rand() question

rand() does not give a gaussian distribution, it has a uniform distribution.

Hrm... rand() does give a uniform distribution.

If yours is broken, use boost::random

This seems to indicate that it is a uniform distribution.

It would seem silly to have it be a gaussian distribution... (And I've never noticed that behavior...)

Okay, well I'm going to kill the guy who told me it gives a gaussian distribution.

if u sum randoms, then u get a normal distribution
example
rand()+rand()+rand()+ ... = normally distributed rands

actually, i believe if you sum ANY random variables--or enough of them anyway, it will be a gaussian (normal) distribution. I think this is the Weak Law of Large Numbers. Like if you flip a coin 100,000 times, the number of heads will follow a normal distribution (actually you don't need to flip it nearly that many times for it to be a normal distribution)

edit: note that all the random variables have to be of an identical distribution for this to be true (assuming it is in the first place)--that is, you can't sum gaussian random variables with uniform and expect it to be normal (makes sense, but worth explicetly saying)

Using rand() will give you uniform distribution, but rand()%100 will not. But we don't want to get in this discussion again, right?

Quote:

Original post by _DarkWIng_ Using rand() will give you uniform distribution, but rand()%100 will not. But we don't want to get in this discussion again, right?

if rand() gives you a uniform distribution, why doesnt rand()%100?

Quote:

Original post by grekster if rand() gives you a uniform distribution, why doesnt rand()%100?

Imagine rand() would return value in range 0 to 127. So if you did rand()%100 you get 2x higher probability of getting a number in 0 to 27 range than in 28 to 99 range. Now, rand() has a range bigger then 127 but I hope you get the idea.

Certain implementations of rand() also have less randomness in the least significant bits and thus a modulo is not a good way to use it. Instead use

range*rand()/(float)INT_MAX

or

range*rand()/static_cast<float>(std::numeric_limits<int>::max())

Actually, it would be much better to use RAND_MAX than the maximum integer value, since the two are usually different.

Enigma

Quote:

Original post by Enigma Actually, it would be much better to use RAND_MAX than the maximum integer value, since the two are usually different. Enigma

Indeed. Sorry about that.

on visual c++ 7.1 i believe RAND_MAX is the same as the highest 16 bit signed number (even though rand() returns a 32 bit number on this platform) ... so it is always good to check what RAND_MAX is on your platform before making too many assumptions ...

as for the distrobution of rand()%100, it should be as close to normal as rand() is, simply because the range is 320 something times greater than 100, so the leftover part will affect your distrobution by a factor of less than 0.3%, which is as close as you are going to get to normal in a pseudo random generator anyway ...

If you tried to generate a random number between 0-25,0000 then things would be different, because, as mentioned the leftovers would cause the low numbers to be twice as likely as the high numbers ...