Guild Wars 2

LittleLepton.8915:

Kenage Achalarus.4276:

myth: after you hit 3000 hours of playing guild wars 2 you will reweave a precursor from a drop. that was non mystic forge,

myth: Busted played 2251 hours on mesmer and 3703 hours total. myth way busted

I don’t really think this is a myth. I don’t know a single person who actually thinks that their number if hours has any bearing on their precursor chances.

Maybe in all fairness, statistically you should have had one drop if you have actually played that many hours. I’ve played more, gone dead broke trying, and I’m still waiting.

Myth #1: People posting information about getting precursors from the forge are posting good data in good faith.

Valandil Dragonhart.2371:

LittleLepton.8915:

Kenage Achalarus.4276:

myth: after you hit 3000 hours of playing guild wars 2 you will reweave a precursor from a drop. that was non mystic forge,

myth: Busted played 2251 hours on mesmer and 3703 hours total. myth way busted

I don’t really think this is a myth. I don’t know a single person who actually thinks that their number if hours has any bearing on their precursor chances.

Maybe in all fairness, statistically you should have had one drop if you have actually played that many hours. I’ve played more, gone dead broke trying, and I’m still waiting.

…. Okay. The problem is that it doesn’t matter how many times you have killed something that had a precursor in the loot table, you still have the same chance as when you first started playing. 1 h of lvl 80 champion kills, say, at 500 playtime hours = 1 h of lvl 80 champion kills at say 5000 playtime hours. It has no bearing whatsoever. If you have the luck, if RNG likes you, etc, then you might get a precursor from a mob, champion, dungeon boss, wvw chest, etc etc. if you haven’t gotten one in 5k hours, that stinks! But it still has no bearing on your likelihood of getting one from the next monster you kill. That’s not a myth. That just is.

Essence Snow.3194:

I think what they are getting at is that it’s not pure rng. It’s modified so that certain occurrences happen more than others.

It’s pure RNG with different probabilities of events. For each weapon type, there are 12 commons (7 pearls, legionnaire, tribal, etched, ceremonial, orrian) which drop with equal probability and collectively drop around 5/6 of the time, 7 named exotics which also drop with equal probability and collectively account for a bit over 16% of drops, and then the precursors – either the one for the weapon type, or one of the two greatswords picked randomly – which pops out somewhere between 1/125 and 1/200th of the time (roughly).

LittleLepton.8915:

…. Okay. The problem is that it doesn’t matter how many times you have killed something that had a precursor in the loot table, you still have the same chance as when you first started playing. 1 h of lvl 80 champion kills, say, at 500 playtime hours = 1 h of lvl 80 champion kills at say 5000 playtime hours. It has no bearing whatsoever. If you have the luck, if RNG likes you, etc, then you might get a precursor from a mob, champion, dungeon boss, wvw chest, etc etc. if you haven’t gotten one in 5k hours, that stinks! But it still has no bearing on your likelihood of getting one from the next monster you kill. That’s not a myth. That just is.

Herein lies the problem…there is no guaranteed reward for time spent. The AP chests are an improvement over what it used to be, which was nothing, but overall ANet punishes you for time spent with ridiculous RNG, which I still think is broken, and DR. Endgame items should NEVER be based on RNG. I don’t think DR was ever about bots but market control and time gating. If it were about bots, why was it not removed when the majority of bot accounts were removed? Because, it was never about bots.

Ensign.2189:

Essence Snow.3194:

I think what they are getting at is that it’s not pure rng. It’s modified so that certain occurrences happen more than others.

It’s pure RNG with different probabilities of events. For each weapon type, there are 12 commons (7 pearls, legionnaire, tribal, etched, ceremonial, orrian) which drop with equal probability and collectively drop around 5/6 of the time, 7 named exotics which also drop with equal probability and collectively account for a bit over 16% of drops, and then the precursors – either the one for the weapon type, or one of the two greatswords picked randomly – which pops out somewhere between 1/125 and 1/200th of the time (roughly).

Perhaps this is common knowledge that I am unaware of, but where did you get these probabilities and much more importantly, how do you know they are right?

oilstorm.1748:

Perhaps this is common knowledge that I am unaware of, but where did you get these probabilities and much more importantly, how do you know they are right?

Probabilities are inferred from a large sample (~8000 exotic forges) of forge output, combining my own data with a couple other sample sets I have high confidence in (such as the 10,000 gold forging streamed and publicized on reddit recently).

