optuna.samplers.NSGAIIISampler

class optuna.samplers.NSGAIIISampler(*, population_size=50, mutation_prob=None, crossover=None, crossover_prob=0.9, swapping_prob=0.5, seed=None, constraints_func=None, reference_points=None, dividing_parameter=3, elite_population_selection_strategy=None, child_generation_strategy=None, after_trial_strategy=None)

Multi-objective sampler using the NSGA-III algorithm.

NSGA-III stands for “Nondominated Sorting Genetic Algorithm III”, which is a modified version of NSGA-II for many objective optimization problem.

Note

When optimizing many objectives, a large fraction of trials may become non-dominated in general due to the curse of dimensionality in the objective space. If possible, consider modeling some objectives as constraints. Constraints can be passed via the constraints_func argument at the sampler initialization. NSGAIISampler, TPESampler, and GPSampler also support constrained multi-objective optimization. Since Bayesian optimization is often sample efficient, it is worth considering TPESampler, or GPSampler for n_trials < 1000.

For further information about NSGA-III, please refer to the following papers:

• An Evolutionary Many-Objective Optimization Algorithm Using Reference-Point-Based Nondominated Sorting Approach, Part I: Solving Problems With Box Constraints

• An Evolutionary Many-Objective Optimization Algorithm Using Reference-Point-Based Nondominated Sorting Approach, Part II: Handling Constraints and Extending to an Adaptive Approach

Parameters:

• reference_points (np.ndarray | None) – A 2 dimension numpy.ndarray with objective dimension columns. Represents a list of reference points which is used to determine who to survive. After non-dominated sort, who out of borderline front are going to survived is determined according to how sparse the closest reference point of each individual is. In the default setting the algorithm uses uniformly spread points to diversify the result. It is also possible to reflect your preferences by giving an arbitrary set of target points since the algorithm prioritizes individuals around reference points.

• dividing_parameter (int) – A parameter to determine the density of default reference points. This parameter determines how many divisions are made between reference points on each axis. The smaller this value is, the less reference points you have. The default value is 3. Note that this parameter is not used when reference_points is not None.

• population_size (int)

• mutation_prob (float | None)

• crossover (BaseCrossover | None)

• crossover_prob (float)

• swapping_prob (float)

• seed (int | None)

• constraints_func (Callable[[FrozenTrial], Sequence[float]] | None)

• elite_population_selection_strategy (Callable[[Study, list[FrozenTrial]], list[FrozenTrial]] | None)

• child_generation_strategy (Callable[[Study, dict[str, BaseDistribution], list[FrozenTrial]], dict[str, Any]] | None)

• after_trial_strategy (Callable[[Study, FrozenTrial, TrialState, Sequence[float] | None], None] | None)

Note

Other parameters than reference_points and dividing_parameter are the same as NSGAIISampler.

Note

Added in v3.2.0 as an experimental feature. The interface may change in newer versions without prior notice. See https://github.com/optuna/optuna/releases/tag/v3.2.0.

Methods

after_trial(study, trial, state, values)	Trial post-processing.
before_trial(study, trial)	Trial pre-processing.
get_parent_population(study, generation)	Get the parent population of the given generation.
get_population(study, generation)	Get the population of the given generation.
get_trial_generation(study, trial)	Get the generation number of the given trial.
infer_relative_search_space(study, trial)	Infer the search space that will be used by relative sampling in the target trial.
reseed_rng()	Reseed sampler's random number generator.
sample_independent(study, trial, param_name, ...)	Sample a parameter for a given distribution.
sample_relative(study, trial, search_space)	Sample parameters in a given search space.
select_parent(study, generation)	Select parent trials from the population for the given generation.

Attributes

population_size

after_trial(study, trial, state, values)

Trial post-processing.

This method is called after the objective function returns and right before the trial is finished and its state is stored.

Note

Added in v2.4.0 as an experimental feature. The interface may change in newer versions without prior notice. See https://github.com/optuna/optuna/releases/tag/v2.4.0.

Parameters:

• study (Study) – Target study object.

• trial (FrozenTrial) – Target trial object. Take a copy before modifying this object.

• state (TrialState) – Resulting trial state.

• values (Sequence[float] | None) – Resulting trial values. Guaranteed to not be None if trial succeeded.

