Observation of the RNG

[Jaytee Bushwacker]

Seeing the posts about how skilling up post 180 in most skills is horrible for some (1000+ combines between some points, and 1 between others) and relativly smooth for others it got me thinking.

To skill up your basically putting stuff in a container and clicking COMBINE. Which is a fancy word for Run the Random Number Generator and see if I make a high enough number to Skill Up. Now since the RNG is legendary at being streaky, and that by and large most people are going to roll low, doing combines that take the largest time between COMBINE presses will skill up slower than someone who clicks COMBINE faster.

Lets take a dice example for a moment, on any two consecutive rolls you are statistically more likely to get a different number on the second roll than you are to duplicate the first roll. So as you can see assuming you fail on combine one, you may still fail on combine two however you are statistically more likely to roll a number that is not calculated as a failure because you already rolled a failure number on roll one.

So the fewer items to combine yields less time between rolls and therefore a slightly higher chance to skill up faster.

Your thoughts?

[Jhales]

My thoughts . . .

As far as dice go, whatever you get on roll #1 has absolutely nothing to do with what you get on roll #2.
The RNG may be different if it does go through streaks. I personally don't think it does, but can't say for certain that it does or does not.

"and that by and large most people are going to roll low"

- I'm not exactly sure what ya mean by that. do "/random 1000" ten thousand times and take the average, and I'm pretty sure you'll end up with a number that's very close to 500.

"doing combines that take the largest time between COMBINE presses will skill up slower than someone who clicks COMBINE faster."

- Well, yes. Since each non-trivial combine gives a chance at a skill-up, the more combines per hour you do, the faster you will skill up. But that has nothing to do with the RNG. If you roll a hundred dice in an hour, and I only roll ten, then you're almost certain to get more 6s than I do.

"Lets take a dice example for a moment, on any two consecutive rolls you are statistically more likely to get a different number on the second roll than you are to duplicate the first roll."

- Yes. There is a 1-in-6 chance you will get the same number again, and a 5-in-6 chance thaty you will get a different number on the second roll. (on a 6-sided die)

"So as you can see assuming you fail on combine one, you may still fail on combine two however you are statistically more likely to roll a number that is not calculated as a failure because you already rolled a failure number on roll one."

- Again, what you get on the first roll has nothing to do with the second roll. Suppose I need to roll a 6 to succeed. No matter what I roll on the first roll, I will still have a 1-in-6 chance of getting a 6 on the second roll. If I roll a 1 on the first roll, then on the second roll I have a 1-in-6 chance of getting a 6. or if I roll a 6 on the first roll, I still have a 1-in-6 chance of getting a 6 on the second roll. What I get on the second roll has nothing to do with the first roll.

"Which is a fancy word for Run the Random Number Generator and see if I make a high enough number to Skill Up"

- I've seen other people talk about having to get a "high" enough number on the RNG to succeed. Some even go so far as to /random 100 until they get a streak of high numbers, and then do their combine. However, I don't believe any of us know the formula they use to calculate success or failure, so actually, getting a low number on the RNG could give you success while a high number gives you failure. If the number that the RNG comes up with is used in the bottom half of a fraction, then having a higher RNG number will actually give you a lower final result. (Of course, for all I know, getting a skill up might require you to get a lower final result.) Well, my point is, since we don't know the formula used for calculating whether or not we get a skill up, we can't say for certain that it is helpful or not to have a high number on the RNG.

[Jaytee Bushwacker]

Let me clarify, lets say you roll a 5 on your first roll of a die. Statistically you have a 1 in 6 chance to roll a 5 any one time. However, your odds of rolling two 5's in a row are signifigantly less than rolling a 5 and then any other number.

So if we assume the RNG has a tendancy to be "streaky", doing more combines within a short amount of time after a success would tend to give a higher chance of a skill up again. The fact we have all seen many times where we get stuck on a number for a long time then get two skill ups withing one or two combines lends creedence to the RNG streakyness.

