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

```{r, setup, message=FALSE}
library(limpidR)
```

## Load and validate

```{r}
db <- load_limpid(quiet = TRUE)
check_database(db)
```

`check_database()` validates identifiers, cross-table event alignment, abundance ranges,
composition closure, coordinates, abundance units and method links.

## Descriptive analysis

```{r}
head(summarise_abundance(db))
head(analyse_morphology(db))
head(analyse_size_distribution(db))
head(analyse_polymers(db))
```

## Modelling without random-row leakage

For a multisite database, random row splitting can place observations from the same lake
in both training and testing sets. `cross_validate_mp()` therefore defaults to grouped
validation by `Lake_ID`.

```{r}
cv <- cross_validate_mp(
  db,
  MP_Mean ~ Season_Global + Lake_Type,
  method = "lognormal_lm",
  group = "Lake_ID"
)
cv$overall
```

## Compositional data

Polymer, morphology and size profiles are compositional. `limpidR` provides closure checks,
CLR transformation and Aitchison distance. Zero handling is explicit through the
`pseudocount` argument.

```{r}
pol <- db$Polymer_Composition
cols <- c("PE_pct", "PP_pct", "PET_PES_pct", "PA_Nylon_pct",
          "PS_EPS_pct", "PVC_pct", "OtherPolymer_pct")
head(clr_transform(pol, cols))
aitchison_distance(pol, cols)
```

## Risk components

`calculate_risk()` intentionally does not ship a universal polymer-hazard weighting scheme.
Supply the abundance reference and any polymer hazard values used in your study, and report
them in the methods section.