optilab.optimizers package

Submodules

optilab.optimizers.cma_es module

CMA-ES optimizer: Covariance Matrix Adaptation Evolution Strategy.

class CmaEs(population_size: int, sigma0: float)[source]

Bases: Optimizer

CMA-ES optimizer: Covariance Matrix Adaptation Evolution Strategy.

__init__(population_size: int, sigma0: float)[source]

Class constructor.

Parameters:

population_size – Size of the population.
sigma0 – Starting value of the sigma,

static _stop_internal(es: CMAEvolutionStrategy) → bool[source]

Check an instance of CMA-ES optimizer for internal stop criteria.

Parameters:

es – An instance of CMA-ES optimizer.

Returns:

If true, the internal stop criteria of CMA-ES have been reached and optimization: should be ended.

static _spawn_cmaes(bounds: Bounds, dim: int, population_size: int, sigma0: float) → CMAEvolutionStrategy[source]

Create a new instance of cma optimizer.

Parameters:

bounds – The bounds of the search area.
dim – The dimensionality of the search area.

Returns:

A new cma optimizer instance.

_stop(es: CMAEvolutionStrategy, log: PointList, population_size: int, call_budget: int, target: float, tolerance: float) → bool[source]

Decide if the optimization should be stopped.

Parameters:

es – CMA-ES instance.
log – Results log.
population_size – The size of the population in one generation.
call_budget – Maximum number of optimized function calls.
target – Global minimum value of the optimized function.
tolerance – Tolerated error value of the optimization.

Returns:

True if the optimization should be stopped.

optimize(function: ObjectiveFunction, bounds: Bounds, call_budget: int, tolerance: float, target: float = 0.0) → PointList[source]

Run a single optimization of provided objective function.

Parameters:

function – Objective function to optimize.
bounds – Search space of the function.
call_budget – Max number of calls to the objective function.
tolerance – Tolerance of y value to count a solution as acceptable.
target – Objective function value target, default 0.

Returns:

Results log from the optimization.

optilab.optimizers.ipop_cma_es module

IPOP-CMA-ES optimizer: CMA-ES with increasing population restarts.

class IpopCmaEs(population_size: int)[source]

Bases: CmaEs

IPOP-CMA-ES optimizer: CMA-ES with increasing population restarts.

optimize(function: ObjectiveFunction, bounds: Bounds, call_budget: int, tolerance: float, target: float = 0.0) → PointList[source]

Run a single optimization of provided objective function.

Parameters:

function – Objective function to optimize.
bounds – Search space of the function.
call_budget – Max number of calls to the objective function.
tolerance – Tolerance of y value to count a solution as acceptable.
target – Objective function value target, default 0.

Returns:

Results log from the optimization.

optilab.optimizers.knn_cma_es module

KNN-CMA-ES optimizer. CMA-ES is enhanced with a KNN metamodel similar to the one from LMM-CMA-ES.

class KnnCmaEs(population_size: int, sigma0: float, num_neighbors: int, buffer_size: int)[source]

Bases: CmaEs

KNN-CMA-ES optimizer. CMA-ES is enhanced with a KNN metamodel similar to the one from LMM-CMA-ES.

__init__(population_size: int, sigma0: float, num_neighbors: int, buffer_size: int)[source]

Class constructor.

Parameters:

population_size – Size of the population.
sigma0 – Starting value of the sigma,
num_neighbors – Number of neighbors used by KNN metamodel.
buffer_size – Number of last evaluated points provided to KNN metamodel.

optimize(function: ObjectiveFunction, bounds: Bounds, call_budget: int, tolerance: float, target: float = 0.0) → PointList[source]

Run a single optimization of provided objective function.

Parameters:

function – Objective function to optimize.
bounds – Search space of the function.
call_budget – Max number of calls to the objective function.
tolerance – Tolerance of y value to count a solution as acceptable.
target – Objective function value target, default 0.

Returns:

Results log from the optimization.

optilab.optimizers.knn_ipop_cma_es module

KNN-IPOP-CMA-ES optimizer. IPOP-CMA-ES is enhanced with a KNN metamodel similar to the one from LMM-CMA-ES.

class KnnIpopCmaEs(population_size: int, num_neighbors: int, buffer_size: int)[source]

Bases: CmaEs

KNN-IPOP-CMA-ES optimizer: CMA-ES with increasing population restarts and with KNN metamodel similar to LMM-CMA-ES.

__init__(population_size: int, num_neighbors: int, buffer_size: int)[source]

Class constructor.

Parameters:

population_size – Starting size of the population.
num_neighbors – Number of neighbors used by KNN metamodel.
buffer_size – Number of last evaluated points provided to KNN metamodel.

optimize(function: ObjectiveFunction, bounds: Bounds, call_budget: int, tolerance: float, target: float = 0.0) → PointList[source]

Run a single optimization of provided objective function.

Parameters:

function – Objective function to optimize.
bounds – Search space of the function.
call_budget – Max number of calls to the objective function.
tolerance – Tolerance of y value to count a solution as acceptable.
target – Objective function value target, default 0.

Returns:

Results log from the optimization.

optilab.optimizers.lmm_cma_es module

LMM-CMA-ES optimizer: CMA-ES with local polynomial regression metamodel.

class LmmCmaEs(population_size: int, sigma0: float, polynomial_dim: int)[source]

Bases: CmaEs

LMM-CMA-ES optimizer: CMA-ES with local polynomial regression metamodel.

__init__(population_size: int, sigma0: float, polynomial_dim: int)[source]

Class constructor.

Parameters:

population_size – Size of the population.
sigma0 – Starting value of the sigma,
polynomial_dim – Dimension of the polynomial regression.

optimize(function: ObjectiveFunction, bounds: Bounds, call_budget: int, tolerance: float, target: float = 0.0) → PointList[source]

Run a single optimization of provided objective function.

