auto_cmpt_freq: Bayesian Gaussian Process regression with Stan: automatically...

In Tan-Furukawa/MCMCfreq: Bayesian Gaussian Process regression with Stan: computation of frequency by MCMC

Description

WAIC is determined by rho, sigma, data and M. So, we can explicitly show WAIC = WAIC(rho,sigam,data,M). If M and data is fixed, WAIC is the function only having rho and sigma as variables. The algorithm for finding the minimum value of WAIC is as follows. for(rho in 1:rho_max); if(WAIC(rho+1,ex_sigma) is larger than WAIC(rho,ex_sigma)); then return cmpt_freq(data,rho,ex_sigma) The details are ######

Usage

auto_cmpt_freq(data, M = as.integer(50), delta = 1/25, max_rho = 10,
  ex_sigma = 3, stan_seed = as.integer(1234),
  stan_chains = as.integer(3), stan_warmup = as.integer(300),
  stan_thin = as.integer(1), stan_iter = as.integer(1300),
  stan_max_treedepth = as.integer(10))

Arguments

data	a vector of one dimentional data.
M	integer; degree of discretization computation. x axis of frequency plot is divided into M grids. Defaults to M = 50, and larger M give accurate frequency plot, but the computation takes much time.
delta	numeric beetween 0 and 1 (defaults to 1/25); the size of no-data area in frequency plot. In order to stabilize computation, both ends of the graph have no data points. The size is expressed by delta times size of plot area in the x-axis direction.
max_rho	numeric > 1; See Description
ex_sigma	expected value of sigma > 0; See Description
stan_seed	see ?stan in library(rstan)
stan_chains	see ?stan in library(rstan)
stan_warmup	see ?stan in library(rstan)
stan_thin	see ?stan in library(rstan)
stan_iter	see ?stan in library(rstan)
stan_max_treedepth	see ?stan in library(rstan)

Value

list data having follow 5 components:

• $WAIC: See Description in in cmpt_freq

• $p: data.frame of the MCMC result. $age_X is x coodinates and $p_mean is estimate. $p_0025 means that the probability that ture frequency is smaller than $p_0025 is 0.025. $p_025, $p_075 and $p_0975 is defined in the same way as $p_0025. $p_n_eff and $p_Rhat is index to check convergence. In this function, if $p_Rhat > 1.1, then regard the MCMC to converge

• $data: same as the input data, used in freq_graph

• $rho: same as the input rho

• $sigma: same as the input sigma

Examples

d <- Osayama
e <- auto_cmpt_freq(d)
freq_graph(e)

Tan-Furukawa/MCMCfreq documentation built on Feb. 7, 2020, 10:25 a.m.