genetic_optimizer

Optimization using Genetic Algorithms (GA)

The genetic_optimizer module provides optimizers and wrappers for GA.

As for now, I didn’t try a convincing Genetic Algorithm library. This is why it is built-in here. Feel free to propose a copyleft library on GitHub!

Created on Thu Jun 10 21:20:47 2021.

@author: HugoFara

class leggedsnake.genetic_optimizer.GeneticOptimization(dna: list[Any], fitness: Callable[[...], tuple[float, list[tuple[float, float]]]], prob: float = 0.07, **kwargs: Any)

Bases: object

__init__(dna: list[Any], fitness: Callable[[...], tuple[float, list[tuple[float, float]]]], prob: float = 0.07, **kwargs: Any) → None

Parameters:

dna (list)
fitness (callable)
prob (float or tuple[float])
kwargs (dict) –
Other useful parameters for the optimization.

max_popint, default=11
Maximum number of individuals. The default is 11.

max_genetic_distfloat, default=.7
Maximum genetic distance, before individuals cannot reproduce (separated species). The default is .7.

startnstopbool, default=False
Ability to close program without loosing population. If True, we verify at initialization the existence of a data file. Population is saved every int(250 / max_pop) iterations. The default is False.

fitness_argstuple
Keyword arguments to send to the fitness function. The default is None (no argument sent).

verboseint
Level of verbosity. 0: no verbose, do not print anything. 1: show a progress bar. 2: complete report for each turn. The default is 1.

birth(par1: list[Any], par2: list[Any]) → list[Any]

Return a new individual with par1 and par2 as parents (two sequences).

Child are generated by a uniform crossover followed by a random “resetting” mutation of each gene. The resetting is a normal law.

Initial positions come from one of the two parents randomly.

Parameters:

par1 (list[float, tuple of float, tuple of tuple of float]) – Dna of first parent.
par2 (list[float, tuple of float, tuple of tuple of float]) – Dna of second parent.

Returns:

child – Dna of the child.

Return type:

list[float, tuple of float, tuple of tuple of float]

dna: list[Any]

evaluate_individual(dna: list[Any], fitness_args: tuple[Any, ...] | None) → tuple[float, list[tuple[float, float]]]

Simple evaluation for a single individual.

Parameters:

dna (list[float, tuple of float, tuple of tuple of float]) – List of the individuals’ DNAs
fitness_args (tuple) – Additional arguments to pass to the fitness function. Usually the initial positions of the joints.

Returns:

Score then initial coordinates.

Return type:

tuple

See also

evaluate_individual: same function but on a single DNA.

fitness: Callable[[...], tuple[float, list[tuple[float, float]]]]

iters: int

kwargs: dict[str, Any]

make_children(parents: list[list[Any]], max_genetic_dist: float = inf) → list[list[Any]]

max_pop: int

pop: list[list[Any]]

prob: float

reduce_population() → list[list[Any]]

Reduce the population down to max_pop.

Returns:: new_population – At most self.max_pop individuals, sorted by score.
Return type:: list of dna

run(iters: int, processes: int = 1) → list[Agent]

Optimization by genetic algorithm (GA).

Parameters:

iters (int) – Number of iterations.
processes (int, default=1) – Number of processes that will evaluate the linkages.

Returns:

List of Agent(score, dimensions, init_positions) sorted by score in descending order. Compatible with pylinkage’s chain_optimizers pipeline.

Return type:

list[Agent]

select_parents(verbose: bool = True) → list[list[Any]]: Selection 1/4 of the population as parents.

startnstop: str | bool

verbosity: int

leggedsnake.genetic_optimizer.agents_to_ensemble(agents: Sequence[Agent], linkage: Any) → Ensemble

Wrap a list of Agents in a pylinkage Ensemble.

Provides .rank(), .top(), .filter(), .filter_by_score() and numpy-style indexing over optimization results. The template linkage is only used for topology metadata; batch simulation via Ensemble.simulate() is not supported for Walker-based mechanisms.

Parameters:

agents (sequence of Agent) – Optimization results (e.g. from genetic_algorithm_optimization).
linkage (Walker, Mechanism, or Linkage) – Template. A Walker is converted to its underlying Mechanism.

Returns:

One member per agent, with scores={"score": agent.score}.

Return type:

Ensemble

leggedsnake.genetic_optimizer.genetic_algorithm_optimization(eval_func: ~typing.Callable[[...], float], linkage: ~typing.Any, center: ~collections.abc.Sequence[float] | None = None, bounds: tuple[~collections.abc.Sequence[float], ~collections.abc.Sequence[float]] | None = None, order_relation: ~typing.Callable[[float, float], float] = <built-in function max>, max_pop: int = 30, iters: int = 100, prob: float = 0.07, max_genetic_dist: float = 10.0, processes: int = 1, startnstop: str | bool = False, verbose: bool = True, **kwargs: ~typing.Any) → Ensemble

Genetic algorithm optimization with the standard pylinkage interface.

This wrapper bridges leggedsnake’s GeneticOptimization to pylinkage’s optimizer contract, making it usable with chain_optimizers and interchangeable with PSO, DE, etc.

The evaluation function receives (linkage, dimensions, init_positions) and returns a scalar score — exactly like pylinkage optimizers.

Parameters:

eval_func (callable) – Evaluation function with signature (linkage, dimensions, init_positions) -> float.
linkage (Walker or Linkage) – The mechanism to optimize. Must provide get_constraints(), set_constraints(), get_coords(), set_coords().
center (sequence of float, optional) – Initial dimensions. If None, read from linkage. chain_optimizers injects the previous stage’s best here.
bounds (tuple of (lower, upper), optional) – Not directly used by the GA, but accepted for API compatibility. When provided, initial random children are clamped to these bounds.
order_relation (callable, optional) – max (default) for maximization, min for minimization.
max_pop (int) – Maximum population size. Default 30.
iters (int) – Number of generations. Default 100.
prob (float) – Mutation standard deviation. Default 0.07.
max_genetic_dist (float) – Speciation threshold. Default 10.0.
processes (int) – Number of parallel processes for evaluation. Default 1.
startnstop (str or bool) – Path to checkpoint file, or False to disable. Default False.
verbose (bool) – Show progress bar. Default True.

Returns:

Population wrapped in a pylinkage Ensemble (one member per candidate, columnar scores). Ranking is already applied: the member at index 0 is the best under order_relation. Iterate, slice, or call .top() / .rank() / .filter_by_score() to drill down. Use ensemble[i].score / .dimensions / .initial_positions to access fields — or call ensemble[i].to_agent() for the legacy tuple shape.

Return type:

Ensemble

leggedsnake.genetic_optimizer.kwargs_switcher(arg_name: str, kwargs: dict[str, Any], default: Any = None) → Any: Simple function to return the good element from a kwargs dict.

leggedsnake.genetic_optimizer.load_population(file_path: str) → list[list[Any]]: Return a population from a given file.

leggedsnake.genetic_optimizer.save_population(file_path: str, population: list[list[Any]], verbose: bool = False, data_descriptors: dict[str, Any] | None = None) → None

Save the population to a json file.

Parameters:

file_path (str) – Path of the file to write to.
population (list) – Sequence of dna
verbose (bool) – Enable or not verbosity (outputs success).
data_descriptors (dict) – Any additional value you want to save for the current generation.