Package 'PUcopulaSynth'

Title: Partition-of-Unity Copula Fitting and Synthesis for R
Description: Fit multivariate distributions using a Partition-of-Unity copula dependence structure, estimate marginals, and generate synthetic data with factor pre/post-processing.
Authors: Andreas Mändle [aut, cre]
Maintainer: Andreas Mändle <[email protected]>
License: MIT + file LICENSE
Version: 0.1.0
Built: 2026-05-10 09:13:05 UTC
Source: https://github.com/amaendle/PUcopulaSynth

Help Index

Estimate marginal models

Description

Fits logspline marginals for numeric/ordered variables, and empirical probability tables for binary or trivial variables. Optional k-NN smoothing is applied to numeric columns.

Usage

estimateMarginals(
  data,
  method = "spline",
  k = NULL,
  lbound = NULL,
  ubound = NULL
)

Arguments

data

Preprocessed data.frame
method

Character; for numeric and ordered factors (currently "spline")
k

Numeric scalar, vector, or named list for k-NN smoothing
lbound

lower boundary of distribution passed to logspline::logspline().
rbound

upper boundary of distribution passed to logspline::logspline().

Value

A named list of marginal models (each element has qfun)

Fit a PUcopula model

Description

Fits a PUcopula model on a (preprocessed) data matrix with optional rank-binning and jitter for numeric variables.

Usage

fitPUcopula(
  data,
  driver_strength_factor = 0.5,
  bin_size = 3,
  jitter = FALSE,
  family = "binom"
)

Arguments

data

Preprocessed data.frame (e.g., preprocessData()$data)
driver_strength_factor

Numeric scalar or vector in (0,1\] used to scale rows per variable
bin_size

Numeric scalar, vector, or named list with bin sizes
jitter

FALSE, numeric (single) or named list mapping variables to jitter factors
family

PUcopula family, e.g. "binom" or "nbinom"

Value

A PUcopula::PUCopula model

Generate synthetic data

Description

Combines a fitted PUcopula and marginal models to produce a synthetic data.frame. Optionally restores original factor structure, names and classes.

Usage

generateSynthetic(
  n,
  copula,
  marginals,
  original_levels = NULL,
  original_varnames = NULL,
  original_classes = NULL
)

Arguments

n

Integer, number of rows to generate
copula

A PUcopula model
marginals

List of marginals from estimateMarginals()
original_levels

Optional preprocessData()$original_levels
original_varnames

Optional vector of original column names
original_classes

Optional named vector of original classes

Value

Synthetic data.frame

k-NN smoother for numeric vectors (no DataSHIELD thresholds)

Description

k-NN smoother for numeric vectors (no DataSHIELD thresholds)

Usage

knnsmoother(x, k = 3)

Arguments

x

Numeric vector (NAs allowed).
k

Integer neighbours in \[1, N-1] for non-missing values.

Value

Numeric vector of same length with smoothed non-missing entries.

Restore original factor structure after synthesis

Description

Restore original factor structure after synthesis

Usage

postprocessData(data, cat_dummy_levels)

Arguments

data

A data.frame containing dummy/encoded columns
cat_dummy_levels

The dummies element returned by preprocessData()

Value

A data.frame with factors restored and columns ordered like input

Preprocess data before copula fitting

Description

Converts multi-level unordered factors to dummy variables and tags remaining factor columns with '.oriname' while storing levels.

Usage

preprocessData(data)

Arguments

data

A data.frame

Value

A list with data (processed) and original_levels (dummies + oriname)

Save original names and classes

Description

Save original names and classes

Usage

save_original_varnames(data)

save_original_classes(data)

Arguments

data

A data.frame

Value

Character vector (names) / Named character vector (classes)

Draw from a fitted PUcopula

Description

Draw from a fitted PUcopula

Usage

simulateCopula(model, n)

Arguments

model

A PUcopula model from fitPUcopula()
n

Number of rows

Value

Matrix of U(0,1) draws