Source code for metaheuristic_designer.initializers.composite_initializer

from typing import Iterable, Optional
from copy import copy

import numpy as np
from ..initializer import Initializer
from ..encoding import Encoding
from ..population import Population
from ..utils import IntVector, RealVector, RNGLike, VectorLike




class CompositeInitializer(Initializer):
    def __init__(
        self,
        dimension: int,
        initializers: Iterable[Initializer],
        weights: Optional[RealVector] = None,
        population_size: Optional[int] = None,
        encoding: Encoding = None,
        rng: Optional[RNGLike] = None,
    ):
        if population_size is None:
            population_size = initializers[0].population_size

        super().__init__(dimension, population_size, encoding=encoding, rng=rng)

        initializers = copy(initializers)
        for inits in initializers:
            inits.population_size = population_size

        self.initializers = initializers
        if weights is None:
            weights = np.ones(len(initializers))
        else:
            weights = np.asarray(weights)
        weights = weights / np.sum(weights)
        self.weights = weights



    def generate_random(self) -> VectorLike:
        """Generate a random individual from one initializers chosen at random.

        Returns
        -------
        VectorLike
            A 1-D array generated by the fallback initializer.
        """

        idx = self.rng.choice(len(self.initializers), p=self.weights)
        return self.initializers[idx].generate_random()




    def generate_individual(self) -> VectorLike:
        """Generate an individual from one of the initializers chosen at random.

        Returns
        -------
        ndarray
            A 1-D array representing the individual.
        """

        idx = self.rng.choice(len(self.initializers), p=self.weights)
        return self.initializers[idx].generate_individual()




    def generate_population(self, n_individuals=None):
        """Generate a population from individuals chosen at random from the initializers.

        Parameters
        ----------
        objfunc: ObjectiveFunc
            Objective function that will be propagated to each individual.
        n_individual: int, optional
            Number of individuals to generate

        Returns
        -------
        generated_population: Population
            Newly generated population.
        """

        if n_individuals is None:
            n_individuals = self.population_size

        n_initializers = len(self.initializers)

        # Generate all populations from which to choose
        populations = [i.generate_population(n_individuals=n_individuals) for i in self.initializers]

        # Randomly choose individuals from each population
        population_tensor = np.asarray([p.genotype_matrix for p in populations])
        choice_idx = self.rng.choice(n_initializers, n_individuals, p=self.weights)
        chosen_population = population_tensor[choice_idx, np.arange(n_individuals), :]

        return Population(genotype_matrix=chosen_population, encoding=self.encoding)






class FixedCompositeInitializer(Initializer):
    def __init__(
        self,
        dimension: int,
        initializers: Iterable[Initializer],
        amounts: Optional[IntVector] = None,
        population_size: Optional[int] = None,
        encoding: Encoding = None,
        rng: Optional[RNGLike] = None,
    ):
        if population_size is None:
            population_size = np.sum(amounts)

        super().__init__(dimension, population_size, encoding=encoding, rng=rng)

        initializers = copy(initializers)
        for inits in initializers:
            inits.population_size = population_size

        self.initializers = initializers
        self.init_counter = 0

        if amounts is None:
            amounts = np.ones(len(initializers))
        self.amounts = amounts
        self._amounts_cumsum = np.cumsum(amounts)



    def generate_random(self) -> VectorLike:
        """Generate a random individual from one of the initializers chosen deterministically.

        Returns
        -------
        VectorLike
            A 1-D array generated by the fallback initializer.
        """

        idx = np.searchsorted(self._amounts_cumsum, self.init_counter, side="right")
        self.init_counter = (self.init_counter + 1) % np.sum(self.amounts)

        return self.initializers[idx].generate_random()




    def generate_individual(self) -> VectorLike:
        """Generate an individual from one of the initializers chosen deterministically.

        Returns
        -------
        ndarray
            A 1-D array representing the individual.
        """

        idx = np.searchsorted(self._amounts_cumsum, self.init_counter, side="right")
        self.init_counter = (self.init_counter + 1) % np.sum(self.amounts)

        return self.initializers[idx].generate_individual()




    def generate_population(self, n_individuals=None):
        """Generate a population from individuals chosen at random from the initializers.

        Parameters
        ----------
        objfunc: ObjectiveFunc
            Objective function that will be propagated to each individual.
        n_individual: int, optional
            Number of individuals to generate

        Returns
        -------
        generated_population: Population
            Newly generated population.
        """

        if n_individuals is None:
            n_individuals = self.population_size

        n_initializers = len(self.initializers)

        # Generate all populations from which to choose
        populations = [i.generate_population(n_individuals=n_individuals) for i in self.initializers]

        # Choose individuals as indicated by the amount attribute
        population_tensor = np.asarray([p.genotype_matrix for p in populations])
        cyclic_amounts = np.repeat(np.arange(n_initializers), self.amounts)
        choice_idx = np.resize(cyclic_amounts, n_individuals)
        chosen_population = population_tensor[choice_idx, np.arange(n_individuals), :]

        return Population(genotype_matrix=chosen_population, encoding=self.encoding)