Fits the universal-kriging model used by em_gp(). Inputs are already
log-transformed single-year observations; the response is log relative
biomass. Training and hyperparameter selection must occur outside run_mse().
Rows from the same underlying OM iteration must stay together when creating
training and validation sets. This function does not select a design or tune
hyperparameters, and never accesses an operating model.
Usage
fit_em_gp(
data,
log_status,
features = c("log_cpue_aw", "log_cpue_ss", "log_catch_total"),
corr_scale = 0.5,
nugget = 0.2,
scaler = NULL,
target = NULL,
catch_component = NULL
)Arguments
- data
Data frame containing the named, already transformed features.
- log_status
Finite numeric log-status response, one per design row.
- features
Names of the supported single-year CPUE/catch features.
- corr_scale
Positive common inverse squared length-scale coefficient.
- nugget
Non-negative covariance nugget relative to GP variance.
- scaler
Optional list with named centre, spread, and features. Estimate it from training data only; omitted means use the supplied design rows.
- target
Optional
mse_target()describing the training response.- catch_component
Optional catch-component name used in training.
Value
A fitted GP list accepted by em_gp().