Robustness of classification under input perturbation — classification

For a given PedonRecord, perturb a chosen list of horizon attributes by a configured fractional amount, re-classify under the requested system, and report how often the classification $rsg_or_order (or full $name) matches the unperturbed baseline.

Usage

classification_robustness(
  pedon,
  system = c("wrb2022", "sibcs", "usda"),
  level = c("order", "name"),
  n = 50L,
  perturbations = NULL,
  seed = 42L
)

Arguments

pedon: A PedonRecord.
system: One of "wrb2022", "sibcs", "usda".
level: Either "order" (compare $rsg_or_order) or "name" (compare full classification name).
n: Number of Monte-Carlo perturbed runs (default 50).
perturbations: Named list. Each name is a horizon column; each element is a function taking the original value and returning a perturbed value. NA-tolerant.
seed: Random seed for reproducibility.

Value

A list with elements baseline (the unperturbed classification name), n (number of MC runs), robustness (fraction of perturbed runs matching baseline), flipped_to (table of alternative classifications when the perturbation flipped the result).

Details

Default perturbation panel:

clay_pct: ±5
sand_pct: ±5
silt_pct: ±5
ph_h2o: ±0.2 absolute
oc_pct: ±10

Examples

if (FALSE) { # \dontrun{
p <- make_ferralsol_canonical()
classification_robustness(p, system = "wrb2022", n = 50)
#> $baseline    : "Ferralsols"
#> $robustness  : 0.96  (48 / 50 perturbed runs landed on Ferralsols)
#> $flipped_to  : table(c("Cambisols" = 1, "Acrisols" = 1))
} # }