plot.IBIS {UNCOVER}

Plot various outputs of IBIS

Description

Allows visualisation of many aspects of IBIS, including co-variate, posterior and diagnostic plots.

Usage

## S3 method for class 'IBIS'
plot(x, type = "samples", plot_var = NULL, diagnostic_x_axis = "full", ...)

Arguments

x

Object of class "IBIS"

type

Can be one of; "samples" for posterior visualisation, "fitted" for co-variate visualisation or "diagnostics" for diagnostic plots. See details. Defaults to "samples".

plot_var

Vector specifying which columns (or associated logistic regression coefficients) of the co-variate matrix should be plotted. Does not apply when type=="diagnostics". Defaults to all columns being selected.

diagnostic_x_axis

Only applies if ⁠"type=="diagnostics"⁠. Either "full" (default) for all observations indices to be plotted on the x-axis or "minimal" for only some observations indices to be plotted on the x-axis.

...

Arguments to be passed to methods

Details

If type=="samples", the resulting plot will be a ggpairs plot giving the coefficient densities in the diagonal, points plots of the posterior samples in the lower triangle and contour plots in the upper triangle.

If ⁠"type=="fitted"⁠, the resulting plot will be a ggpairs plot. The diagonal entries will be two density plots, one for training data predicted to have response 0 by the model (red) and one for training data predicted to have response 1 by the model (green). The off-diagonal elements are scatter-plots of the observations, given a label according to their actual response and a colour scale based on their predicted response. If length(plot_var)==1 then the co-variate variable is plotted against it's index and a density plot is not provided. If length(plot_var)==1 then the density plot and the scatter-plot are combined. If a predicted class (0 or

1. contains less than two data points the density will not be plotted.

If "type==diagnostics", the resulting plot will be a combination of three plots; one tracking the log Bayesian evidence as observations are added to the posterior, one tracking the effective sample size of the particles for each step of the SMC sampler and one tracking the acceptance rate of the Metropolis-Hastings step when a resample-move is triggered. See Emerson and Aslett (2023) and Chopin (2002) for more details. Multiple Metropolis-Hastings steps can be performed when a resample-move step is triggered, and so for the acceptance rate plot observations are suffixed with "." and then the index of current Metropolis-Hastings step. For example the x-axis label for the acceptance rate of the 2nd Metropolis-Hastings step which was triggered by adding observation 1 to the posterior would be labelled "1.2". When the training data for the "IBIS" object created is large setting diagnostic_x_axis=="minimal" is recommended as it gives a more visually appealing output.

Value

No return value, called for side effects

References

• Emerson, S.R. and Aslett, L.J.M. (2023). Joint cohort and prediction modelling through graphical structure analysis (to be released)

• Chopin, N. (2002). A sequential particle filter method for static models. Biometrika, 89(3), 539-552, doi:10.1093/biomet/89.3.539

See Also

IBIS.logreg()

Examples

require(graphics)
# First we generate a co-variate matrix X and binary response vector y
CM <- matrix(rnorm(200),100,2)
rv <- sample(0:1,100,replace=TRUE)

# Now we can obtain 1000 samples from the posterior from a standard
# multivariate normal prior and plot the results
out <- IBIS.logreg(X = CM,y = rv)
plot(out,type = "samples")
plot(out,type = "fitted")
plot(out,type = "diagnostics",diagnostic_x_axis = "minimal")

# If we only wanted to view the second co-variate
plot(out,type = "samples",plot_var = 2)
plot(out,type = "fitted",plot_var = 2)