Skip to contents

Leave-one-out cross-validation via Rasmussen & Williams (2006) shortcut

Source: R/loo_cv.R
loo_cv.Rd

Computes closed-form LOO-CV predictions at NOAA sites using the block version of Rasmussen & Williams (2006) eq. 5.12. For site k with observation block B_k, the LOO predictive mean and covariance are computed from the precision matrix without refitting.

Usage

loo_cv(model, ...)

# S3 method for class 'evfuse_model'
loo_cv(model, loo_source = NULL, ...)

# S3 method for class 'evfuse_naive_model'
loo_cv(model, ...)

Arguments

model

A fitted model (evfuse_model or evfuse_naive_model).

...

Additional arguments (currently unused).

loo_source

Which data source to evaluate LOO-CV at (default: first source in source_params, typically "NOAA").

Value

An evfuse_loo object with components:

loo_mean

Matrix (n_sites x 3) of LOO predictive means.

loo_cov

List of 3x3 LOO predictive covariance matrices.

observed

Matrix (n_sites x 3) of observed Stage 1 MLEs.

sites

Data frame of site metadata.

loo_lpd

Vector of per-site log predictive densities.