What's an efficient way to generate N unique vectors of length M (each element a random variable drawn from its own arbitrary distribution pmfm) so that each vector satisfies two rules:
- Elements are unique
- Elements are integers bounded in the interval (0,M]
For context- I'm performing a Monte Carlo simulation relying on M competitors' rankings after a contest as input, but want to consider only realistic outcomes by modeling the likelihood of each one's placement based on a measure of their skill.
Edit: In this context, I suppose the RVs that compose each vector are not really independent, giving rise to the constraints. In that case, maybe I need to perform Gibbs sampling from an M-dimensional joint pmf. I would need to somehow define such a joint pmf to account for constraints. However, this introduces memory issues since M can be as large as 37.