Half the man I used to be ... (maybe a quarter of the elf actually)

Discussion in 'Berserker' started by ARCHIVED-Dak-Dod, Sep 19, 2005.

  1. ARCHIVED-CherobylJoe Guest

    Gurg
    And yes, this was the calculation for EQ, as well, and yes, there was no check for equipping 2 imbueds in EQ, either...the rationale we were given in EQ for having no check was "the cost of imbuing a single DW is the same as that for imbuing a 2H or 1H, therefore crafters couldn't afford to sell DW imbueds at a competative price"
    Do you have a link/general place to point me to for that quote? Not trying to be a nig but I never heard it before; its astonishing that a SOE developer would express such a candid (not mention lucid) thought! (btw this is not sarcasm directed at you ;-) just bet its a cool read)
    BTW you guys are starting to scare me with the statistical analysis; its like memory lane back to grad school.
  2. ARCHIVED-Sokolov Guest

    That's true, as it amounts to the same thing. And actually Neg. Binomial should've generated 1 in 158 too, so I must've missed a digit somewhere.
  3. ARCHIVED-GurgTheBashur Guest

    TBH, Joe, I can't remember if that's how a developer put it, or if that's the argument some smart boy player made that everyone else said "uhh....that's RIGHT, isn't it?"...it's been years since I touched the original Evercrack, having spent quite a bit of time in withdrawl, trying to kick the habit.



    And Solokov and Pin, you're BOTH still ignoring the same thing I overlooked in the opposite direction. The question isn't "what are the odds of coming up with a string of 200 slots entirely lacking the variable we expect to see 5 times in that period?" It is "what are the odds of coming up with a substring of 200 spaces entirely missing an incidence of a variable we expect to see five times in that period in a master string of only 5000 total slots AND without having any subsequent or preceding "overly close together" incidents of the expected variable to maintain the overall odds of incidence of the variable"

    Much less having it happen TWICE in that statistical sampling of under 5000 incidents.



    Statistically, this is the difference of testing, say, incidents of cancer...if you expect an incidence of 2.5% overall for cancer-type diseases, and you take random samplings of 200 people, yeah, your chances of finding 200 consecutive samples without any incedents of cancer is 1 in about 156. If you take a random sampling of 5000 people who maintain that 2.5% overall ratio, your chances of finding 200 consecutive samples without an incidence of cancer are MUCH lower. Doing it TWICE in the same small sampling borders on "impossible".


    Oi, this is one of the reasons I hated statistical analysis classes....same data could be used to come up with 8 different answers, each one "legitimate" and "correct", simply off overlooking one factor that changes exactly what you're looking at.
    Message Edited by GurgTheBashur on 09-30-2005 04:43 PM
  4. ARCHIVED-Pin StNeedles Guest

    Umm. What are you on about?

    The chance of picking a sequence of 200 samples, and there not being a proc in there is 1 in 158 if the probability is 0.025 for each hit (I'll ignore that it's still actually 0.0125). It doesn't matter how large the sample is, it's 1 in 158.

    The chance of finding at least 1 such sequence in 4180 samples is... basically 1 in 1 (you're almost guaranteed to do so).

    The chance of finding 2, non-overlapping sequences in 4180 samples is... also close enough to 1 for me to wager a lot of plat on finding them.
  5. ARCHIVED-GurgTheBashur Guest

    Let me put this in simple English for you, Pins:

    You guys are working negative binomial distribution for the odds of finding a string of 200 variables in which a variable which SHOULD appear one in 40 incidents is missing in a given 4910 variable string (at least Solokov is, I'm suspecting you didn't work it at all, after what you had to say about "and working it for 4910 hits makes it almost a certainty"...HIS workings took that into account already...the odds of doing so in a randomly generated sample of 200 swings are quite a bit lower, which is where I made MY mistake, working it as an inverted exponential probablilty curve for a sample of 200).

    But my point is that working it for finding the string of 200 variables which fulfill conditions inside a randomly generated string of 4910 variables, and then NOT HAVING IT BALANCE by having other strings where it is "over-concentrated", and THEN having had a randomly generated string of 4910 in which TWO SEPERATE 200 variable strings fulfilled these conditions is not anywhere NEAR on that high a frequency scale.


    It's possible to generate a 4910 variable string in which lies a 200 string variable totally lacking a variable which occurs once in it. It's possible on the order of 160 to one against...exactly the same odds you'd have of having a string of 200 in which the same variable occurred TEN times in any randomly generated string of 4910 variables, total.


