tauopt.engine.strategies.random_

Uniform-random strategy – the robustness ladder’s floor.

Deliberately stdlib-only (requires no extras): the never-fail guarantee (pin P6) rests on this rung being importable and crash-proof in ANY environment. Linear constraints are honored by rejection (a capped scan; an effectively-empty feasible region raises honestly – experiment creation already probes for that).

Module Contents

Classes

RandomStrategy

Uniform-random strategy (the module docstring carries the floor contract). The final rung of both never-fail ladders (pin P6), never routed directly by the default policy; stdlib-only, so it requires no extras and stays importable in any environment.

API

class tauopt.engine.strategies.random_.RandomStrategy

Uniform-random strategy (the module docstring carries the floor contract). The final rung of both never-fail ladders (pin P6), never routed directly by the default policy; stdlib-only, so it requires no extras and stays importable in any environment.

name

‘random’

requires: tuple[str, ...]

()

accepted_params: frozenset[str]

‘frozenset(…)’

suggest(state: tauopt.engine.state.ExperimentState, cfg: tauopt.engine.config.OptimizerConfig, n: int = 1) → list[tauopt.engine.strategies.base.SuggestionDraft]

Return the next n decoded, linearly-feasible uniform draws from a stdlib random.Random seeded from (state, cfg), with the state’s draw counts mixed in so a fixed explicit seed still progresses across asks (pin P4). Raises ValueError when linear constraints reject 100,000 consecutive draws (feasible region too small to sample).

suggest_unique(state: tauopt.engine.state.ExperimentState, cfg: tauopt.engine.config.OptimizerConfig, n: int = 1, exclude: frozenset[str] = frozenset()) → list[tauopt.engine.strategies.base.SuggestionDraft]

The dedup repair path: advance to the next draw past collisions, at most SEQUENCE_REPAIR_FACTOR * n examined candidates.