使用遗传算法解决旅行商问题

旅行商问题，简单地说，即一个商人要找一条通过n个城市的最短巡回。

开始产生的随机路径：

遗传算法的其中一个结果

对个体采用字符编码的方式。如［３，１，４，２，５］表示开始从第３个结点开始，然后第１个结点，依次而行。为了避免产生非法解，对个体进行变因时，实质是交换个体中的两个数字的位置。例如［４，１，３，２，５］即为一个新的个体。变异时，例如对于个体ａ＝［３，１，４，２，５］，ｂ＝［５，２，４，１，３］，如果选择ａ的第２和第３个位置和ｂ的第２和第３个位置进行交换。首先作出在ａ中而不在ｂ［２，４］中的字符，即ｃ＝［３，１，５］，然后在ｃ中第１个空位插入ｂ［２，４］，即生成新的个休［３，２，４，１，５］，从而不会有重复字符，避免了非法解的产生。

'''
Created on 2009-10-29
@author: Administrator
'''
import sys, random
from math import sqrt
from pickle import *

PIL_SUPPORT = None
try:
   from PIL import Image, ImageDraw, ImageFont
   PIL_SUPPORT = True
except:
   PIL_SUPPORT = False 

def cartesian_matrix(coords):
   """ A distance matrix """
   matrix = {}
   for i, (x1, y1) in enumerate(coords):
      for j, (x2, y2) in enumerate(coords):
         dx, dy = x1 - x2, y1 - y2
         dist = sqrt(dx * dx + dy * dy)
         matrix[i, j] = dist
   return matrix 

def tour_length(matrix, tour):
   """ Returns the total length of the tour """
   total = 0
   num_cities = len(tour)
   for i in range(num_cities):
      j = (i + 1) % num_cities
      city_i = tour[i]
      city_j = tour[j]
      total += matrix[city_i, city_j]
   return total

def write_tour_to_img(coords, tour, img_file):
   """ The function to plot the graph """
   padding = 20
   coords = [(x + padding, y + padding) for (x, y) in coords]
   maxx, maxy = 0, 0
   for x, y in coords:
      maxx = max(x, maxx)
      maxy = max(y, maxy)
   maxx += padding
   maxy += padding
   img = Image.new("RGB", (int(maxx), int(maxy)), color=(255, 255, 255))
   font = ImageFont.load_default()
   d = ImageDraw.Draw(img);
   num_cities = len(tour)
   for i in range(num_cities):
      j = (i + 1) % num_cities
      city_i = tour[i]
      city_j = tour[j]
      x1, y1 = coords[city_i]
      x2, y2 = coords[city_j]
      d.line((int(x1), int(y1), int(x2), int(y2)), fill=(0, 0, 0))
      d.text((int(x1) + 7, int(y1) - 5), str(i), font=font, fill=(32, 32, 32)) 

   for x, y in coords:
      x, y = int(x), int(y)
      d.ellipse((x - 5, y - 5, x + 5, y + 5), outline=(0, 0, 0), fill=(196, 196, 196))
   del d
   img.save(img_file, "PNG")
   print "The plot was saved into the %s file." % (img_file,)  

cm = []
coords = [] 

def eval_func(chromosome):
   """ The evaluation function """
   global cm
   return tour_length(cm, chromosome) 

def cities_random(cities, xmax=800, ymax=600):
   """ get random cities/positions """
   coords = []
   for i in xrange(cities):
      x = random.randint(0, xmax)
      y = random.randint(0, ymax)
      coords.append((float(x), float(y)))
   return coords

#Individuals
class Individual:
    score = 0
    length = 30
    seperator = ' '
    def __init__(self, chromosome=None, length=30):
        self.chromosome = chromosome or self._makechromosome()
        self.length = length
        self.score = 0  # set during evaluation  

    def _makechromosome(self):
        "makes a chromosome from randomly selected alleles."
        chromosome = []
        lst = [i for i in xrange(self.length)]
        for i in xrange(self.length):
            choice = random.choice(lst)
            lst.remove(choice)
            chromosome.append(choice)
        return chromosome

    def evaluate(self, optimum=None):
        self.score = eval_func(self.chromosome)
        
    def crossover(self, other):
        left, right = self._pickpivots()         
        p1 = Individual()
        p2 = Individual()   
        c1 = [ c for c in self.chromosome if c not in other.chromosome[left:right + 1]]
        p1.chromosome = c1[:left] + other.chromosome[left:right + 1] + c1[left:]
        c2 = [ c for c in other.chromosome if c not in self.chromosome[left:right + 1]]
        p2.chromosome = c2[:left] + self.chromosome[left:right + 1] + c2[left:]
        return p1, p2

    def mutate(self):
        "swap two element"
        left, right = self._pickpivots()      
        temp = self.chromosome[left]
        self.chromosome[left] = self.chromosome[right]
        self.chromosome[right] = temp   

