plot.MTDest: Plot method for MTDest objects

Description

Produces plots for an MTDest object. By default, it shows in sequence: (i) barplot of oscillations by relevant lag, (ii) barplot of mixture weights $\lambda_j$ (including lam0 if > 0), and (iii) graphs of pj (one graph for each lag in Lambda). If EM diagnostics are available, a convergence panel (log-likelihood variation per update) is shown last. When type is specified, only the requested plot is drawn.

Usage

# S3 method for MTDest
plot(x, type, main, ylim, col = "gray70", border = NA, pj_index = 1, ...)

Value

If type is "oscillation" or "lambdas", invisibly returns the numeric vector that was plotted. If type = "pj", invisibly returns the selected transition matrix pj(x)[[pj_index]]. If type = "convergence", invisibly returns the numeric vector deltaLogLik. If type is missing, invisibly returns the list produced by plot.MTD(m) (typically with components oscillation

and lambdas); if deltaLogLik is available, the returned list also includes deltaLogLik.

Arguments

x: An object of class "MTDest".
type: If type is missing, oscillation, lambdas, pj graphs and—if available—the convergence plots are shown sequentially (press Enter to proceed). Else, type is a character string indicating what to plot: "oscillation", "lambdas", "pj" or "convergence".
main: Optional main title. When type is missing, panel-specific defaults are used and main is ignored.
ylim: Optional y-axis limits. When type is missing, panel-specific defaults are used and ylim is ignored.
col: Bar fill color (passed to barplot). Defaults to "gray70".
border: Bar border color (for bar plots). Defaults to NA.
pj_index: Integer index of pj(x) to plot when type = "pj" (default 1).
...: Further graphical parameters: passed to barplot() when type %in% c("oscillation","lambdas"), to plot.igraph() when type = "pj", and to plot() when type = "convergence". Ignored when type is missing.

Details

For type = "oscillation", the function delegates to plot.MTD(), which returns $\delta_j = \lambda_j \max_{b,c} d_{TV}(p_j(\cdot|b), p_j(\cdot|c))$ for each lag in lags(x), and draws a bar plot named by the lags.

For type = "lambdas", the function delegates to plot.MTD(), it plots the mixture weights $\lambda_j$ by lag. If lam0 > 0, the weight for the independent component is included and labeled "0".

For type = "pj", the function delegates to plot.MTD(), it draws the directed, weighted graph of the transition matrices pj taken from pj(x). Vertices correspond to the states states(x). A directed edge a -> b carries weight p_j(b | a). Edge widths and edge labels are proportional to the transition probabilities (labels shown in the plot are rounded to two decimals). By default, self-loops (a -> a) are not drawn. The self-loop probability at a state a can be inferred as $$1 - \sum_{b: b \ne a} p_j(b \mid a).$$ For type = "pj", a specific matrix can be selected via pj_index (e.g., pj_index = 2 plots pj(x)[[2]]). In automatic mode (when type is missing), graphs for all pj matrices are shown sequentially.

If EM iteration diagnostics are available (i.e., the object was fitted with iter = TRUE and length(deltaLogLik) > 0), a convergence panel showing the log-likelihood variation per update is displayed automatically when type is missing. You can also request it explicitly with type = "convergence".

Examples

Run this code

if (FALSE) {
set.seed(1)
M <- MTDmodel(Lambda = c(1, 3), A = c(0, 1), lam0 = 0.05)
X <- perfectSample(M, N = 300)
fit <- MTDest(X, S = c(1, 3), init = coef(M), iter = TRUE)

plot(fit)
plot(fit, type = "pj", pj_index = 2)
plot(fit, type = "convergence")
}