StartNewsamples: Start new model fits

Description

Fit a hierarchical or a fixed-effect model, using Bayeisan optimisation. We use a specific type of pMCMC algorithm, the DE-MCMC. This particular sampling method includes crossover and two different migration operators. The migration operators are similar to random-walk algorithm. They wouold be less efficient to find the target parameter space, if been used alone.

Usage

StartNewsamples(
  data,
  prior = NULL,
  nmc = 200,
  thin = 1,
  nchain = NULL,
  report = 100,
  rp = 0.001,
  gammamult = 2.38,
  pm_Hu = 0.05,
  pm_BT = 0.05,
  block = TRUE,
  ncore = 1,
  add = FALSE,
  is_old = FALSE
)
run(
  samples,
  nmc = 500,
  thin = 1,
  report = 100,
  rp = 0.001,
  gammamult = 2.38,
  pm_Hu = 0,
  pm_BT = 0,
  block = TRUE,
  ncore = 1,
  add = FALSE,
  is_old = TRUE
)

Arguments

data: data model instance(s)
prior: prior objects. For hierarchical model, this must be a list with three sets of prior distributions. Each is respectively named, "pprior", "location", and "scale".
nmc: number of Monte Carlo samples
thin: thinning length
nchain: number of chains
report: progress report interval
rp: tuning parameter 1
gammamult: tuning parameter 2. This is the step size.
pm_Hu: probability of migration type 0 (Hu & Tsui, 2010)
pm_BT: probability of migration type 1 (Turner et al., 2013)
block: Only for hierarchical modeling. A Boolean switch for update one parameter at a time
ncore: Only for non-hierarchical, fixed-effect models with many subjects.
add: Boolean whether to add new samples
is_old: start sampling from a DMI or fit samples
samples: posterior samples.