I'm trying to quickly sample from a near-uniform discrete probability distribution exactly once without calculating the entire CDF. Here's the algorithm.


  • $N,$ the number of elements to sample from
  • $p : \{n \in \mathbb{N} | n < N\} \to \mathbb{R}[0,1],$ the probability distribution function
  • $M : \mathbb{R} = \max_{\forall n < N} p(n)$, the maximum probability

The algorithm is as follows (pseudocode):

define select(N, p, M):
    LET i := a random natural number < N
        return i
        return select(N, p, M)

The distribution that this algorithm selects is

$$\mathbf{q} = \frac{1}{N}\sum_{i=0}^N\left(\frac{p_i}{M}\mathbf{e_i} + \left(1-\frac{p_i}{M}\right)\mathbf q\right)$$

Solving for $\mathbf q$:

$$N\mathbf{q} = \frac{1}{M}\mathbf p + N\mathbf q - \frac{1}{M}\mathbf q$$

$$\mathbf p = \mathbf q$$

And the time complexity is

$$t = \frac{1}{n}\sum_{i=0}^n \left(\frac{p_i}{M} \cdot 1 + (1 + t)\left(1-\frac{p_i}{M}\right)\right)$$

$$Nt = \frac{1}{M} + N(1+t) - \frac{t+1}{M}$$

$$0 = N - \frac{t}{M}$$ $$t = MN$$

So for a near-uniform distribution, where $M = O(N^{-1})$ then the algorithm is $O(1)$.

My question is: is there a name for this algorithm?


Yes; that's called rejection sampling.

  • $\begingroup$ Thanks! For some reason I couldn't find any fast algorithms for this purpose anywhere. Everyone was suggesting finding a CDF $\endgroup$ – k_g Jul 23 '16 at 18:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.