Gaussian RNG issue in PoE (both games)

You are so correct that RNG is currently impossible (although physics is getting closer with white noise generation as a seed).
Gaussian is just so easy to create pseudo-random numbers that nearly every coder uses it, hence why I could do the code for it so easily.
The problem is any RNG will create a bell curve and approx 25% of the bell curve will fall outside of the desired effect. So for example I started the thread with I was using Gaussian RNG code to generate an average of 50 with a variation of + or - 10. This should (if it was truly random) only create numbers between 40-60 but 25% are higher or lower. To get rid of these you just put in a basic (lol literally almost B.A.S.I.C.) loop that if result greater than 60 or less than 40 make a new ‘random’ number to ‘cut the sides’ off the bell curve.

I have literally quit this games predecessor due to constantly getting map sustain on the low outliers side of the bell curve, for 18 months (and that’s a lot of lost revenue considering I have enough lifetime spend on Xbox to have got multiple EA keys).

GGG coders it’s time to add a filter to your Gaussian rng code ;)

I guess the point of this post is to show how smart you are (because you learnt how standard deviation works in normal distribution?), not discuss something about the game. Or it's to complain how less loot you got than you deserved? :)

"

Suchka_777#4336 wrote:

I guess the point of this post is to show how smart you are (because you learnt how standard deviation works in normal distribution?), not discuss something about the game. Or it's to complain how less loot you got than you deserved? :)

Mmmm maybe, but OP has started an interesting start point for a deeper look into "how it REALLY works", for me at least. The distribution simply implies the more you play (or more quantity you use) the more likely your "luck" is going to fall in the upper +ve deviation more often/sooner hence why even a little quantity is so strong.

You cannot really change this deviation without forcibly rerolling when the numbers fall outside, that would be worse that it is now. I personally like a bit of "good luck" - "bad luck". It is all the other RNG factors on top of this that convolute.

This is only one of many factors worth exploring. When you know a system you can find the sweet spot of IIR, IIQ, rare monsters, magic....etc. It is essentially control theory. If we map the integer variables at each point we can build a model but there are too many unknowns. So many unknowns in fact it would be like trying to decrypt 8k encryption with an ipad.

We meet more RNG after RNG. What I mean by that is 1) rolling maps has XX chance to roll XX mod 2) number of Rare monsters 3) number of modifiers on rares etc. etc. These monster have XX base chance at XX level to drop Tier 1-5 blar blar blar.

As we don't know at what rates or what factors or mods or a great deal after "drop rates" we need access to the source code to know. That ain't gonna happen so meh.

Yep, thanks smarty pants :) I don't think I can be arsed trying to figure out all that noise.

"

Soulscar1#4896 wrote:

The distribution simply implies the more you play (or more quantity you use) the more likely your "luck" is going to fall in the upper +ve deviation more often/sooner hence why even a little quantity is so strong.

What you are talking about? Expected value you get never change no matter how long you play. You got something completely wrong, which is common case when someone (OP) is trying to sound smart and say some stupid things in the foggiest way possible (otherwise, noone will believe that he is saying something smart).

"

Timbo Zero#8289 wrote:

@GGG I am, with relative confidence, sure you are using Gaussian RNG generation

Can you share any of the data you gathered to achieve that confidence?

"

Soulscar1#4896 wrote:

"

Suchka_777#4336 wrote:

I guess the point of this post is to show how smart you are (because you learnt how standard deviation works in normal distribution?), not discuss something about the game. Or it's to complain how less loot you got than you deserved? :)

Mmmm maybe, but OP has started an interesting start point for a deeper look into "how it REALLY works", for me at least. The distribution simply implies the more you play (or more quantity you use) the more likely your "luck" is going to fall in the upper +ve deviation more often/sooner hence why even a little quantity is so strong.

You cannot really change this deviation without forcibly rerolling when the numbers fall outside, that would be worse that it is now. I personally like a bit of "good luck" - "bad luck". It is all the other RNG factors on top of this that convolute.

This is only one of many factors worth exploring. When you know a system you can find the sweet spot of IIR, IIQ, rare monsters, magic....etc. It is essentially control theory. If we map the integer variables at each point we can build a model but there are too many unknowns. So many unknowns in fact it would be like trying to decrypt 8k encryption with an ipad.

We meet more RNG after RNG. What I mean by that is 1) rolling maps has XX chance to roll XX mod 2) number of Rare monsters 3) number of modifiers on rares etc. etc. These monster have XX base chance at XX level to drop Tier 1-5 blar blar blar.

As we don't know at what rates or what factors or mods or a great deal after "drop rates" we need access to the source code to know. That ain't gonna happen so meh.

Yep, thanks smarty pants :) I don't think I can be arsed trying to figure out all that noise.

