plot.Results_IMIFA {IMIFA}

Plotting output and parameters of inferential interest for IMIFA and related models

Description

Plotting output and parameters of inferential interest for IMIFA and related models

Usage

## S3 method for class 'Results_IMIFA'
plot(x,
     plot.meth = c("all", "correlation", "density", "errors", "GQ",
                   "means", "parallel.coords", "trace", "zlabels"),
     param = c("means", "scores", "loadings", "uniquenesses",
               "pis", "alpha", "discount"),
     g = NULL,
     mat = TRUE,
     zlabels = NULL,
     heat.map = TRUE,
     show.last = FALSE,
     palette = NULL,
     ind = NULL,
     fac = NULL,
     by.fac = FALSE,
     type = c("h", "n", "p", "l"),
     intervals = TRUE,
     common = TRUE,
     partial = FALSE,
     titles = TRUE,
     transparency = 0.75,
     ...)

Arguments

x

An object of class "Results_IMIFA" generated by get_IMIFA_results.

plot.meth

The type of plot to be produced for the param of interest, where correlation refers to ACF/PACF plots, means refers to posterior means, density, trace and parallel.coords are self-explanatory. "all" in this case, the default, refers to "trace", "density", "means", and "correlation". "parallel.coords" is only available when param is one of "means", "loadings", or "uniquenesses" - note that this method applies a small amount of horizontal jitter to avoid overplotting.

Special types of plots which don't require a param are:

"GQ"

for plotting the posterior summaries of the numbers of clusters/factors, if available.

"zlabels"

for plotting clustering uncertainties - in four different ways (incl. the posterior confusion matrix) - if clustering has taken place, with or without the clustering labels being supplied via the zlabels argument. If available, the average similarity matrix, reordered according to the MAP labels, is shown as a 5-th plot.

"errors"

for conducting posterior predictive checking of the appropriateness of the fitted model by visualising the posterior predictive reconstruction error (PPRE) &/or histograms comparing the data to replicate draws from the posterior distribution &/or error metrics quantifying the difference between the estimated and empirical covariance matrices. The type of plot(s) produced depends on how the error.metrics argument was supplied to get_IMIFA_results and what parameters were stored.

The argument g can be used to cycle through the available plots in each case. ind can also be used to govern which variable is shown for the 2-nd plot.

param

The parameter of interest for any of the following plot.meth options: all, trace, density, means, correlation. The param must have been stored when mcmc_IMIFA was initially ran. Includes pis for methods where clustering takes place, and allows posterior inference on alpha (for the "IMFA", "IMIFA", "OMFA", and "OMIFA" methods) and discount (for the "IMFA" and "IMIFA" methods). Otherwise "means", "scores", "loadings", and "uniquenesses" can be plotted.

g

Optional argument that allows specification of exactly which cluster the plot of interest is to be produced for. If not supplied, the user will be prompted to cycle through plots for all clusters. Also functions as an index for which plot to return when plot.meth is GQ, zlabels, or errors in much the same way.

mat

Logical indicating whether a matplot is produced (defaults to TRUE). If given as FALSE, ind is invoked.

zlabels

The true labels can be supplied if they are known. If this is not supplied, the function uses the labels that were supplied, if any, to get_IMIFA_results. Only relevant when plot.meth = "zlabels". When explicitly supplied, misclassified observations are highlighted in the first type of uncertainty plot (otherwise observations whose uncertainty exceed the inverse of the number of clusters are highlighted). For the second type of uncertainty plot, when zlabels are explicitly supplied, the uncertainty of misclassified observations is marked by vertical lines on the profile plot.

heat.map

A logical which controls plotting posterior mean loadings or posterior mean scores as a heatmap, or else as something akin to link{plot(..., type="h")}. Only relevant if param = "loadings" (in which case the default is TRUE) or param = "scores" (in which case the default is FALSE). Heatmaps are produced with the aid of mat2cols and plot_cols.

show.last

A logical indicator which defaults to FALSE, but when TRUE replaces any instance of the posterior mean with the last valid sample. Only relevant when param is one of "means" "scores", "loadings", "uniquenesses", or "pis" and plot.meth is one of "all" or "means". Also relevant for "means", "loadings" and "uniquenesses" when plot.meth is "parallel.coords". When TRUE, this has the effect of forcing intervals to be FALSE.

palette

An optional colour palette to be supplied if overwriting the default palette set inside the function by viridis is desired. It makes little sense to a supply a palette when plot.meth="all" and param is one of "scores" or "loadings".

ind

Either a single number indicating which variable to plot when param is one of means or uniquenesses (or plot.meth="errors"), or which cluster to plot if param is pis. If scores are plotted, a vector of length two giving which observation and factor to plot; if loadings are plotted, a vector of length two giving which variable and factor to plot. Will be recycled to length 2 if necessary. Also governs which two factors are displayed on posterior mean plots of the "scores" when heat.map is FALSE; otherwise only relevant when mat is FALSE.

