Skip to contents

Compute applicability-domain scores for new samples

Source: R/predict.R
predict_applicability.Rd

Uses the squared Mahalanobis distance in the soilVAE latent space, compared against the chi-squared threshold at df = 16, alpha = 0.05 (thr95 ~26.3). Latent statistics (mu, Sigma) are read from the per-property metrics JSON saved during training.

Usage

predict_applicability(
  df,
  family_id,
  prop,
  model_dir = getOption("autoSpectra.model_dir", "models")
)

Arguments

df

Data frame with Soil_ID and spectral columns

family_id

Family identifier

prop

Soil property name (determines which model's latent space is used)

model_dir

Root model directory

Value

Data frame with columns Soil_ID, mahal_dist, thr95, in_domain