PickStuck: Which chains get stuck

In ggdmc: Cognitive Models

Description

Calculate each chain separately for the mean (across many MCMC iterations) of posterior log-likelihood. If the difference of the means and the median (across chains) of the mean of posterior is greater than the cut, chains are considered stuck. The default value for cut is 10. unstick manually removes stuck chains from posterior samples.

Usage

PickStuck(x, hyper = FALSE, cut = 10, start = 1, end = NA,
  verbose = FALSE, digits = 2)

Arguments

x	posterior samples
hyper	whether x are hierarhcial samples
cut	a criterion deciding if a chain is stuck.
start	start to evaluate from which iteration.
end	end at which iteration for evaeuation.
verbose	a boolean switch to print more information
digits	print how many digits. Default is 2

Value

PickStuck gives an index vector; unstick gives a DMC sample.

Examples

model <- BuildModel(
p.map     = list(A = "1", B = "1", t0 = "1", mean_v = "M", sd_v = "1", st0 = "1"),
match.map = list(M = list(s1 = 1, s2 = 2)),
factors   = list(S = c("s1", "s2")),
constants = c(st0 = 0, sd_v = 1),
responses = c("r1", "r2"),
type      = "norm")
p.vector <- c(A = .75, B = .25, t0 = .2, mean_v.true = 2.5, mean_v.false = 1.5)

p.prior <- BuildPrior(
  dists = c("tnorm", "tnorm", "beta", "tnorm", "tnorm"),
  p1    = c(A = .3, B = .3, t0 = 1, mean_v.true = 1, mean_v.false = 0),
  p2    = c(1, 1,   1, 3, 3),
  lower = c(0,  0,  0, NA, NA),
  upper = c(NA,NA,  1, NA, NA))

## Not run: 
dat <- simulate(model, 30, ps = p.vector)
dmi <- BuildDMI(dat, model)
sam <- run(StartNewsamples(dmi, p.prior))
bad <- PickStuck(sam)

## End(Not run)

ggdmc documentation built on May 2, 2019, 9:59 a.m.