PWI's randomizers
Comments
-
YesDevoted - Lost City wrote: »Yay, someone that understands. I'm sure the code pw runs takes drastic performance hits to improve the cost. The line "The laziest approach to get it done is probably the way it actually works" is exactly how I feel but replace laziest with cheapest.
.Elenacostel - Heavens Tear wrote: »Before we continue to theorize about the correctness of the random number generator(s) used in the game, it should first be measured.
...
3) Take those thousands of Hits and Crits and feed the expected crit % and the entire sequence into a statistical pattern recognition program.
If anyone wants to do this, I can do one or two simple randomness tests (covariance and maybe another) that might give us an indication. If you do just PM me (I might not check this thread, and it seems to be growing fast... I have work to do).Klizzahrd - Lost City wrote: »rofl the sad thing is that I DO notice when random is working well....cause everybody gets something during the bh, which at least for my experiences is odd since I notice that normally a person is left out or just gets one or two really crappy items (and I usually feel bad for them
)
The only other thing I can think of is that a bh/HH just isn't long enough for everything to even out. If you have a coin and flip it 10 times it often will look like it favors 1 side over the other. Flip 100...200...1000 times and the results are more even.
As Asterelle also pointed out, what humans think is random is actually very un-random. Uniform distribution of drops in a BH would be cause for suspicion, not the opposite. And you're absolutely right on the second case.
---
lol Aadi 0 -
Delia - Harshlands wrote: »If anyone wants to do this, I can do one or two simple randomness tests (covariance and maybe another) that might give us an indication. If you do just PM me (I might not check this thread, and it seems to be growing fast... I have work to do).
Running tests can be fun. Something else you can do is drop many arrows individually and have a squad of 2 with random distribution pick them up and keep track of who gets the pick ups.
Wikipedia has a few simple tests for gauging randomness
http://en.wikipedia.org/wiki/Statistical_randomness#Tests
As well as a good discussion of how to gauge the certainty of an observed probability
http://en.wikipedia.org/wiki/Checking_if_a_coin_is_fair[SIGPIC][/SIGPIC]
Refining Simulator - aster.ohmydays.net/pw/refiningsimulator.html (don't use IE)
Genie Calculator - aster.ohmydays.net/pw/geniecalculator.html - (don't use IE)
Socket Calculator - aster.ohmydays.net/pw/socketcalculator.html0 -
Firstly some of you people ragging on each other are dam funny.....
I dont think the sea dragon test would be much good because it doesnt drop anything does it? Why not take 2 barbs who take ages to get killed say level 40 each, to kill level 30 mobs for the test and check damge output, damge taken and drops per each player. Then run the test again with a level 10 barb in the squad, then run it again with a level 60 barb in the squad. Then all the different randoms could be checked.
I would suggest that there would be common randomization algorithms applied to each separate section, because a computer running a slightly different formula for each aspect (damage done/taken, drops and distribution of drops, differences between char/mob levels, types of mobs - weak, sacrifical etc etc) might just blow up.
But taking even the same algorithm and running it simultaneously for each aspect would generate infinte results, although generally within set (but wide) parameters (a slight dichotemy there - so perhaps only seemingly infinite).
I forgot to mention that when I find grinding doesnt result in many drops, I let myself get killed by the mobs Im grinding and often the drops improve....0 -
Of course, clusters are to be expected, but my concern is with how they take place the majority of the time I land a crit. Looking at your uniform chart, one can see that the clusters pretty much make up the minority of the data. From my experience though, when I land a crit, another will follow much more often than not.
Here, let me try to explain what I see as the flaw ~
Every attack should have a x% chance to crit, and a 100 - x % chance to not crit.
Okay, so at x = 10% crit, the chances of landing 2 successive criticals is x * x, which is 1% (1/100).
To land 3 successive critical hits, the chance is x * x * x = 0.1% (1/1000).
Yet this happens quite often, even at the lower levels of crit rate where the chances of successive crits are extremely low.
Doesn't this kind of show that whatever their algorithm may be, it's not the way it should be? - as in making it so that every hit has an x % to crit and a 100 - x % to be normal.0 -
Asterelle - Sanctuary wrote: »Running tests can be fun. Something else you can do is drop many arrows individually and have a squad of 2 with random distribution pick them up and keep track of who gets the pick ups.
Wikipedia has a few simple tests for gauging randomness
http://en.wikipedia.org/wiki/Statistical_randomness#Tests
As well as a good discussion of how to gauge the certainty of an observed probability
http://en.wikipedia.org/wiki/Checking_if_a_coin_is_fair
Yeah - when I said "run" I meant do the calculations if someone else does the punching and keeping track of the numbers. I would if I had time to, but, alas, college. I haven't even had time to get on to do my daily quests this week, let alone run tests
. If I get some time, maybe I can, but that's a really big if.