As to having to get a "high" number, thats relative. They could use 1-10 out of 32,000 as the number needed for success. Who knows? But it really doesnt matter, the idea is that a specific subset of the RNG is used in a formula to determine if a skill up is attained.

Again assuming the RNG tends to be streaky, it may be more likely you will get a series of skill ups within a short period after that.

[kiztent]

That's not the craziest idea I've ever seen. Depending on how the RNG works, it might even be true.

[Silound]

Well, in a sense, it can be. No matter what you are combining, the more randoms you do in any set amount of time, the more chances for the desired range you will have. From a programmers point of view, what Jaytee has suggested would be reasonable, and quite probable, since you are using a nested function call of the same function over and over. However, no one has yet to determine the seed of the randomization algorithm, much less solve the dead horse mystery of serverwide RNG.

For those who dont know, the SWRNG (Server Wide Random Number Generator) theory is that all players doing anything related to the RNG roll the "same" theoretical generator, therefore the more people doing it, the less of a chance *you* will skillup instead of someone else. Contradicting theory is that the RNG is based on a totally per character basis.

Problem is, Jaytee's theory would backfire *if* the SWRNG theroy was true, it's all one mystry to us.

Silound
Minstel of Prexus

[Itek]

Argh...

"you need to roll high" ... "or you may need to roll low" ...

No.

Trust me, no.

Again, no.

Why? Because you might need to roll "high or low" (1-10 or 31991-32000) or "middle" (15995-16005)

Or, and this is something people tend to forget, some hideously complex formula that would be too obscure to print. (if indeed there is truth to the "serverwide RNG" or the "the number of people in the zone influcences combines" ... heck the success formula COULD be tied to the number of bats alive in GFay at the time...) Computers are VERY FAST at adding numbers. What would take you a minute to work out long hand would take the computer a millionth of a second.

Here's an old "MUD" formula. It took your "donation" to the "wishing well" and checked to see if you got a "wish" and added a random stat point.

if "donation"
then ("donation value" + "donation previous" == "donation previous");
if "donation pervious" >= (5000 + /ran 10000 + /ran 10000)
then "wish" AND "donation previous" == 0

so if you donated it added your donation value (in copper) to 0 or the total of your previous donation, then compared that value against a weighted bell curve between 5002 and 25000

Someone figured out that if you seeded the donation with a BIG 6000 copper initial donation (or one immediately after a "wish") then kept flinging in 1 copper at a time you were pretty likely to get a "wish" within 100 to 200 copper.

Times between trials relates to RNG only as it relates to how quickly you can submit for a RNG answer in comparison to the rate at which the RNG changes. If you can submit /ran's faster than the RNG can change you will get the same two in a row. (This is probably humanly impossible, but it IS why they slowed down how quickly you could request /rand in the game.)

The RNG is streaky, from a human observational point of view. It's only pseudo-random, but let's not forget that REAL RNGs (like dice) are streaky as well.

Black came up 3 times in a row on roulette. Should you bet the streak will continue or that it will break? If anyone had the answer Las Vegas would be a ghost town an hour later. You got above 90 on the last 3 /rand's ... should you try your tradeskill trophy attempt right now? Or wait?

Here's a little food for thought....

What if the skills have a "last use - decay" function built into them? I have a jewelry skill of 117 and come back a few months later and try to skill up again... 5 stacks and no skill, and huge failure numbers, nearly twice the expectation. Then a point of skill... next stack sees 3 skill points and no failures. This could be due to "streaking" of the RNG, or I could simply have needed to "repair" my decayed skill.

There are those that say "oh, that would be too difficult for them to remember the last time you had tried every skill on every character on every server."

Abashi (remember him?) once said about why we couldn't have "High Exp but Only Do Once" quests that it was impossible "because we don't have the ability to store information like that on characters."

Faction.

How many factions are there in the game? Your standing modifier for "actions" is added and stored, permenently, on your character.

(hint: computers are REALLY REALLY GOOD at storing large amounts of data and recalling them at high speed, almost as if the were DESIGNED to do so)

It's random, semi-random, or pseudo-random. Take your pick. Cross your fingers, cast spells till you get a fizzle, leave a tradeskill book open, ... do what you have to do. But it's, in the long run, not going to help much more than the guy with a rabbit's foot hanging off his cowboy hat trying to beat the slot machine.

