ctgan

The ctgan package provides an R interface to CTGAN, a GAN-based data synthesizer. The package enables one to create synthetic samples of confidential or proprietary datasets for sharing. For more details and use cases, see the papers in the References section.

Installation

You can install the development version from GitHub with:

# install.packages("remotes")
remotes::install_github("kasaai/ctgan")

Example

A quick example:

library(ctgan)

# Install dependencies before first usage
# install_ctgan()

synthesizer <- ctgan()
synthesizer %>% 
  fit(iris, epochs = 20)
#> Epoch 1, Loss G: 1.1087, Loss D: -0.0124
#> Epoch 2, Loss G: 1.1455, Loss D: 0.0002
#> Epoch 3, Loss G: 1.1319, Loss D: -0.0197
#> Epoch 4, Loss G: 1.1054, Loss D: 0.0126
#> Epoch 5, Loss G: 1.0720, Loss D: -0.0240
#> Epoch 6, Loss G: 1.0652, Loss D: -0.0669
#> Epoch 7, Loss G: 1.0788, Loss D: -0.0513
#> Epoch 8, Loss G: 1.0766, Loss D: -0.0393
#> Epoch 9, Loss G: 1.0464, Loss D: -0.0116
#> Epoch 10, Loss G: 1.0297, Loss D: 0.0042
#> Epoch 11, Loss G: 1.0110, Loss D: -0.0482
#> Epoch 12, Loss G: 0.9828, Loss D: -0.0413
#> Epoch 13, Loss G: 0.9493, Loss D: -0.0031
#> Epoch 14, Loss G: 1.0067, Loss D: -0.0687
#> Epoch 15, Loss G: 0.9857, Loss D: -0.0215
#> Epoch 16, Loss G: 0.9618, Loss D: 0.0044
#> Epoch 17, Loss G: 0.9280, Loss D: 0.0061
#> Epoch 18, Loss G: 0.8579, Loss D: -0.0139
#> Epoch 19, Loss G: 0.8993, Loss D: 0.0400
#> Epoch 20, Loss G: 0.8277, Loss D: 0.0320
synthesizer %>% 
  ctgan_sample() %>% 
  # Dataset-specific post-processing
  dplyr::mutate_if(is.numeric, ~ pmax(.x, 0.1))
#> # A tibble: 100 x 5
#>    Sepal.Length Sepal.Width Petal.Length Petal.Width Species   
#>           <dbl>       <dbl>        <dbl>       <dbl> <chr>     
#>  1         4.67        3.49        3.97         1.94 virginica 
#>  2         7.75        3.14        3.87         1.96 setosa    
#>  3         5.84        3.68        4.81         2.46 versicolor
#>  4         5.03        2.90        2.74         2.38 virginica 
#>  5         5.01        4.39        0.423        1.83 versicolor
#>  6         5.74        3.33        3.20         2.58 virginica 
#>  7         4.13        3.15        2.88         3.24 setosa    
#>  8         6.83        2.81        3.25         3.60 setosa    
#>  9         5.21        3.92        6.06         2.15 setosa    
#> 10         4.23        3.98        2.81         2.52 virginica 
#> # … with 90 more rows

This generated dataset can then be shared, but one can also serialize and share the trained synthesizer:

model_dir <- tempdir()
synthesizer %>% 
  ctgan_save(model_dir)

ctgan_load(model_dir)
#> A CTGAN Model

References

If you use ctgan, please consider citing the original work,

• Lei Xu, Maria Skoularidou, Alfredo Cuesta-Infante, Kalyan Veeramachaneni. Modeling Tabular data using Conditional GAN. NeurIPS, 2019.

and the work implementing the R package,

• Kevin Kuo. Generative Synthesis of Insurance Datasets. arXiv:1912.02423, 2019.

@inproceedings{xu2019modeling,
  title={Modeling Tabular data using Conditional GAN},
  author={Xu, Lei and Skoularidou, Maria and Cuesta-Infante, Alfredo and Veeramachaneni, Kalyan},
  booktitle={Advances in Neural Information Processing Systems},
  year={2019}
}

@misc{kuo2019generative,
    title={Generative Synthesis of Insurance Datasets},
    author={Kevin Kuo},
    year={2019},
    eprint={1912.02423},
    archivePrefix={arXiv},
    primaryClass={stat.AP}
}