a couple of the game formulas seem out of whack, particularly although perhaps not exlusively

missiles/sdi

recently i took 62 nuke missiles to my country maintaining 85% or more sdi throughout, 22 succeeded. The improbability of that occuring i can only assume that something relating to the formula of probabilities is off quite substantially

is this a known difference, or am i supposedly just increadibly unlucky?

You're incredibly unlucky that so many got through. You're incredibly lucky that SOL would pick someone with 85% SDI as their nuke target.

Assuming your SDI was at 85% the whole time for simplicity, what we have is a Bernoulli process with n=62 trials and p=.15 as the success rate. The probability for this is governed by the binomial distribution. The probability of there being exactly 22 successes in 62 trials is 5.2649E-4 %, and the probability of there being 22 or MORE successes in 62 trials is 6.4481E-4 %.

So yeah, you got screwed or there is something else at play.

N.B.: I used the normal approximation to the binomial distribution, i.e. N(np,np(1-p)), so the numbers may be *slightly* off

this happned with another guy in laf i saw. A shorter term amount of missiles though...

i'm pretty sure we've tested them fairly thoroughly

I bet if you nuked qzjul's country and saw how many were actually defended this fluff would get fixed like pronto. So do it! ;) ;)

hmn but i have 0 SDI points so i wouldn't really notice...

Keep in mind that as you get hit with more missiles, your SDI drops. So you not only have what Ozz described above, but you also incrementally increase the likelihood of a missile passing your SDI every time one does not (i.e. those 44 that hit your SDI)

Just because you have %85 SDI does not mean you're going to block %85 of the missiles fired at you. It just means you have a %85 chance of blocking every missile. If the first bounces and takes %1 of your SDI away, you will then have an %84 chance of blocking the next, %83 of blocking the next, %82, etc etc, etc.

44 missiles will decrease your SDI by appx %20-%25 or so if I remember correctly.

But yeah, it's just some bad luck too.

l3arn to r3ad NOW3P :-P

"i took 62 nuke missiles to my country maintaining 85% or more sdi throughout"

THROUGHOUT

whether that is true, I do not know, but that is what he claimed :-P If anything the probability of it happening is lower than my numbers

Yea, SDI is fubar'd in ffa as well, i have 80% plus on my attacked countries,and at LEAST 1 of 3 get thru...the formula is skeeked BAD..

your sdi has 80% chance to stop those missiles so .......... prolly just bad luck.

I expanded on the thought to help him understand the point, Ozz.

SDI seems to be exactly the same as before - sometimes it'll stop 10:10, sometimes it'll stop 6:10 - all depends on random luck though.

I understand that, in fact I provided the probabilities of the occurrenes of such things and how to calculate them :-P

Our own internal review concluded that the only issue would be with the random number generator. If any statistically minded people want data to test it, I would be happy to provide it.

I would be happy to confirm that your random number generator is working if you provide me with a sizable set of samples, but the best solution would be to post the relevant code.

It is very hard to believe that this could be caused by a bad random number generator. If people consistently think this is happening my guess is it is bad code.

For the record, I am not necessarily saying there is ANY problem. Things much more unlikely than 5.2649E-4 % happen every day.

return (mt_rand()/mt_getrandmax());

is all it is =/

where those functions are here:

http://ca3.php.net/...l/en/function.mt-rand.php

Shouldn't there be some code to seed the generator somewhere? to stop repeat sequences from cropping up.

I haven't tested this as thoroughly as qz and slag have, but I'm inclined to agree with them. I've looked at the formula, and short of the random number generator not being random, the math looks solid.

but this is probably the most common formula "bug" that gets brought to my attention. and not just by SS. :p

we could potentially run some more thorough analytics, but that would have to come down the line

Are we talking a record of missile success rates with SDI percentages, or just a statistical breakdown of the random number generator itself?

-Fooglmog
Guy with no clue.

Wait a second... why are you using mt_getrandmax() in your "return (mt_rand()/mt_getrandmax());" statement?

Isn't that consistently going to give you a return with incredibly small exponents? Don't you just want to do "return (mt_rand(1, 100)/100);"?

I set up a test page at http://inklined.net/test/. It's just a sample of 20 mt_rand(1, 100)/mt_getrandmax() returns. To be fair, it's all just random number, and I don't know how the random number influence your formula but I can't imagine why you'd want exponents in the E-8 or larger range.

EDIT: I made an error above, I forgot that the arguments added to mt_rand() were mine... that page no longer shows what I said it does.

-Fooglmog
Guy with no clue.

The code for checking success or failure is laughably simple, but I just made a table to log certain data for further testing. We love you guys that much!

mt_rand() belongs to the integers in [0, mt_getrandmax()]

therefore, mt_rand()/mt_getrandmax() belongs to [0,1]

(mt_rand(1, 100)/100) isn't quite what we want because SDI defense is more finely grained than integers.

Ah, stupid mistake from me. Sorry. I added the arguments to mt_rand() when I was playing with the code and forgot they were my addition instead of exactly what qz had posted.

Could you still post the section of code dealing with missile success? I'd like to see it in its entirety if I could.

-Fooglmog
Guy with no clue.

