Initial Release of POTNet

Highlights

• This release introduces POTNet, a computationally efficient deep generative model designed for the synthetic generation of potentially mixed-type tabular data (continuous, discrete, and categorical) using the marginally-penalized Wasserstein loss. POTNet provides flexibility and enhanced performance for a variety of applications.

---

Changelog

New Features:

• Added support for saving and loading training checkpoints with the option:

  save_checkpoint=True

• Introduced functionality to save and load the entire model for easier reuse:

  model.save('filename')
  model = load_model('filename')

• Enhanced the network architecture for improved performance and robustness.
• Added support for custom importance weights to allocate sample-specific weights during training using:

  model.fit(weights=sample_weights)

---

Acknowledgments

Special thanks to Yuanqing Wang (@Lin-xs) for contributions to organizing the package directory.