---
title: "LOO Influence and Predictive Model Comparison"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{LOO Influence and Predictive Model Comparison}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(collapse = TRUE, comment = "#>")
```

The LOO display layer converts existing `gp3bayes_psis_loo`,
`gp3bayes_loo_comparison`, and `gp3bayes_loo_weights` objects into explicit
tables and publication-oriented figures.

```{r, eval=FALSE}
loo_result <- compute_psis_loo(fit, cores = 1)

loo_summary_table(loo_result)
loo_diagnostic_table(loo_result)
plot_loo_influence(loo_result)
```

For multiple prespecified models:

```{r, eval=FALSE}
comparison <- compare_psis_loo(
  list(reference = fit_reference, sensitivity = fit_sensitivity),
  cores = 1
)

weights <- compute_loo_model_weights(comparison, method = "stacking")

model_comparison_table(comparison)
model_weights_table(weights)

plot_model_comparison(comparison)
plot_model_weights(weights)
```

ELPD differences and predictive weights are retained as descriptive predictive
evidence. The package does not promote the highest-ranked model to an
automatically preferred substantive model.