Showing Posts For Dumb Woob.9415:
The drop rate is .4% on the wiki drop rate.
you need 4000 chests to get a 100% chance to get one weapon with the common coffers.
For any n independent events with outcome x having probability less than 1, in order to ensure (100% chance, or probability of 1) that P(x)=1, n would have to be infinite.
This concept can be better simplified by flipping a coin. If I flip a coin twice, I can expect, on average, to get one ‘heads’ result. However, there is a .25 chance of two flips not being heads. Three coins, you say? Well, the chance of no heads is still .125. Four coins? .0625 and so on. For any finite number of coins, that probability of a negative result will always be greater than zero.
Edit: it should be obvious given the context, but I should specify that n is a positive integer.
Maybe I’m misunderstanding, but isn’t that part of what Zhaitaffy is trying to do?
I would agree with all of that except for the implied assumption that accuracy translates to parity. It doesn’t. Just because we know precisely how much better or worse one team is relative to another does not mean that we will get equitable (and therefore enjoyable) matches. All that it means is that we will know more accurately in advance how bad the mismatches will be.
Interesting points, and easy to see once you put it that way. I agree that knowing the true ratings distribution won’t necessarily lead to parity. However, knowing the true ratings might reveal some kind of differential in the ratings under the (more relatively) closed matchup system and this matchup system. I can only speculate when it comes to the outcomes of knowing this closer approximation of the true rankings on the average “excitingness” of matchups with a low-variance version of this matchmaking system. Would make for a pretty cool academic journal paper though.
Personally, I don’t think we can come up with 17 matches every week that will be void of mismatch—the real debate is finding the method to minimize such mismatches while still allowing for appropriate ranking/rating mobility. I don’t have those answers, which is why it’s thoroughly interesting to have these kinds of discussions.
2 points here
1) stop acting like you’re a kittening graduate at Harvard.
2) your sentence is impossible to follow because it lacks good grammar.
What makes you think that I’m a Harvard graduate? I’m not, and it’s not something I’m trying to represent.
And, yes. I probably could word a couple of those sentences differently (I just use the word just just too many times, justly). Though I don’t know that it’s impossible to follow…
I get the sense that you’re trying to be snarky for no particular reason.
Math and RNG are the best fun destroyers ever.
Two points:
1) Don’t talk bad about math. Though I concdede that I am a bit odd for doing math in my free time for fun, I contend that everybody likes math; just those who think they don’t just haven’t realized it yet.
2) What would be less fun: total certainty or a lot of uncertainty? A very interesting question, since you claim that total certainty is the only thing that is not un-fun. Unless, of course, you mean some artifacts associated with the implementation of a pseudo-RNG in this game are un-fun (which you should say instead—words mean things after all).
If you’re good at math tell me what numbers you want to make analysis of whether glicko was accurate or not and i’ll pull them from millenium and put them in excel for you.
I don’t know what you mean by this. Even if I did, I suspect I would to need to make some assumptions (and thus a weaker argument) in order to support one position or another.
What I can tell you is the previous system of matchups had reduced match variety to the extent that you could not easily compare servers by rating—which is why people ended up using tiers to talk about server rank. Specifically, you could only compare servers that were close together in rating—because they were the ones who played each other.
The current matchmaking system will bring more match variety and make corrections to the ratings structure to make the ratings of the servers more “accurate” or closer to their theoretical true values. Now, would we see perfectly accurate ratings? Of course not—we don’t have infinitely many matches to play. But this system over the course of 5-10 matches will probably better seed the servers. Then, the variation of the matchmaking rating can be reduced to make better matchups on average, with greater match-to-match variety.
Is this painful in the near term? Absolutely. It’s an artifact of the previous matchmaking system. However, I’m willing to bet that our patience will be rewarded.
dagnabbit…if it stays this way….We’ll,(FA)….will probably be playing TC and DB yet again, lol.
The sad part is that because of RNG we have no way of knowing if this is true.
A server can pull together, fight against all odds, over come all obstacles and at the last moment snatch victory from the jaws of defeat! Only to be punished next week by RNG’ing….
It’s pretty much impossible to have a better than 1/2 probability at a rematch…I think SoR-BG-JQ had the highest chance because of the massive rating difference and we were only about 1/3
What are these tiers you are talking about, and why do you think they are arguments/inputs for the stochastic function that determines a server’s matchmaking rating?
Oh, and you can find the blog post that talks about the nuts and bolts of the matchmaking system here:
https://www.guildwars2.com/en/news/big-changes-coming-to-wvw-matchups/
You can see exactly how they are determined.
