Module fanova_bench 

Forest-fANOVA importance, for cross-checking the in-house PED-ANOVA ranking.

The M5.4 / D21 oracle helper (docs/design/09-implementation.md §8.2). It fits the [fanova] crate’s random-forest fANOVA to a study’s completed trials and returns a per-parameter importance ranking — an estimator independent of atune’s own PED-ANOVA (a random-forest variance decomposition versus a closed-form Parzen-density divergence), so agreement on the ranking is real evidence rather than a tautology. The absolute values differ by construction; only the ordering is meant to line up.

The encoding mirrors what PED-ANOVA folds onto its grid, so both estimators see the same features: a numeric parameter contributes its value, a boolean 0.0/1.0, and a categorical its choice index as an f64 (the same wasm_fanova_calculate feature construction optuna-dashboard uses). The forest is seeded (FOREST_SEED) so the ranking is deterministic run to run.

Constants

FOREST_SEED

The fixed random-forest seed, so a run’s ranking is reproducible.

Functions

fanova_ranking

The forest-fANOVA importance ranking of a study’s parameters, most important first.