Return type:

None

before_trial(study, trial)

Trial pre-processing.

This method is called before the objective function is called and right after the trial is instantiated. More precisely, this method is called during trial initialization, just before the infer_relative_search_space() call. In other words, it is responsible for pre-processing that should be done before inferring the search space.

Note

Added in v3.3.0 as an experimental feature. The interface may change in newer versions without prior notice. See https://github.com/optuna/optuna/releases/tag/v3.3.0.

Parameters:

• study (Study) – Target study object.

• trial (FrozenTrial) – Target trial object.

Return type:

None

get_parent_population(study, generation)

Get the parent population of the given generation.

This method caches the parent population in the study’s system attributes.

Parameters:

• study (Study) – Target study object.

• generation (int) – Target generation number.

Returns:

List of parent frozen trials. If generation == 0, returns an empty list.

Return type:

list[FrozenTrial]

get_population(study, generation)

Get the population of the given generation.

Parameters:

• study (Study) – Target study object.

• generation (int) – Target generation number.

Returns:

List of frozen trials in the given generation.

Return type:

list[FrozenTrial]

get_trial_generation(study, trial)

Get the generation number of the given trial.

This method returns the generation number of the specified trial. If the generation number is not set in the trial’s system attributes, it will calculate and set the generation number.

The current generation number depends on the maximum generation number of all completed trials.

Parameters:

• study (Study) – Study object which trial belongs to.

• trial (FrozenTrial) – Trial object to get the generation number.

Returns:

Generation number of the given trial.

Return type:

int

infer_relative_search_space(study, trial)

Infer the search space that will be used by relative sampling in the target trial.

This method is called right before sample_relative() method, and the search space returned by this method is passed to it. The parameters not contained in the search space will be sampled by using sample_independent() method.

Parameters:

• study (Study) – Target study object.

• trial (FrozenTrial) – Target trial object. Take a copy before modifying this object.

Returns:

A dictionary containing the parameter names and parameter’s distributions.

Return type:

dict[str, BaseDistribution]

See also

Please refer to intersection_search_space() as an implementation of infer_relative_search_space().

reseed_rng()

Reseed sampler’s random number generator.

This method is called by the Study instance if trials are executed in parallel with the option n_jobs>1. In that case, the sampler instance will be replicated including the state of the random number generator, and they may suggest the same values. To prevent this issue, this method assigns a different seed to each random number generator.

Return type:

None

sample_independent(study, trial, param_name, param_distribution)

Sample a parameter for a given distribution.

This method is called only for the parameters not contained in the search space returned by sample_relative() method. This method is suitable for sampling algorithms that do not use relationship between parameters such as random sampling and TPE.

Note

The failed trials are ignored by any build-in samplers when they sample new parameters. Thus, failed trials are regarded as deleted in the samplers’ perspective.

Parameters:

• study (Study) – Target study object.

• trial (FrozenTrial) – Target trial object. Take a copy before modifying this object.

• param_name (str) – Name of the sampled parameter.

• param_distribution (BaseDistribution) – Distribution object that specifies a prior and/or scale of the sampling algorithm.

Returns:

A parameter value.

Return type:

Any

sample_relative(study, trial, search_space)

Sample parameters in a given search space.

This method is called once at the beginning of each trial, i.e., right before the evaluation of the objective function. This method is suitable for sampling algorithms that use relationship between parameters such as Gaussian Process and CMA-ES.

Note

The failed trials are ignored by any build-in samplers when they sample new parameters. Thus, failed trials are regarded as deleted in the samplers’ perspective.

Parameters:

• study (Study) – Target study object.

• trial (FrozenTrial) – Target trial object. Take a copy before modifying this object.

• search_space (dict[str, BaseDistribution]) – The search space returned by infer_relative_search_space().

Returns:

A dictionary containing the parameter names and the values.

Return type:

dict[str, Any]

select_parent(study, generation)

Select parent trials from the population for the given generation.

This method is called once per generation to select parents from the population of the current generation.

Output of this function is cached in the study system attributes.

This method must be implemented in a subclass to define the specific selection strategy.

Parameters:

• study (Study) – Target study object.

• generation (int) – Target generation number.

Returns:

List of parent frozen trials.

Return type:

list[FrozenTrial]