I know they are ‘right’ in as much as a data set of that size tells you a lot about the credible range of parameter values; for example I can tell you that based upon that data set I can say with 99.9% confidence that the precursor rate is no better than 1 in 113, and with 99.996% confidence that it’s no better than 1 in 100.

As for there being 12 commons of each type, and that the forge only kicks out 7 named, non-precursor exotics of each weapon type; that should be common knowledge, and if it isn’t someone should add it to the wiki.

Ensign.2189:

oilstorm.1748:

Perhaps this is common knowledge that I am unaware of, but where did you get these probabilities and much more importantly, how do you know they are right?

Probabilities are inferred from a large sample (~8000 exotic forges) of forge output, combining my own data with a couple other sample sets I have high confidence in (such as the 10,000 gold forging streamed and publicized on reddit recently).

I know they are ‘right’ in as much as a data set of that size tells you a lot about the credible range of parameter values; for example I can tell you that based upon that data set I can say with 99.9% confidence that the precursor rate is no better than 1 in 113, and with 99.996% confidence that it’s no better than 1 in 100.

As for there being 12 commons of each type, and that the forge only kicks out 7 named, non-precursor exotics of each weapon type; that should be common knowledge, and if it isn’t someone should add it to the wiki.

Interesting. So if I am understanding you correctly, you are arguing that the mean sample proportion is binomially distributed and therefore the mean population proportion is also a binomial distribution? And that since your mean sample proportion times the sample size is less than (1-mean sample proportion) times your sample size, you invoked the Central Limit Thrm, right? I completely agree with your analysis at this point and your work is (mostly) spot on. Nicely done.

However, how do you know the mean population proportion and the population variance that allows you to do a mean proportion comparison test? Similiarly, at what power where your tests carried out? I am asking because your post lacked the rigor that I am looking for (I am naturally a curious person). I know it is not common for people to understand most of the terminology, jargon and whatnot of this subject, but please feel free to be more technical.

Lastly, how do you know that the common exotics drop with equal probability?

FACT: Unless you are a Dev and have first hand knowledge with the game code you know absolutely ZERO about the RNG and/or probabilities of getting anything.
The drop rate of anything usable by a level 80 toon is a joke.

One Prarie Outpost.4860:

FACT: Unless you are a Dev and have first hand knowledge with the game code you know absolutely ZERO about the RNG and/or probabilities of getting anything.

Anyone that has a background in statistics can determine the probabilities for events to occur from a set of data. We can definitely say we know more than absolutely zero about the probabilities of getting anything. You don’t need to be a Dev to figure this out.

oilstorm.1748:

you are arguing that the mean sample proportion is binomially distributed and therefore the mean population proportion is also a binomial distribution?

In a classical sense, for a given population mean, the mean sample proportion will be a binomially distributed random variable. The population mean is not a random variable; it is a (unknown) constant.

oilstorm.1748:

And that since your mean sample proportion times the sample size is less than (1-mean sample proportion) times your sample size, you invoked the Central Limit Thrm, right?

Not at all; in fact I find that Gaussian approximations do a poor job when you’re looking at events where Pr(x) is close to 0 or 1 as the true distribution has significant skew.

I took a prior of a uniform distribution on [0,1] for Pr(x) and each data point is a Bernoulli trial; this is easy to compute explicitly, as the posterior distribution is simply a beta distribution. Integration over the beta distribution generates credible intervals for my belief about the true value of Pr(x).

oilstorm.1748:

However, how do you know the mean population proportion and the population variance that allows you to do a mean proportion comparison test? Similiarly, at what power where your tests carried out?

I’m working in an inference framework and estimating model parameters not performing hypothesis tests.

Sample comparisons pop right out of the framework as it’s more or less built to calculate Pr(data|prior), which functions very similarly to a classical T-test.

oilstorm.1748:

Lastly, how do you know that the common exotics drop with equal probability?

A Monte Carlo equivalent of a Pearson’s Chi-Squared test; I also checked it for autocorrelations at one point as a sanity check on the RNG algorithm. There’s nothing in there to imply there’s anything other than random selection between 12 items with equal probability.

I heard shooting the tail of the Shatterer guarantees a precursor drop.

So I did, with my ranger. Longbow’s got to come good somewhere, right?

Nah, wrong.

Giggled a bit at myself, though, but wish I’d checked to see if anyone else was doing it as well.

