---
title: "Using priorsense with Stan"
vignette: >
  %\VignetteIndexEntry{Using priorsense with Stan}
  %\VignetteEngine{quarto::html}
  %\VignetteEncoding{UTF-8}
---


```{r}
#| include: false
ggplot2::theme_set(bayesplot::theme_default(base_family = "sans"))
options(priorsense.plot_help_text = FALSE)
```

```{r}
#| message: false
#| warning: false
library(posterior)
library(priorsense)
```

`priorsense` is compatible with models fit with either `rstan` and `cmdstanr`.
To use `priorsense` with a Stan model, the log prior and log likelihood
evaluations should be added to the model code.

Consider the univariate normal model with unknown mu and sigma available
via`example_powerscale_model("univariate_normal")`. In Stan, `lprior` and
`log_lik` variables can be defined as below. By also defining separate
`lprior_mu` and `lprior_sigma` variables, it will be possible to check the
sensitivity for each prior separately.


```{r}
model <- example_powerscale_model("univariate_normal")
```

```{r}
#| echo: false
#| results: asis
cat("```stan\n")
cat(model$model_code)
cat("```")
```

We first fit the model using Stan:
```{r}
#| echo: false
#| message: false
#| warning: false
fit <- readRDS(system.file("extdata", "univariate_normal_rstan.RDS", package = "priorsense"))

```

```{r}
#| eval: false
#| message: false
#| warning: false
fit <- rstan::stan(
  model_code = model$model_code,
  data = model$data,
  refresh = FALSE,
  seed = 123
)
```

Then the `priorsense` functions will work as usual.

```{r}
powerscale_sensitivity(fit)
```

```{r}
powerscale_sensitivity(fit, prior_selection = "sigma")
```

```{r}
powerscale_sensitivity(fit, prior_selection = "mu")
```

```{r}
#| message: false
#| warning: false
#| fig-width: 6
#| fig-height: 4
powerscale_plot_dens(fit)
```


In some cases, setting `moment_match = TRUE` will improve
unreliable estimates at the cost of some further computation. This
requires the [`iwmm` package](https://github.com/topipa/iwmm).