus posting probabilities too late to be seen by all interested players.

anyway:

here are the estimated matchup probababilities for the upcoming matchups. “na1” and “eu1” files show the probabilities of getting any single opponent, “na2” and “eu2” files show the probabilities of getting a particular pair of opponents. the last two files are the estimated rating outcomes for the current matchups, assuming that final scores don’t meaningfully differ from today’s scores.

here are the estimated matchup probababilities for the upcoming matchups. “na1” and “eu1” files show the probabilities of getting any single opponent, “na2” and “eu2” files show the probabilities of getting a particular pair of opponents. the last two files are the estimated rating outcomes for the current matchups, assuming that final scores don’t meaningfully differ from today’s scores.

here are the estimated matchup probababilities for the upcoming matchups. “na1” and “eu1” files show the probabilities of getting any single opponent, “na2” and “eu2” files show the probabilities of getting a particular pair of opponents. the last two files are the estimated rating outcomes for the current matchups, assuming that final scores don’t meaningfully differ from today’s scores.

I want to see FA and SoS(my current server) matched up. I have a few peeps that need to be taught a lesson….

ugh….JQ needs to go away.

Aww, not enough chance of getting NSP. They’re killing it in T4 need to move up so we can face them.

Simple. Just get some people to post here asking for NSP to get glicko adjusted.

Chaba.5410:

Simple. Just get some people to post here asking for NSP to get glicko adjusted.

It’s hapenning…

Sorry about missing last week, my PC was being repaired and I don’t have the programs I need for this installed on another computer.

For this week, I started the calculations but then I saw the 10/7 Glicko adjustment post (https://forum-en.gw2archive.eu/forum/game/wuv/Glicko-Temporary-Manual-Adjustments-10-7). Yak’s Bend is going from a +150 adjustment to +75?

Oops. I had no idea there was a +150 adjustment in place. Apparently it has been there since 9/16. So all the NA numbers I’ve posted since 9/15 have been wrong.

So, here are this week’s numbers. I don’t know if the NA numbers are right or not (as usual, ArenaNet has said the adjustments are “temporary” but hasn’t said how long “temporary” is. Is YB going back to the default of +150 this week? Or are they staying at +75? If they stay at +75 does that mean NSP and DB are staying at +150/-150?

I am assuming that the NSP and DB adjustments are being removed, since NSP is now solidly in T3 and DB is now solidly in T4 (last place in the ratings), as ArenaNet intended. Continuing +150 for NSP will put them solidly in T2 (with a chance to get into T1), and continuing -150 for DB will give them a 1455 rating which is just insulting.

And what about YB? Continuing +75 for them would give them a matchup rating of 2046, which is higher than the #1 server. While I believe that ArenaNet thinks YB is strong enough to stay in tier 1, I don’t think they believe YB should be the highest-rated server in NA. So I’m guessing the +75 is going away this week as well.

That means, I am applying no adjustments at all this week. I am going to assume that the 10/7 adjustments were for that week only (even though the 9/16 adjustment turned out to be “until further notice”). I would really prefer for ArenaNet to be clearer about this kind of thing.

Snowreap.5174:

I am assuming that the NSP and DB adjustments are being removed, since NSP is now solidly in T3 and DB is now solidly in T4 (last place in the ratings), as ArenaNet intended. Continuing +150 for NSP will put them solidly in T2 (with a chance to get into T1), and continuing -150 for DB will give them a 1455 rating which is just insulting.

And what about YB? Continuing +75 for them would give them a matchup rating of 2046, which is higher than the #1 server. While I believe that ArenaNet thinks YB is strong enough to stay in tier 1, I don’t think they believe YB should be the highest-rated server in NA. So I’m guessing the +75 is going away this week as well.

From what I see happening lately, everytime they do these manual adjustments they keep them live until said otherwise (or until relinked)… At least this is what happened to CD +200 adjustment last links and the YB first adjustment of +150. Pretty sure NSP still have +150, DB have -150 and YB have +75.

yeah, that was going to be my assumption as well, that “temporary” really means “until further notice”. Actually I am expecting ArenaNet to post today or tomorrow that the adjustments are being removed; it really doesn’t make sense to continue them.

