RNG is not random.

[Itek]

Lickity-

Yep. Which is why I was hopeful I wouldn't have to break out the "beat a dead horse" stick... again... but it seems the poster doesn't grok the math as well as s/he should.

Kieroth_whiteleaf-

the one thing I *would* like to see on the issue is that a lot of people have seemed to notice short-term streakiness - I've noticed it a bit and I see other people commenting on this. I'd be interested to know if any devs have tried running a good sized test sample (like a billion numbers or so) in batches of 1000 or so, and see if the distributions average out in the short term for the most part.

Um...

Humans are shockingly bad at noticing patterns. They commonly do NOT see patterns where there ARE patterns and almost ALWAYS see them where there are NONE. Over and over and over again.

One never... let me be clear... absoluetly NEVER EVER EVER... never checks for randomness in SMALL batches. It's the OPPOSITE of how one checks for randomness.

I roll 3 dice and get an 18. Yippee. I roll 3 dice and get another 18 ... WOW this character will rock... I roll 3 dice and get ... an 18? Gosh my dice must be broken...

No. My dice are NOT broken. I got a 1 in 216 result 3 times in a row. Very unlikely. 10 million to one or so.

Have 10 million D+D characters been rolled? Yeah that's a certainty.

On the other hand I once entered a game where one of the other players -assured- us that he had actually rolled 6 18's for his character. Um, sure whatever you say. (18 Wisdom on a D+D, not even AD+D character isn't all that useful for a Fighter. Anything after 3 18's was pretty much just window dressing.) But let's face it. If the EQ characters depended on a /random stat roll .... people would just roll new characters till they got the max points anyway. Which is what the developers found out VERY early on for MUDs so almost everyone has moved to "assign points" methods to prevent hideous lag due to character re-creation. Gosh... wait... that would mean.... the EQ devs actually understand.... *gasp* ... game design THEORY??!? *shock and horror*

Has "black" come up on the roulette wheel the last 7 spins? That's about 200 to one. Everyone should bet on "red" next right? No because the wheel isn't broken and the next spin black is still nearly a 50/50 chance. (Again, we need to not delve TOO far into specifics of RealLife gaming/gambling as the mods close the threads that do nearly INSTANTLY. This single example is to give a concrete case to those for whom probability isn't a past-time like myself.)

Streaks are a HUMAN phenomenon. You can't tell when one is about to start, you can't tell when one is about to end, and most of the time when you think you are ON a streak you are actually WRONG. (read the quote above)

1 5 3 6 7 9 8 3 4 5 6 1 6 7 3 2 9 0 0 0

That looks like I hit a streak at the end doesn't it? Nope it's pretty random (for a human rather than computer generated list anyway) and the trailing 0's just bring it closer to "average" as it was pretty out of whack (17 in a row with no 0's) moments before.

Here's a rule of thumb.

"If you think you notice streaks in random events....
.... stay the bloody Plane of Hate out of casinos."

Or...

How to make a fortune gambling...

1) Start with a small fortune and a gaming license

2) Depend on the fact that 90% of people are BAD at math

3) Depend on the laws of probability

4) Do not offer beatable games

5) Hire good looking wait-staff

6) Understand that booze is cheap

7) Place the cashier in the MIDDLE of the floor not next to the door

"never give a sucker an even break" - P.T. Barnum

[Anubsin]

This is more in response to the first post i had a couple huge no skillup streaks in JC on my way to 250 but i had just as many improbable skillup streaks. and example 195-196 was well over 300 combines but 197-198 was only one combine go figure (this was done on 2 different nights so no real "streaks".

[Malathos]

... actually, it isn't truly random. It can't be. The RNG is churning out numbers according to a formula - which may or may not be modified by various other factors - but a formula just the same. IF we could replicate the EXACT environment that the RNG operates in and duplicate ALL influencing factors AND kick it off at the same point as the real RNG, then they should produce identical streams of numbers.

BUT. And it's a fairly big BUT, that doesn't stop everything that Itek says being true. There is virtually nothing in the world that is truly random (except for maybe some human decisions and maybe some things down at quantum level) - there are a heck of a lot of things that are non-deterministic however. Which means - as good as random until we invent some REALLY BIG computers so we can predict what will happen next, and please can we have some more money in our budget ... and one of those really cool expresso machines

The RNG is one of these things.
IF we had enough computing power to analyse a large (and by large I mean, say 10 to the power 20 or so results) number of results, then there may be the slightest hint of a pattern emerging. However, the number of factors influencing the RNG and the number of times it is called in between successive skill up attempts mean that the results you will get are as good as being truly random - or pseudo random as Itek correctly calls it.

