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

```{r, include = FALSE}
knitr::opts_chunk$set(collapse = TRUE, comment = "#>",
                      fig.width = 7, fig.height = 5, dpi = 96)
```

```{r setup}
library(logcumulant)
data(reliability_datasets)
```

## Why log-cumulants?

For positive-support data the Mellin transform plays the role that the Fourier
or Laplace transform plays on the whole line. Differentiating the Mellin
characteristic function at its central point yields the **log-cumulants**
\(\kappa_1, \kappa_2, \ldots\): the cumulants of \(\log X\). These quantities
are natural shape descriptors for reliability distributions and behave well
under the multiplicative structure typical of lifetime data.

The package turns these descriptors into (i) diagnostic **diagrams** and (ii)
formal **goodness-of-fit tests**.

## A first analysis

We use the classic ball-bearing fatigue-life dataset bundled with the package.

```{r}
bb <- reliability_datasets$BallBearing
length(bb)
summary(bb)
```

The quickest entry point is `plot_lc()`, which draws the log-cumulant diagram
with a bootstrap cloud of the sample estimate:

```{r}
plot_lc(bb, B = 200)
```

The sample point sits near the Weibull and Gamma loci, suggesting a
light-tailed model.

## Comparing candidate models

`gof_compare_all()` fits all six families and reports the three \(T^2\)
statistics, the Anderson--Darling and Cramer--von Mises tests, and the AIC. The
parametric bootstrap is recommended for the p-values:

```{r}
gof_compare_all(bb, use_bootstrap = TRUE, B = 199, seed = 1)
```

## Where to go next

- `vignette("diagrams")` explains the three diagnostic diagrams.
- `vignette("gof-tests")` covers the tests and why the bootstrap is needed.
- `vignette("simulation")` reproduces the size and power studies.