Most of this is my opinion, some of it is computer science theory, probability theory, economics of gambling theory, and the code example is a poorly remembered snipet smashed in with a few million lines of pseudo-C++ code remembered from about a decade ago. Your milage probably will vary. *grin*

I heartily wish someone would come up with a proveable way to outsmart the "hell levels" of tradeskills. For Innoruuk's sake the tailoring and smithing sub-communities would fall over themselves to reward the person who finds the way. If I am wrong.... no one will be more pleased.

[Jaytee Bushwacker]

Is it a way to beat it? Certainly not

It was just my observation that with the "streaky" ness of the RNG, and using lower item combines to attempt to fit more combines into a "streak" you may have more success. Of course you might just end up combining more attempts into a "bad" streak :roll:

[Zdapus]

I have currently have 7 different trade skills over 244 which includes tinkering. In each of the seven skills I have achieved 241 thru 243 within 6 to 20 combines. In my last skilling up attempt in pottery I even had all three of these skill ups visible in my chat window at the same time.

I have also seen this drastic skill up rate increase in some of the other skill up data posted on this board.

It makes me wonder if there is a total combine counter that over rides range to compensate for those nasty runs without skill ups. Also if 240 is the magic number to reach the 252 skill camp with a 5% with a geerlok, what purpose would this serve?

[Taraddar]

Humans are actually horrible when it comes to random. We innately don't do well with random. We look for patterns in everything we see and when we can come up with patterns we assume something can't be random. If you ask someone to pick random numbers from their head they are very much NOT random since they will almost never have the same number occure twice in a row.

Take flipping a coin. Probablity says that we have 50-50 shot of it being heads or tails and that if we flipped a coin 1000 times it should have about 500 heads and 500 tails. What that doesn't mean is that we will alternate heads/tails back and forth. That is compleatly unrandom.

Take your dice example. Yes the statical likelyhood of getting 2 6's in a row is 1 in 36 so alot less likely than getting a single six but the odds of rolling a second 6 after rolling the first 6 is still 1 in 6. The exact same thing as it would be if you hadn't rolled at all or if you had rolled a 5 before. The results of the first roll, if it even exsists tells us abosuletly nothing new about the odds of another roll. Yes it's a bit wierd but the odds of 2 6's in a row is small but the odds of getting a 6 is always exactly the same every time you roll the dice. Doesn't matter if it's the first time you've ever rolled the dice or if it's the 500th 6 you've just rolled in a row it's still 1 in 6 that you'll get another 6.

As far as the idea of rand being streaky we have no Idea and no real way to exploit that even if it was. While it is possible that a particular seed for rand may produce more high numbers than low numbers depending on the function they use we have no way of knowing. We don't have access to the code for the rand and there are too many questions left unanswered to even begin testing it. Is rand handled on an individual level, a zone level, a machine(talking about a single physical machine out of the many that make up a "server" like ct that might host multiple zones) or even accross the entire "server". Is random just ran for /rand and skill ups or is it even used for things like how much you hit for and any number of other things. Even if we somehow managed to deturmine all those we are still left trying to setup a test with controled inputs when any other person on the server can potentially add uncontroled input on anything but a personal level for rand. Not to mention that the seed for rand almost assuredly changes much to quickly for us to be able to derturmine a favorable seed let alone exploit it. Alot of rand systems work of something constantly changing even as simple as the system time in some form or another.

Best thing you can do to get skill ups is get any applicable stats as high as possible and just hit combine as many times as possible. Beyond that there are just too many questions left open including some not even touched on already like does recipe have any effect on skill up rate, does success vs fail have an effect on skill up rate. Those are just 2 commonly discussed ones.

[Thorvari]

Could the Streakyness of the RNG be explained if it takes a new seed per tick? This could explain why some people see 2-3 skillups in a row, especially if they are combining quickly.