“Stupid bearbow ranger, huhhuhhuh…”

Belzebu.3912:

khani.4786:

snip

I’m in a similar boat.
1800 hours since day 1 of the game, almost 12k AP, 6 lvl 80’s, every LS done, all JPs done, rank 80 in WvW, I kill world bosses daily (not all bosses though), and a total of 0 precursors, but at least I got a rabbid ascended weapon box from WvW.

so to complain as well. I have 4k hours spent since day one, almost 15k ap , 7 Level 80’s, all ls achievment and stuff done, about rank 240 in wvw, I finished fractals scale 80 including maw, I lighted up liadri… dude who cares legendary are crap cause they don’t hsow off accomplishment but stupid grind .. just ignore em like I do I don’t have one even with double the time played I don’t have a legendary or ever dropped e prcursor

Yeah, I got some friends too with 4k hours without any precursor drop. I’m personally at 3k and sitting on 3 precursor drops (outside the forge; include the forge and that makes 6). (pseudo)-RNG is (pseudo)-random.

I got a precursor from the Warrior gear box you buy for 5 laurels, it was the rifle precursor, i had no idea one could get a precursor from that.

But you can get precursors from http://wiki.guildwars2.com/wiki/Warrior_Gear those.

Ensign.2189:

In a classical sense, for a given population mean, the mean sample proportion will be a binomially distributed random variable. The population mean is not a random variable; it is a (unknown) constant.

Weird. You used classical inference statistics to define your population mean, then used Bayesian statistics to make your inferences. I prefer the Bayesian approach, but to each their own.

Ensign.2189:

Not at all; in fact I find that Gaussian approximations do a poor job when you’re looking at events where Pr(x) is close to 0 or 1 as the true distribution has significant skew.

Uhhh… wut? This would be significant news to the statistical/mathematical community if such a find where true. Gaussian approximations are used frequently for binomial random variables. Recall that the binomial distribution converges in distribution to the normal distribution as our proportion goes to zero and our sample size goes to infinity. A quick google such provided a much more rigorous help of this notion; http://math.stackexchange.com/questions/435574/convergence-of-binomial-to-normal

Ensign.2189:

I took a prior of a uniform distribution on [0,1] for Pr(x) and each data point is a Bernoulli trial; this is easy to compute explicitly, as the posterior distribution is simply a beta distribution. Integration over the beta distribution generates credible intervals for my belief about the true value of Pr(x).

Integration over the beta dist. would only give you the cdf of the beta. So… of course the true value of our population proportion will be there. My guess is that you were attempting to explain how you found the posterior distribution through integration, but this is a hairy process that takes a bit. Then you still have to check for bias and sufficiency. I usually just use the proportionality of the prior and the proiri. That is if I have to find a Bayes estimator; I would much rather find the MME or MLE. But regardless, why did you not just find the expected value of your found beta?

Ensign.2189:

I’m working in an inference framework and estimating model parameters not performing hypothesis tests.

If its been a while since youve studied studied statistical convergence or are not in the math feild, I could totally understand why you think that point estimates do not need power tests. Most people just ask a computer to find some 10,000 sample means and call it good, but we are not all that lucky. Power tests are essential in finding the probability of type II errors and how large our sample sizes need to be to avoid those errors.

Elbegast.6970:

Zommoros is actually an ArenaNet employee named ‘Phil’ who actively sits at a pc with a ‘drop precursor’ button. If he likes the way you look, he may hit the button.

Totally calling him Phil from now on!

BTW, RNG is a killer. I’ve spent roughly 1800 gold on exotics staves to throw into Phil and got a grand total of zero precursors from him. I guess he doesn’t like my look

oilstorm.1748:

Weird.

I worked entirely in a Bayesian framework, but you asked a classical question so I gave you a classical answer.

oilstorm.1748:

Uhhh… wut?…Recall that the binomial distribution converges in distribution to the normal distribution as our proportion goes to zero and our sample size goes to infinity.

It converges as your sample size goes to infinity; our sample is sadly not infinite, so there will be some error introduced by using a Gaussian approximation. The rate at which the binomial distribution converges depends upon the true probability; if it is close to 0.5, it will converge very quickly, if it is very close to 1 or 0, it converges very slowly.

Since the probability of getting a precursor is very low (<1%), the distribution converges very slowly and is still significantly non-normal at a sample size of ~6000.

With the collection of more data, eventually the sample size will be large enough that a Gaussian would be a good approximation for the actual distribution. We just aren’t there yet.

oilstorm.1748:

Power tests are essential in finding the probability of type II errors and how large our sample sizes need to be to avoid those errors.

…Which would be really important if I was performing hypothesis tests?