Edit: I was thinking covariance since it measures whether previous values depend on each other or not, it should be useful for the "string of crits" theory.0 -
Of course, clusters are to be expected, but my concern is with how they take place the majority of the time I land a crit. Looking at your uniform chart, one can see that the clusters pretty much make up the minority of the data. From my experience though, when I land a crit, another will follow much more often than not.
Strings of crits are definitely noticeable but so is a cloud that looks like a rabbit.
Having a feeling or impression is just not as good as doing a real trial and recording data.
You can prove your theory if you can show that a graph of the number of occurrences of a crit string of length x in a significant sample deviates from a graph of (crit rate)^x[SIGPIC][/SIGPIC]
Refining Simulator - aster.ohmydays.net/pw/refiningsimulator.html (don't use IE)
Genie Calculator - aster.ohmydays.net/pw/geniecalculator.html - (don't use IE)
Socket Calculator - aster.ohmydays.net/pw/socketcalculator.html0 -
Sorry, I made an edit to my previous post to make it a bit more clear >.<''0
-
It's actually a wonderful field, basically created single-handedly by one of the mostly unknown intellectual greats of the last century - Claude Shannon. His work is central to pretty much all modern communications theory (networking, cell phones, etc). And he came up with it all in the 1930s and 1940s before the first digital computer was ever invented. His seminal work, A Mathematical Theory of Communcation, is so succinct and easy to read that it's still required reading for anyone in the field of information theory and can be followed by most with just a college-level math background.Aadi - Lost City wrote: »Statistical pattern recognition programs...graphs of quasi- and uniform random scatter...
...Archers b:surrender
Anyway, he refers to randomness as entropy (not to be confused with thermodynamic entropy, but similar in concept). It turns out the more random a sequence is, the harder it is to compress with something like zip or rar. So a simple way to test a sequence of numbers for randomness is to just compress it and compare the resulting file size to a known random sequence of same length.
Brael came up with a pretty good way to do this. He just recorded a video of his fights using FRAPS, then reviewed it afterwards slowed down and recorded the numbers.Delia - Harshlands wrote: »Yeah - when I said "run" I meant do the calculations if someone else does the punching and keeping track of the numbers. I would if I had time to, but, alas, college. I haven't even had time to get on to do my daily quests this week, let alone run tests. If I get some time, maybe I can, but that's a really big if.0 -
Every attack should have a x% chance to crit, and a 100 - x % chance to not crit.
Okay, so at x = 10% crit, the chances of landing 2 successive criticals is x * x, which is 1% (1/100).
To land 3 successive critical hits, the chance is x * x * x = 0.1% (1/1000).
Yet this happens quite often, even at the lower levels of crit rate where the chances of successive crits are extremely low.
I remember heavy TT days where I would fire some 15,000 arrows. A 1/1000 occurrence would have occurred 15 times that day so I agree that is pretty often.
Again real data is more meaningful. Why not do as elena says and punch the ancient sea dragon for 5 mins and record crits / nonhits? It is obvious from the damage log which is which.[SIGPIC][/SIGPIC]
Refining Simulator - aster.ohmydays.net/pw/refiningsimulator.html (don't use IE)
Genie Calculator - aster.ohmydays.net/pw/geniecalculator.html - (don't use IE)
Socket Calculator - aster.ohmydays.net/pw/socketcalculator.html0 -
Because I don't know what/where the ancient sea dragon is and couldn't find its location in the database b:cute.0
-
Because I don't know what/where the ancient sea dragon is and couldn't find its location in the database b:cute.
http://www.pwdatabase.com/pwi/mob/24902
It's basically the games largest punching bag. Great for testing damage.[SIGPIC][/SIGPIC]
Refining Simulator - aster.ohmydays.net/pw/refiningsimulator.html (don't use IE)
Genie Calculator - aster.ohmydays.net/pw/geniecalculator.html - (don't use IE)
Socket Calculator - aster.ohmydays.net/pw/socketcalculator.html0 -
I dont think the sea dragon test would be much good because it doesnt drop anything does it? Why not take 2 barbs who take ages to get killed say level 40 each, to kill level 30 mobs for the test and check damge output, damge taken and drops per each player. Then run the test again with a level 10 barb in the squad, then run it again with a level 60 barb in the squad. Then all the different randoms could be checked.
It's not about collecting the drops of the sea dragon. It's testing crit % based on your attacks. You can obtain a numbers a lot more quickly this way than, say, measuring who gets what drops.
My thanx for the information regarding A Mathematical Theory of Communcation, Sir Solandri. I will have to check it out.0 -
just lol b:byeI post in forums. This one and others. That's why I post.0
-
"Of course, humans tend to view random sequences of events as non random. Humans actually expect a lot more variation in random streams than there actually is. So I'm not opposed to thinking this is just the case."
