plot.nmix: Plotting for Bayesian Inference on Univariate Normal Mixtures

In Nmix: Bayesian Inference on Univariate Normal Mixtures

Plotting for Bayesian Inference on Univariate Normal Mixtures

Description

Plotting of various information from nmix object on current graphics device

Usage

## S3 method for class 'nmix'
plot(x, which=1:5, separate=FALSE, plugin=FALSE, offset=1, nsamp=50, 
	equi=TRUE, allsort=TRUE, trued=NULL, ...) 

Arguments

x	nmix object, as output by Nmix function
which	integer vector, specifying which of several available plots are required, see 'Details' below
separate	logical, if TRUE opens a fresh default device for each new plot, otherwise prompts before overwriting a previous plot
plugin	logical, should plug-in estimator of density, computed from posterior means of parameters, be superimposed on density plot in darkgreen, in the case which contains 1?
offset	t numeric, vertical displacement between plotted traces, in the case which contains 2.
nsamp	integer, number of posterior samples plotted, in the case which contains 3.
equi	logical, should thinning of posterior density samples be equi-spaced, rather than random, in the case which contains 3
allsort	logical, should observations be sorted before making posterior clusters plot, in the case which contains 4
trued	vectorised function defining a probability density function to be superimposed in blue on density plots, in the cases which contains 1 and/or 3
...	additional arguments to plot

Details

If which includes 1, a 2-panel plot of which: the first is a histogram of the data, and if z has a component den (Nmix output option D), superimposed plots of the posterior density estimates, conditional on k=1,2,...,6 and unconditionally (in red); and the second a barplot of the estimated posterior distribution of k.

If which includes 2, a multiple trace plot of various statistics for a thinned subsample of the MCMC run, after burn-in. The statistics are the (named) components of the list z$traces that are numerical vectors, some or all of (i) the number of components k (Nmix output option k), (ii) the entropy (Nmix output option e), and (iii) the deviance (Nmix output option d), of the current sample. The traces may be of different lengths, the horizontal scales in the plot are adjusted to span the length of the (post burn-in) MCMC run, regardless of these differences.

If which includes 3 (and Nmix output option p is present), a thinned sample of size nsamp from the posterior distribution of the density function, computed from a thinned sample of (weight, mean, sd) values generated in the posterior simulation.

If which includes 4 (and Nmix output option a is present), an image plot showing the posterior probabilities that the corresponding observations are in the same mixture component. Not recommended for large n.

If which includes 5 (and Nmix output option C is present), a 4-panel plot displaying Bayersian classifications based on the fitted model, analogous to Fig. 10 in the Richardson and Green paper. The 4 panels corresponding to conditioning on the 4 values of k most probable according to the posterior (among those for which the necessary posterior sample information has been recorded (see argument k1k2 of Nmix), and excepting k=1).

Value

NULL, invisibly; plot method for class nmix. Function called for its side effect of plotting selected information about the fitted posterior distribution and sampler performance from x on the current graphics device

Author(s)

Peter J. Green

References

Richardson and Green (J. R. Statist. Soc. B, 1997, 59, 731-792; see also the correction in J. R. Statist. Soc. B, 1998, 60, 661)

Examples

data(galx)
z<-Nmix('galx',nsweep=10000,nburnin=1000,out="d")
plot(z,1:2)

Nmix documentation built on Sept. 20, 2023, 1:06 a.m.