Enum SaasboInferenceMethod

Namespace

OptunaSharp.Samplers

Assembly

OptunaSharp.dll

The posterior-inference strategy SAASBO uses to fit the SAAS GP each trial (Eriksson & Jankowiak 2021, §4.2). The choice trades inference fidelity for runtime and does not change the SAAS function prior itself.

public enum SaasboInferenceMethod

Fields

Nuts = 0

The No-U-Turn Sampler (§4.2.1) — the paper's default and "gold standard". It draws a bag of L posterior hyperparameter samples and averages expected improvement over them, which the paper finds significantly more robust than MAP (§5.2) because averaging over posterior modes mitigates EI's pull toward the domain boundary. It is the more expensive option (Hamiltonian Monte Carlo over an O(N³D) gradient per leapfrog step).

Map = 1

Maximum a posteriori (§4.2.2 / §A.2) — the faster point-estimate alternative. It optimizes the log posterior with Adam for a few pre-selected shrinkage values τ_s and selects one by leave-one-out predictive likelihood, yielding a single hyperparameter estimate rather than a sample bag. It needs no α hyperparameter but, per the paper, is less robust than NUTS.