(edited by Dumb Woob.9415)
Math and RNG are the best fun destroyers ever.
Two points:
1) Don’t talk bad about math. Though I concdede that I am a bit odd for doing math in my free time for fun, I contend that everybody likes math; just those who think they don’t just haven’t realized it yet.
2) What would be less fun: total certainty or a lot of uncertainty? A very interesting question, since you claim that total certainty is the only thing that is not un-fun. Unless, of course, you mean some artifacts associated with the implementation of a pseudo-RNG in this game are un-fun (which you should say instead—words mean things after all).
This was the matchup I wanted to see—I suspect Dragonbrand will be stiffer competition than most will think!
Good luck; and have fun!
Cheers,
Chris (Blackgate)
I mean, WvW is just a casual game mode.
This is where I stopped reading and started skimming.
You might be able to argue for a simple ranking that ships servers up and down based on results instead of a rating system. You might even be doing this, but you’re doing a bad job of it.
GVG ought to be sPVP.
It doesn’t make much sense to me how it could reasonably be a part of WvW…guilds who want to do GVG must make poor decisions in WVW to just set up matches, which is strange to me.
Lower ranked servers that are winning or have won this week, meaning they have a positive result (greater than .5) in the given matchup.
Note that this doesn’t even include servers that have done better than the rating differentials would expect.
EU:
6 Kodsah def. 5 Desolation
10 Riverside def. 8 Piken Square
19 Ruins of Surmia def. 18 Fort Ranik (virtually drawn)
16 Drakkar Lake def. 14 Aurora Glade
NA (In progress):
11 Crystal Desert def. 8 Kaineng
11 Crystal Desert def. 9 Yak’s Bend
23 Ferguson’s Crossing def. 20 Devona’s Rest
Cheers to that. Was a good fight this last week against you guys; a .506 – .494 result. Very close. Odds are we’ll get the same match again, but I would like to see a bit of fresh matchups. A BG-JQ-DB matchup would be interesting indeed—I suspect that Dragonbrand is better than their rating would suggest.
Looking forward to the rematch with a beverage in hand!
Cheers,
Chris (Blackgate)
exaudic
This is more sad than funny.
Here’s what the top-10 list of possible Ehmry Bay matchups looked like just before the current match started:
6.511720 Sea of Sorrows Crystal Desert
5.399440 Yak’s Bend Crystal Desert
5.019430 Borlis Pass Anvil Rock
4.857220 Sea of Sorrows Stormbluff Isle
4.484480 Borlis Pass Darkhaven
4.180100 Yak’s Bend Stormbluff Isle
4.031770 Kaineng Crystal Desert
3.206880 Kaineng Stormbluff Isle
2.965830 Yak’s Bend Sea of Sorrows
2.794180 Maguuma Crystal DesertHere’s what it looks like now:
8.674800 Borlis Pass Anvil Rock
5.203620 Borlis Pass Darkhaven
4.158190 Borlis Pass Sorrow’s Furnace
4.044620 Kaineng Crystal Desert
3.993170 Yak’s Bend Crystal Desert
3.547160 Maguuma Crystal Desert
3.122490 Yak’s Bend Kaineng
3.052280 Anvil Rock Darkhaven
3.046610 Sea of Sorrows Crystal Desert
2.837190 Maguuma KainengNote that in the first set, only 2 out of the top 10 potential matchups had either BP or AR as opponents.
Now 4 out of the top 10 potential matchups have either BP or AR as opponents, and the likelihood of getting one or the other has gone way up.
Yes, it’s true that the random numbers might save us next week, but it’s half as likely that we’ll get a good matchup next week as it was this week. This is why so many players on EB feel like we were screwed over by the RNG. It’s not just that we got a bad roll, but that all the servers around us got good rolls, and the combination has about doubled the chances we’ll get a bad roll next week.
-ken
You’re going from a .950 chance of no rematch to a .913 chance of no rematch.
Okay. This is mostly explained by your rating decrease this week. Again, n=1 is bad luck, not an inescapable black hole..but really whining about the dealer getting Blackjack is kind of silly, yes?
Oh well.
I’m a bit confused. Doesn’t Arenanet already publish this kind of stuff? Aside from the guild networking bit..
