From the original SynthMaker forum (2004–2014), with its members’ names and dates.
gregh ·
here is a random number generator to output floats in the range 0 to 1. However, the output does not appear to be uniformly distributed as I expected. Instead it is biased toward the upper bound.
If anyone knows why this is happening I'd appreciate knowing too!
thanks, greg
http://www.yousendit.com/download/UW15U ... bTgwTVE9PQ
oddson ·
You're taking the absolute value of a cosine.
The slope is steap near zero and shallow near 1. So it spends much more if its 'time' near 1 than near zero.
gregh ·
Well it's a few years since i finished my phd and that was in neuroscience rather than maths so i guess i'm pretty rusty. I was assuming that each value of the cosine was equally probable so that mapping from the relatively random floats to the cosine would normalise the values to -1,1 and that the absolute value would be taken as the sqrt of the square. Hence it would be okay. But obviously not 🙂
oddson ·
A triangle wave function would give you an even distribution... but the rand numbers should already be even so I don't know what it's for.
I saw Tuz used this as a normalizing scheme and I thought it was a cool way to get a biased distribution.
The other thing is you're always seeding the lower RNG with zero as it takes an integer seed and you're having a normalized float being coerced to an integer (and therefore it's virtually always going to truncate to zero).
gregh ·
Thanks for your time on this.
I was looking for more random numbers than the current integer function offers and I wanted them as floats on 0,1
greg
oddson ·
gregh ·
oddson wrote:
thanks, that's what i first had (though with the button) until i thought of the idea of generating more than 32768 of them by doing the division between two sets of 32768
oddson ·
Yeah... I figured that was the intent. It won't generate numbers random to the full resolution of single floats - but it's good enough for most uses.
gregh ·
oddson wrote: Yeah... I figured that was the intent. It won't generate numbers random to the full resolution of single floats - but it's good enough for most uses.
yep, it'll do - i'll also look at coding something. I need to learn that anyway.
infuzion ·
Tip: Try to avoid division where you can; tends to be expensive CPU cycle-wise. [old link]
gregh ·
thanks for the tip re cpu
Wrote something here? To have a post removed or your name hidden, write to [email protected].
SynthMaker