---
title: "Migrating from synthpop"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Migrating from synthpop}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

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

## Why switch?

`synthpop` excels at disclosure-controlled individual-level microdata but
lacks joint distribution modeling via copulas. `rsdv` uses a Gaussian copula
to preserve inter-column correlations.

## Side-by-side comparison

### synthpop workflow

```r
library(synthpop)
synth <- syn(adult_income[, c("age", "occupation", "income")])
synthetic_data <- synth$syn
```

### rsdv workflow

```{r, message=FALSE}
library(rsdv)

set.seed(42)

meta  <- metadata(adult_income) |>
  set_column_type("age",        "numerical") |>
  set_column_type("occupation", "categorical") |>
  set_column_type("income",     "categorical")

syn   <- gaussian_copula_synthesizer(meta)
syn   <- fit(syn, adult_income)
synthetic_data <- sample(syn, n = nrow(adult_income))
```

## Key differences

| Feature | synthpop | rsdv |
|---|---|---|
| Correlation modeling | CART-based sequential | Gaussian copula over all column types |
| Column constraints | Limited | Equality, inequality, fixed combos, custom |
| Conditional sampling | Via predictor order | `sample_conditions()` on categorical values |
| Quality metrics | Built-in utility measures | KS, TVD, correlation & contingency similarity, ML efficacy |
| Diagnostics | None | Validity report (ranges, categories, key uniqueness) |
| Privacy metrics | None | NNDR, attribute disclosure risk |
| Python interop | No | API-compatible with SDV |