fg-data-synthetic

Why we included this project

Synthetic data is a workaround teams reach for when real customer records are too sensitive to move around, and this package covers the two shapes of data that come up most often: tabular rows and time series. It draws on several generative models, including GAN-based architectures, and offers a Gaussian Mixture option that runs without a GPU, so you can get quick results without waiting on hardware. A bundled Streamlit app lowers the barrier further, letting analysts generate data through a UI instead of writing synthesizer code by hand. Because the output keeps the statistical structure of the source data while stripping identifiable information, it works well for sharing datasets with partners, building demos, or padding out a thin training set.