As other people have said, making 400 EVBs without a skillup is unlucky, but still a random (read pseudo random) result. If it took everyone who was trying to raise skills using EVBs 400 attempts per skillup, then we might be able to comment on the skillup formulae around that area. I've gone through a hundred EVBs for one skillup and then got 3 skillups from the next hundred - it's random.

As a slight derail on the subject of casinos. Read a great book Called the Newtonian Casino. It details the efforts of a group of students with computers hidden in shoes and the like, that statistically analysed some of the casino wheels in Vegas. They found that some wheels were out of balance - and hence, less pseudo random that most others. They used computing power to try and predict - or at least shift the odds more in their favour for their gambling on the imbalanced wheels. The casinos in Vegas now regularly check all wheels for imbalance

[Arghargh]

"A single player can only get 480 skill points post 190. The norm for 480 skill points would be one run of 120 failures in a row. It's not realistic to assume that there have been millions of post 190 skillups in EQ's history."

2,000,000 skill ups / 480 skill-ups per player = 4,167 players. Given that EQ has had hundreds of thousands of players, I think it extremely likely that there have in fact been millions of post 190 skill ups.

Arghargh Grumble, Darkblood
65 Ogre Shaman of Rallos Zek

205 Alchemy
250 Baking
250 Blacksmithing
250 Brewing
200 Fishing
200 Fletching
250 Jewelry
250 Pottery
187 Tailoring

[Andara Lan`Gstrom]

:dedhorse:

That's all that needs to be said.

[Abazagaroth]

Itek:

That developer stated that they HAD to run the system for billions of cycles to get a perfectly random DISTRIBUTION.

That is not the same thing as a perfectly "Random Number Generator".

The RNG used by everquest is streaky. The best proof of streakiness is when something improbable based on the odds happens an unusual number of times over the short term. Find a very trivial combine, and track when the failures occur. The odds of getting a failure based on what actually occurs in practice is much greater in the few combines directly AFTER another failure. This is also true of fizzles (actually, this is even worse with fizzles, leading me to believe there is something else working there, because you rarely fizzle, but when you do, you have a *very* high probability of fizzling twice in a row).

The RNG isn;t broken, its just not a RNG, because such a thing doesn't actually exist in a computer, because all such systems rely on algorithms programmed by people to act in a certain way.

Personally, I think the RNG acts the way it does because maybe there are multiple algorithms running in order to conserve CPU cycles (very possible) so that when doing many combines in a row in a short period of time you are having the same seed number dumped into the same RNG algorithm and thus getting a greater chance of getting similar results, rather than getting a new seed value for every single calculation.

The RNG isn't perfectly random. It is very streaky. 400 combines without a skillup isn't the result of the RNG programming related streakiness though. That's just bad luck, which does in fact happen.

[Cantatus]

*sigh*

Guys, if you think something is a "dead horse," posted too frequently, or just an issue that you get annoyed with, no one is making you post in it. There are several things in this thread I could easily deem as being unacceptable or, at least, on the brink of unacceptable and very rude. If you find a thread that annoys you and think you are going to post something with an attitude, close the thread and move on.

The sad thing is these comments are coming from regular posted that should know better.

Consider this a warning.

[Akishka]

My statistics professor once had the entire class generate a string of 10 'random' heads vs tails outcomes. He then summerized the data. He pointed out that not a single person wrote all heads or all tails. This is because most people believe that if its actually random, its going to be pretty much even on the heads and tails and its not going to show a pattern. Which is true BUT like many people said, you can't use small sample sizes. When he had his computer sample 1000 sets of 10, he got 2 that were all of one kind. And there is always the chance that you can run a sample size of billions of numbers and still get all heads (or 0's or whatever). Because randomness really doesn't care what the proceeding number is. Yes, you start getting into some pretty insane probablilities but that doesn't mean that it is impossible.

[Taraddar]

That developer stated that they HAD to run the system for billions of cycles to get a perfectly random DISTRIBUTION.

Thats the thing even if you are dealing with a truely random event you still have to do a very large sample set to get a good test and good distrubution. The larger the set the better. The fact that you can notice apparent patterns on the small scale is not an indication that it isn't random. In fact if you think about it so many randoms are used for so many different things for so many different people odds are that your 2 combine rolls for skill ups are 100's of rands appart with other peoples inbetween.

Simple example is flip a coin 10 times and record heads or tails. Repeate that process 10 times. If you got 5/5 every time it wouldn't be an indication that the random is working right. Infact it would be more of an indication that it wasn't working right. You expect to get some streaks and end up with 4/6 6/4 or even 10/0. Take all 100 combined together and it should be pretty close to 50/50.

