---
title: "Examples: AutoKC"
author: "Philip I. Pavlik Jr."
date: "2026-06-07"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Examples: AutoKC}
  %\VignetteEngine{knitr::rmarkdown}
  \usepackage[utf8]{inputenc}
---

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

The complete AutoKC 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()
```

# AutoKC model search

```{r autokc-search}
mnames <- c(
  "IRT",
  "Faculty",
  "Log Full autoKC",
  "Log Simple PFA",
  "Log Full PFA",
  "Log Full PFA full autoKC additive",
  "Log Full PFA Faculty additive ",
  "Log Simple PFA Faculty interactive ",
  "Log Simple PFA full autoKC interactive",
  "Log Full PFA simple autoKC interactive",
  "Log Simple PFA simple autoKC interactive"
)
r2s <- data.frame(name = mnames, r2s = NA)
compl <- list(
  c("Anon.Student.Id", "KC..Default."),
  c("Anon.Student.Id", "KC..Default.", "Anon.Student.Id", "Anon.Student.Id"),
  c("Anon.Student.Id", "KC..Default.", "KC..Default.", "KC..Default."),
  c("Anon.Student.Id", "KC..Default.", "KC..Default.", "KC..Default."),
  c("Anon.Student.Id", "KC..Default.", "KC..Default.", "KC..Default."),
  c("Anon.Student.Id", "KC..Default.", "KC..Default.", "KC..Default.",
    "KC..Default.", "KC..Default."),
  c("Anon.Student.Id", "KC..Default.", "KC..Default.", "Anon.Student.Id",
    "KC..Default.", "Anon.Student.Id"),
  c("Anon.Student.Id", "KC..Default.", "KC..Default.", "Anon.Student.Id",
    "KC..Default.", "Anon.Student.Id"),
  c("Anon.Student.Id", "KC..Default.", "KC..Default.", "KC..Default.",
    "KC..Default.", "KC..Default."),
  c("Anon.Student.Id", "KC..Default.", "KC..Default.", "KC..Default.",
    "KC..Default.", "KC..Default."),
  c("Anon.Student.Id", "KC..Default.", "KC..Default.", "KC..Default.",
    "KC..Default.", "KC..Default.")
)
featl <- list(
  c("intercept", "intercept"),
  c("intercept", "intercept", "logfail", "logsuc"),
  c("intercept", "intercept", "logfail$", "logsuc$"),
  c("intercept", "intercept", "logfail", "logsuc"),
  c("intercept", "intercept", "logfail$", "logsuc$"),
  c("intercept", "intercept", "logfail$", "logfail$", "logsuc$", "logsuc$"),
  c("intercept", "intercept", "logfail$", "logfail", "logsuc$", "logsuc"),
  c("intercept", "intercept", "logfail", "logfail", "logsuc", "logsuc"),
  c("intercept", "intercept", "logfail", "logfail$", "logsuc", "logsuc$"),
  c("intercept", "intercept", "logfail$", "logfail", "logsuc$", "logsuc"),
  c("intercept", "intercept", "logfail", "logfail", "logsuc", "logsuc")
)
connl <- list(
  c("+"),
  c("+", "+", "+"),
  c("+", "+", "+"),
  c("+", "+", "+"),
  c("+", "+", "+"),
  c("+", "+", "+", "+", "+"),
  c("+", "+", "+", "+", "+"),
  c("+", "+", "*", "+", "*"),
  c("+", "+", "*", "+", "*"),
  c("+", "+", "*", "+", "*"),
  c("+", "+", "*", "+", "*")
)
autol <- list(
  c(0, 0),
  c(0, 0, 0, 0),
  c(0, 0, 40, 40),
  c(0, 0, 0, 0),
  c(0, 0, 0, 0),
  c(0, 0, 0, 40, 0, 40),
  c(0, 0, 0, 0, 0, 0),
  c(0, 0, 0, 0, 0, 0),
  c(0, 0, 0, 40, 0, 40),
  c(0, 0, 0, 40, 0, 40),
  c(0, 0, 0, 40, 0, 40)
)

for (i in seq_along(compl)) {
  modelob <- LKT(
    data = val,
    components = compl[[i]],
    features = featl[[i]],
    connectors = connl[[i]],
    autoKC = autol[[i]],
    verbose = TRUE
  )
  cat(" R2 =  ", modelob$r2, "\n")
  r2s$r2s[i] <- modelob$r2
}
check_true("AutoKC model count", nrow(r2s) == 11)
check_close("AutoKC first model R2", r2s$r2s[1], 0.17526, tolerance = 1e-4)
check_close("AutoKC best listed R2", max(r2s$r2s, na.rm = TRUE), 0.336243, tolerance = 1e-3)
```

```{r autokc-plot}
r2s$name <- factor(r2s$name, levels = rev(mnames))
ggplot(r2s, aes(name, r2s)) +
  geom_bar(stat = "identity") +
  xlab("Model Version") +
  ylab("R-squared Gain") +
  coord_flip() +
  theme(text = element_text(size = 12))
```

```{r autokc-cluster-sensitivity}
mnames <- seq(2, 71, 10)
for (i in c(3, 6)) {
  r2s <- data.frame(name = mnames, r2s = NA, r2sr = NA)
  j <- 1
  for (k in mnames) {
    j <- j + 1
    modelob <- LKT(
      data = val,
      components = compl[[i]],
      features = featl[[i]],
      connectors = connl[[i]],
      autoKC = k * (autol[[i]] > 0),
      verbose = FALSE
    )
    cat(" R2 =  ", modelob$r2, "\n")
    r2s$r2s[j - 1] <- modelob$r2

    modelob <- LKT(
      data = val,
      components = compl[[i]],
      features = featl[[i]],
      connectors = connl[[i]],
      autoKC = k * (autol[[i]] > 0),
      verbose = FALSE,
      autoKCcont = rep("rand", length(featl[[i]]))
    )
    cat(" R2 =  ", modelob$r2, "\n")
    r2s$r2sr[j - 1] <- modelob$r2
  }

  r2s$name <- factor(r2s$name, levels = mnames)
  ggplot(r2s, aes(name, group = 1)) +
    geom_line(aes(y = r2s)) +
    geom_line(aes(y = r2sr), linetype = "twodash") +
    scale_x_discrete(breaks = seq(from = 2, to = 71, by = 5)) +
    xlab("autoKC Clusters") +
    ylab("McFadden's R-squared Gain") +
    theme(text = element_text(size = 16)) +
    geom_point(aes(y = r2s)) +
    geom_point(aes(y = r2sr))
}
check_true("AutoKC cluster sensitivity completed", exists("r2s") && nrow(r2s) == length(mnames))
check_true("AutoKC cluster sensitivity finite R2", all(is.finite(r2s$r2s)))
```