BASTIONfit_C: BASTION Regression using Rcpp

In rayisaacalan/BASTION: Bayesian Additive Spanning Tree's Input / Output For Non-Parametric Regression

BASTION Regression using Rcpp

Description

BASTION Regression using Rcpp

Usage

BASTIONfit_C(Y, graph, init_vals, hyperpars, MCMC, BURNIN, THIN, seed = NULL)

Arguments

Y	A numeric vector of responses with length equal to the number of vertices in 'graph'
graph	An object of class 'graph' from the igraph package. The graph must have weights for each edge
init_vals	A named list with the following components: \itemsigmasq_yA numeric value \itemmuA list of length M where each element is a vector of length 1
hyperpars	A named list with the following components: \itemk_maxAn integer, the maximum number of clusters for each weak learner \itemMAn integer, the number of weak learners \itemlambda_kA numeric value \itemlambda_sA numeric value \itemnuA numeric value \itemsigmasq_muA numeric value
MCMC	An integer, the number of iterations for the MCMC
BURNIN	An integer, the number of initial iterations for the MCMC before recording output
THIN	An integer, specifies the thinning interval
seed	Optional, value to feed to R's set.seed function

Value

A named list with the following components:

mu_out	List of fitted values for mu for each weak learner at each output interval
sigmasq_y_out	List of sigmasq_y at each output interval
cluster_out	List of cluster membership of each vertex for each weak learner at each output interval
log_post_out	List of the log of the posterior probability at each output interval

Examples

test_coords = data.frame(x = c(0,1,1,0), y = c(1,0,1,0)) # observed locations
test_graph = constructGraph(test_coords, 3)
# Not run: plot this graph using ggraph package
# ggraph(test_graph, layout = test_coords) + geom_node_point() + geom_edge_link0()
n = igraph::vcount(test_graph)
z = c(1,2,3,4) # response

M = 3      # number of weak learners
k_max = 2   # maximum number of clusters per weak learner
mu = list() # initial values of mu
cluster = matrix(1, nrow = n, ncol = M)  # initial cluster memberships
for(m in 1:M) {
  mu[[m]] = c(0)
}

init_val = list()
init_val[['mu']] = mu
init_val[['sigmasq_y']] = 1


# find lambda_s
nu = 3; q = 0.9
quant = qchisq(1-q, nu)
lambda_s = quant * var(z) / nu

hyperpar = c()
hyperpar['sigmasq_mu'] = (0.5/(2*sqrt(M)))^2
hyperpar['lambda_s'] = lambda_s
hyperpar['nu'] = nu
hyperpar['lambda_k'] = 4
hyperpar['M'] = M
hyperpar['k_max'] = k_max

# MCMC parameters
# number of posterior samples = (MCMC - BURNIN) / THIN
MCMC = 100    # MCMC iterations
BURNIN = 50  # burnin period length
THIN = 5        # thinning intervals

BASTIONfit_C(z, test_graph, init_val, hyperpar, MCMC, BURNIN, THIN, seed = 12345)

rayisaacalan/BASTION documentation built on April 27, 2023, 2:06 p.m.