pdmlabs.optimization.mango_adapter

Mango optimizer adapters for PdMLabs.

Provides MangoAdapter (Bayesian, default) and MangoRandomAdapter (pure random search), both wrapping the built-in pdmlabs.mango.Tuner. The @scheduler.parallel decorator is applied inside the adapter so experiment execute() methods define a raw (**params) -> float objective with no Mango-specific decoration.

Classes

MangoAdapter()	Adapter for the built-in Mango Bayesian optimizer.
MangoRandomAdapter()	Adapter for Mango in pure random-search mode.

class pdmlabs.optimization.mango_adapter.MangoAdapter

Bases: BaseOptimizerAdapter

Adapter for the built-in Mango Bayesian optimizer.

Wraps pdmlabs.mango.Tuner with optimizer='Bayesian' (Gaussian Process surrogate + UCB acquisition).

maximize(param_space: dict, objective_fn, n_iterations: int, n_jobs: int, initial_random: int, constraint_fn=None) → dict

Maximize via Mango, applying scheduler.parallel internally.

minimize(param_space: dict, objective_fn, n_iterations: int, n_jobs: int, initial_random: int, constraint_fn=None) → dict

Minimize via Mango’s native tuner.minimize().

supports_categorical: bool = True

class pdmlabs.optimization.mango_adapter.MangoRandomAdapter

Bases: MangoAdapter

Adapter for Mango in pure random-search mode.

Sets optimizer='Random' in the Mango conf_dict, bypassing the Gaussian Process surrogate and sampling hyperparameters uniformly.