Skip to contents

Build a physics gate from a PIML profile fit

Source: R/active_physics_gate.R
al_physics_gate_piml.Rd

Convenience helper for constructing the physics_gate argument of al_query() and al_loop() from an existing PIML fit — either the parametric profile (piml_profile_fit()) or the Neural ODE variant (piml_neural_ode_fit()).

Usage

al_physics_gate_piml(
  profile_fit,
  safety_factor = 1.2,
  lower = NULL,
  upper = NULL
)

Arguments

profile_fit

A edaphos_piml_profile or edaphos_piml_neural_ode object.

safety_factor

Numeric (>= 1). 1 = strict envelope; 1.2 = allow 20 % slack on each side.

lower

Optional hard lower bound in the target units (e.g. 0 for mass fractions). If supplied, overrides the lower end of the envelope.

upper

Optional hard upper bound.

Value

A function function(candidates, predicted_mean) suitable for al_query(..., physics_gate = <this>).

Details

The gate rejects any candidate whose model-predicted target value falls outside a physically plausible envelope derived from the PIML fit. For the parametric ODE the envelope is [min(y0, y_inf), max(y0, y_inf)]; for the Neural ODE it is the range of predictions between z = 0 and the deepest training depth (extrapolated to 2 * max(depth) as a conservative floor). Both are widened by safety_factor.