Expand description
Superforecaster pipeline — LHS sampling → polynomial chaos expansion (PCE) → Sobol’ sensitivity indices → Bayesian optimization.
Mirrors v26’s mc.superforecaster: cheaply explore the parameter box
with Latin Hypercube Sampling, build a polynomial surrogate (PCE) for
variance-based sensitivity analysis, then refine the optimum with the
GP/EI optimizer from crate::bayesian.
The PCE uses probabilists’ Hermite polynomials (the orthogonal basis of
the standard normal), so Sobol’ indices are analytic in the coefficients:
S_i = Σ_{α: αᵢ>0} c_α² / Σ_α c_α².
Structs§
- Pce
- Polynomial chaos expansion — a Hermite-polynomial surrogate with analytic Sobol’ sensitivity indices.
- Superforecaster
Result - Result of a superforecaster run.
Functions§
- latin_
hypercube - Latin Hypercube Sampling over a box.
- superforecaster
- Run the full superforecaster pipeline.