The point is of you pick 5 numbers fom 0 to 100 then "1,1,1,1,1" is the same random as "35,82, 22, 75, 53" and MMO's use resources wise solutions cause theyre MMO's and computers have their limits so get used to it.0 -
A pseudo-random number generator is intended to approximate a uniform distribution where each number drawn is as equally likely as the next number, therefore drawing the sequence "1,1,1,1,1" has a likelihood of (1/100)^5 = 1.0000E-10 whereas drawing a specific sequence of five unique numbers is (1/100 x 1/99 x 1/98 x 1/97 x 1/96) = 1.1069E-10. Those are both random, but to say they have the same probability ("is the same random as") would be wrong.Pervera - Lost City wrote: »The point is of you pick 5 numbers fom [1] to 100 then "1,1,1,1,1" is the same random as "35,82, 22, 75, 53" ...
Note: I changed your range from 0-100 to 1-100 just to simplify things, but the principle is the same if the size of the sample space is 100 or 101 outcomes.0 -
Thelas_Carr - Sanctuary wrote: »A pseudo-random number generator is intended to approximate a uniform distribution where each number drawn is as equally likely as the next number, therefore drawing the sequence "1,1,1,1,1" has a likelihood of (1/100)^5 = 1.0000E-10 whereas drawing a specific sequence of five unique numbers is (1/100 x 1/99 x 1/98 x 1/97 x 1/96) = 1.1069E-10. Those are both random, but to say they have the same probability ("is the same random as") would be wrong.
What?
...............0 -
I was trying to say that drawing the same number five times in a row is less likely (i.e., not the same) as drawing five different numbers, but I did the second probability without replacement, which in retrospect isn't apples and apples.Elenacostel - Heavens Tear wrote: »What?
...............
0 -
Thelas_Carr - Sanctuary wrote: »I was trying to say that drawing the same number five times in a row is less likely (i.e., not the same) as drawing five different numbers, but I did the second probability without replacement, which in retrospect isn't apples and apples.
So thus they are the same, then.0 -
Michael_dark - Lost City wrote: »Someone said something somewhere like if that something happens that it won't work because something someone said that sometimes something doesn't work...
QUOTE]
Lol, sounds like a GM response I've heard in a few games now b:laugh
You should make that your sig! (Logic explained) = )
On Topic: I can remember having similar discussions on a different game even ten years ago. Fact is no program works perfect, and as previously stated; peoples perception of randomness sadly differs from the meaning of randomn. I won't argue it does always seem like you hit "hot & cold" streaks when it comes to gettings drops or crits (etc.,etc.), but I would be amazed if any of us without access to the codes could accurately predict these trends.
That being said, I'm no expert. If you can figure out a way to forecast these things,more power to ya and best of luck0 -
For a given random string yes, but for conditional probabilities (where you say the 2nd trial is 1, GIVEN the first trial was already 1, etc.) no... a string of 1s when the expected value is 50 in a range from 1 to 100, is very small probability compared to a 1 followed by a number other than 1.Elenacostel - Heavens Tear wrote: »So thus they are the same, then.
It's the same as saying that flipping 5 heads in a row is the same likelihood as flipping 3 heads and 2 tails (it's not the same probability, flipping 5 heads is much less likely to happen). That relates to seeing a string of critical hits when the probability of a single crit is only 5% (or some other low number not anywhere near 50%). I'll try and provide a more correct example when I get out of class. -_^0 -
You have a problem with language.
Flipping 5 heads in a row is the same probability as flipping 3 heads followed by 2 tails.
Likewise the sequence 1,1,1,1,1 is the same likelihood as any other sequence.
If you ignore order then there are more permutations that are composed of 3 heads and 2 tails but you specifically compared "sequences" which have an order.[SIGPIC][/SIGPIC]
Refining Simulator - aster.ohmydays.net/pw/refiningsimulator.html (don't use IE)
Genie Calculator - aster.ohmydays.net/pw/geniecalculator.html - (don't use IE)
Socket Calculator - aster.ohmydays.net/pw/socketcalculator.html0 -
0
-
Lulz... you have a problem with statistics! b:chuckleAsterelle - Sanctuary wrote: »You have a problem with language.
Flipping 5 heads in a row is the same probability as flipping 3 heads followed by 2 tails.
Likewise the sequence 1,1,1,1,1 is the same likelihood as any other sequence.
If you ignore order then there are more permutations that are composed of 3 heads and 2 tails but you specifically compared "sequences" which have an order.
Probablility of getting the number 1 five times in a row when each number is a random draw from 1 to 100 = (1/100)^5 = 1E-10; near impossible
Probability of getting five different numbers = 99/100 x 98/100 x 97/100 x 96/100 = 0.9035; pretty darned likely!
