---
title: "Bayesian analysis with Stan"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Bayesian analysis with Stan}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.width = 7,
  fig.height = 4.5,
  dpi = 96,
  message = FALSE,
  warning = FALSE
)
```

Bayesian fits use the **same Stan likelihood** as MLE but sample from the
posterior with `rstan::sampling`.

## Before you model

Brief exploratory context for the larynx data (see *Getting started* for full
EDA):

```{r setup}
library(spsurv)
library(generics)
library(KMsurv)
library(survival)
library(ggplot2)
data(larynx)
larynx$stage <- factor(larynx$stage)
```

```{r eda-km, fig.cap = "Kaplan-Meier by stage (unadjusted)."}
km_stage <- survfit(Surv(time, delta) ~ stage, data = larynx)
km_long <- data.frame(
  time = km_stage$time,
  surv = km_stage$surv,
  stage = rep(levels(larynx$stage), km_stage$strata)
)
ggplot(km_long, aes(x = time, y = surv, color = stage)) +
  geom_step(linewidth = 0.6) +
  labs(x = "Time (years)", y = "Survival probability", color = "Stage") +
  theme_bw() +
  theme(legend.position = "bottom")
```

### What to decide

- Stage separation motivates the same formula used in the getting-started vignette.
- Small sample → posteriors may be wide; priors matter more.

## Fit and summarize

```{r fit-bayes, cache = TRUE}
fit <- bpph(
  Surv(time, delta) ~ age + stage,
  degree = 5,
  data = larynx,
  approach = "bayes",
  iter = 400,
  warmup = 200,
  chains = 1,
  cores = 1,
  init = 0,
  priors = list(
    beta = c("normal(0,4)"),
    gamma = c("lognormal(0,4)")
  )
)
summary(fit)
```

```{r fit-mle}
fit_mle <- bpph(
  Surv(time, delta) ~ age + stage,
  degree = 5,
  data = larynx,
  approach = "mle",
  init = 0
)
```

## Visual check — posterior forest plot

```{r forest-bayes, fig.cap = "Posterior medians with 95% HPD intervals (exponentiated)."}
td_b <- tidy(fit, conf.int = TRUE, exponentiate = TRUE)
hpd <- credint(fit, prob = 0.95, type = "HPD")
td_b$hpd.low <- exp(hpd[, 1])
td_b$hpd.high <- exp(hpd[, 2])
td_b$term <- factor(td_b$term, levels = rev(td_b$term))

ggplot(td_b, aes(x = estimate, y = term)) +
  geom_vline(xintercept = 1, linetype = "dashed", color = "grey50") +
  geom_pointrange(aes(xmin = hpd.low, xmax = hpd.high), linewidth = 0.4, color = "steelblue") +
  labs(x = "Hazard ratio (posterior median)", y = NULL, title = "95% HPD intervals") +
  theme_bw()
```

## Visual check — MLE vs Bayes

```{r mle-bayes-compare, fig.cap = "MLE 95% CI vs Bayesian 95% HPD (exponentiated)."}
td_m <- tidy(fit_mle, conf.int = TRUE, exponentiate = TRUE)
cmp <- merge(
  td_m[, c("term", "estimate", "conf.low", "conf.high")],
  td_b[, c("term", "estimate", "hpd.low", "hpd.high")],
  by = "term",
  suffixes = c("_mle", "_bayes")
)
cmp_long <- rbind(
  data.frame(term = cmp$term, method = "MLE", low = cmp$conf.low, high = cmp$conf.high,
             est = cmp$estimate_mle),
  data.frame(term = cmp$term, method = "Bayes", low = cmp$hpd.low, high = cmp$hpd.high,
             est = cmp$estimate_bayes)
)
cmp_long$term <- factor(cmp_long$term, levels = rev(unique(cmp$term)))

ggplot(cmp_long, aes(x = est, y = term, color = method)) +
  geom_vline(xintercept = 1, linetype = "dashed", color = "grey50") +
  geom_pointrange(aes(xmin = low, xmax = high), position = position_dodge(width = 0.5), linewidth = 0.4) +
  labs(x = "Hazard ratio", y = NULL, color = NULL) +
  theme_bw()
```

### Why this is insightful

Posterior intervals show whether data overwhelm priors. Agreement between MLE and
Bayes supports data-driven conclusions; wide HPDs flag weak information.

### What to decide

- Posterior ≈ MLE → conclusion robust to prior choice.
- Wide HPD + low ESS → do not report precise HRs; tune Stan (`adapt_delta`, more
  `iter`).
- WAIC/DIC for comparing models **within the same family** at the same degree.

## Credible intervals and criteria

```{r credint}
credint(fit, prob = 0.95, type = "HPD")
```

```{r criteria}
sm <- summary(fit)
c(DIC = sm$dic, WAIC = sm$waic, LPML = sm$lpml)
```

## Convergence (optional deep dive)

Install `bayesplot` for trace and pairs plots:

```{r stanfit, eval = FALSE}
library(bayesplot)
mcmc_pairs(fit$posterior$beta, off_diag_fun = "hex")
```

Always inspect divergences, split R-hat, and ESS before interpreting results.

## tidybayes posterior draws

Install **posterior** and **tidybayes** for long-format draws and interval plots.
After `library(spsurv)`, `spread_draws()` and `tidy_draws()` dispatch on Bayes
fits when **tidybayes** is loaded.

```{r tidybayes, eval = requireNamespace("tidybayes", quietly = TRUE) && requireNamespace("posterior", quietly = TRUE)}
library(tidybayes)
dr <- as_draws_df.spbp(fit)
spread_draws(fit, `beta[age]`)
```

Draw-level survival curves:

```{r spread-surv, eval = requireNamespace("tidybayes", quietly = TRUE)}
head(spread_surv_draws.spbp(fit, times = c(1, 2, 3), newdata = larynx[1, ]))
```

See `vignette("tidymodels", package = "spsurv")` for workflows and parsnip
engines.

## Survival prediction

```{r predict-bayes}
pr <- predict(fit, times = seq(0, max(larynx$time), length.out = 50))
head(pr)
```

See the *Survival prediction and ggplot* vignette for ribbon workflows.