In anticipation of such a post, I’ve run the calculations without any adjustments. I’d like to avoid (a) having to run the calculations over again and (b) having to monitor these forums for the next 30 hours to see if Arenanet says anything.

Here’s what the NA numbers would look like if the 10/7 adjustments stay in effect this week.

Snowreap.5174:

I am assuming that the NSP and DB adjustments are being removed, since NSP is now solidly in T3 and DB is now solidly in T4 (last place in the ratings), as ArenaNet intended. Continuing +150 for NSP will put them solidly in T2 (with a chance to get into T1), and continuing -150 for DB will give them a 1455 rating which is just insulting.

I think you may be misunderstanding how the “temporary” adjustment works. “Continuing” doesn’t mean DB will get -150 additional on top of their new rating at the end of the match. The rating changes for servers that have been adjusted “temporarily” don’t exhibit that kind of behavior you are describing. Seems more like a one-time adjustment on a rating on X date that can be removed/changed later with some complex calculations that would affect rating at the end of next match. They do have all the data of the historical matches and can probably re-run predicted rating on past matches without the temporary adjustment.

I don’t know… After seeing NSP rolling a T2 match in FA place like when they had a 150 rating difference feels simply like they are just adding the rating at the time of RNG roll.

Jeknar.6184:

I don’t know… After seeing NSP rolling a T2 match in FA place like when they had a 150 rating difference feels simply like they are just adding the rating at the time of RNG roll.

When they adjust a server they adjust it at the end of a match up no?

Jeknar.6184:

I don’t know… After seeing NSP rolling a T2 match in FA place like when they had a 150 rating difference feels simply like they are just adding the rating at the time of RNG roll.

That’s possible. That wouldn’t be reflected in published ratings then. But what confuses me about it though is how YB’s “temporary adjustment” was changed to +75. YB’s showing a published rating that puts them as rank 3 server pretty solid in T1 so why would their adjustment need to be reduced then?

MaLeVoLenT.8129:

When they adjust a server they adjust it at the end of a match up no?

Yeah, but after the ratings calculations from the matchup result are ended. So instead of rolling with the rating that resulted from the matchup, they add the adjustment rating prior the RNG roll to define the new matchups. If NSP match ended right now, they would gain 31 rating going from 1734 to 1765. Then to roll the matchups they would be using 1915 (1765+150) as NSP instead.

At least that’s how I belive it work right now as we’ve been seeing too many low % matchups happening to servers that had their ratings adjusted.

Chaba.5410:

That’s possible. That wouldn’t be reflected in published ratings then. But what confuses me about it though is how YB’s “temporary adjustment” was changed to +75. YB’s showing a published rating that puts them as rank 3 server pretty solid in T1 so why would their adjustment need to be reduced then?

Beats me… Maybe they wanted YB to still have chances of rolling T2 (#RotateTheDream), or maybe they just wanted to YB not be the highest rated server as +150 would put them far ahead of the other 2 T1 servers even with less than 2000 actual rating.
YB rolled green in a T1 matchup twice, which means that the RNG rolls ended with YB being the highest rated server both times and it wasn’t even close to T1 server ratings.

Jeknar.6184:

Maybe they wanted YB to still have chances of rolling T2 (#RotateTheDream)

Which YB wouldn’t even need +75 to their published rating on the RNG roll to do….

Shouldn’t the chance of HoD and SoS rolling tier 4 (or 3) be almost identical since they are only 3 points apart on glicko? Those numbers show HoD with a 66% chance of rolling tier 4, and SoS around 30%?

HoD and SoS ended up only 3 points apart, but for the purposes of the calculations they were estimated to end up 35 points apart, based on how the scores looked on Thursday.

the actual matchups are done based on the final ratings at the end of the matchup, but at that point nobody will be interested in probabilities. Hence the need to estimate what we think the ratings will be at the end, based on how the scores look on Thursday. Normally the scores on Friday aren’t very different from what they were on Thursday, percentage-wise, but if a server makes a last-day push things can change and that will throw off the numbers.

Here is what the probabilities would have looked like if the estimated ratings had been exactly right (which they never are):

ArenaNet has explained pretty clearly how the “temporary” adjustments are applied. They don’t affect the published ratings, they are just silently added/subtracted for the purposes of making matchups.

