Example #1
0
def inversion_mutation(individual, mutation_rate):
    result = Individual(individual.dna, individual.mutation_operator, individual.crossover_operator)
    for swap1 in range(individual.__len__()):
        if random.random() < mutation_rate:
            swap2 = int(random.random() * individual.__len__())
            a_list = result.dna[min(swap1, swap2):max(swap1, swap2) + 1]
            a_list = a_list[::-1]
            for i in range(min(swap1, swap2), max(swap1, swap2) + 1):
                result.dna[i] = a_list[i - min(swap1, swap2)]
    return result
Example #2
0
def swap_mutation(individual, mutation_rate):
    result = Individual(individual.dna, individual.mutation_operator, individual.crossover_operator)
    for swap1 in range(individual.__len__()):
        if random.random() < mutation_rate:
            swap2 = int(random.random() * individual.__len__())
            city1 = result.dna[swap1]
            city2 = result.dna[swap2]
            result.dna[swap1] = city2
            result.dna[swap2] = city1
    return result
Example #3
0
def scramble_mutation(individual, mutation_rate):
    result = Individual(individual.dna, individual.mutation_operator, individual.crossover_operator)
    for swap1 in range(individual.__len__()):
        if random.random() < mutation_rate:
            swap2 = int(random.random() * individual.__len__())
            a_list = []
            for i in range(min(swap1, swap2), max(swap1, swap2) + 1):
                a_list.append(i)
            random.shuffle(a_list)
            cop = list(result.dna)
            for i in range(min(swap1, swap2), max(swap1, swap2) + 1):
                result.dna[i] = cop[a_list[i - min(swap1, swap2)]]
    return result
Example #4
0
def cycle_crossover(parent1, parent2):
    child = []
    p1 = []
    p2 = []

    # 从第一个个体中随机选择一段城市,拼接第二个个体中所有不重复的城市,生成的新的个体
    geneA = int(random.random() * len(parent1.dna))
    dic = {}
    dic1 = {}
    for i in range(len(parent1.dna)):
        dic1[parent1.dna[i]] = parent2.dna[i]
    dic[parent1.dna[geneA]] = parent2.dna[geneA]
    k = parent2.dna[geneA]
    while k != parent1.dna[geneA]:
        if k is dic1[k]:
            break
        dic[k] = dic1[k]
        k = dic1[k]
    for i in range(len(parent1.dna)):
        if parent1.dna[i] in dic.keys():
            child.append(parent1.dna[i])
        else:
            child.append(parent2.dna[i])
    return Individual(child, parent1.mutation_operator,
                      parent1.crossover_operator)
Example #5
0
 def initial_population(self, pop_size):
     population = Population(self.alg.selection_strategy)
     for i in range(0, pop_size):
         population.append(
             Individual(random.sample(self.city_list, len(self.city_list)),
                        self.alg.mutation_operator,
                        self.alg.crossover_operator))
     return population
Example #6
0
def insert_mutation(individual, mutation_rate):
    result = Individual(individual.dna, individual.mutation_operator, individual.crossover_operator)
    for insert1 in range(individual.__len__()):
        if random.random() < mutation_rate:
            insert2 = int(random.random() * individual.__len__())
            if insert1 == insert2:
                return result
            city1 = result.dna[insert1]
            city2 = result.dna[insert2]
            insert3 = min(insert1, insert2)
            insert4 = max(insert1, insert2)
            for i in range(insert3 + 1, insert4):
                result.dna[i + 1] = result.dna[i]
            if insert1 > insert2:
                result.dna[insert2 + 1] = city1
            else:
                result.dna[insert1 + 1] = city2
    return result
Example #7
0
def order_crossover(parent1, parent2):
    child = []
    p1 = []
    p2 = []

    # 从第一个个体中随机选择一段城市,拼接第二个个体中所有不重复的城市,生成的新的个体
    geneA = int(random.random() * len(parent1.dna))
    geneB = int(random.random() * len(parent2.dna))
    start = min(geneA, geneB)
    end = max(geneA, geneB)
    for i in range(start, end + 1):
        p1.append(parent1.dna[i])
    for i in range(end + 1, end + len(parent1.dna) + 1):
        cou = i
        if cou >= len(parent1.dna):
            cou -= len(parent1.dna)
        if parent2.dna[cou] not in p1:
            p2.append(parent2.dna[cou])
    child = p2[len(parent2.dna) - end:] + p1 + p2[:len(parent2.dna) - end]
    return Individual(child, parent1.mutation_operator,
                      parent1.crossover_operator)
Example #8
0
def PBX_crossover(parent1, parent2):
    child = []

    # 从第一个个体中随机选择一段城市,拼接第二个个体中所有不重复的城市,生成的新的个体
    geneA = int(random.random() * len(parent1.dna))
    geneB = int(random.random() * len(parent2.dna))
    start = min(geneA, geneB)
    end = max(geneA, geneB)
    dic = {}
    child = parent1.dna
    q1 = []
    q2 = []
    nums = []
    for i in range(start, end + 1):
        dic[parent2.dna[i]] = parent1.dna[i]
        q1.append(parent1.dna[i])
        q2.append(parent2.dna[i])
        nums.append(i)

    for i in range(len(q2)):
        if q2[i] not in q1:
            k = q2[i]
            dict = {}
            while k in q2:
                if k is dic[k] or k in dict.keys():
                    break
                dict[k] = dic[k]
                k = dic[k]
            flag = True
            if k in dic.keys() and k is dic[k]:
                flag = False
            if k in parent2.dna and flag is True:
                t = parent2.dna.index(k)
                child[t] = q2[i]
                nums.append(i)
    for i in range(len(child)):
        if i not in nums:
            child[i] = parent2.dna[i]
    return Individual(child, parent1.mutation_operator,
                      parent1.crossover_operator)