def evolve(gen=1000, population=None, pop=50):
"""runs fern genetic algorithm and returns final population"""
if population is None:
r = fernpy.region2()
r[0], r[1] = fernpy.interval(-pi, pi), fernpy.interval(-50.0, 50.0)
r.set_uniform( fernpy.interval(-pi, pi) )
population = [fernpy.fern2(r, 3) for i in range(pop)]
randomize=True
else:
pop = len(population)
randomize=False
for index, individual in enumerate(population):
population[index].set_node_type_chance(node_type_chance)
population[index].set_mutation_type_chance(mutation_type_chance_fork,
mutation_type_chance_leaf)
if randomize:
for index, individual in enumerate(population):
population[index].randomize(100)
#else:
# for index, individual in enumerate(population):
# population[index].randomize(50)
job_server = pp.Server(ppservers=())
jobs = []
subfuncs = ()
packages = ("numpy", "fernpy")
template = pp.Template(job_server, fitness, subfuncs, packages)
max_fitness = [0]*gen
median_fitness = [0]*gen
min_fitness = [0]*gen
for generation in range(gen):
#generating new data for each generation helps encourage
#a healthy genetic variation
numbers, classes = generate_data(10000)
#evaluate ferns
pop_fitness = np.array([0.0]*pop)
for index, individual in enumerate(population):
if len(jobs) < pop: #only happens in first generation
jobs.append( template.submit(individual, numbers, classes) )
else:
jobs[index] = template.submit(individual, numbers, classes)
for index, result in enumerate(jobs):
pop_fitness[index] = result()
#record max and median fitness
max_fitness_index = sp.argmax(pop_fitness)
max_fitness[generation] = pop_fitness[max_fitness_index]
median_fitness[generation] = median(pop_fitness)
min_fitness[generation] = min(pop_fitness)
#control ratio of max to median fitness and normalize
#fitness_mean = np.mean(pop_fitness);
#if max_fitness[generation] != fitness_mean: #median_fitness[generation]:
# a = log(fitness_ratio) / (log(max_fitness[generation]) - log(fitness_mean))
# pop_fitness = (pop_fitness - fitness_mean)**a + fitness_mean
sys.stdout.write("\rgen %i" % generation)
sys.stdout.flush()
if generation == gen-1:
sys.stdout.write("\n")
break #skip breeding on last step
pop_fitness /= sum(pop_fitness)
#select parents and breed new population
new_population = []
new_population.append( population[max_fitness_index] ) #elitism
for i in range(1, pop):
new_population.append(fernpy.fern2( population[select(pop_fitness)] ))
if rand.random() < crossover_rate:
new_population[-1].crossover( population[select(pop_fitness)] )
if rand.random() < mutation_rate:
new_population[-1].mutate()
population = new_population
h = 1.0*np.array(range(-40, 40))
theta = path(h)
plot.figure()
plot.plot(theta, h)
a = plot.gca()
a.set_xlim([-pi, pi])
plot.show()
fplt.plot(population[max_fitness_index], "static_satellite.png")
datafile = open("static_satellite.dat", "wb")
pickler = pickle.Pickler(datafile)
pickler.dump(population)
pickler.dump(pop_fitness)
datafile.close()
plot.figure()
plot.plot(range(gen), max_fitness, '+', color='green') #plot fitness progression
plot.plot(range(gen), median_fitness, 'x', color='blue')
plot.plot(range(gen), min_fitness, '.', color = 'red')
plot.show()
return population, pop_fitness
def evolve(gen=1000, population=None, pop=50):
"""runs fern genetic algorithm and returns final population"""
if population is None:
r = fernpy.region1()
r[0] = fernpy.interval(-20.0, 20.0)
population = [fernpy.fern1(r, 3) for i in range(pop)]
randomize = True
else:
pop = len(population)
randomize = False
for index, individual in enumerate(population):
population[index].set_node_type_chance(node_type_chance)
population[index].set_mutation_type_chance(mutation_type_chance_fork,
mutation_type_chance_leaf)
if randomize:
population[index].randomize(15)
job_server = pp.Server(ppservers=())
jobs = []
subfuncs = (simulate, f)
packages = ("time", "scipy", "scipy.integrate", "numpy", "math", "fernpy")
template = pp.Template(job_server, fitness, subfuncs, packages)
max_fitness = [0]*gen
median_fitness = [0]*gen
min_fitness = [0]*gen
#state0 = [random_state() for i in range(5)] #5 simulations per evaluation
for generation in range(gen):
state0 = [-15, -7, 7, 15] #4 simulations per evaluation
optimal_fitness = fitness(OptimalController(), state0)
#evaluate ferns
pop_fitness = numpy.array([0.0]*pop)
#state0 = [random_state() for i in range(10)] #simulations per evaluation
for index, individual in enumerate(population):
if len(jobs) < pop: #only happens in first generation
jobs.append( template.submit(individual, state0) )
else:
jobs[index] = template.submit(individual, state0)
for index, result in enumerate(jobs):
pop_fitness[index] = result() / optimal_fitness