    def _pickpivots(self):
        left = random.randint(0, self.length - 2)
        right = random.randint(left, self.length - 1)
        return left, right    

    def __repr__(self):
        "returns string representation of self"
        return '<%s chromosome="%s" score=%s>' % \
               (self.__class__.__name__,
                self.seperator.join(map(str, self.chromosome)), self.score)  

    def copy(self):
        twin = self.__class__(self.chromosome[:])
        twin.score = self.score
        return twin
     

    def __cmp__(self, other):
        return cmp(self.score, other.score)

class Environment:
    size = 0
    def __init__(self, population=None, size=100, maxgenerations=1000, newindividualrate=0.6,
                 crossover_rate=0.90, mutation_rate=0.1):       
        self.size = size      
        self.population = self._makepopulation()       
        self.maxgenerations = maxgenerations
        self.newindividualrate = newindividualrate  
        self.crossover_rate = crossover_rate
        self.mutation_rate = mutation_rate       
        for individual in self.population:           
            individual.evaluate()      
        self.generation = 0
        self.minscore = sys.maxint       
        self.minindividual = None
        #self._printpopulation()
        if PIL_SUPPORT:
            write_tour_to_img(coords, self.population[0].chromosome, "TSPstart.png")
        else:
            print "No PIL detected,can not plot the graph"

    def _makepopulation(self):
        return [Individual() for i in range(0, self.size)]

    def run(self):
        for i in range(1, self.maxgenerations + 1):
            print "Generation no:" + str(i)
            for j in range(0, self.size):
                self.population[j].evaluate()
                curscore = self.population[j].score
                if curscore < self.minscore:
                    self.minscore = curscore
                    self.minindividual = self.population[j]           
            print "Best individual:", self.minindividual
            if random.random() < self.crossover_rate:                            
                children = []               
                newindividual = int(self.newindividualrate * self.size / 2)                               
                for i in range(0, newindividual):
                    selected1 = self._selectrank()                   
                    selected2 = self._selectrank()                   
                    parent1 = self.population[selected1]
                    parent2 = self.population[selected2]
                    child1, child2 = parent1.crossover(parent2)
                    child1.evaluate()
                    child2.evaluate()
                    children.append(child1)
                    children.append(child2)
                for i in range(0, newindividual):
                    #replce with child                   
                    totalscore = 0
                    for k in range(0, self.size):
                        totalscore += self.population[k].score
                    randscore = random.random()                   
                    addscore = 0
                    for j in range(0, self.size):
                        addscore += (self.population[j].score / totalscore)                       
                        if addscore >= randscore:                                                                                 
                            self.population[j] = children[i]
                            break               
            if random.random() < self.mutation_rate:               
                selected = self._select()
                self.population[selected].mutate()
        #end loop
        for i in range(0, self.size):
                self.population[i].evaluate()
                curscore = self.population[i].score
                if curscore < self.minscore:
                    self.minscore = curscore
                    self.minindividual = self.population[i]
        print "Result."
        print self.minindividual
        #self._printpopulation()

    def _select(self):       
        totalscore = 0
        for i in range(0, self.size):
            totalscore += self.population[i].score
        randscore = random.random()*(self.size - 1)
        addscore = 0
        selected = 0
        for i in range(0, self.size):
            addscore += (1 - self.population[i].score / totalscore)
            if addscore >= randscore:
                selected = i
                break
        return selected

    def _selectrank(self, choosebest=0.9):
        self.population.sort()
        if random.random() < choosebest:
            return random.randint(0, self.size * self.newindividualrate)
        else:
            return random.randint(self.size * self.newindividualrate, self.size - 1)

    def _printpopulation(self):
        for i in range(0, self.size):
            print "Individual ", i, self.population[i]      

def main_run():   
    global cm, coords
    #get cities's coords
    coords = [(553.0, 132.0), (167.0, 381.0), (395.0, 51.0), (416.0, 101.0), (575.0, 351.0), (243.0, 347.0), (218.0, 415.0), (247.0, 86.0), (404.0, 286.0), (209.0, 462.0), (23.0, 204.0), (54.0, 492.0), (350.0, 405.0), (689.0, 78.0), (729.0, 543.0), (429.0, 371.0), (472.0, 64.0), (316.0, 570.0), (297.0, 570.0), (261.0, 198.0), (559.0, 272.0), (684.0, 224.0), (13.0, 548.0), (146.0, 93.0), (125.0, 558.0), (499.0, 414.0), (510.0, 503.0), (55.0, 234.0), (454.0, 144.0), (381.0, 0.0)]

    cm = cartesian_matrix(coords)
    ev = Environment()
    ev.run()
    if PIL_SUPPORT:
        write_tour_to_img(coords, ev.minindividual.chromosome, "TSPresult.png")
    else:
        print "No PIL detected,can not plot the graph"   
if __name__ == "__main__":   
    main_run()