def create_offspring(self, ind):
delta = 0.001 * np.random.normal(0, 2, self.gen_size)
new_genome = ind.genome + delta
new_genome = np.abs(new_genome) / sum(new_genome)
new_ind = Individual(genome=new_genome, fitness=np.nan)
new_ind.fitness = self.f.calculate(new_ind.genome)
self.f.count += 1
return new_ind
def eval_genomes(genomes, config):
for genome_id, genome in genomes:
net = neat.nn.FeedForwardNetwork.create(genome, config)
individual = Individual()
individual.neat_network = net
genome.fitness = individual.fitness()
def select_individuals(self, count):
selected = list()
while len(selected) < count:
round_participants = sample(self.individuals, PARTICIPATION_CNT)
winner = Individual()
winner.fitness = 0
for participant in round_participants:
if participant.fitness + participant.bonus > winner.fitness + winner.bonus and participant not in selected:
winner = deepcopy(participant)
selected.append(winner)
return selected
def mutate_LHC(self, network):
'''
Local Hill-Climbing, 2-opt
'''
# Chance to change the means of travel for some gene
if Individual.budget != None and randint(0, 100) < 10: #10% chance
i = randint(0, len(self.genes) - 1)
if self.genes[i] == SolutionGenerator.FLY:
self.genes[i] = (self.genes[i][0], SolutionGenerator.DRIVE)
else:
self.genes[i] = (self.genes[i][0], SolutionGenerator.FLY)
self.used_budget = self.all_used_budget(network)
self.reduce(network)
self.__fitness = self.fitness(network)
# Chance to mutate via local hill climbing
if randint(0, 100) < 5: # 5% chance
maxiters = 100 # Max iterations, iterations rarely go above 10
iters = 0
improved = True
best = self # initial best sequence
size = len(best.genes)
while improved and iters < maxiters:
improved = False
for i in range(0, size - 3):
for j in range(i + 1, size):
# New individual created, genes swapped
newGuy = Individual(best.genes[:], network)
newGuy.genes[i], newGuy.genes[j] = newGuy.genes[
j], newGuy.genes[i]
# Calculate the gain
gain = newGuy.fitness(network) - best.fitness(network)
if gain > 0:
best = newGuy
improved = True
break # return to while
else:
break
iters += 1
# Genes are set to new local optima
self.genes = best.genes
self.__fitness = self.fitness(network)