Parameters:

function – Objective function to optimize.
bounds – Search space of the function.
call_budget – Max number of calls to the objective function.
tolerance – Tolerance of y value to count a solution as acceptable.
target – Objective function value target, default 0.

Returns:

Results log from the optimization.

optilab.optimizers.lmm_ipop_cma_es module

LMM-IPOP-CMA-ES optimizer: IPOP-CMA-ES with local polynomial regression metamodel. It’s a fusion of IPOP-CMA-ES and LMM-CMA-ES.

class LmmIpopCmaEs(population_size: int, polynomial_dim: int)[source]

Bases: CmaEs

LMM-IPOP-CMA-ES optimizer: IPOP-CMA-ES with local polynomial regression metamodel. It’s a fusion of IPOP-CMA-ES and LMM-CMA-ES.

__init__(population_size: int, polynomial_dim: int)[source]

Class constructor.

Parameters:

population_size – Size of the population.
polynomial_dim – Dimension of the polynomial regression.

optimize(function: ObjectiveFunction, bounds: Bounds, call_budget: int, tolerance: float, target: float = 0.0) → PointList[source]

Run a single optimization of provided objective function.

Parameters:

function – Objective function to optimize.
bounds – Search space of the function.
call_budget – Max number of calls to the objective function.
tolerance – Tolerance of y value to count a solution as acceptable.
target – Objective function value target, default 0.

Returns:

Results log from the optimization.

optilab.optimizers.optimizer module

Base class for optimizers.

class Optimizer(name: str, population_size: int, hyperparameters: Dict[str, Any])[source]

Bases: object

Base class for optimizers.

__init__(name: str, population_size: int, hyperparameters: Dict[str, Any]) → None[source]

Class constructor.

Parameters:

name – Name of this optimizer.
population_size – Size of the population.
hyperparameters – Dictionary with the hyperparameters of the optimizer.

static _stop_budget(log: PointList, population_size: int, call_budget: int) → bool[source]

Check if the call budget will allow for another algorithm generation.

Parameters:

log – Log with all points evaluated so far.
population_size – Number of points in a single generation.
call_budget – Number of allowed point evaluations.

Returns:

If true, the call budget has been expended and another generation: cannot be evaluated.

static _stop_target_found(log: PointList, target: float, tolerance: float) → bool[source]

Check if the optimal function value has been found.

Parameters:

log – Log with all points evaluated so far.
target – Global optimum value of the optimized function.
tolerance – Allowed error value from global optimum value.

Returns:

If true, the global optimum has been found.

_stop_external(log: PointList, population_size: int, call_budget: int, target: float, tolerance: float) → bool[source]

Check if the external stop criteria has been met, i.e. if the budget has been expended or the the global optimum has been found.

Parameters:

log – Log with all points evaluated so far.
population_size – Number of points in a single generation.
call_budget – Number of allowed point evaluations.
target – Global optimum value of the optimized function.
tolerance – Allowed error value from global optimum value.

Returns:

If true, the external stop criteria has been met and the optimization: should be stopped.

optimize(function: ObjectiveFunction, bounds: Bounds, call_budget: int, tolerance: float, target: float = 0.0) → PointList[source]

Run a single optimization of provided objective function.

Parameters:

function – Objective function to optimize.
bounds – Search space of the function.
call_budget – Max number of calls to the objective function.
tolerance – Tolerance of y value to count a solution as acceptable.
target – Objective function value target, default 0.

Returns:

Results log from the optimization.

run_optimization(num_runs: int, function: ObjectiveFunction, bounds: Bounds, call_budget: int, tolerance: float, target: float = 0.0, *, num_processes: int = 1) → OptimizationRun[source]

Optimize a provided objective function.

Parameters:

num_runs – Number of optimization runs to perform.
function – Objective function to optimize.
bounds – Search space of the function.
call_budget – Max number of calls to the objective function.
tolerance – Tolerance of y value to count a solution as acceptable.
target – Objective function value target, default 0.
num_processes (int) – Number of concurrent processes to use to speed up the optimization. By default only one is used.

Returns:

Metadata of optimization run.

optilab.optimizers.top_half_knn_ipop_cma_es module

Top-half KNN-IPOP-CMA-ES optimizer.

IPOP-CMA-ES enhanced with a KNN-based top-half metamodel: each generation, the surrogate estimates all candidates and only the best half is evaluated with the real objective function.

class TopHalfKnnIpopCmaEs(population_size: int, num_neighbors: int, buffer_size: int)[source]

Bases: CmaEs

Top-half KNN-IPOP-CMA-ES optimizer.

Uses the TopHalfMetamodel with a KNN surrogate: every generation the surrogate pre-screens all lambda candidates and only the best mu = lambda / 2 are evaluated with the real objective. Penalised values for the remaining candidates guarantee that CMA-ES updates its state exclusively from real evaluations.

__init__(population_size: int, num_neighbors: int, buffer_size: int)[source]

Class constructor.

Parameters:

population_size – Starting size of the population.
num_neighbors – Number of neighbours used by the KNN surrogate.
buffer_size – Number of last evaluated points provided to the KNN surrogate for training.

optimize(function: ObjectiveFunction, bounds: Bounds, call_budget: int, tolerance: float, target: float = 0.0) → PointList[source]

Run a single optimisation of the provided objective function.

Parameters:

function – Objective function to optimise.
bounds – Search space of the function.
call_budget – Max number of calls to the objective function.
tolerance – Tolerance of y value to accept a solution.
target – Objective function value target, default 0.

Returns:

Results log from the optimisation.

Module contents

Submodule containing optimizers.