The point is, i believe, the servers that were matched with a higher tier servers will likely gain score (due to being matched with a higher score servers) despite loosing. Servers that were moved down a tier will likely gain score due to outmatching the lower tier servers. Servers that stayed in the same tier will likely remain with unchanged (in any meaningful way) score. That means that the gap will widen from the previous weeks for the servers that stayed unchanged. That in turns means that the random component for the “unchanged” servers will have to be higher to move up or down, which is less likely. Meaning that the chances of t5 staying the same way will increase after this week.
Your statements seem to be unclear, but I’ll try to address what you mean rather than what you’re saying. Servers matched against higher-rated servers seem to be gaining in rating. This is part of the adjustment from the artificial ratings caused by the former matchmaking system.
Even if your conjecture (given without proof, by the way—for shame) about the probability of the 5th matchup this week compared to the probability of the 5th matchup next week is true (which I’m willing to accept, given the way that Ehmry Bay is playing, even though the 12-13-14 matchup is most likely for you next week) for the first week of matchmaking, how can you argue that such a system will only result in less varied matchups as time becomes larger? This adjustment process takes more than a single match to make.
By looking at new WvW score (Millenium courtesy) and after the drama effects from last Friday resets regarding to North America, until now it only benefits 5 servers:
Sea of Sorrows ……………………. (moved 2 tiers up, from previews tier)
Stormbluff Isle ……………………. (moved 2 tiers up, from previews tier)
Crystal Desert …………………….. (remains same tier, but other 2 servers went down 1 tier )
Northern Shiverpeaks ……… (moved 1 tier up, from previews tier)
Fergusson’s Crossing …………. (moved 1 tier up, from previews tier)Common denominator involving those worlds: all are fighting servers coming from above tiers and their borderland is RED!
Regarding to tier 5, after 12 weeks in a row, my homeworld Borlis Pass matchup against Anvil Rock + Ehmry Bay (old tier 4).
Ebay population playing WvW is the double of any other servers in this tier (AR + BP). Add this to their experience in previews tier 4 and our threesome matchup end week after week in tie rounds. Finally, last week they earn the right to move up to tier 4, due to Anvil Rock sacrifice in fight less to let them go, both servers red and blue expected Stormbluff Island coming down a tier to refresh this stagnant tier 5.
Unfortunately, last Friday reset result to push back Ehmry Bay even when they start to smell tier 4 and ….. end up with a deception for all the 3 servers envolved! We think this is a forced marriage done by ArenaNet and most of the players can’t stand another matchup like this, a few are leaving GW2 while others return to PVE or doing large breaks from this game until find another with plenty of fun (this one is losing theirs)!
In resume, this week reset, can you ArenaNet give us hope and refresh tier 5? THAT WOULD BE AWESOME THANK YOU
“We cannot get out” tier 5 image, after 13 weeks:
Something tells me you have basically no idea how the current matchmaking system works. Here’s a hint: get away from thinking about a tiered matchmaking structure and start thinking about ratings matchups.
My server, too, drew the same opponents as last week; I can see how that might be frustrating. But hey—that happens. It’s probable that you’ll get new opponents next week.
I love this thread so much.
Your right. Give it some time and lets see what happens. gliko or no gliko that doesn’t change the fact that a T4 server is no where near a T2 server. You think that gliko was holding that server back or that it was holding a higher server from falling that isn’t the case. Becasue free transfers are what caused the servers to be the way they are not the system. The real scores were already developed, any score from now on will just go back to how it was before eventually.
You should get away from thinking about these “tiers,” artifacts of the old matchmaking system that had little mobility. Instead, consider rankings, or better—ratings.
Analogous to your statement: Yes—a 1300-rated server should be much worse than a 1750-rated server. But when that 1300-rated server gets a .4-.6 result against the 1700-rated server (much better than what Glicko would expect), you have to think that there is some correcting to do in the rating scheme. To say the real ratings are already known is to not understand how Glicko (and other Elo-type) ratings systems work, regardless of how we arrived at the current disparity between servers..
Edit: To clarify, I’m not pulling this example out of nowhere. What my matchup describes is between the 20 seed and the 8 seed in EU.
BG noobie guardian here.
Had loads of fun this evening defending Dreadfall Bay—against both JQ and SOR at the same time! That last portal bomb into the Keep might have worked out if it was just 50 distance or so further into the keep—though it was a great effort.
Looking forward to similar fights in the future!
The argument that the new system will cause server ratings to move closer together seems to be based on the idea that higher ranked servers generally bleed points to the lower ranked server in their matchup.
