Skip to content

Optimizer

Optimizer state and info dataclasses.

BudgetInfo dataclass #

BudgetInfo(
    max_cost_total: float | None = None,
    used_cost_budget: float = 0.0,
    max_evaluations: int | None = None,
    used_evaluations: int = 0,
)

Information about the budget of an optimizer.

clone #

clone() -> BudgetInfo

Create a copy of the budget info.

Source code in neps/state/optimizer.py 
21
22
23
def clone(self) -> BudgetInfo:
    """Create a copy of the budget info."""
    return replace(self)

OptimizationState dataclass #

OptimizationState(
    budget: BudgetInfo | None,
    seed_snapshot: SeedSnapshot,
    shared_state: dict[str, Any] | None,
)

The current state of an optimizer.

budget instance-attribute #

budget: BudgetInfo | None

Information regarind the budget used by the optimization trajectory.

seed_snapshot instance-attribute #

seed_snapshot: SeedSnapshot

The state of the random number generators at the time of the last sample.

shared_state instance-attribute #

shared_state: dict[str, Any] | None

Any information the optimizer wants to store between calls to sample and post evaluations.

For example, an optimizer may wish to store running totals here or various other bits of information that may be expensive to recompute.

Right now there's no support for tensors/arrays and almost no optimizer uses this feature. Only cost-cooling uses information out of .budget.

Please reach out to @eddiebergman if you have a use case for this so we can make it more robust.