pbtf: Proximal Bayesian Trend Filtering

In qhengncsu/BNRPxMCMC: Bayesian l1 Trend Filtering via Proximal MCMC

Description

Proximal Bayesian Trend Filtering

Usage

pbtf(
  y,
  x = NULL,
  restrictions = NULL,
  lambda = NULL,
  k = 2,
  delta = 1,
  N.burn = 5000,
  N.sim = 10000,
  t = 5,
  MH_correct = F,
  alpha = NULL,
  alpha_grid = NULL,
  K = 10,
  s = 0.001,
  r = 0.001
)

Arguments

y	normal observations (vector)
x	predictor (vector); if null, will use 1:m by default
restrictions	a collection of shape restrictions, including 'increasing','decreasing','convex','concave'
lambda	lambda parameter of Moreau envelope
k	order of the desired piecewise polynomial; 1 linear, 2 quadratic, 3 cubic; k=0 or k>=4 not recommended
delta	percentage of maximum step size to use, default is 1
N.burn	number of MCMC iterations burned
N.sim	number of MCMC samples returned
t	number of theta sampling steps to carry out before sigma2, default is 5
MH_correct	whether to use MH correction or not
alpha	regularization parameter; if NULL, it is determined by cross validation heuristic
alpha_grid	a grid of alpha candidates to select from; we recommend including 0
K	conduct K-fold-cross validation to find the optimal alpha
s	the shape parameter for inverse gamma prior of sigma2
r	the rate parameter for inverse gamma prior of sigma2

Examples

x=seq(0,4*pi,len=100)
y=x+sin(x)+rnorm(100)
out_pbsrtf <- pbtf(y,x,'increasing',k=2,alpha_grid=seq(0,10,len=1001))
plot(out_pbsrtf)

qhengncsu/BNRPxMCMC documentation built on Dec. 31, 2020, 2:10 a.m.