diagPlot {mcmcOutput}

Diagnostic graphics for class mcmcOutput

Description

Display trace plots and density plots for the chains in the MCMC output. Each chain is plotted with a different colour.

Usage

diagPlot(object, params, howMany, chains,
  maxRows=4, RhatBad=1.05, precision=c("MCEpc", "n.eff"), ask=NULL, ...)

tracePlot(object, layout=c(3,3), ask=NULL, ...)
densityPlot(object, layout=c(3,3), ask=NULL, ...)
acfPlot(object, lag.max=NULL, layout=c(3,3), ask=NULL, ...)

Arguments

object

An object of any class with MCMC output that can be coerced to class mcmcOutput.

params

An optional vector of column numbers or names; names are partially matched, so params="alpha" will match all of alpha, alpha0, alpha[1], alphanew; negative indices indicate parameters to exclude, positive and negative indices cannot be mixed.

howMany

How many draws per chain to plot; if negative, the draws at the end of the chains will be plotted; default is to plot all.

chains

Which chains to plot, a numeric vector; default is to plot all.

maxRows

Maximum number of rows to display in one window; each row consists of a trace plot and a density plot for one parameter.

RhatBad

Threshold for Rhat; parameters with Rhat > RhatBad are highlighted in red.

precision

The statistic to use for the precision, displayed above the density plot.

layout

a length-2 vector with the maximum number of rows and columns to display in the plotting frame.

lag.max

Maximum lag at which to calculate the acf.

ask

If TRUE and the number of parameters to plot is greater than maxRows, the user will be prompted before the next page of output is displayed. The default is to ask if the plotting device is the screen, not if it is a file.

...

Additional graphical parameters.

Value

Return nothing, used for their plotting side effects.

Author(s)

Mike Meredith

See Also

crosscorrPlot, postPlot for a histogram and summary statistics.

Examples

# Create a fake mcmcOutput object:
tmp <- cbind(
  mu0 = rnorm(3000),         # normal, mean zero
  mu10 = rnorm(3000, rep(9:11, each=1000), 1),
                             # normal, mean 10, but poor mixing
  sigma=rlnorm(3000),        # non-negative, skewed
  `prob[1,1]` = rbeta(3000, 4, 4),  # probability, central mode
  `prob[1,2]` = 0.3,         # constant
  `prob[2,1]` = rbeta(3000, 1, 3),  # probability, mode = 0
  N = rpois(3000, rep(c(24, 18, 18), each=1000)),
                             # large integers (no zeros), poor mixing
  n = rpois(3000, 2),        # small integers (some zeros)
  allNA = NA,                # all values NA
  someNA = suppressWarnings(log(rnorm(3000, 2, 2))),
                             # some NaNs
  const1 = rep(1, 3000),     # all values = 1
  const3.2 = rep(10/3, 3000))# all values the same but not integer
( fake <- mcmcOutput(tmp, nChains = 3) )
summary(fake)
diagPlot(fake)
diagPlot(fake, params=3:6, main="params = 3:6")
diagPlot(fake, params=c("mu", "prob"), main="params = c('mu', 'prob')")
diagPlot(fake, params=c("mu", "prob"), howMany=200, main="howMany = 200")
diagPlot(fake, params=c("mu", "prob"), howMany=50, main="howMany = 50")
diagPlot(fake, params=c("mu", "prob"), howMany=-200, main="howMany = -200")
diagPlot(fake, params=c("mu", "prob"), chains=1:2, main="chains = 1:2")
diagPlot(fake, params=c("mu", "prob"), chains=2, main="chains = 2") # 1 chain -> no Rhat

tracePlot(fake, layout=c(2,2))
densityPlot(fake, xlab="value")
acfPlot(fake, lag.max=10, lwd=2)

# Use diagPlot with an mcmc.list object
data(mcmcListExample)
diagPlot(mcmcListExample)
diagPlot(mcmcListExample, main="example", params=1:3, precision="n.eff")