predict variable (BDF-SSL framework, legacy version) — predict

Selection of best model for agiven variable based on evaluation metric from a set of models and prediction for target spectra

Usage

predict_variable(
  target_data,
  model_folder,
  eval_model_folder = model_folder,
  prefix = "cubist_",
  variable_ = "CORG",
  metric = "test.rmse",
  maximise = F,
  restrict = T,
  diss_limit = 2.5,
  manual = NA
)

Arguments

prefix: Model names prefix, usually defines model type
metric: evaluation statistic which is used to select best model
maximise: Should the model with the highest metric be selected? E.g., for R2 set to True. Default is False.
taget_data: Nested tibble with spc data at different pre-processing steps
variable: Variable for which the best model is to be searched
restrict_spc: If True only regard models for which spc preprocessing sets are available. If False, search in all models for best candidate and return ERROR when spc set for best candidate is not available