For this to be the case, the expectation would have to be for lower ranked servers to do better than Glicko expects them to. Whether or not this is true is dependent on how the scores actually correlate with the rankings, so the parameters in Glicko 2 alone alone aren’t enough to prove anything. You would need to analyse past results.
Very good! What seems to be embedded in this problem is there exists some difference in the rating distribution between the previous and current systems—an artifact of the restricted nature of the previous matchmaking system. It follows (without proof) that a wider variety of matches will move the ratings to this true^ rating distribution. We can, of course, include such assumptions in a proof like this; these assumptions are actually most likely necessary.The question becomes: will this movement in ratings lead to better matches, on average?
^It isn’t actually the true distribution because we would need a very large amount of time and an unlimited matchmaking system to get that information into Glicko. Not to mention changes that occur to each realm over time.
Has nobody pointed out that (1) and (2) cited by OP are both identical rating systems? Just the spread of ratings among opponents are higher, on average.
You should be talking about matchup systems, not rating systems, I think.
This empty debate bores me.
Someone please prove or disprove (mathematically using the parameters in Glicko 2) the following statement:
Let the current time t =0. Let X={x1, x2, .. , xn} be the set of realms in the Glicko rating system. As t becomes large, the average rating difference between any two xk in X will be lower under the current matchup system compared to the previous matchup system.
Don’t despair! If this continues, SBI will move up 60 Elo points after this round. Honest. Since you’re so much lower rated than your opponents, you don’t have to score as many points to get a positive rating. It’s sort of like how a draw in chess can result in a rating change if the opponents are rated much different from each other.
yeh yeh really funny if a t3 goes against t7.
the t7 is gonna rage quit and WvW lose more ppls VERY GOOD IDEA!
(leave the game not just WvW)
Ah but this would be a mistake. A server at a vast rating disadvantage (to use your example, a 10 Seed versus a 20 Seed, it makes little sense to use tiers now that there is match-to-match mobility) can come out ahead in rating even if it scores less than 0.5 of the points against the higher rated server. That is, there is also pressure on the 10 Seeded server to dominate the match; if the 20 Seeded server is able to put up enough of a struggle (gaining valuable experience against a better team, as well) it might be able to move up a spot—19th Seed. This is the motor behind rank mobility with the newer system.
This guy gets it.
Lower seeded servers could still gain rating even while getting blown out. The math would then re-adjust the following week with higher probability of finding accurate matches in the newly-adjusted rating spread.
Essentially, this should fix glicko’s tendency to only absorb rating points from within your own tier and lead to much more volatility in the positions.
Oh you’re nice.
To provide an example outside of our complicated system (and then apply it to that system), consider an FIDE rated chess game between an 1800 rated and 2200 rated player. If they draw (1/2 – 1/2), the 1800 rated player gains 6.4 points!
In GW2, results are between 0 and 1 depending on the ratio of your points and the total of your and your opponent’s points. The higher the rating differential, the less you have to score (perhaps even as low as 0.25—a 3-1 defeat) to get a positive rating change.
Very interesting stuff.
We dominated last week against Whiteside and Arborstone too. How does it come that AR got up of tier and we got to T9
The “Digging Deeper” section of this page should help you understand the matchmaking system.
https://www.guildwars2.com/en/news/big-changes-coming-to-wvw-matchups/
Consider that you are a chess player: you will sometimes be matched against weaker players. You’re expected to win. If you draw, your rating will decline. When you get matched against a higher rated player, fight your best for a draw—you might not win but you can still gain rating.
yeh yeh really funny if a t3 goes against t7.
the t7 is gonna rage quit and WvW lose more ppls VERY GOOD IDEA!
(leave the game not just WvW)
Ah but this would be a mistake. A server at a vast rating disadvantage (to use your example, a 10 Seed versus a 20 Seed, it makes little sense to use tiers now that there is match-to-match mobility) can come out ahead in rating even if it scores less than 0.5 of the points against the higher rated server. That is, there is also pressure on the 10 Seeded server to dominate the match; if the 20 Seeded server is able to put up enough of a struggle (gaining valuable experience against a better team, as well) it might be able to move up a spot—19th Seed. This is the motor behind rank mobility with the newer system.
Ah, skill. Tough to define.
The economist in me tends to try to measure it in terms of decision making. Here, I use the term decision very loosely: any action you take in game—even inaction—is a decision. When I talk about positions, I am talking about the exact circumstance at that point in time: skill cooldowns, relative positions of friendly or foe players, NPCs, etc. and environmental circumstances.
