alogos._optimization.ea.parameters

Default parameters for an evolutionary algorithm.

Detailed object descriptions

alogos._optimization.ea.parameters.default_parameters

Default parameters for an evolutionary algorithm.

Parameters:

  • population_size (int, default=100) – This is the population size of the main population during initialization and during the run in surivor selection.

  • offspring_size (int, default=100) – This is the size of the offspring population used in parent selection, crossover and mutation.

  • verbose (bool or int, default=False) – Possible values:

    • If False or 0, no output is printed.

    • If True or 1, concise status messages about the run are printed.

    • If 2 or larger, detailed status messages about the run are printed.

  • database_on (bool, default=False) – If True, data created during the run is stored in a SQLite database, which slows down the algorithm but allows later analysis and visualization via the database attribute of the algorithm object.

  • database_location (str, default=”:memory:”) – Filepath for the SQLite database. Special case: ":memory:" leads to the use of a faster but impermanent in-memory database instead of creating a file on disk.

  • generation_model (str, default=”overlapping”) – Choice of the pool of individuals that the survivor selection operator acts on.

    Possible values:

    • "overlapping": Survivors are selected from the combined parent and offspring population.

    • "non_overlapping": Survivors are selected only from the offspring population. The individuals of the parent population are lost if they are not also part of the offspring population.

  • parent_selection_operator (str, default=”truncation”) – Parent selection operator. The available options put a different selection pressure on the individuals of the search and therefore can lead to faster or slower convergence, which in turn influences how likely it is to get stuck in a local optimium.

    Possible values:

    • "uniform": Random choice of individuals independent of their fitness.

    • "truncation": Deterministically uses the best individuals.

    • "tournament": Randomly chooses some number of individuals determined by parent_selection_tournament_size and lets them compete for being selected by simply chosing the one with the best fitness.

    • "rank_proportional": Randomly choose individuals from a probability distribution calculated by the rank of each individual.

    • "fitness_proportional": Randomly choose individuals from a probability distribution calculated by the fitness of each individual. High fitness individuals can have a very high chance of being selected, possibly leading to premature convergence.

  • parent_selection_tournament_size (int, default=3) – Number of individuals sampled in each tournament. A higher value means higher selection pressure, because individuals with low fitness have a smaller chance to win in any tournament.

  • parent_selection_rank_slope (float, default=1.5) – Slope used for linear scaling in rank-proportional selection. A higher value means that individuals with a good rank get a higher probability of being selected and therefore higher selection pressure.

  • survivor_selection_operator (str, default=”tournament”) – Survivor selection operator. See description and options for parent_selection_operator, because the same methods are available for parent and survivor selection, though they may act on a different number of individuals.

  • survivor_selection_tournament_size (int, default=3) – See description for parent_selection_tournament_size.

  • survivor_selection_rank_slope (float, default=1.5) – See description for parent_selection_rank_slope.

  • gen_to_phe_cache_lookup_on (bool, default=True) – If True, results of genotype-to-phenotype mappings are stored in a cache and can be reused instead of being recalculated when a the same genotype appears again.

  • gen_to_phe_cache_size (int, default=200) – Number of entries that can be stored in the LRU cache for genotype-to-phenotype mappings.

  • phe_to_fit_cache_lookup_on (bool, default=True) – If True, results of phenotype-to-fitness mappings are stored in a cache and can be reused instead of being recalculated when a the same phenotype appears again. If the objective function is computationally demanding this can reduce the search time considerably!

  • phe_to_fit_cache_size (int, default=200) – Number of entries that can be stored in the LRU cache for phenotype-to-fitness mappings.

  • phe_to_fit_database_lookup_on (bool, default=True) – If True, results of phenotype-to-fitness mappings are looked up in the database if 1) a database is active and 2) the phenotype is not present in the cache or there is no cache.

  • max_generations (int, default=None) – Stop criterion that halts the search after a given number of generations.

  • max_fitness_evaluations (int, default=None) – Stop criterion that halts the search after a given number of fitness evaluations with the objective function. Note that the actual number of fitness evaluations can be slightly higher, because all individuals of a generation are evaluated before halting.

  • max_runtime_in_seconds (int, default=None) – Stop criterion that halts the search after a given number of seconds.

  • max_or_min_fitness (float, default=None) – Stop criterion that halts the search if the fitness of the best individual exceeds a given threshold. If the objective is minimization, the search halts when the fitness gets lower than the given value. If the objective is maximization, the search halts when the fitness gets higher than the given value.