Predictive posterior from a Neural-ODE deep ensemble — predict.edaphos_piml_neural_ode

Evaluates every ensemble member at the requested depths and returns either the raw K × length(newdepths) matrix of member-wise predictions or a tidy (mean, sd, lower, upper) summary corresponding to a symmetric central credible interval computed directly from the empirical ensemble distribution.

Usage

# S3 method for class 'edaphos_piml_neural_ode_ensemble'
predict(
  object,
  newdepths,
  interval = NULL,
  include_obs_noise = FALSE,
  seed = NULL,
  ...
)

Arguments

object: An edaphos_piml_neural_ode_ensemble.
newdepths: Numeric vector of depths.
interval: Optional numeric in (0, 1) — when supplied, returns a summary data frame with a symmetric central credible interval at the requested level. When NULL (default) returns the raw K × length(newdepths) matrix.
include_obs_noise: Logical — when TRUE, adds residual Gaussian noise (with SD equal to the pooled member-wise training RMSE) to every predictive draw so the interval represents the predictive distribution of a future observation rather than the uncertainty on the mean function alone. Default FALSE.
seed: Optional integer — RNG seed for the observation-noise draw. Only consulted when include_obs_noise = TRUE.
...: Unused.

Value

Either a numeric matrix (when interval is NULL) or a data frame with columns depth, mean, sd, lower, upper.