1.3. Selection

Table of Contents

• Selection

  • Custom functions

  • Parent selection

    • Tournament selection

    • Random selection

    • Top N selection (elites)

    • Roulette selection

  • Parent combination

    • Two parents

    • Two parents shuffled

    • N-gram parents

  • Survivor selection

    • All (default)

    • Random survivors

Simple functions for encapsulating performing population selection.

1.3.1. Custom functions

Custom selection functions can be used in islands.

def selection_random(individuals : list, n : int = 4, island=None) -> list:
   return random.choice(individuals, size=n).tolist()

Note the following in the above example:

• The function takes a list of individuals

• The function returns a list of individuals

• The functions takes the island as a param

• The custom parameters have default values defined

• All attributes and methods of the Island object are accessible

Important: Parent selection functions yield lists of individuals instead of returning them.

def selection_parents_two(individuals : list, n : int = 4, island=None) -> list:
   for parent_1, parent_2 in zip(individuals[::2], individuals[1::2]):
      yield [parent_1, parent_2]

Survivor selection is similar to elite selection.

def selection_survivors_random(individuals : list, n : int = 4, island=None) -> list:
   return random.choice(individuals, size=n).tolist()

1.3.2. Parent selection

1.3.2.1. Tournament selection

natural_selection.genetic_algorithms.operators.selection.selection_tournament(individuals: list, n: int = 4, tournament_size: int = 5, with_replacement: bool = True, island=None) → list

Classic tournament selection. Given a number of selection rounds (n), select a random list of individuals of tournament_size and select the top individual from the random selection.

Parameters

• individuals (list) – A list of Individuals.

• n (int) – The number of tournaments to run, effectively the number of selected individuals to return (default = 4).

• tournament_size (int) – The number of random individuals to select during each tournament (default = 5).

• with_replacement (bool) – Whether the sample is with or without replacement (default = True).

• island (Island) – The Island calling the method (default = None).

Returns

Top n Individuals from tournaments.

Return type

list

natural_selection.genetic_algorithms.operators.selection.selection_tournament_unique(individuals: list, n: int = 4, tournament_size: int = 5, with_replacement: bool = True, max_step: int = 100, island=None) → list

Classic tournament selection but ensures a unique list of selected individuals. Given a number of selection rounds (n), select a random list of individuals of tournament_size and select the top individual from the random selection.

Parameters

• individuals (list) – A list of Individuals.

• n (int) – The number of tournaments to run, effectively the number of selected individuals to return (default = 4).

• tournament_size (int) – The number of random individuals to select during each tournament (default = 5).

• max_step (int) – In the unlikely event that a unique list of size n can not be achieved, break out of the loop after this amount of steps (default = 100).

• with_replacement (bool) – Whether the sample is with or without replacement (default = True).

• island (Island) – The Island calling the method (default = None).

Returns

Top n Individuals from tournaments.

Return type

list

1.3.2.2. Random selection

natural_selection.genetic_algorithms.operators.selection.selection_random(individuals: list, n: int = 4, with_replacement: bool = True, island=None) → list

Completely random selection.

Parameters

• individuals (list) – A list of Individuals.

• n (int) – Number to select (default = 4).

• with_replacement (bool) – Whether the sample is with or without replacement (default = True).

• island (Island) – The Island calling the method (default = None).

Returns

Random n Individuals.

Return type

list

1.3.2.3. Top N selection (elites)

natural_selection.genetic_algorithms.operators.selection.selection_elites_top_n(individuals: list, n: int = 4, desc: bool = True, island=None) → list

A Classic top N selection function, sorted on fitness.

Parameters

• individuals (list) – A list of Individuals.

• n (int) – Number to select (default = 4).

• desc (bool) – In descending order, only used if Island is None, else maximise_function overrides (default = True).

• island (Island) – The Island calling the method (default = None).

Returns

Top n Individuals.

Return type

list

1.3.2.4. Roulette selection

natural_selection.genetic_algorithms.operators.selection.selection_roulette(individuals: list, n: int = 4, with_replacement: bool = True, island=None) → list

Classic roulette wheel selection, or also known as Fitness proportionate selection. This method selects individuals based on their fitness in proportion to the whole population.

Note

Can only be used for function maximisation.

Parameters

• individuals (list) – A list of Individuals.

• n (int) – The number of selected individuals to return (default = 4).

• with_replacement (bool) – Whether the sample is with or without replacement (default = True).

• island (Island) – The Island calling the method (default = None).

Raises

ValueError – If the island is set for function minimisation.

Returns

n Individuals from roulette selection.

Return type

list

natural_selection.genetic_algorithms.operators.selection.selection_almost_roulette_minimisation(individuals: list, n: int = 4, with_replacement: bool = True, island=None) → list

Almost roulette wheel selection selection but for function minimisation. This method selects individuals based on their fitness in proportion to the whole population.

Note

Can only be used for function minimisation.

Parameters

• individuals (list) – A list of Individuals.

• n (int) – The number of selected individuals to return (default = 4).

• with_replacement (bool) – Whether the sample is with or without replacement (default = True).

• island (Island) – The Island calling the method (default = None).

Raises

ValueError – If the island is set for function maximisation.

Returns

n Individuals from roulette selection.

Return type

list

1.3.3. Parent combination

1.3.3.1. Two parents

natural_selection.genetic_algorithms.operators.selection.selection_parents_two(individuals: list, island=None)

Simple function to select two parents at a time, sequentially. Parental selection always yields.

Parameters

• individuals (list) – A list of Individuals, specifically the selected “elites”.

• island (Island) – The Island calling the method (default = None).

Yields

list – Containing the two individuals selected for crossover.

1.3.3.2. Two parents shuffled

natural_selection.genetic_algorithms.operators.selection.selection_parents_two_shuffled(individuals: list, island=None)

Simple function to select two parents at a time, randomly shuffled. Parental selection always yields.

Parameters

• individuals (list) – A list of Individuals, specifically the selected “elites”.

• island (Island) – The Island calling the method (default = None).

Yields

list – Containing the two individuals selected for crossover.

1.3.3.3. N-gram parents

natural_selection.genetic_algorithms.operators.selection.selection_parents_n_gram(individuals: list, n: int = 2, island=None)

Simple function to select two parents at a time, sequentially, and yields n individuals at a time.

Yielded result is a n-gram generated result. The list l = [a,b,c,d] with n = 2 will yield [a,b], [b,c], [c,d].

Parameters

• individuals (list) – A list of Individuals, specifically the selected “elites”.

• n (int) – N size of grams (default = 2).

• island (Island) – The Island calling the method (default = None).

Yields

list – Containing n individuals selected for crossover.

1.3.4. Survivor selection

1.3.4.1. All (default)

natural_selection.genetic_algorithms.operators.selection.selection_survivors_all(individuals: list, island=None) → list

Simply passes on the individuals as is.

Parameters

• individuals (list) – A list of Individuals.

• island (Island) – The Island calling the method (default = None).

Returns

All individuals.

Return type

list

1.3.4.2. Random survivors

natural_selection.genetic_algorithms.operators.selection.selection_survivors_random(individuals: list, n: int = 4, island=None) → list

Selects a list of random individuals to survive.

Parameters

• individuals (list) – A list of Individuals.

• n (int) – Number of survivors (default = 4).

• island (Island) – The Island calling the method (default = None).

Returns

Surviving individuals.

Return type

list