One thing that I have seen people post about doing is to try and take advantage of a streak. For example, when I recently successfully combined my 6th shawl(Triv 200ish) with a skill of 158. I set up the combine, put my mouse over the combine button, and typed "/random 0 1000" in the chat bar and pressed enter. When a tick happened, I would press up arrow, enter then repeat and see what rolls I got. The first tick I got ~200, ~300. The second tick I got ~850, ~900, and promptly clicked combine. I succeeded, probably from luck. Its only my experience, but it does explain some of the streakyness. (I did find a corner where I wouldn't spam anyone)

To tell when a tick happens, use up some mana, and wait for it to regen. When the numbers change(use a UI with numbers), that is an indicator of a tick.

[DeegerStormcrow]

I'm not sure I understand you. If you are saying:

Combine as fast as you can becuase the faster you combine the better chance you have of hitting a good streak.

This is wrong, because while it's true that you might take advantage of a good streak, you'll also be hit by the bad streaks. To refrase, no matter how good/bad the RNG is, your pace of combines matters not at all if you keep a constant rate.

However, if you are saying:

Do combines slowly until you get a success, then do as many as you can as quickly as you can.

This is possibly true. This would happen only if the RNG was "server wide" (otherwise your pace again means nothing). And if it's server-wide, then I think it's probably false because a server creates TONS of RNs between your combines and any streakyness would be lost.

Personally, I think the supposed streakyness of the RNG is just our perception. Has anybody done an analysis using /random?

[Jhales]

I compiled the following data from GM (250) tailors who had posted lists of how many combine attempts they made to get each skill-up point. (Thanks to all those who posted that information). I only took the data from charts that went up to 250 skill. I disregarded posts that only covered a a small sample of skill-ups and were called "my terrible run today" or "the RNG loves me today". None of the data comes from any skill-ups under 150.

Just to make it clear, this is an example of one of the posts I used:
http://mboards.eqtraders.com/phpBB2/...pic.php?t=8025

The data I looked at over several posts included 8757 attempted combines, and 453 skill-ups in those combines. That's an average of about 19.33 attempts per skill-up. (5.17% percent of all combines resulted in a skill-up)

Now, if the RNG really is more streaky than random chance would allow, then whenever a player gets a skill-up, he would have an improved chance of getting a skill up on the next combine. Some of you have stated this seems to be the case.

Of those 453 skill-ups I looked at, 20 of them came on the next combine attempt right after a skill-up. (They were skill-ups on back-to-back combines.) That's an average of 22.65 attempts per skill-up. (4.42% of combines immediately following a skill-up, themselves resulted in skill-ups).

I know this isn't enough data to make an extremely firm conclusion, but it does tend to indicate that a combine made right after a skill-up has about the same chance to itself give a skill-up as any other combine. (In fact, it indicates that you have a slightly less chance of skilling-up on the combine immediately following a skill-up, but the difference is very slight and is likely just because of the small amount of data.)

One flaw of this analysis is that it doesn't take into account that a significant amount of time may have passed between combines. I don't believe this is a major flaw though, as most people tend to gather a lot of materials and then do several combine attempts at once. (Now that I think of it, I should have looked at the brewing posts, as many of them would have gone from 150-248 in one sitting, but too late now heh.)

Also of note, and this is using an even smaller set of data. Of the 20 back-to-back skills ups, there was 1 instance of back-to-back-to-back skill-ups (3 of them right in a row). So, of all 20 back-to-back skill-ups, 5% of them were back-to-back-to-back skill-ups, which fits right in with what the average would predict.

My conclusion from this data is that skill-ups do not come in streaks anymore than they would be expected to. I may go back and compile more of the posted lists when I have more time, to get a more accurate and correct model. Or I may not =)

Comments?

Jhales, of E'ci

[Itek]

Again...

Itek (Me) said:

Times between trials relates to RNG only as it relates to how quickly you can submit for a RNG answer in comparison to the rate at which the RNG changes. If you can submit /ran's faster than the RNG can change you will get the same two in a row. (This is probably humanly impossible, but it IS why they slowed down how quickly you could request /rand in the game.)


