Sztuczna_inteligencja_gr_13/bin/Classess/Travel.py

204 lines
7.1 KiB
Python
Raw Normal View History

2021-05-23 22:32:00 +02:00
import queue
2021-05-24 16:43:07 +02:00
from itertools import permutations, islice
from math import sqrt
import random
from resources.Globals import NUMBER_OF_INDIVIDUALS_FOR_DUEL, NUMBER_OF_POINTS_PERMUTATION, PERCENT_OF_MUTATION, \
PERCENT_OF_OUTGOING_INDIVIDUALS
2021-05-23 22:32:00 +02:00
class Travel:
def __init__(self):
self.points_coord = []
self.points_map = {}
def genetic_algorithm(travel_map):
2021-05-24 16:43:07 +02:00
population = []
2021-05-23 22:32:00 +02:00
road_map = list(travel_map.keys())
2021-05-24 16:43:07 +02:00
points_permutation = list(map(list, islice(permutations(road_map), NUMBER_OF_POINTS_PERMUTATION)))
# Generate the first population
for i in range(0, len(points_permutation)):
road = points_permutation[i]
priority = adaptation_function(points_permutation[i], travel_map)
population.append((priority, road))
while len(population) < 10000:
parent1, parent2 = tournament_selection(population)
child = edge_recombination_crossover(parent1[1], parent2[1])
child_priority = adaptation_function(child, travel_map)
population.append((child_priority, child))
mutation_function(population, travel_map)
population.sort(key=lambda x: x[0], reverse=True)
return population[0]
def adaptation_function(list_points, travel_map):
index_of_point = 0
distance = 0
while True:
if index_of_point < (-len(list_points)):
return round((1 / distance) * 1000000)
if index_of_point == (len(list_points) - 1):
x1 = travel_map.get(list_points[index_of_point])[0]
y1 = travel_map.get(list_points[index_of_point])[1]
x2 = travel_map.get(list_points[-len(list_points)])[0]
y2 = travel_map.get(list_points[-len(list_points)])[1]
index_of_point = -len(list_points) - 1
else:
x1 = travel_map.get(list_points[index_of_point])[0]
y1 = travel_map.get(list_points[index_of_point])[1]
x2 = travel_map.get(list_points[index_of_point + 1])[0]
y2 = travel_map.get(list_points[index_of_point + 1])[1]
index_of_point += 1
distance += sqrt((x2 - x1) ** 2 + (y2 - y1) ** 2)
def tournament_selection(population):
individuals_for_duel1 = []
individuals_for_duel2 = []
population_length = len(population)
while True:
if len(individuals_for_duel1) == NUMBER_OF_INDIVIDUALS_FOR_DUEL and len(individuals_for_duel2) == NUMBER_OF_INDIVIDUALS_FOR_DUEL:
break
if len(individuals_for_duel1) != NUMBER_OF_INDIVIDUALS_FOR_DUEL:
index1 = random.randint(0, population_length - 1)
candidate_for_duel1 = population[index1]
if candidate_for_duel1 not in individuals_for_duel1:
individuals_for_duel1.append(candidate_for_duel1)
if len(individuals_for_duel2) != NUMBER_OF_INDIVIDUALS_FOR_DUEL:
index2 = random.randint(0, population_length - 1)
candidate_for_duel2 = population[index2]
if candidate_for_duel2 not in individuals_for_duel1 and candidate_for_duel2 not in individuals_for_duel2:
individuals_for_duel2.append(candidate_for_duel2)
winner_of_duel1 = max(individuals_for_duel1, key=lambda x: x[0])
winner_of_duel2 = max(individuals_for_duel2, key=lambda x: x[0])
return winner_of_duel1, winner_of_duel2
def edge_recombination_crossover(parent1, parent2):
dict_of_neighbors = generate_dict_of_neighbors(parent1, parent2)
gen_index = random.randint(0, len(parent1) - 1)
gen = parent1[gen_index]
child = []
while True:
child.append(gen)
if len(child) == len(parent1):
return child
for key in dict_of_neighbors.keys():
if gen in dict_of_neighbors[key]:
dict_of_neighbors[key].remove(gen)
if not dict_of_neighbors[gen]:
while True:
# new_gen = random.randint(parent1[0], parent1[-1])
new_gen_index = random.randint(0, len(parent1) - 1)
new_gen = parent1[new_gen_index]
if new_gen not in child:
break
else:
new_gen = dict_of_neighbors[gen][0]
best_neighbor = len(dict_of_neighbors[new_gen])
for neighbor in dict_of_neighbors[gen][1:]:
possible_best_neighbor = len(dict_of_neighbors[neighbor])
if possible_best_neighbor <= best_neighbor:
best_neighbor = possible_best_neighbor
new_gen = neighbor
gen = new_gen
def generate_dict_of_neighbors(parent1, parent2):
dict_of_neighbors = {}
for i in range(0, len(parent1)):
list_of_neighbors = []
element = parent1[i]
if i == 0:
left_neighbor1 = parent1[-1]
right_neighbor1 = parent1[i + 1]
elif i == (len(parent1) - 1):
left_neighbor1 = parent1[i - 1]
right_neighbor1 = parent1[0]
else:
left_neighbor1 = parent1[i - 1]
right_neighbor1 = parent1[i + 1]
list_of_neighbors.extend([left_neighbor1, right_neighbor1])
index = parent2.index(element)
if index == 0:
left_neighbor2 = parent2[-1]
right_neighbor2 = parent2[index + 1]
elif index == (len(parent2) - 1):
left_neighbor2 = parent2[index - 1]
right_neighbor2 = parent2[0]
else:
left_neighbor2 = parent2[index - 1]
right_neighbor2 = parent2[index + 1]
if left_neighbor2 not in list_of_neighbors:
list_of_neighbors.append(left_neighbor2)
if right_neighbor2 not in list_of_neighbors:
list_of_neighbors.append(right_neighbor2)
dict_of_neighbors[element] = list_of_neighbors
return dict_of_neighbors
def mutation_function(population, travel_map):
mutation_percentage = random.random()
if mutation_percentage <= PERCENT_OF_MUTATION:
count_individual_for_mutation = round(len(population) * mutation_percentage)
mutants = set()
for i in range(0, count_individual_for_mutation):
while True:
individual_for_mutation = random.randint(0, len(population) - 1)
if individual_for_mutation not in mutants:
mutants.add(individual_for_mutation)
candidate_mutant = population[individual_for_mutation]
while True:
chromosome1 = random.randint(0, len(candidate_mutant[1]) - 1)
chromosome2 = random.randint(0, len(candidate_mutant[1]) - 1)
if chromosome1 != chromosome2:
candidate_mutant[1][chromosome1], candidate_mutant[1][chromosome2] = candidate_mutant[1][chromosome2], candidate_mutant[1][chromosome1]
candidate_mutant_priority = adaptation_function(candidate_mutant[1], travel_map)
mutant = (candidate_mutant_priority, candidate_mutant[1])
if mutant not in population:
population[individual_for_mutation] = mutant
break
break
2021-05-23 22:32:00 +02:00