Expand description
Bayesian potential-energy wrappers kept separate from distribution families.
This crate deliberately starts with one small, statistically explicit vertical slice: normalized Gaussian priors on predictor coefficients plus the summed, unpenalized compiled likelihood. It contains no sampler, autodiff backend, chain storage, or implicit conversion from fitting penalties to priors.
Modules§
- prelude
- Common imports for the Bayesian vertical slice.
Structs§
- Gaussian
Coefficient Prior - Independent normalized Gaussian priors on predictor coefficients.
- Posterior
Potential - Potential energy
summed weighted NLL - log p(beta). - Workspace
Posterior Potential - Posterior potential with reusable likelihood buffers.
Traits§
- Coefficient
Prior - A normalized prior defined with respect to the flat coefficient vector.