fac

Optional argument that provides an alternative way to specify ind[2] when mat is FALSE and param is one of scores or loadings.

by.fac

Optionally allows (mat)plotting of scores and loadings by factor - i.e. observation(s) (scores) or variable(s) (loadings) for a given factor, respectively, controlled by ind or fac) when set to TRUE. Otherwise all factor(s) are plotted for a given observation or variable when set to FALSE (the default), again controlled by ind or fac. Only relevant when param is one of scores or loadings.

type

The manner in which the plot is to be drawn, as per the type argument to plot.

intervals

Logical indicating whether credible intervals around the posterior mean(s) are to be plotted when is.element(plot.meth, c("all", "means")). Defaults to TRUE, but can only be TRUE when show.last is FALSE.

common

Logical indicating whether plots with plot.meth="means" (or the corresponding plots for plot.meth="all") when param is one of "means", "scores", "loadings", or "uniquenesses" are calibrated to a common scale based on the range of the param parameters across all clusters (defaults to TRUE, and only relevant when there are clusters). Otherwise, the only the range corresponding to the image being plotted is used to determine the scale.

Note that this affects the "loadings" and "scores" plots regardless of the value of heat.map. An exception is the "scores" plots when plot.meth="means" and heat.map is FALSE, in which case common defaults to FALSE.

partial

Logical indicating whether plots of type "correlation" use the PACF. The default, FALSE, ensures the ACF is used. Only relevant when plot.meth = "all", otherwise both plots are produced when plot.meth = "correlation".

titles

Logical indicating whether default plot titles are to be used (TRUE), or suppressed (FALSE).

transparency

A factor in [0, 1] modifying the opacity for overplotted lines. Defaults to 0.75, unless semi-transparency is not supported. Only relevant when palette is not supplied, otherwise the supplied palette must already be adjusted for transparency.

...

Other arguments typically passed to plot or the breaks argument to mat2cols and heat_legend when heatmaps are plotted.

Value

The desired plot with appropriate output and summary statistics printed to the console screen.

Note

Supplying the argument zlabels does not have the same effect of reordering the sampled parameters as it does if supplied directly to get_IMIFA_results.

When mat is TRUE and by.fac is FALSE (both defaults), the convention for dealing with overplotting for trace and density plots when param is either scores or loadings is to plot the last factor first, such that the first factor appears 'on top'.

Author(s)

Keefe Murphy - <[e-mail]>

References

Murphy, K., Viroli, C., and Gormley, I. C. (2020) Infinite mixtures of infinite factor analysers, Bayesian Analysis, 15(3): 937-963. <doi:10.1214/19-BA1179>.

See Also

mcmc_IMIFA, get_IMIFA_results, mat2cols, plot_cols

Examples

# See the vignette associated with the package for more graphical examples:
# vignette("IMIFA", package = "IMIFA")

# data(olive)
# simIMIFA <- mcmc_IMIFA(olive, method="IMIFA")
# resIMIFA <- get_IMIFA_results(simIMIFA, z.avgsim=TRUE)

# Examine the posterior distribution(s) of the number(s) of clusters (G) &/or latent factors (Q)
# For the IM(I)FA and OM(I)FA methods, this also plots the trace of the active/non-empty clusters
# plot(resIMIFA, plot.meth="GQ")
# plot(resIMIFA, plot.meth="GQ", g=2)

# Plot clustering uncertainty (and, if available, the similarity matrix)
# plot(resIMIFA, plot.meth="zlabels", zlabels=olive$area)

# Visualise the posterior predictive reconstruction error
# plot(resIMIFA, plot.meth="errors", g=1)

# Compare histograms of the data vs. replicate draw from the posterior for the 1st variable
# plot(resIMIFA, plot.meth="errors", g=2, ind=1)

# Visualise empirical vs. estimated covariance error metrics
# plot(resIMIFA, plot.meth="errors", g=3)

# Look at the trace, density, posterior mean, and correlation of various parameters of interest
# plot(resIMIFA, plot.meth="all", param="means", g=1)
# plot(resIMIFA, plot.meth="all", param="means", g=1, ind=2)
# plot(resIMIFA, plot.meth="trace", param="scores")
# plot(resIMIFA, plot.meth="trace", param="scores", by.fac=TRUE)
# plot(resIMIFA, plot.meth="mean", param="loadings", g=1)
# plot(resIMIFA, plot.meth="mean", param="loadings", g=1, heat.map=FALSE)
# plot(resIMIFA, plot.meth="parallel.coords", param="uniquenesses")
# plot(resIMIFA, plot.meth="density", param="pis", intervals=FALSE, partial=TRUE)
# plot(resIMIFA, plot.meth="all", param="alpha")
# plot(resIMIFA, plot.meth="all", param="discount")