Author: Marcel Neunhoeffer

Status: Published on CRAN

Downloads: 8,300+ (check current count on CRAN logs)

Overview

RGAN is an R package that enables social scientists and researchers to train Generative Adversarial Networks without needing Python or PyTorch expertise. Built on the torch library in R (native implementation, not through Python), RGAN makes deep generative models accessible to the R community.

Features

  • Multiple GAN Variants:

    • Original GAN
    • Wasserstein GAN (WGAN)
    • f-WGAN

  • Post-Processing Methods:

    • Discriminator Rejection Sampling
    • Post-GAN Boosting

  • Native R Implementation: Built on torch for R, no Python dependencies required

Installation

# Install from CRAN
install.packages("RGAN")

# Or install development version from GitHub
# devtools::install_github("mneunhoe/RGAN")

Why RGAN?

Many social scientists work primarily in R and face barriers when trying to use deep learning methods typically implemented in Python. RGAN bridges this gap by providing a native R implementation of GANs, enabling researchers to:

  • Generate synthetic data for privacy-preserving data sharing
  • Augment small datasets
  • Explore generative modeling without switching languages