So the code you provided for sampling a number in [0,1] in php is fine as long as you call mt_srand() before you call it, otherwise you are always getting the same sequence of random numbers.

I assume the rest of the code looks like:

isMIssileBlocked() {
SDI = getSDI%()
randNum = getRand(0,1)*100
if((1.0 - SDI) > randNum)
return true
else
return false
}

?

this centering is bullfluff for the record....

They don't need mt_srand()... mr_rand() uses a random seed by default... in effect, the seed is random unless you use mt_srand() with arguments.

And making up code isn't really helpful... if we're going to confirm that the code works, we need to see what it is, not make up what it could be.

-Fooglmog
Guy with no clue.

http://php.net/.../en/function.mt-srand.php


Note: As of PHP 4.2.0, there is no need to seed the random number generator with srand() or mt_srand() as this is now done automatically.


whoops :-P





suck mah balls

http://slagpit.earthempires.com/sditest

Have fun.

If someone has access to minitab or something like that, it would be really easy to look through those statistics.

I've tested the rng extensively for "randomness" (lack of pattern) and distribution and it is fine for purposes of this game. (Good enough to test at least 1 in 500 events which is beyond the scope of this game).

"Assuming your SDI was at 85% the whole time for simplicity". There's your problem. in fact it drops faster than you think it does and that greatly throw's off your test I think.

your normal approx actually understates the probability somewhat but at n=65 it's shouldn't have that great of an impact.

First of all I tested the numbers myself and got similar results, but my estimate is not incorrect. SS claims that his SDI stayed above 85% THE ENTIRE TIME. Whether or not that is true I cannot say, but that is what he said.

=P

wouldn't it have to do with the % of land the nuke takes away if successful and wouldn't that raise SDI?

"SS claims that his SDI stayed above 85%".
missed that.
Well if that is true than your argument is correct.

It's entirely possible that the issue isn't the formula testing missile success + rng but something relating to the SDI% as displayed vs what is actually being used.

Just a random thought

SDI display percentage uses the exact same formula as the SDI calculated percentage; there is no problem there.

I suspect the "SS claims that his SDI stayed above 85%". is incorrect.

Either that or he's just unlucky ;)

Technically speaking somebody with 98% SDI could be missile killed without a single defend.... though it's unlikely ;)

8269 data points so far... I assume that nobody did anything with it?

I've done some Slagpit, it all works out as far as I can see. Here's how the numbers break down with 8360 data points.

Average SDI: 0.34536566
Overall Success Rate: 0.65450413
Variance: 0.00013021
Average Random Variable: 0.50094538

As for random variable distribution, it is as follows:
Expected: 836
0.0-0.1: 832 Variance: -4
0.1-0.2: 834 Variance: -2
0.2-0.3: 837 Variance: +1
0.3-0.4: 843 Variance: +7
0.4-0.5: 793 Variance: -43
0.5-0.6: 871 Variance: +35
0.6-0.7: 809 Variance: -27
0.7-0.8: 853 Variance: +17
0.8-0.9: 852 Variance: +16
0.9-1.0: 836 Variance: 0

As you can see, the largest variance from the expected takes place between 0.4 and 0.5... a variance of 5.14%. This is high, but not unreasonably so given the sample size.

I also did a breakdown in 500 hit increments, to make sure that there weren't time-based statistical anomalies. Obviously smaller sample sizes mean greater error rates, however, there was never more than a 3.5% difference between bounce rates predicted bounce rates based on SDI. The random variable never deviated more then 3.5% from the expected 0.5 average either.

In other words, as far as I can tell, everything checks out.

-Fooglmog
Guy with no clue.

sounds like it all checks out

nifty info

qz.. all you have to do is look at my country to see considering the nm loss it would of been impossible to go bellow 82-83% even without buying sdi which i did consistantly between hits

your basic answer is that i am just unlucky to an extent that should be impossible...

I never said anything about impossible; it's more than possible

we have mechanisms in place now that can tell us that exact type of data, so if it ever happens again, we'll be able to model it very effectively.

so let's drop the issue until then, and then look at the data next time :p

so next set. :P

bah there isn't any issue. unlikely fluff happens all the time. what are the odds that we would exist and be talking on these forums. A fluffton less than your little missile bullfluff =]

If you won the lottery, would you conclude that there was something wrong with how they generated the winning numbers?

Yes. But I wouldn't tell anyone.

-Fooglmog
Guy with no clue.

Slag: I'd really question their methods for sure, as I've never actually bought a lottery ticket.


disclaimer: well I guess I was forced to contribute to a work pool once or twice, but that doesn't really count, as I was certain we'd lose that time and I'd be unlikely to see that much anyway.

at here its 1:15m chance to get jackpot on lottery.

and remember if you live in north america your odds of dying in a car crash are roughly 1 in 15,000
:P

so after you've died in a car crash 1000 times, then you have a good chance at winning the lottery =D

Also, I presume that's per year or something martian, not per time you drive...

at italy its worse to get jackpot on lottery, 1:650mish so very very very small.

I think it's per life-time qz ;)

As in, 1 in every 15,000 people die in a car crash... rather than, each year (or worse, each time you drive) you have a 1 in 15,000 chance of dying.

Assuming that the number is correct.

-Fooglmog
Guy with no clue.