#record and then normalize fitness
max_fitness_index = scipy.argmax(pop_fitness)
max_fitness[generation] = pop_fitness[max_fitness_index]
median_fitness[generation] = median(pop_fitness)
min_fitness[generation] = min(pop_fitness)
#control ratio of max to median fitness and normalize
#fitness_mean = np.mean(pop_fitness);
#if max_fitness[generation] != median_fitness[generation]: #fitness_mean:
# a = log(fitness_ratio) / (log(max_fitness[generation]) - log(median_fitness[generation]))
# pop_fitness = (pop_fitness - median_fitness[generation])**a + median_fitness[generation]
sys.stdout.write("\rgen %i" % generation)
sys.stdout.flush()
if generation == gen-1:
sys.stdout.write("\n")
break #skip breeding on last step
pop_fitness /= sum(pop_fitness)
#select parents and breed new population
new_population = []
new_population.append( population[max_fitness_index] ) #elitism
for i in range(1, pop):
new_population.append(fernpy.fern1( population[select(pop_fitness)] ))
if random.random() < crossover_rate:
new_population[-1].crossover( population[select(pop_fitness)] )
if random.random() < mutation_rate:
new_population[-1].mutate()
population = new_population
##### plots ########
fplt.plot(population[max_fitness_index], "velocity_control.png")
#print population[max_fitness_index]
datafile = open("velocity_control.dat", "wb")
pickler = pickle.Pickler(datafile)
pickler.dump(population)
pickler.dump(pop_fitness)
datafile.close()
plot_sim( population[ max_fitness_index ] ) #plot best solution
plot.figure()
plot.plot(range(gen), max_fitness, '+', color='green') #plot fitness progression
plot.plot(range(gen), median_fitness, 'x', color='blue')
plot.plot(range(gen), min_fitness, '.', color = 'red')
plot.show()
return population, pop_fitness
def evolve(gen=500, population=None, pop=50):
"""runs fern genetic algorithm and returns final population"""
if population is None:
r = fernpy.region1()
r.set_uniform( fernpy.interval(-3.0, 3.0) )
population = [fernpy.fern1(r, 2) for i in range(pop)]
randomize = True
else:
pop = len(population)
randomize = False
for index, individual in enumerate(population):
population[index].set_node_type_chance(node_type_chance)
population[index].set_mutation_type_chance(mutation_type_chance_fork,
mutation_type_chance_leaf)
population[index].randomize(15)
job_server = pp.Server(ppservers=())
jobs = []
subfuncs = ()
packages = ("numpy", "fernpy",)
max_fitness = [0]*gen
median_fitness = [0]*gen
min_fitness = [0]*gen
for generation in range(gen):
#generating new data for each generation helps encourage
#a healthy genetic variation
numbers, classes = generate_data()
#evaluate ferns
pop_fitness = numpy.array([0.0]*pop)
if len(jobs) < pop:
for index, individual in enumerate(population):
jobs.append(job_server.submit(fitness,
(individual, numbers, classes,),
subfuncs, packages))
else:
for index, individual in enumerate(population):
jobs[index] = job_server.submit(fitness,
(individual, numbers, classes,),
subfuncs, packages)
for index, result in enumerate(jobs):
pop_fitness[index] = result()
#record and then normalize fitness
max_fitness_index = sp.argmax(pop_fitness)
max_fitness[generation] = pop_fitness[max_fitness_index]
median_fitness[generation] = median(pop_fitness)
min_fitness[generation] = min(pop_fitness)
#control ratio of max to median fitness and normalize
#fitness_mean = numpy.mean(pop_fitness);
#if max_fitness[generation] != fitness_mean:
# a = log(fitness_ratio) / (log(max_fitness[generation]) - log(fitness_mean))
# pop_fitness = (pop_fitness - fitness_mean)**a + fitness_mean
sys.stdout.write("\rgen %i" % generation)
sys.stdout.flush()
if generation == gen-1:
sys.stdout.write("\n")
break #skip breeding on last step
pop_fitness /= sum(pop_fitness)
#select parents and breed new population
new_population = []
new_population.append( population[max_fitness_index] ) #elitism
for i in range(1, pop):
new_population.append(fernpy.fern1( population[select(pop_fitness)] ))
if random.random() < crossover_rate:
new_population[-1].crossover( population[select(pop_fitness)] )
if random.random() < mutation_rate:
new_population[-1].mutate()
population = new_population
fplt.plot(population[max_fitness_index], "classify_fern2.png", 7)
datafile = open("classify_fern2.dat", "wb")
pickler = pickle.Pickler(datafile)
pickler.dump(population)
pickler.dump(pop_fitness)
datafile.close()
plot.figure()
plot.plot(range(gen), max_fitness, '+', color='green') #plot fitness progression
plot.plot(range(gen), median_fitness, 'x', color='blue')
plot.plot(range(gen), min_fitness, '.', color = 'red')
plot.xlabel("Generation")
plot.ylabel("Fitness (# correct)")
plot.show()
return population, pop_fitness