Probability Distribution

> I would like to know how to coose elements from an

> array(or similar data structure)

So you access an element by an integer index?

>according to a

> probability distribution. For example a higher

> probability of choosing the elements nearer the

> start.

>

> Any tips or hints would be greatly appreciated

You start from a uniform random number generator, which presumably generates a double in [0, 1[ uniformly. From this you can derive a random int in [0, N[ according to any distribution, by dividing [0, 1[ into N intervals. You look in which interval each new random number falls; if it is in the n-th interval your random index is n.

Now, if you make the N intervals equal, each index has a 100/N % probability, and the result is again uniformly distributed.

If you make the N intervals unequal, you get a different result distribution. The length of each interval corresponds to the chance that the next uniform random double falls within it.

For example, with N = 4, intervals [0, 0.25[, [0.25, 0.5[ , ... yields a uniformly random int in [0, 4[. Intervals [0, 0.4[, [0.4, 0.7[ , [0.7, 0.9[ , [0.9, 1[ (numbers chosen pseudo-arbitrarily) yields something like what you wanted for example: 40% for 0, 30% for 1, 20% for 2, 10% for 3.

So you'll need to first calculate, given N and the desired result distribution, what you want the lengths of each interval to be.

Lokoa at 2007-7-15 3:42:52 >

> So you'll need to first calculate, given N and the

> desired result distribution, what you want the

> lengths of each interval to be.

Have you any idea how I would go about calculating this result distribution? I've had a look into probability distributions but I'm struggling. I would like to have the first interval to be the largest and the intervals to get smaller, not necessarily in a linear fashion. I can't hard code this because the size of my set is dynamic.

Thanks in advance

scathacha at 2007-7-15 3:42:52 >