HMMbvs: Bayesian Variable Selection for HMM/MSM model

In mliang4/HMMbvs: Bayesian Variable Selection for HMM/MSM model

Bayesian Variable Selection for HMM/MSM model

Description

Performs Bayesian variable selection for multistate Markov (MSM) and hidden Markov models (HMM) to identify factors associated with transitions between (latent) discrete states in the presence of potential measurement error.

Usage

HMMbvs_R( data = NULL, tcova = NULL, tforce = NULL, ecova = NULL, 
eforce = NULL, standardize = NULL, model = "HMM", init = "baseline", 
initvalue = NULL, iter = 5000, v1 = 5, v2 = 1, a = 1, b = 9, 
thin = 10, thin_hidden = 10)

Arguments

datmat	A long format matrix of observed data. Must contain at least three columns representing the subject ID, observed state, and observation time for each observation.
tcova	A named list or a vector representing covariates that may affect one of both directions of the transition
ecova	A named list or a vector representing covariates that may affect one of both directions of the emission
tforce	A named list representing covariates that are forced into the model
eforce	A named list representing covariates that are forced into the model
standarize	A vector of variable names to standardize
model	Either "HMM" or "MSM" indicating which model to run (Default = "HMM")
init	"baseline", "warmstart" or "manual", indicating MCMC initialization method (Default = "baseline")
initvalue	A named list indicating the value for each regression term to initialize
iter	Number of MCMC iterations (Default = 5000)
v1	Prior variance of the regression coefficients (Default = 5)
v2	Proposal variance of the regression coefficients (Default = 1)
a	Hyperparameter in beta prior for inclusion probability (Default = 1)
b	Hyperparameter in beta prior for inclusion probability (Default = 9)
thin	The number of MCMC samples are thinned to (Default = 10)
thin_hidden	The number of further thinning the simulated hidden states. For example, thin = 10, thin_hidden = 10 indicates the simulated hidden states are save every 100 iterations. (Default = 10)

Value

HMMbvs returns a list including the MCMC samples of the coefficients and inclusion parameters, samples of the hidden chains, and the log-likelihood for each iteration.

Author(s)

Mingrui Liang, Matt Koslovsky, Marina Vannucci

Maintainer: Mingrui Liang <[e-mail]>

References

Mingrui Liang, Matthew D. Koslovsky, Emily T. Hebert, Darla E. Kendzor, Michael S. Businelle, Marina Vannucci - Bayesian Variable Selection for Binary Longitudinal Data with Measurement Error: An Application to mHealth Data

Examples

# Generating simulation data

set.seed(125)

df <- sim_data( ind = 150, window = 30, lambda_0 = 0.5, mu_0 = 0.5, 
emisP0 = 0.95, emisP1 = 0.95, pt = 30, pe = 20, bval = c(0.7,1.0,1.5), 
biased = TRUE )

# Sampling for the HMM model

hmmout <- HMMbvs_R( data = df$DATA, tcova = c( "tVar1", "tVar2", "tVar3", 
"tVar4", "tVar5", "tVar6", "tVar7", "tVar8", "tVar9", "tVar10", "tVar11", 
"tVar12", "tVar13", "tVar14", "tVar15" ), ecova = c( "eVar1", "eVar2", 
"eVar3", "eVar4", "eVar5", "eVar6", "eVar7" ), tforce = NULL, eforce = NULL, 
standardize = NULL, model = "HMM", init = "baseline", initvalue = NULL, 
iter = 25000, v1 = 5, v2 = 1, a = 1, b = 9, thin = 10, thin_hidden = 10)
  
# Accessing convergence and goodness-of-fit

convergence( output = hmmout, file = "convergence.pdf")

ypost <- ppc( output = hmmout, file = "ppc_plot.pdf", type = "sum_one", 
burnin = 1000, postsample = NULL)

# Posterior inference of selection and estimation

out <- selection( output = hmmout, burnin = 1000, cred = 0.95, 
threshold = 0.5, trueGamma = NULL, plotting = TRUE, file = "mppi.pdf")

mliang4/HMMbvs documentation built on April 17, 2025, 8:21 p.m.