GPDPQuantReg: Fits GPDPQuantReg

In opardo/GPDPQuantReg: Bayesian and Nonparametric Quantile Regression Modelling

Description

Runs a MCMC algorithm to fit a quantile regression model, using Gaussian and Dirichlet Processes. Returns a GPDP_MCMC object with thinned chains. Scales data in the process so the induced correlation makes sense, and unscales at the end. Take it into account for the initial parameters.

Usage

GPDPQuantReg(
  formula, data, p = 0.5, c_DP = 2, d_DP = 1, c_lambda = 2,
  d_lambda = 0.5, alpha = sqrt(nrow(data)), M = zero_function,
  mcit = 3e4, burn = 1e4, thin = 10
)

Arguments

formula	formula object with the dependent and independent variables
data	data frame
p	real between (0,1), probability corresponding to the estimated quantile
c_DP	real > 0, shape parameter for the DP's base distribution
d_DP	real > 0, scale parameter fot the DP's base distribution
c_lambda	real > 0, shape parameter for the GP's lambda distribution
d_lambda	real > 0, scale parameter fot the GP's lambda distribution
alpha	real > 0, DP's concentration parameter
M	function, a priori estimation for the final function
mcit	integer > 0, number of MCMC algorithm's valid chains
burn	integer > 0, number of MCMC algorithm's first burned chains
thin	integer > 0 and < mcit, MCMC algorithm's thinning

Value

GPDP_MCMC object, with the MCMC algorithm's chains

Author(s)

Omar Pardo ([e-mail])

Examples

m <- 35
x <- sort(sample(seq(-15, 15, 0.005), m))
f_x <- function(x) return((1/40) * x^2 - (1/20) * x - 2)
data <- data.frame(x = x, y = f_x(x) + rnorm(m, 0, 1))
GPDP_MCMC <- fit_GPDPQuantReg(y ~ x, data, p = 0.250)

opardo/GPDPQuantReg documentation built on Nov. 24, 2019, 3:28 a.m.