    It's possible to randomly generate a string of 4910 variables in which there rests two seperate strings of 200 variables in which is totally absent a variable which should occur once in every 40 slots, on average...but the odds are MUCH higher against this happening.

    It's even possible to generate a string of 4910 variables in which a string of 200 variables rests which is totally absent any incidence of a variable that should occur once in 40 swings WITHOUT having strings of 200 variables wherein the incidence of the variable in question does not occur 7 or 8 times to offset it...but the odds of this are right in the middle of the last two situations.

    It is NOT, however, on the order of believable that the "randomly generated" string of 4910 variables that you provided was the one which "just happened" to contain TWO SEPERATE strings that SINGLY have a 1-160 chance of occurring in the given data sample size AND failed to have any OTHER strings with "overconcentration" balancing them.
  6. ARCHIVED-Pin StNeedles Guest

    Try using maths/stats/probability insteda, because English doesn't work too well for you. Using your numbers...

    The probability of finding at least 1 string of 200 variables with 0 instances within a 4910 variable string is 0.99999999999989405746282329361658 (in other words, it WILL happen).

    The probability of finding at least 2 strings of 200 variables with 0 instances within a 4910 variable string is 0.99999999999671887140854699430051 (in other words, it WILL happen).

    You'd have to find a string of at least 500 hits without a proc before it would be improbable.
  7. ARCHIVED-GurgTheBashur Guest

    And you're still missing what I'm saying, pin....

    Your odds of getting the number 12345 in a 5 digit randomly generated number are 1 in 100K
    Your odds of getting two sequences of 12345 in an 11 digit number are much lower, desite the fact you can get 12345*12345, *1234512345, and 1234512345* as possible numbers (giving you a total possible of 30 different numbers in which it will happen out of the total possible combinations)...on the order of 30 in 100 million, or 1 in 3 million.

    Solokov's oriinal figure was looking for the odds of having no incidence of a 40 to 1 overall odds variable in a random 200 variable string...essentially looking for "40" to be absent in a 200 space linear array that is randomly generated, with possibilities "ONE" through "FOURTY"

    When he did the Hypergeometric Distribution problem, he took into account the size of the actual string generated, to figure the odds of finding a 200 variable string INSIDE such a string missing the variable "FOURTY", and got 1 in 180.

    When he calculated the odds of it ocurring twice, he got 1 in 45K+ change, inside a randomly generated string of 4910 total variables...but he STILL ignored the lack of "clusters" anywhere else in the number to balance the overall odds.

    Probability does say you'll occasionally get a string of 200 variables totally lacking a variable which should appear 5 times in that space..it even says it'll happen (more rarely) within a string nested within a larger string...or even twice, seperately, within two seperate strings nested in a larger string. If you take a large enough sampling as your larger string, it's near certain to happen...but that string has to be large enough to contain 45K+ nested 200 variable strings of its own, by his figure, to make that a certainty. HOWEVER, the reciprocal odds ALSO pertain, and say that you will ALSO find a sting of 200 within that same number whereing there are 10 incidents of that variable in the nested string of 200 inside the larger string...twice, if that larger string is large enough to contain the 45K plus nested strings to produce a certitude of two seperate strings holding zero incidents...and YOUR sample holds NO balancing "clusters" to offset.
  8. ARCHIVED-Pin StNeedles Guest

    No. You're missing what you are saying.
    The odds of getting the number 12345 in a 100 digit randomly generated number are 999 in 1000. Try again.
  9. ARCHIVED-GurgTheBashur Guest

    Not quite, but close enough, the odds of getting it ONCE in a 100 digit string are high...because the odds of it happening with any single 5 digit string are 1 in 100K, and there are 95 potential 5 slot strings in a 100 digit string (assuming you allow for overlapping strings as potentials, which they are, on a purely randomly generated 100 digit string)...but by specifying that they have to be seperate strings, or in looking for arrays that otherwise require seperate strings within the larger string, you reduce odds.

    What I'm saying is that if you simply say "look for the odds of finding a string of 200 variables without an incident of the variable "FOURTY" in a randomly generated string of 4910 variables", then you're going to have to count a string of 201 variables without a single incidence of the variable "FOURTY" as two incidences, though they overlap...THIS particular situation is no more improbable than having the 200 consecutive variable string missing any incidence of "FOURTY", for all practical purposes (because the odds of that last spot disqualifying any given string are 1 in 40). In other words, odds are IF you find a string of 200 consecutive variables that firts the bill, you'll find a bunch of them, overlapping.

