This might be a really elementary question, but I'm not sure what it means. I have a density function f(x). How do I sample a value from f? For known distributions there are functions in R which do it for you (e.g. runif, rnorm, etc.) but how do I generate a random number using my own density?
-
1$\begingroup$ Why the down votes? This is highly nontrivial (see my answer) $\endgroup$– Igor RivinMar 29, 2012 at 17:43
-
$\begingroup$ I agree with that, it had also intrigued me, and I was a bit confused to ask ;) $\endgroup$– AminMar 29, 2012 at 21:49
-
$\begingroup$ Read chapter 3.4.1 from Knuth's The Art of Computer Programming. $\endgroup$– Zsbán AmbrusMar 30, 2012 at 6:28
3 Answers
In the simple case that $X$ is a real valued random variable, the first thing I would reach for is the inverse-cdf method, especially since you have mentioned "runif" which gives draws from a uniform distribution.
There is a pretty extensive literature on ways to sample from a variety of distributions, with names like Gibbs sampling, Metropolis-Hastings, slice samplers, perfect samplers, etc. A Google search of any of these should bring up a wealth of info. Did you want something more specific?
I am not going to answer the philosophical question of "what does it mean", but for the practical question, there is the Ziggurat method of Marsaglia to generate a sample from your favorite distribution. Read all about it.
I suppose a definition of a random sample would be a sequence of numbers {$a_{n}$} such that, for any measurable set S we have $\sum_{1}^{n}\chi_{S}(a_{i})/n\rightarrow\mu(S)$ as $n\rightarrow\infty$, where $\mu(S):=\int_{S}f(y)dy$, with $f(y)$ your density.
How to generate such a sequence is another, much more involved question; there's the inverse CDF technique, various algorithms like Metropolis Hastings or Gibbs sampling (for higher dimensional densities), etc. etc.