Note that while the adjustments do not appear in the official ratings, they do appear in the predicted ratings I post, because that’s the way my tools work. Arenanet discards the ‘adjusted’ ratings immediately after matchups are done, but I keep them so it’s clear to everyone what ratings I used for the calculations. But I do not carry them over week to week; every week I start with the official ratings posted on the leaderboard, estimate the new ratings based on current scores, then I manually apply whatever adjustment I have reason to expect will be in effect for the next matchups.

Despite the lack of any announcement, it does appear that the adjustments were removed last week. The NA matchups we got don’t appear to have been possible if the adjustments had still been in place.

Here are the predicted probabilities and ratings for the upcoming matchups, assuming no adjustments.

Let’s not forget this friday is relinking day.

Too late, already forgot.

If we get sufficiently advanced notice of the new linkings I can re-run the calculations. I am guessing we will not.

Turns out the new linkings are the same as far as the calculations are concerned. Only the host servers affect the calculations, and since the list of host servers didn’t change the numbers turned out to be fine as-is.

Here are the numbers for this week.

is it just me, or does the WvW leaderboard not have updated ratings for the NA servers?

https://leaderboards.guildwars2.com/en/na/wvwmain

NA says “Next Update 11/4/16 6:03 PM PST” which is very overdue. All the NA ratings are exactly the same as they were on 11/3. EU appears to be fine.

I haven’t actually played GW2 in a long time, so there are a lot of things I haven’t kept up with. I just read something about skirmish “victory points” and “war score” now contributing to Glicko rating adjustments? And I saw something else that says that victory points are not available via any API?

Is all this true? Because if it is, it means I haven’t been correctly calculating new ratings the past few weeks, and that I cannot calculate new ratings going forward because the GW2 API does not have the needed information (i.e. this is not simply a matter of writing some code to handle things a new way), which in turn means the probabilities will be wrong.

In which case, calculating and posting these numbers is a colossal waste of time (even more than it was before, I mean). So unless I get me some education quick, I think that means the end of my probability estimates.

Here’s the important bits if someone else wants to attempt to take up this task (I had to make all these files .txt files to attach them, so remove the .txt extension after downloading).

I use the calc.cmd file to make things easier on myself. It’s not necessary, it just saves some typing. basically you invoke calc.cmd and tell it whether you want numbers for NA or EU, the date of the file containing leaderboard ratings, and today’s date. It will then do the rest, calling gw2match.exe for most of the grunt work.

I use sed to delete all the linked worlds from the leaderboard file, because linked worlds don’t matter for anything related to scores or ratings. I included the sed instruction files that do this, if you are interested. sed itself is available from numerous sources; I use the GnuWin32 version myself but just about any version is fine.

Most of the work is done by a program called gw2match.exe, which is written in C# (source attached as Program.cs). It has very few comments so it’s not the easiest thing to follow, but I didn’t expect to have to ever explain how it worked to anybody. You will also need a .NET JSON parser. These are commonly available so I didn’t bother to attach mine.

gw2match has lots of modes, most of which are for experimentation and irrelevant. The important modes for calculating probabilities are mode 0, mode 1 and mode 2.

Mode 0 calculates new ratings based on 2 inputs: previous ratings and current scores. Current scores normally come from the GW2 API but they can be read from a file if you want to see what the rating outcome would have been if scores had been different, or if the scores you want to use aren’t currently available from the API.

Mode 1 calculates probabilities of getting a particular world as one of your opponents. It needs a ratings file as input because ratings affect the probabilities. Mode 1 produces the “na1” and “eu1” files.

Mode 2 calculates probabilities of getting a particular pair of opponents. It also needs a ratings file as input. Mode 2 produces the “na2” and “eu2” files.

If somebody wants to calculate probabilities going forward, the first step is to figure out how skirmish scoring works for rating updates. Then either arrange for the needed scores to be in a file, or figure out how to derive what you need from the API. Then make any changes needed to the Mode 0 code (possibly no changes at all if you can generate a score file), and the rest should work as is.

Snowreap, thank you for doing these. The victory points are a different scoring method.

I think MOS keeps track of victory points outside of the API and shows predicted ratings based on it because MOS doesn’t show predicted ratings until after the end of the first skirmish. The other website, http://wvwmatchups.azurewebsites.net shows some probabilities but doesn’t advertise how they arrive at them.