crossover.py

def crossover(parents, img_shape, n_individuals=8):
    new_population = numpy.empty(shape=(n_individuals, functools.reduce(operator.mul, img_shape)), dtype=numpy.uint8)
    
    #Previous parents (best elements).
    new_population[0:parents.shape[0], :] = parents

    # Getting how many offspring to be generated. If the population size is 8 and number of parents mating is 4, then number of offspring to be generated is 4.
    num_newly_generated = n_individuals-parents.shape[0]
    # Getting all possible permutations of the selected parents.
    parents_permutations = list(itertools.permutations(iterable=numpy.arange(0, parents.shape[0]), r=2))
    # Randomly selecting the parents permutations to generate the offspring.
    selected_permutations = random.sample(range(len(parents_permutations)), 
                                          num_newly_generated)
    
    comb_idx = parents.shape[0]
    for comb in range(len(selected_permutations)):
        # Generating the offspring using the permutations previously selected randmly.
        selected_comb_idx = selected_permutations[comb]
        selected_comb = parents_permutations[selected_comb_idx]
        
        # Applying crossover by exchanging half of the genes between two parents.
        half_size = numpy.int32(new_population.shape[1]/2)
        new_population[comb_idx+comb, 0:half_size] = parents[selected_comb[0], 
                                                             0:half_size]
        new_population[comb_idx+comb, half_size:] =  parents[selected_comb[1], 
                                                             half_size:]
    
    return new_population