$rand bug? - mIRC Discussion Forums

hi,

don't know if this can be fixed but I did a test on the $rand identifier and it seems to pick the first and last number only half of the time compared to the others... can this be fixed somehow or can I get around it?

There is no such thing as a "random number" A computer uses a PRNG (pseudo random number generator) to try and create random numbers. But computers do not understand the concept of randomness, all the random numbers are predictable if you know the formula to use. The only real way to fix this is to use a better PRNG, people have suggested others that are better, but Khaled has never actually used them, so I doubt he will change his mind now...

That might be an issue but is it such a bad thing? Randomness does not mean that each number should be drawn out of the hat the same number of times. As codemastr stated, computers don't know about randomness. They can't really because they don't know how long you are going to draw random numbers for. Therefore the likelyhood of all numbers getting drawn the same number of times is very slim indeed.

Actually that is the exact point - over a large enough scale all numbers should come out around the same number of times. Of course, in general when it comes to mIRC scripting the randomness of $rand() is perfectly acceptable (perhaps when using only 3-6 random characters it might be noticeable).

Well it's just annoying that when I use $rand(1,3) it returns 2 half of the time. I know computers have a hard time generating random numbers, but I didn't know mIRC's algorithm is so messed up. Another weird thing is that this 'bug' doesn't occur when using letters instead of numbers.

To test this and see if the allegations are true, I wrote this alias to run a series of 100,000 random runs:

Code:


alias rand.count {
  var %1, %2, %3, %4, %5
  var %i = 1
  while (%i &lt; 100000) {
    inc $+(%,$r(1,5))
    inc %i
  }
  echo -ati2 ** 1( $+ %1 $+ ) 2( $+ %2 $+ ) 3( $+ %3 $+ ) 4( $+ %4 $+ ) 5( $+ %5 $+ )
}

Here are the results of several runs of this alias.

** 1(12458) 2(24842) 3(25015) 4(25157) 5(12527)
** 1(12876) 2(24764) 3(25271) 4(24720) 5(12368)
** 1(12422) 2(25034) 3(25186) 4(24885) 5(12472)
** 1(12600) 2(24789) 3(25104) 4(25070) 5(12436)
** 1(12632) 2(25110) 3(24968) 4(25071) 5(12218)
** 1(12517) 2(25157) 3(25078) 4(24769) 5(12478)
** 1(12599) 2(25137) 3(24835) 4(24956) 5(12472)
** 1(12457) 2(25135) 3(25054) 4(24914) 5(12439)
** 1(12666) 2(25126) 3(25196) 4(24663) 5(12348)
** 1(12630) 2(25098) 3(24893) 4(24997) 5(12381)

From this, we can determine that the first and last DO show up half the number of times as the others. 2-4 are quite evenly distributed. That means that if you wanted to derive a "more truly random number", take $rand(1-(N-1), counting N as a 1. So if I need a random number from 1 to 3, I can use something $rr(1,3) - like this:

Code:


alias rr {
  var %n = $rand($1,$calc($2 + 1))
  if (%n == $calc($2 + 1)) return $1
  return %n
)

Substituting $rr for $r in the above alias uses the new RealRand alias and returns the following:

** 1(19939) 2(19977) 3(19938) 4(20143) 5(20002)
** 1(19971) 2(19956) 3(20202) 4(20162) 5(19708)
** 1(19813) 2(20021) 3(20046) 4(19967) 5(20152)
** 1(20190) 2(20199) 3(19999) 4(19823) 5(19788)
** 1(19808) 2(20136) 3(19971) 4(19991) 5(20093)
** 1(20027) 2(20198) 3(19994) 4(19891) 5(19889)
** 1(19928) 2(19954) 3(20049) 4(20106) 5(19962)
** 1(20020) 2(20014) 3(20107) 4(19976) 5(19882)
** 1(19998) 2(20161) 3(19927) 4(19971) 5(19942)
** 1(20078) 2(19899) 3(19991) 4(20139) 5(19892)

Hmm, I just posted about this myself and only found this thread now (sorry

)

The thing is that this effect was not present in 6.0 and 6.01, so I would consider it a bug and not a result of computers being unable to produce random numbers.

The formula in the $rand function was changed and it was broken... Change it back please

the bug is/was also present in 5.91

As has been pointed out before, computer generated random numbers are not random.

The only way to be as close to random as possible, would be to use some sort of equation that took the input of radio wave noise from stars, and turned it into a number, yet that itself would still be a somewhat 'psuedo' generated number.