---
title: "Examples: buildLKTModel Default KC Feature Search"
author: "Philip I. Pavlik Jr."
date: "2026-06-23"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Examples: buildLKTModel Default KC Feature Search}
  %\VignetteEngine{knitr::rmarkdown}
  \usepackage[utf8]{inputenc}
---

```{r setup}
source("lkt-vignette-setup.R")
knitr::opts_chunk$set(eval = run_expensive_vignette)
```

This full-data feature search is intentionally not evaluated during routine
CRAN checks. Set `LKT_RUN_EXPENSIVE_VIGNETTES=true` to execute it when
rendering.

# Load data

```{r bundled-data}
val <- prepare_largeraw_sample()
```

# Search Default KC Features

```{r build-lkt-default-kc-feature-search}
search_features <- c("recency", "ppes", "linesuc", "linefail")

default_kc_feature_search <- buildLKTModel(
  data = val,
  allcomponents = "KC..Default.",
  allfeatures = search_features,
  forv = 0,
  bacv = 0,
  forward = TRUE,
  backward = FALSE,
  maxitv = 10,
  verbose = FALSE
)

trace_default_kc <- default_kc_feature_search[[1]]
fit_default_kc <- default_kc_feature_search[[2]]
last_trace <- trace_default_kc[nrow(trace_default_kc), ]
selected_features <- as.character(trace_default_kc$feat[trace_default_kc$feat != "none"])

final_bic <- last_trace$params * log(length(fit_default_kc$prediction)) -
  2 * fit_default_kc$loglike
final_aic <- last_trace$params * 2 - 2 * fit_default_kc$loglike

check_true("default KC feature search accepted at least one feature", nrow(trace_default_kc) > 1)
check_true("default KC feature search returned final model", !is.null(fit_default_kc$coefs))
check_true(
  "default KC selected features came from requested search set",
  all(selected_features %in% search_features)
)
check_close("default KC final BIC matches trace", final_bic, last_trace$BIC, tolerance = 5e-2)
check_close("default KC final AIC matches trace", final_aic, last_trace$AIC, tolerance = 5e-2)

cat("\nFinal selected nonlinear feature parameters:\n")
model_spec <- as.data.frame(fit_default_kc$model_specification[[1]])
nonlinear_columns <- c("component", "feature", "para", "parb", "parc", "pard", "pare")
parameter_columns <- c("para", "parb", "parc", "pard", "pare")
present_parameter_columns <- intersect(parameter_columns, names(model_spec))
nonlinear_spec <- unique(model_spec[
  rowSums(!is.na(model_spec[, present_parameter_columns, drop = FALSE])) > 0,
  intersect(nonlinear_columns, names(model_spec)),
  drop = FALSE
])
print(nonlinear_spec, row.names = FALSE)

cat("\nFinal logistic regression coefficients:\n")
coefficient_columns <- c(
  "coefficient_name", "feature", "component", "component_level", "coefficient"
)
print(model_spec[, intersect(coefficient_columns, names(model_spec)), drop = FALSE],
      row.names = FALSE)
```