A single-objective benchmark problem.

Configuration for one objective in a grouped TLP problem.

A multi-objective benchmark problem.

HOLA single-objective adapter with configurable strategy.

Result of a multi-objective optimization run.

Generic pymoo single-objective adapter.

Thread-safe CSV result store with resume support.

Result of a single-objective optimization run.

HOLA multi-objective adapter. Each objective is its own priority group (the group defaults to the field name), so all objectives are Pareto-r

A grouped TLP benchmark problem (HOLA-specific).

Configuration for a benchmark run.

Wraps a MultiObjectiveProblem for pymoo.

Iterative Grid Refinement. Starts with a grid search over the full domain, finds the best point, shrinks the search region around it, and rep

NSGA-II via Optuna for multi-objective optimization.

Tree-structured Parzen Estimator via Optuna.

MOEA/D via pymoo.

NSGA-II via pymoo.

Random search with twice the iteration budget. Delegates to HOLA's random strategy but runs 2 * budget trials. This is a calibration baseline

Wraps a SingleObjectiveProblem for pymoo.

Result of running a child command with bounded output capture.

Output of a single capped stream read: the captured (lossy UTF-8) text and whether the source exceeded MAX_CAPTURE_BYTES and was truncated.

A reader that yields `total` bytes of 'a' without ever blocking, used to exercise the capture cap. Tracks how many bytes were actually consumed so a t

The decision an exec-mode worker makes after a command completes.

A test space with arbitrary dimensionality mapping to a `Vec<f64>`.

A space whose `from_unit_cube` rejects the strategy's raw sample (any vector other than the exact midpoint) but accepts the `vec![0.5; n]` fallback. T

A space whose `from_unit_cube` rejects the strategy's raw sample (here, any vector other than the exact midpoint) but accepts the `vec![0.5; n]` fallb

Interface for multi-objective optimizers.

Extension trait for strategies that can refit their internal distribution from historical trial data. This is separate from `Strategy` to: - Not burd

Defines the structure and validity rules for a hyperparameter search space. Every optimization problem starts with a `SampleSpace` that describes wha

A bijective transformation between a linear internal space and the user-facing actual space. `forward` maps internal → actual; `inverse` maps actual

Interface for single-objective optimizers.

Optional extension of [`SampleSpace`] that provides a bijection to the unit hypercube `[0, 1]^n`. Strategies like [`RandomStrategy`](crate::strategie

A search algorithm that proposes candidate configurations and learns from results. The two associated types wire the strategy into the rest of the sy