I'm learning Monte-Carlo approach in sampling. There I faced with ways of how to draw samples from given distribution. But can you give me an example of a distribution which can not be trivially simulated as normal or binomial distribution?
Asked
Active
Viewed 136 times
2
-
4It's still not quite clear what your last sentence is asking. Do you mean "cannot be *as* trivially simulated as normal or binomial"? What's your trivial means of simulating a binomial? What makes something trivial or non-trivial more generally? – Glen_b Jun 02 '15 at 10:46
-
von Mises-Fisher is also such an example, [cf. this previous post on Cross Validated](http://stats.stackexchange.com/questions/156729/sampling-from-von-mises-fisher-distribution-in-python/157177#157177). – mic Jun 18 '15 at 12:39
2 Answers
5
Drawing gamma random numbers usually requires rejection sampling, it is less trivial.
I assume that you refer to trivial if the CDF is invertible, or conversion from the uniform to the target distribution can be resolved with thresholds.
spdrnl
- 2,017
- 8
- 11
4
In this earlier Cross Validated question, a density defined as $$h_β(r)∝(1−w_{\mu,τ}(r))f_{β_0}(r)+w_{\mu,τ}(r)g_{ϵ,σ}(r)$$ is proposed, with a non-trivial simulation solution.
In my class, I usually give the benchmark density target $$h(x)\propto \{1+\sin^2(2x)+\sin^4(4x)\}\exp\{-x(1+\cos^2(4x)+\cos^4(2x))\}$$ to simulate. You can make similar examples by piling up complex but upper bounded terms.