The first one is pretty obvious... there are 100 numbers, but you can only choose 1 on each draw thus 1/100 x 1/100 x 1/100 x 1/100 x 1/100.
The second one you don't care what the first number is... any number from 1 to 100. The second number can't be the same as the first so there are 99 choices, but there are still 100 possible outcomes, so probabiltiy of that is 99/100. The third number cna't be the same as the first or second so number of choices is reduced to 98 and so on.
So yeah, getting a string of crits suggests something unexpected, not just "good luck".0 -
Drop this maths for a while and call your logic. Just look at sequence as at 5 sepearate numbers and you'll see that hitting 1 has same chance as any other number.
Besides that I was talking about ppl perception not probablity, when ppl see sequence of 1,1,1,1... theyre sure thats not random while it can be.
EDIT:
If you wanna se it more clear what I was talking about then make it 0-1000 and take hit in 0-99 as hit in 0, 100-199 as in 1 etc. Is not easier to hit 1,1,1... now?
I guess it's the same, just looks different. 0 -
Thelas_Carr - Sanctuary wrote: »Probability of getting five different numbers = 99/100 x 98/100 x 97/100 x 96/100 = 0.9035; pretty darned likely!
Pervera wasn't talking about 5 numbers different from each other. She was talking about the numbers "35,82, 22, 75, 53".
In that order. Only those numbers.
The chance of obtaining those 5 numbers, in that order, is the same chance as obtaining "1, 1, 1, 1, 1".0 -
Ahhh... didn't get that from the original post... yes, then agreed if it's not just any 5 "different" numbers but only that exact sequence, then odds are the sameElenacostel - Heavens Tear wrote: »Pervera wasn't talking about 5 numbers different from each other. She was talking about the numbers "35,82, 22, 75, 53".
In that order. Only those numbers.
The chance of obtaining those 5 numbers, in that order, is the same chance as obtaining "1, 1, 1, 1, 1".
Edit: Pervera said: Drop this maths for a while and call your logic. -- Pseudo-random numbers and their sequences are all about math; why would you suggest to steer away from math on a math topic?! O_o0 -
Actually, even though I love fiddling with math and statistics to figure things out in games, I tend to agree that logic, not math, is probably more useful in this case.Thelas_Carr - Sanctuary wrote: »Edit: Pervera said: Drop this maths for a while and call your logic. -- Pseudo-random numbers and their sequences are all about math; why would you suggest to steer away from math on a math topic?! O_o
Thinking about the problem logically, the server has a RNG that is receiving thousands of hits per second from all the players on the server for all sorts of reasons. Regardless of what formula is being used, every time it is called, the random seed changes.
Even if a flaw exists in the formula of the RNG, there is no possible way to predict how many times it will be called between any two events in your game. Thus, the results you will recieve from the RNG effectively contain true entropy. Attempting to find patterns based on certain events happening "close" together thus is just not possible. You can't know if your two events were actually 50 events apart or 2000 events apart, and the distance apart will always be changing randomly, since the actions of other players create true entropy in the system.[SIGPIC][/SIGPIC]
PWI Merchanting Guides: warrenwolfy.wordpress.com0 -
Two guys are standing in line to buy lottery tickets. The guy in back peeks at the guy in front and sees that he's picked the numbers 1, 2, 3, 4, 5, and 6. He says, "Are you crazy? Those numbers will never win!" b:chuckleThelas_Carr - Sanctuary wrote: »Ahhh... didn't get that from the original post... yes, then agreed if it's not just any 5 "different" numbers but only that exact sequence, then odds are the same0 -
I understand the idea that random numbers can exist in clusters. it just seems that PWI's luck based events seem too organized in neat clusters. Here's what I mean
1) throw a jar pennies on the ground( truer to randomness)
2) drop a few pennies into every other pan-handler's hat( seems more like PWI's luck system).and bro...fyi this isn't a story.
All people are idiodic but some are more idiotic than others.0 -
Untamedf - Lost City wrote: »
Just another day in the Customer Service ticket box.[SIGPIC][/SIGPIC]0
Categories
- All Categories
- 181.9K PWI
- 699 Official Announcements
- 2 Rules of Conduct
- 264 Cabbage Patch Notes
- 61K General Discussion
- 1.5K Quality Corner
- 11.1K Suggestion Box
- 77.4K Archosaur City
- 3.5K Cash Shop Huddle
- 14.3K Server Symposium
- 18.1K Dungeons & Tactics
- 2K The Crafting Nook
- 4.9K Guild Banter
- 6.6K The Trading Post
- 28K Class Discussion
- 1.9K Arigora Colosseum
- 78 TW & Cross Server Battles
- 337 Nation Wars
- 8.2K Off-Topic Discussion
- 3.7K The Fanatics Forum
- 207 Screenshots and Videos
- 22.8K Support Desk