saealib.GA#

class saealib.GA(crossover, mutation, parent_selection, survivor_selection, *, duplicate_elimination=None, variation_execution='batch', integer_crossover=None, integer_mutation=None, categorical_crossover=None, categorical_mutation=None)[source]#

Bases: AskTellAlgorithm

Genetic Algorithm class.

Parameters:

crossover (Crossover)
mutation (Mutation)
parent_selection (ParentSelection)
survivor_selection (SurvivorSelection)
duplicate_elimination (DuplicateElimination | None)
variation_execution (Literal['batch', 'sequential'] | dict[str, Literal['batch', 'sequential']])
integer_crossover (Crossover | None)
integer_mutation (Mutation | None)
categorical_crossover (Crossover | None)
categorical_mutation (Mutation | None)

crossover#

Crossover operator for continuous dimensions.

Type:: Crossover

mutation#

Mutation operator for continuous dimensions.

Type:: Mutation

parent_selection#

Parent selection operator.

Type:: ParentSelection

survivor_selection#

Survivor selection operator.

Type:: SurvivorSelection

integer_crossover#

Crossover operator for integer dimensions.

Type:: Crossover

integer_mutation#

Mutation operator for integer dimensions.

Type:: Mutation

categorical_crossover#

Crossover operator for categorical dimensions.

Type:: Crossover

categorical_mutation#

Mutation operator for categorical dimensions.

Type:: Mutation

duplicate_elimination#

When set, offspring that duplicate any member of the current population are replaced by re-generated candidates (up to max_retries attempts). None disables duplicate elimination (default behaviour).

Type:: DuplicateElimination or None

variation_execution#

Execution mode for crossover and mutation. A string applies to both; a dict may set "crossover" and "mutation" independently, with omitted keys defaulting to "batch". In batch mode, all gate decisions and operations complete before any post-operation hook runs, so a stateful hook cannot influence a later individual’s operation. Sequential mode completes each individual’s operation and hook before starting the next, preserving the pre-batching behavior byte-for-byte. Mixed-variable problems now use per-type batch operations by default, which changes their historical random-number sequence; select sequential mode to preserve it. This is a semantic choice, not only a performance setting.

Type:: {“batch”, “sequential”} or dict

Methods

`__init__`	Initialize GA (Genetic Algorithm) class.
`ask`	Generate offspring via crossover and mutation.
`contract`	Return the genetic-algorithm contract.
`create_pareto_archive`	Create a ParetoArchive with the correct direction for the problem.
`get_required_attrs`	Return algorithm-specific attributes (GA needs none beyond the defaults).
`tell`	Update the population using (μ+λ) survivor selection.

Method Details

GA.__init__(crossover, mutation, parent_selection, survivor_selection, *, duplicate_elimination=None, variation_execution='batch', integer_crossover=None, integer_mutation=None, categorical_crossover=None, categorical_mutation=None)[source]#

Initialize GA (Genetic Algorithm) class.

Parameters:

crossover (Crossover) – Crossover operator for continuous dimensions.
mutation (Mutation) – Mutation operator for continuous dimensions.
parent_selection (ParentSelection) – Parent selection operator.
survivor_selection (SurvivorSelection) – Survivor selection operator.
duplicate_elimination (DuplicateElimination, optional) – When provided, offspring that duplicate any member of the current population are replaced by re-generated candidates. None (default) disables duplicate elimination.
variation_execution ({"batch", "sequential"} or dict, optional) – Execution mode for crossover and mutation. A string applies to both operators. A dict may contain "crossover" and "mutation" independently; an omitted key defaults to "batch". In batch mode, every individual’s gate decision and operation is computed before any post_crossover or post_mutation hook runs for any individual in that call. A stateful hook may observe earlier hooks in the subsequent post-processing pass, but cannot influence a later individual’s operation. In sequential mode, each individual’s complete operation-then-hook cycle finishes before the next begins, matching the library’s pre-batching behavior byte-for-byte. This is a semantic choice, not only a performance setting. Mixed-variable problems use per-type batch operations by default, changing their historical random-number sequence; use sequential mode to preserve it.
integer_crossover (Crossover, optional) – Crossover operator for integer dimensions. Defaults to CrossoverIntegerSBX with the same rate as crossover.
integer_mutation (Mutation, optional) – Mutation operator for integer dimensions. Defaults to MutationIntegerUniform with the same rate as mutation.
categorical_crossover (Crossover, optional) – Crossover operator for categorical dimensions. Defaults to CrossoverCategorical with the same rate as crossover.
categorical_mutation (Mutation, optional) – Mutation operator for categorical dimensions. Defaults to MutationCategorical with the same rate as mutation.

GA.ask(request, state)[source]#

Generate offspring via crossover and mutation.

Parameters:

request (ProposalRequest) – Request-specific offspring count.
state (StateView) – Read-only algorithm state view.

Return type:

ProposalBatch

GA.contract()[source]#

Return the genetic-algorithm contract.

Return type:: ComponentContract

GA.create_pareto_archive(attrs, init_capacity, problem)#

Create a ParetoArchive with the correct direction for the problem.

Parameters:

attrs (list[PopulationAttribute])
init_capacity (int)
problem (Problem)

Return type:

ParetoArchive

GA.get_required_attrs(problem)[source]#

Return algorithm-specific attributes (GA needs none beyond the defaults).

Parameters:: problem (Problem)
Return type:: list[PopulationAttribute]

GA.tell(feedback, state)[source]#

Update the population using (μ+λ) survivor selection.

Parameters:

feedback (FeedbackBatch) – Feedback delivered for the current proposal.
state (StateView) – Read-only algorithm state view.

Return type:

StatePatch