Can we all please RE-READ that until it's clear?

jaytee replied:

==Is it a way to beat it? Certainly not

==It was just my observation that with the "streaky" ness of the RNG, and using lower item combines to attempt to fit more combines into a "streak" you may have more success. Of course you might just end up combining more attempts into a "bad" streak

Um, what you are saying is... wait till you get the RNG to spit out a "success range number" and then submit as many trials as possible before the number changes.... Go back and read what I wrote ...

zdapus replied:

==I have currently have 7 different trade skills over 244 which includes tinkering. In each of the seven skills I have achieved 241 thru 243 within 6 to 20 combines. In my last skilling up attempt in pottery I even had all three of these skill ups visible in my chat window at the same time.

So, again, multiple trials before the "seed" changes (or at least changes a great deal) could be influencing the outcome... Someone said that if you can submit /ran requests before the seed changes you get the same result...

thorvari replied:

==Could the Streakyness of the RNG be explained if it takes a new seed per tick? This could explain why some people see 2-3 skillups in a row, especially if they are combining quickly.

==The first tick I got ~200, ~300. The second tick I got ~850, ~900, and promptly clicked combine. I succeeded, probably from luck. Its only my experience, but it does explain some of the streakyness. (I did find a corner where I wouldn't spam anyone)

This would tend to show that the seed of the RNG changes on tick rather than on the sub-millisecond time the clock speed of the processor. Somehow I doubt it. Why? Cause you don't always miss or hit or do the same damage on dual wield / double attack. (Some one else pointed out, as I did, that humans observe chaotic events BADLY.) But if true would support the "do quick combines when the time is right" idea. But we still need a way to prove the idea, and a way to detect AHEAD of time when the time WILL BE right.

deeger replied:

==Do combines slowly until you get a success, then do as many as you can as quickly as you can.

==This is possibly true. This would happen only if the RNG was "server wide"

No, this would happen if the RNG seed changes only on tick.

....

Here's the deal folks... You can't predict the RNG. Well, you can but it take some hefty calculations and some VERY precise knowledge of the formula, the base seed number, and a detailed history of the most recent run of "determined /ran numbers" (guy went to jail because he tried to use specific knowledge of how to predict computer developed KENO numbers to cheat.)

Test it. Find someone who two-boxes. Have them sit in the same zone with two newbie tailors. Have them put two spiderling silk in a kit each. Do the /random 1000 hotkey on both till they both get >900 twice in a tick and have them both hit combine. Repeat till they trivial threads.

I will wager that they both don't often get >900 twice in a "tick" and I would wager they both don't skill up at the same time, or even the same rate. (which, if EITHER the "streak" or "server wide RNG" were true, they would.)

Wish that it were true. But I can point out a few billion dollar a year companies that survive simply because... it's not.

Again, do whatever makes you THINK works. I wish you all the luck in the world. Some people ARE lucky. (dozens of people win the lottery every year)

More than half of the game is remembering... it's a game. Tradeskills should be fun. Even when you are pulling your hair out desperate to get that last point of skill to trivial a hated recipe. (if I _never_ make another Mino hero brew I will die a happy man)

[Jaytee Bushwacker]

Its fun because Aid Gimil has an earring waiting for me =P Until then I avoided trades like the plague. And really until SoL there really wasnt much worth making imo.

[tlag]

On a programmers view and some math background that is somewhat more substantiel than that horrible conclusions of the original author:

RNG has to have streaks. Like someone said even perfect random events will have streaks. This streaks are however about as likely as changes and there is no way to predict them.

RNG has to be server based to avoid cheating. Server in this aspect refers to a physical computer, which in the case of EQ runs 1 to 3 zones. Although the network traffic on a server would make a wonderful source of enthropy for a random number generator, there is actual no need for this. EQ has tons of events that request for a random number. Even poor pseudo RNG that have some short term correlation, would serve nearly perfect random stuff to a individual player since between 2 requests from a single player some hundred calls from other player and NPC are proceeded.