Benchmark soilKey SiBCS predictions against the Redape gold standard

Runs classify_sibcs on each pedon and compares against the curator-validated reference label (Order / Suborder / Great Group / Subgroup). Returns per-level accuracy and the confusion matrix at the requested granularity.

Usage

benchmark_redape(
  pedons,
  level = c("order", "subordem", "gde_grupo", "subgrupo"),
  verbose = TRUE
)

Arguments

pedons

List of PedonRecord objects (typically from load_redape_pedons).

level

One of "order" (default), "subordem", "gde_grupo", or "subgrupo".

verbose

Print progress (default TRUE).

Value

A list with accuracy, n_compared, confusion, per_class_recall, and the per-pedon predictions table. predictions now also includes columns ref_norm and pred_norm – the canonical comparison keys – for downstream auditing.

v0.9.81 level-aware comparison

Earlier versions accepted the level argument but always used rsg_or_order for the prediction and the order field for the reference, so all four levels reported identical accuracy. v0.9.81 reads the level-specific slots from res$trace (subordem, grande_grupo, subgrupo) and concatenates the matching reference fields, applying SiBCS-aware Portuguese pluralisation so the comparison key matches the predictor's plural Title Case form.