    But if you insist that they be distinctly seperate strings of 200 consecutive variables fulfilling conditions, you up the odds HUGELY against it, to the point that if the odds of it happening once are, say, 1 in 10, for it to happen twice, you have to have room for 20 fully seperated strings of whatever size you're looking for, rather than being able to say "in any string large enough to hold 11 strings that *could* fulfill these conditions, you have a 1 in 10 chance of showing two arrays that fulfill these conditions", which is, strictly speaking, true. That is you have to have room for those 20 seperate condition fulfilling strings for it to be AT ALL possible, otherwise your chance of it happening at all is literally zero.


    And with reciprocating situations being absent, it just increases the odds against it (in other words, if you have a 400 slot variable, your chances of getting one string of 200 in which the 40/1 chance variable is totally absent isn't that high, but getting it WITHOUT having 10 incidences of the 40/1 variable in the REMAINING 200 spaces is astronomical...and if you get, say, the FIRST 200 slots as the string in which the 40/1 incidence variable is missing, odds are pretty much 100% that the NEXT string of 200 to occur WILL have 10 incidences of the 40/1 variable, in order to balance the overall odds of that variable occurring...so the odds of having it missing entirely, and then NOT BALANCING ITSELF by "clustering" in the rest of the sample are rediculous...the odds of having it missing in two FULLY SEPERATED strings within a total sample of 4910, and not having ANY incidence of overt clustering to balance are on the order of the odds of creating a fusion powerplant in your basement with two ounces of copper, and ounce of zinc, and a lemon to work with.)
  10. ARCHIVED--Aonein- Guest

    Little quote i found from Moorgard himself ( game developer ) :
    Moorgard wrote:
    This is why the math in the post above doesn't hold up in practice. It may sound good on paper, but as we've all learned (developers included), formulas don't count. What counts is what happens in the game itself.
  11. ARCHIVED-Lorellia Guest

    I don't post often, but I do follow the boards a lot. This post interested me because I do play a Beserker and I was interested in this long running debate about which is better....

    I think that perhaps both parties are right, but are looking at different things (Please don't flame me if I am wrong :()

    Gleaming strike has a 5% chance to proc normalised for weapon speed, and accounts for Duel wield...

    However... duel wielding 2 weapons with the Gleaming Proc strike might give you the proc twice (I know it works for Sanguine imbued as it's on leggings and shield and I remember seeing them both proc)

    So Each instance of Gleaming strike has the 5% chance to proc, giving effectively twice the proc rate of a single handed weapon if BOTH weapons are imbued...

    This would mean someone with a single imbued weap coupled with a normal weap of the same delay would proc GS at the same rate as a large two hander, validating both arguments....?

    Just a thought that hit me like a bolt of lightning (And that doesn;'t happen often.. just ask any of my guildies ;-) )

    Anyway before I start rambling.... is this a possibility? And has it already been covered...?
  12. ARCHIVED-CherobylJoe Guest

    Probability does say you'll occasionally get a string of 200 variables totally lacking a variable which should appear 5 times in that space..it even says it'll happen (more rarely) within a string nested within a larger string...or even twice, seperately, within two seperate strings nested in a larger string

    This is why there are things called confidence intervals and standard deviations. While its probable that you would see certain behaviour its not certain. Odds are you are seeing a sample set thats fitting several deviations from expected; this could be due to polluted test, insufficient sampl3e size, alteration etc all. /shrug
  13. ARCHIVED-GrayStorm Guest

    Well, all I can say is it's a good thing there's no 4 armed races in this game. Although it would be fun to watch you all go at it trying to calculate how quadruple wield fits into this mess. hehe
    I'm also glad that I'm a ranger and can't even use a 2 hander. So I don't have to stress over which is better.
    I just want to thank you all, since until now I had always assumed that a shorter delay meant more chance to proc poison, etc. Thanks to this thread I now know that longer delay is better for classes like mine that rely on a lot of procs. I now have my first DW weapon and went with the imbued cobalt leafblade, which has a 2.1 sec delay I think. Tonight I'm having a second DW made as I recently acquired another cobalt / lambent stone combo. I'm thinking either another leafblade, or a slashing weapon with high delay.
    Thanks again!
  14. ARCHIVED-Mordock of the Highwynd Guest

    Everybody is in agreement with Skeetar? That a big slow weapon should result in more poison procs than a dual weilded fast weapon?

    It makes sense, since the bow nearly always procs poison.