Accept-Reject sampling Method (并非近似贝叶斯计算采样)

The Accept-Reject method is a classical sampling method which allows one to sample from a distribution which is difficult or impossible to simulate by an inverse transformation. Instead, draws are taken from an instrumental density and accepted with a carefully chosen probability. The resulting draw is a draw from the target density.

Accept-Reject Algorithm

The objective is to sample from a target density \pi(x)=f(x) / K, where x \in \mathbb{R}^{d}f(x) is the unnormalized target density, and K the potentially unknown normalizing constant. Suppose that we can sample from another density h(x) and that there exists a constant c such that  for all x. To obtain a draw from \pi:

  1. Draw a candidate z from h and u from U(0,1), the uniform distribution on the interval (0,1).
  2. If u \leq \frac{f(z)}{c h(z)}, return z.
  3. Otherwise, return to 1.

The expected number of iterations required to accept a draw is c^{-1}. To ensure efficiency, the optimal choice of c is:

c=\sup _{x} \frac{f(x)}{h(x)}

References

  • Chib, S. and E. Greenberg (1995). Understanding the Metropolis Hastings Algorithm. American Statistical Journal 49, 327–335.

  • Robert, C.P., and G. Casella (2004). Monte Carlo Statistical Methods, Second Edition. New York: Springer.

你可能感兴趣的:(ᕦ,机器学习,ᕤ,A\R,sampling,sampling)