Module diffusion 

Diffusion model abstraction for statistical and neural data generation.

Implements two diffusion backends behind a common DiffusionBackend trait:

• Statistical (always available): pure-Rust Langevin-inspired denoising guided by target means, stds, and Cholesky-decomposed correlations.

• Neural (requires neural feature): learned score network trained via denoising score matching. Captures nonlinear cross-column dependencies that parametric models miss. Powered by candle.

Both backends slot into HybridGenerator for blended rule+diffusion output.

Re-exports

pub use backend::*;

pub use hybrid::*;

pub use schedule::*;

pub use statistical::*;

pub use training::*;

pub use utils::*;

Modules

backend

hybrid

Hybrid generator that blends rule-based and diffusion-generated data.

schedule

statistical

Statistical diffusion backend that generates data matching target distributions.

training

Diffusion model training pipeline: fit from column statistics, persist, and evaluate.

utils