LaplacesDemon (version 16.1.0)

RejectionSampling: Rejection Sampling

Description

The RejectionSampling function implements rejection sampling of a target density given a proposal density.

Usage

RejectionSampling(Model, Data, mu, S, df=Inf, logc, n=1000, CPUs=1, Type="PSOCK")

Arguments

Model

This is a model specification function. For more information, see LaplaceApproximation.

Data

This is a list of data. For more information, see LaplaceApproximation.

mu

This is a mean vector $\mu$ for the multivariate normal or multivariate t proposal density.

S

This is a convariance matrix $\mathrm{\Sigma }$ for the multivariate normal or multivariate t proposal density.

df

This is a scalar degrees of freedom parameter $\nu$. It defaults to infinity, in which case the multivariate normal density is used.

logc

This is the logarithm of the rejection sampling constant.

n

This is the number of independent draws to be simulated from the proposal density.

CPUs

This argument accepts an integer that specifies the number of central processing units (CPUs) of the multicore computer or computer cluster. This argument defaults to CPUs=1, in which parallel processing does not occur.

Type

This argument specifies the type of parallel processing to perform, accepting either Type="PSOCK" or Type="MPI".

Value

The RejectionSampling function returns an object of class rejection, which is a matrix of accepted, independent, simulated draws from the target distribution.

Details

Rejection sampling (von Neumann, 1951) is a Monte Carlo method for drawing independent samples from a distribution that is proportional to the target distribution, $f(x)$, given a sampling distribution, $g(x)$, from which samples can readily be drawn, and for which there is a finite constant $c$.

Here, the target distribution, $f(x)$, is the result of the Model function. The sampling distribution, $g(x)$, is either a multivariate normal or multivariate t-distribution. The parameters of $g(x)$ (mu, S, and df) are used to create random draws, $\theta$, of the sampling distribution, $g(x)$. These draws, $\theta$, are used to evaluate the target distribution, $f(x)$, via the Model specification function. The evaluations of the target distribution, sampling distribution, and the constant are used to create a probability of acceptance for each draw, by comparing to a vector of $n$ uniform draws, $u$. Each draw, $\theta$ is accepted if $$u\le \frac{f(\theta |\mathbf{\text{y}})}{cg(\theta )}$$

Before beginning rejection sampling, a goal of the user is to find the bounding constant, $c$, such that $f(\theta |\mathbf{\text{y}})\le cg(\theta )$ for all $\theta$. These are all expressed in logarithms, so the goal is to find $\log f(\theta |\mathbf{\text{y}})-\log g(\theta )\le \log (c)$ for all $\theta$. This is done by maximizing $\log f(\theta |\mathbf{\text{y}})-\log g(\theta )$ over all $\theta$. By using, say, LaplaceApproximation to find the modes of the parameters of interest, and using the resultant LP, the mode of the logarithm of the joint posterior distribution, as $\log (c)$.

The RejectionSampling function performs one iteration of rejection sampling. Rejection sampling is often iterated, then called the rejection sampling algorithm, until a sufficient number or proportion of $\theta$ is accepted. An efficient rejection sampling algorithm has a high acceptance rate. However, rejection sampling becomes less efficient as the model dimension (the number of parameters) increases.

Extensions of rejection sampling include Adaptive Rejection Sampling (ARS) (either derivative-based or derivative-free) and Adaptive Rejection Metropolis Sampling (ARMS), as in Gilks et al. (1995). The random-walk Metropolis algorithm (Metropolis et al., 1953) combined the rejection sampling (a method of Monte Carlo simulation) of von Neumann (1951) with Markov chains.

Parallel processing may be performed when the user specifies CPUs to be greater than one, implying that the specified number of CPUs exists and is available. Parallelization may be performed on a multicore computer or a computer cluster. Either a Simple Network of Workstations (SNOW) or Message Passing Interface (MPI) is used. With small data sets and few samples, parallel processing may be slower, due to computer network communication. With larger data sets and more samples, the user should experience a faster run-time.

This function is similar to the rejectsampling function in the LearnBayes package.

References

Gilks, W.R., Best, N.G., Tan, K.K.C. (1995). "Adaptive Rejection Metropolis Sampling within Gibbs Sampling". Journal of the Royal Statistical Society. Series C (Applied Statistics), Vol. 44, No. 4, p. 455--472.

Metropolis, N., Rosenbluth, A.W., Rosenbluth, M.N., and Teller, E. (1953). "Equation of State Calculations by Fast Computing Machines". Journal of Chemical Physics, 21, p. 1087-1092.

von Neumann, J. (1951). "Various Techniques Used in Connection with Random Digits. Monte Carlo Methods". National Bureau Standards, 12, p. 36--38.

See Also

dmvn, dmvt, IterativeQuadrature, LaplaceApproximation, LaplacesDemon, and VariationalBayes.

Examples

# NOT RUN {
library(LaplacesDemon)
### Suppose an output object of class laplace is called Fit:
#rs <- RejectionSampling(Model, MyData, mu=Fit$Summary1[,1],
#     S=Fit$Covar, df=Inf, logc=Fit$LP.Final, n=1000)
#rs
# }