Now “skill” for me boils down to making the best (or optimal) moves in a given position for any position a player finds him/herself in. Often this might come with a short-term sacrifice for later-term gain or counterplay.
“In chess, there are only three kinds of moves: a good move, a bad move, and the best move.”
Let me give you this link, it will explain some things about statistics.
Large data sets aren’t necessary?
In a roll where certain events can happen 1/1,000 of the time (Source: http://wiki.guildwars2.com/wiki/Lost_Orrian_Jewelry_Box/Drop_rate. All the lodestone are 0.1% or lower) you think 83 rolls is enough to get an accurate distribution? Are you serious?
This is the crux of the difference between salvaging ectos and opening these boxes. Since the success rate is already exceedingly small, to detect a decrease in this rate (which is necessarily also small), you need a much larger sample size to make a statistically significant statement about some kind of change in the rate.
With ectos, the prior rate was about .9 per item and dropped to around .5 per item, making a relatively small sample just fine. However, if we consider the 1/1000 drops from the Lost Orrian Jewelry Box, we would need to open about 3,000 boxes to be 95% assured to get at least one of those drops. As a corollary, we would need a sample slightly higher than that to make any conclusions about a statistically significant change, if there is one.
I appreciate you using my post in the other thread to try to shed light on an issue, but I’m afraid the examples aren’t exactly compatible.
I’m not being mean but 100 isn’t even close to a statistically valid sample (~1000 is a bare minimum, and 2000+ is ideal). RNG really can just be that bad sometimes, especially since it’s not ‘truly’ random and discrepancies can creep in due to the server clock.
While the overall possibility of a change can’t be completely ruled out, it wouldn’t make any sense for it to be directly linked to a jump in TP prices. Luxury goods like ectos don’t move fast enough for the hike to be caused by a supply run. It’s either hoarding or a bank panic (or both).
Please explain, using confidence intervals and levels, why 100 salvages isn’t a valid sample size to determine probability of ecto returns. Show the math, please.
Pretty much this. Without doing the math, I’m willing to bet you could conclude there is some statistically significant difference in ecto salvage yield per item (with alpha being 0.05, for the sake of defining one) with a sample of just 20 or so. This is because we can use the fact that each salvage is an independent and identically distributed event to take advantage of one of the most powerful concepts in inferential statistics: the Central Limit Theorem.
Samples of a couple of thousand are often used to make estimates of things like national-level opinions, but in these cases there are a ton of variables to deal with, making a large sample necessary to minimize sampling error—ensuring that your sample is indeed a representative of the population. Here, though, each rare is exactly the same as all the others (for the purposes of salvaging ectos), meaning there aren’t any other variables to control for in the population—making relatively small samples illustrative of any change (especially if the change is a large one, like how we have observed).
Now, like I said earlier my estimate using the binomial distribution is a touch flawed, but not 16 orders of magnitude flawed.
It is a binomial distribution when you count ectos returned as the success whether it’s 1, 2, or 3. It’s a very safe assumption that the quantity is determined on a secondary loot roll. Otherwise, great post.
Ah I was considering total number of ectos as somehow part of the binary “do you get an ecto?” calculation. Thanks for making this more clear.
Edit: I think I should use a modified example of this to illustrate my students next year the power of the Central Limit Theorem..
(edited by Dumb Woob.9415)
Armor level 80 yellow salvaged with Master Salvage Kit
775 Armors
402 Ectoplasm
634 Rune
1120 Common Materials
99 Rare materials
Drop Rate Ectoplasm 51.87
In the past I had 89 % ration on ecto from rareI would like to ask anyone with a sound knowledge of statistics: what is the percentage chance of this occurring if we assume the 0.9 return rate is unchanged?
Now, this isn’t truly a binomial event, because we can get more than one success per attempt. But we can approximate this as a binomial event, with a .89 probability of getting the desired result. The number of trials here is what should jump out to you to tell you this result is strange if the .89 rate was still there.
If we use the binomial distribution to make this approximation, there is a probability of 4.441 × 10^(-16) that 775 trials would yield 445 or fewer successes given a success rate of .89. That is what leads these folks to the conclusion that the likelihood of the rate being the same (Type I error, here) is extremely small (you probably have a better chance of dying by a lightning strike, and someone at your funeral also dying by a lightning strike in the next year). Typical thresholds for these kinds of conclusions are 0.001, 0.01, or 0.1. This is something like 0.0000000000000004.
Now, like I said earlier my estimate using the binomial distribution is a touch flawed, but not 16 orders of magnitude flawed.