DiffGBM
Copyright (c) 2026, Silas Koemen

This product is a derivative work of Treeffuser, and includes software
developed by Nicolas Beltran-Velez, Alessandro Antonio Grande, and
Achille Nazaret:

    Treeffuser
    Copyright (c) 2024, Nicolas Beltran, Alessandro Antonio Grande,
    and Achille Nazaret
    https://github.com/blei-lab/treeffuser
    Licensed under the MIT License

Treeffuser is described in:

    Beltran-Velez, N., Grande, A. A., Nazaret, A., Kucukelbir, A., and
    Blei, D. (2024). Treeffuser: Probabilistic Predictions via Conditional
    Diffusions with Gradient-Boosted Trees. Advances in Neural Information
    Processing Systems 37. arXiv:2406.07658.

DiffGBM retains Treeffuser's gradient-boosted-tree backbone, SDE module,
feature pipeline, and sample-handling utilities, and extends them with
flow matching on Gaussian paths, conditional-mean residualization,
EDM-style preconditioning, log-sigma time sampling and noise-level
features, loss weighting, and additional samplers. The original MIT
license terms continue to apply to the inherited portions.
