DAbayesSuite: Parallel inference in Bayesian hierarchical model for...

In mapn/DAbayes: Bayesian Statistical Model for Climate-Change Detection and Attribution

Description

This function wraps all the necessary functions in DAbayes package to run the MCMC Suite. By default, this will execute the adaptive MCM algorithm in single core (sequentially), and it can be executed in parallel by changing the number of cores to the number of cores available.

Usage

DAbayesSuite(ensemble, model_runs, control_runs, r_max = NULL,
  theta_intVal = NULL, prior = NULL, AD_proposal = NULL, niter = 20000,
  start_adapting = 50, numCore = 1)

Arguments

ensemble	an n by N matrix of ensembles of measured variable, say temperature increase
model_runs	a list of elements, each of which is model-output of the measured variable (temperature increase) under a specific forcing scenario
control_runs	an n by L0 matrix, where n is the number of grid cells, and L0 is the number of control runs
r_max	an integer representing the maximum number of empirical orthogonal functions
theta_intVal	a list of 3 elements containing initial values for parameters in the MCMC algorithm
prior	a list of 6 elements in prior distributions for corresponding parameters
AD_proposal	a list of 6 elements containing quantities to adjust mean and covariance in the proposal distribution
niter	an integer containing the total number of MCMC interations
start_adapting	an integer specifiying when to adapt proposal in the MCMC algorithm
numCore	an integer specifying how many cores are used in parallel computing

Value

a list of r elements, where r is the number of EOFs specified, and each element is again a list of 6 elements containing posterior quantities of parameters, log-likelihood, chisq statistics, and prior:

beta: a matrix holds the posterior samples for the parameter beta with each row corresponding to each beta

logsigma: a vector holds the posterior samples for the parameter log of sigma

lambda: a vector holds the posterior samples for the parameter lambda

loglik: a vector holds the log-likelihood evaluated with updated parameters

chisq: a vector holds chisquare statistics for residual consistency test

prior: a vector holds the prior density evaluated with updated parameters

Author(s)

Pulong Ma <[e-mail]>

Examples

## Not run:  
## Run Gibbs sampling with Metropolis-Hastings algorithm in the 
# Bayesian statistical model
data(ensemble_temperature)
data(GCM_runs)
data(GCM_control_run)

Res <- DAbayesSuite(ensemble=ensemble_temperature, model_runs=GCM_runs, 
control_runs=GCM_control_run, r_max=2, theta_intVal=NULL, prior=NULL,
AD_proposal=NULL, niter=20000, start_adapting=100, numCore=1)

## diagnostic plots for posterior samples, and more sophisticated
# diagnostic plots can be obtained with coda package
# take the first element in Res for example
# plot only first 2 beta's
plot(Res[[1]]$beta[1, ], xlab="Iteration", ylab="beta_1",  type="l")
plot(Res[[1]]$beta[2, ], xlab="Iteration", ylab="beta_2", type="l")
plot(Res[[1]]$logsigma, xlab="Iteration", ylab="logsigma", type="l")
# plot only first 3 lambda's
plot(Res[[1]]$lambda[1, ], xlab="Iteration", ylab="lambda1", type="l")
plot(Res[[1]]$lambda[2, ], xlab="Iteration", ylab="lambda2", type="l")
plot(Res[[1]]$lambda[3, ], xlab="Iteration", ylab="lambda3", type="l")

## save plots 
burnin <- 5000
plotMCMC(Res, N, n, dir=getwd(), burnin)

## End(Not run)

mapn/DAbayes documentation built on May 21, 2019, 11:26 a.m.