TabDDPM Synthetic Data Generator

Generates synthetic tabular data using denoising diffusion probabilistic models (DDPM). This is a state-of-the-art generative approach for tabular data.

Usage

tabddpm_generate(
  data,
  target,
  n_synthetic,
  n_steps = 1000L,
  hidden_dim = 256L,
  epochs = 1000L,
  batch_size = 32L
)

Arguments

data

Training data (data.frame)

target

Target column name

n_synthetic

Number of synthetic samples to generate

n_steps

Number of diffusion steps (default: 1000)

hidden_dim

Hidden layer dimension (default: 256)

epochs

Training epochs (default: 1000)

batch_size

Batch size (default: 32)

Value

A data.frame of synthetic samples

Details

TabDDPM requires the 'torch' package. The model learns the data distribution through a denoising process and can generate realistic synthetic samples.

References

Kotelnikov et al. (2023). TabDDPM: Modelling Tabular Data with Diffusion Models.

Examples

if (FALSE) { # \dontrun{
if (requireNamespace("torch", quietly = TRUE)) {
  synthetic <- tabddpm_generate(
    data = training_data,
    target = "outcome",
    n_synthetic = 100
  )
}
} # }