Exactly right but the average is the mean (50) and the deviation is 10 hence a ‘luck’ potential of + or - 10 and that gives us a range of 40(bad luck) to 60(good luck).
Curbing outliers that a pseudo RNG code outputs (clipping the edges of the bell curve) means no one gets excessively good or bad luck, your magic find (quantity would multiply the items) and rarity(would change the value of the item) would still work on a pseudo RNG value just with a different mana value.
I have ofc (I could data mine the values if they are in the client side but suspect they are on the server side) give actual numbers but an example suffices for the conversation.

"

Jadian#0111 wrote:

"

Timbo Zero#8289 wrote:

@GGG I am, with relative confidence, sure you are using Gaussian RNG generation

Can you share any of the data you gathered to achieve that confidence?

I could but not publically and most definitely not here as it would breach code of conduct.
You of course, to have asked such a question, have a foundation in advanced mathematics and computer coding so can also find the evidence for your self.

"

Suchka_777#4336 wrote:

"

Soulscar1#4896 wrote:

The distribution simply implies the more you play (or more quantity you use) the more likely your "luck" is going to fall in the upper +ve deviation more often/sooner hence why even a little quantity is so strong.

What you are talking about? Expected value you get never change no matter how long you play. You got something completely wrong, which is common case when someone (OP) is trying to sound smart and say some stupid things in the foggiest way possible (otherwise, noone will believe that he is saying something smart).

Would you care to elaborate on why you think I am not smart enough to say what you say is smart sounding but stupid please ;)

I have used some exceptionally simple maths and basic C programming code to prove my point, and that I understand the issues. And left a challenge to GGG support to disprove my work.

This thread is super frustrating because so many terms are being used incorrectly.

First, the default distribution for rand in most languages is uniform, not gaussian. Every discrete value has an equal chance of happening. To *accidentally* use a gaussian distribution requires a phenomenal order of luck or ineptitude.

Secondly, using the modulo operator on a number that doesn't evenly divide by whatever the defined max is will give some unfairness. Usually the number when it's an integer range is from 0 to (inclusive) 2^16-1, other ceilings and floors are common though. However, since that number is 65535, and you are using a modulo of 21, you are giving a small favor of rolling numbers in the range [0..4], given these parameters.

Third, you generated a handful of numbers from a website for generating a list of gaussian numbers, and you misunderstand that the normal distribution does not shunt values, you can have a value that is several standard deviations above or below the mean, but it is still within possibility. Your small sample size for example has an average of ~47.7, and a standard deviation of ~9.4
That means approximately 99% of results are going to be between approx 19.5 and 75.9 if this is really a normal distribution, (it's probably a polynomial distribution if your generator is only giving discrete instead of continuous output, but that's a quibble.) That's already outside the 40 +/- 20.

Fourth, if you tested that c code you wrote (or maybe found by asking google's AI, https://i.imgur.com/ozcOkO9.png ) maybe did something like, modify it a bit to generate 100k numbers, and check the histogram, you'd get a pretty good idea that this would be a very even distribution. If you wanted people to believe you, you could have asked google, "c random number between 40 and 60 on the gaussian distribution" and got the following result. Note that it takes a lot more deliberate arithmetic to get a number on a gaussian distribution, and you use the mean and standard deviation as parameters to that function.
https://i.imgur.com/9HoiX2O.png

"

Timbo Zero#8289 wrote:

"

Jadian#0111 wrote:

"

Timbo Zero#8289 wrote:

@GGG I am, with relative confidence, sure you are using Gaussian RNG generation

Can you share any of the data you gathered to achieve that confidence?

I could but not publically and most definitely not here as it would breach code of conduct.
You of course, to have asked such a question, have a foundation in advanced mathematics and computer coding so can also find the evidence for your self.

Indeed, I could find evidence myself. Without breaching code of conduct. Loot isn't the only thing RNG is used for, after all.

Damage is random, and damage numbers are shown when fighting bosses. Now, I lack the motivation, but if you wanted proof you could use a skill with high damage variance, and track the damage numbers. You'd need to set up the character for it--no conditional bonuses on low life or full life, for example--as well as accounting for the enemy's resistance.

That would be a good sign that a Normal Distribution is being used where you might otherwise expect a Uniform Distribuion. Without cause like that, I'd assume a less reliable RNG would be used in places where strict randomness isn't required, like visual effects.



Even if a Normal Distribution is being used though, I don't see how that would lead to people getting "disgustingly bad" or "disgustingly good" RNG. Unless they were regularly getting rolls that aren't within 1-2 deviations and all to the same side of the expected mean, it would all just average out. Just like how the average of the number set in the first post is ~47.7--barely quantifible as bad luck.

There were instances of extreme luck, but if an instance of extreme bad luck is indistinguishable from regular bad luck...your suggestion just sounds like getting rid of instances of extremely good luck.