---
title: "Getting started with selSurv"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Getting started with selSurv}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

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

`selSurv` separates Cox-Lasso fitting from the inferential question asked after selection. The main user-facing functions correspond to different inferential targets rather than interchangeable ways to compute the same confidence interval.

## A common Cox-Lasso fit

The following small example is used only to illustrate the interface.

```{r data}
set.seed(2026)
n <- 120
p <- 6
X <- matrix(rnorm(n * p), nrow = n, ncol = p)
colnames(X) <- paste0("x", seq_len(p))
beta <- c(0.8, -0.6, 0.4, 0, 0, 0)
T <- rexp(n, rate = 0.06 * exp(drop(X %*% beta)))
C <- rexp(n, rate = 0.04)
time <- pmin(T, C)
status <- as.integer(T <= C)
```

```{r fit}
library(selSurv)
fit <- coxlasso_fit(X, time, status, lambda_choice = "1se")
fit
```


```{r api-smoke, include=FALSE}
db_smoke <- coxlasso_debias(fit)
db_boot_smoke <- coxlasso_boot(
  fit, inference = "debiased", type = c("wild", "efron"),
  B = 3L, conf_level = 0.95, seed = 2026
)
fixed_smoke <- coxlasso_boot(
  fit, inference = "conditional",
  B = 3L, conf_level = 0.95, seed = 2026
)
split_smoke <- coxlasso_split(X, time, status, seed = 2026)
reselected_smoke <- coxlasso_boot_reselect(
  X, time, status, B = 3L, conf_level = 0.95,
  methods = c("fixed_support", "reselect_fixed_lambda", "reselect_cv_min"),
  seed = 2026, nfolds = 3L, standardize = TRUE
)
```

## Choose the inferential target

After fitting the Cox-Lasso, the main routes are:

```{r routes, eval=FALSE}
# Full-model coefficient target: debiased Wald inference
db <- coxlasso_debias(fit)

# Full-model coefficient target: wild bootstrap and Efron's bootstrap
db_boot <- coxlasso_boot(
  fit,
  inference = "debiased",
  type = c("wild", "efron"),
  B = 500,
  conf_level = 0.95,
  seed = 2026
)

# Selected-submodel target: observed support held fixed
fixed <- coxlasso_boot(
  fit,
  inference = "conditional",
  B = 500,
  conf_level = 0.95,
  seed = 2026
)

# Selection and inference on disjoint samples
split <- coxlasso_split(X, time, status, seed = 2026)

# Procedure-level re-selection
reselected <- coxlasso_boot_reselect(
  X, time, status,
  B = 500,
  conf_level = 0.95,
  methods = c("fixed_support", "reselect_fixed_lambda", "reselect_cv_min"),
  seed = 2026
)
```

These calls should not be interpreted as interchangeable confidence intervals. Debiased inference addresses a full-model coefficient target, fixed-support model-based bootstrap inference addresses coefficients of the selected working model, sample splitting uses a selected model obtained from an independent selection sample, and re-selection describes the repeated data-adaptive procedure.

## Inspecting results

```{r summaries, eval=FALSE}
summary(db)
summary(db_boot)
summary(fixed)
summary(split)

rs <- summary(reselected)
rs$diagnostics
rs$intervals
```

For publication-scale bootstrap analyses, use substantially larger `B` than values used in package smoke tests and inspect method-specific diagnostics together with confidence intervals.