rsdv (version 0.2.0)

gaussian_copula_synthesizer: Create a Gaussian Copula synthesizer

Description

Fits a single Gaussian copula over all modeled columns. Numerical columns use a fitted parametric marginal (see default_distribution); categorical and boolean columns are embedded into the copula via their cumulative-frequency intervals, so cross-column dependence (numeric vs. categorical, categorical vs. categorical) is preserved.

Usage

gaussian_copula_synthesizer(
  metadata,
  enforce_min_max = TRUE,
  numerical_distributions = list(),
  default_distribution = "auto"
)

Value

An unfitted gaussian_copula_synthesizer object.

Arguments

metadata

An rsdv_metadata object.

enforce_min_max

Logical. Clamp sampled numerical values to the observed range. Default TRUE.

numerical_distributions

Optional named character vector/list mapping numerical column names to a distribution in "norm", "beta", "gamma", "truncnorm", "uniform", or "auto".

default_distribution

Distribution used for numerical columns not named in numerical_distributions. "auto" (default) selects the best-fitting family per column by Kolmogorov-Smirnov distance.

Examples

# \donttest{
meta <- metadata(adult_income) |>
  set_column_type("age", "numerical") |>
  set_column_type("occupation", "categorical")
syn <- gaussian_copula_synthesizer(meta, default_distribution = "auto")
syn <- fit(syn, adult_income)
# }