Batch prediction from a frozen observation-to-status GP

Uses the same mean and predictive-variance equations as the annual EM. intervals=FALSE avoids the quadratic covariance calculation during offline screening. It never changes the predictive median. No refitting occurs.

Usage

predict_em_gp(model, newdata, intervals = TRUE, chunk_size = 1000L)

Arguments

model

A fitted GP as returned by fit_em_gp().

newdata

Data frame of already log-transformed single-year inputs.

intervals

Compute predictive intervals including the nugget.

chunk_size

Maximum number of prediction rows per matrix operation.

Value

Data frame with status, lower_95, upper_95, sd_log, and marginal outside_training_range flags. Interval columns are NA if not requested.