Predict from a Fitted DBN Model

Generates predictions from a fitted DBN model. With no arguments (or PPD-only arguments), returns posterior predictive replications of the observed data for model checking. With forecasting arguments (H, S, summary), propagates the estimated dynamics forward to produce multi-step-ahead forecasts.

Usage

# S3 method for class 'dbn'
predict(object, ...)

Arguments

object

A fitted dbn object returned by dbn().

...

Model-specific arguments:

H

Forecast horizon (number of steps ahead). Triggers forecasting mode.

S, draws

Number of posterior draws to use for forecasting.

summary

If "mean", return the pointwise mean across draws. If NULL (default), return the full array of simulated trajectories.

ndraws

Number of draws for posterior predictive checks (default: 100).

seed

Random seed for reproducibility.

Value

For forecasting (H specified): an array of dimensions [n_row, n_col, p, H] (if summary = "mean") or [n_row, n_col, p, H, draws]. For posterior predictive checks: a list of class "dbn_ppd" containing replicated datasets.

Details

Posterior predictive distribution (default): Simulates new datasets from the fitted model to compare against the observed data. Use plot_ppc_ecdf() or plot_ppc_density() for visual checks.

Forecasting: Pass H (horizon), S or draws (number of simulations), and optionally summary = "mean" to average across draws. The forecast propagates \(A_t\), \(B_t\), and process noise forward from the final observed time point.

See also

dbn(), posterior_predict_dbn(), plot_ppc_ecdf(), plot_ppc_density()

Examples

# \donttest{
sim <- simulate_dynamic_dbn(n = 6, time = 10, seed = 6886)
fit <- dbn(sim$Z, model = "dynamic", family = "gaussian",
           nscan = 200, burn = 100, verbose = FALSE)

# forecasting: 5 steps ahead
fc <- predict(fit, H = 5, S = 50, summary = "mean")
dim(fc)
#> [1] 6 6 2 5

# posterior predictive check
ppd <- predict(fit, ndraws = 10)
# }