Skip to contents

Causal-driven AL via Neural Operator disagreement (Pilar 8 x Pilar 5)

Source: R/bridges_active_learning_v3.R
al_query_neural_operator.Rd

Ranks candidate sites by the disagreement between a Neural Operator's predicted depth profile and a classical Pilar 2 pedogenetic ODE's predicted profile, normalised by the NO's perturbation-spread uncertainty.

Usage

al_query_neural_operator(
  no_fit,
  ode_fit,
  pool_covariates,
  pool_depths = NULL,
  n_select = 5L,
  n_pert = 20L,
  pert_sd = 0.1,
  seed = NULL
)

Arguments

no_fit

A fit from no_deeponet_fit() or no_fno_fit().

ode_fit

A fit from piml_profile_fit() or piml_profile_fit_bayesian().

pool_covariates

Matrix of per-site summary covariates (n_pool x p_in) matching the branch input of no_fit.

pool_depths

Optional depths to evaluate at; defaults to the training depths of no_fit.

n_select

Integer; number of candidates to return.

n_pert

Integer; number of Gaussian-noise perturbations per candidate to estimate NO uncertainty.

pert_sd

Numeric; standard deviation of the Gaussian perturbation (in covariate-standard-deviation units).

seed

Optional RNG seed.

Value

A data frame of class edaphos_al_neural_operator_query sorted by descending score.