[Mystyqe]

Hmm...

Seems to me that people are confusing randomness with probability. As Abazagaroth said, there is no such thing as a RNG because even at the highest levels (NSA) said RNG is based on a human generated formula.

The algorithm may be difficult, it may use a large subset of data (The NYC phonebook to use one famous example) but in the end - it is not random.

However, as was correctly pointed out by others using the simple coin flip scenario, no matter how many consecutive tails or heads have been flipped in a row, the chance of either on the next flip is still 1 in 2.

What * I * have noticed is that successive or multiple skill ups seem to occur at regular intervals. For example, I'll get many successful combines on a near trivial product without a skill up then get two in a row. A set amount of time later the process repeats itself. And again.

This seems to indicate that skill ups are based (at least in some part) on on something other than a a RNG.

[Zaephod]

I think it was said that the RNG has a perfectly normal distribution over very large sets but is extremely streaky in small sets.

[Jarak_akn]

I think it was said that the RNG has a perfectly normal distribution over very large sets but is extremely streaky in small sets.

Humm that would describe the following sequence of 1s and 0s :

1111111111000000000011111111110000000000

Normal distribution (50:50) but rather streaky :-P

Reasuring ?

Jarak.

[Itzena]

Random numbers tend to be streaky anyway, by definition. An evenly distributed & smooth selection (of whatever) = Predictable (and therefore <> Random).

(Of course, an evenly distributed and smooth selection is one of the possible results of a RNG...but only one of many, many others that are in varying degrees of streakyness)

[Ngreth Thergn]

What * I * have noticed is that successive or multiple skill ups seem to occur at regular intervals. For example, I'll get many successful combines on a near trivial product without a skill up then get two in a row. A set amount of time later the process repeats itself. And again.

This seems to indicate that skill ups are based (at least in some part) on on something other than a a RNG.

And look. The Developers are NOT lying to us. If they do not want something known they just will not tell us, they are not lying.

It is based on computer based random numbers. A computer based random number is a formula off a table. Maybe the table that EQ uses is a bit streaky, but that does NOT mean that there is some code that specifically goes "make these streaks" it is more likely that the table that the RNG uses is a problem (and unfortunately not very changeable)

The other problem though is, for every person that shows the pattern you are saying, there is another person that shows a DIFFERENT pattern, and another person that sows pure randomness, and another person that shows a third pattern.

Really, truly... skill ups and successes are as random as possible from a computer. Your chance for a combine at a certain skill is the same as the last combine at the same skill. Your chance of a skill up is the same as the last chance you had using the same "trivial" and still at the same skill and the same trade skill.

[Itek]

*Itek nods at Cantalus and Ngreth*

Exactly.

1) They didn't -have- to run it for a billion to "get it to be random" they tested it over an unreasonably large data set to PROVE it was sufficiently random.

2) Consider the phrase "sufficiently random"

3) Again, remember, you couldn't do anything about it if it WAS broken.

4) And, again, humans are SHOCKINGLY bad at detecting patterns.

I'm not posting to beat the dead horse. I'm posting to inform people who seem to be bad at math that, seriously, they make us programmers STUDY the math.

MINIMUM math requirements to get a B.S. in C.S.

Calc One
Calc Two
Calc Three
Linear Algebra (that's DOWN from the prior Diff. Eq.)
Descrete Math Structures
Probability and Statistics for Engineers (what? probability? in a 380 level engineering class? ... perhaps a degreed programmer MIGHT understand the BASIC THEORY of probability? ... you think? maybe?)

Yes, p-RNG uses an algorithim to determine the outcome ....

but since it's got more effective "sides" than you can shake a stick at it's actually BETTER then simply using a dice or coin-flip.

(The math gets hairy, and I say that while my class discusses the integration of parameterized curves in polar coordinates, but it works out that any half decent /random, like the one on my Casio fx-9750g+ graphing calculator, is much more "random" than a dice. Seriously, take the /random output and wash it to a 1 to 6 outcome and compare. It takes a few million data points but the end result is... the chip beats the dice. Every time.)

Random events are counter-intuitive for most people who expect a deterministic (Newtonian/Mechanics) universe. Explaining reality of probalistic (quantum/chaotic) systems is .... hard.

Belief != Reality

"I've noticed it's streaky"

Yep, and that's EXPECTED. Which is why probability only works in the LONG run, and why it's not broken. Probablistic systems are only broken when they violate expectation in the LONG run. Really, I'm not making it up. (What could I possibly gain? SOE isn't paying me to de-bunk the "broken RNG" myth.)