Random

Random is the mandatory “floor” baseline: it assigns each candidate a uniform random score, completely ignoring the surrogate’s predictions (mean or uncertainty). Any acquisition strategy that does not clearly beat Random is not adding value over blind selection, which is why every acquisition-strategy comparison should include it.

Scores are drawn from a per-round RNG seeded by (seed, round), so results are reproducible for a given seed while decorrelating across rounds.

class alf_tools.optimizer.acquisition_functions.random_acquisition.RandomAcquisition(seed)

Bases: AcquisitionFunction

Random acquisition function — the mandatory “floor” baseline.

Assigns each candidate a uniform random score, completely ignoring the surrogate’s predictions (mean or uncertainty). Any acquisition strategy that does not clearly beat Random is not adding value over blind selection, which is why every comparison should include this floor.

A fresh RNG is derived per round from (seed, state.round_metrics.round) so that draws decorrelate across rounds (no two rounds pick candidates in the same relative order) while remaining fully reproducible for a given seed.