def start_cycle(self, one_by_one=False):
"""Start the cycle.
Keyword arguments:
one_by_one -- evaluate the genomes one by one if True (default False)
"""
if one_by_one:
# play each genome in a game alone.
for generation in range(self.generations):
# retrieve genome list and call evaluation function for each one.
genome_list = NEAT.GetGenomeList(self.population)
best_fitness = 0
print("generation", generation + 1, ":")
print("testing " + str(self.params.PopulationSize) +
" genomes : ")
i = 0
for genome in genome_list:
i += 1
print(i, end=" ")
net = NEAT.NeuralNetwork()
genome.BuildPhenotype(net)
fitness = self.evaluate(genome, generation + 1, i)
if best_fitness < fitness:
best_fitness = fitness
genome.SetFitness(fitness)
# print best fitness and advance to the next generation
print("best fitness : ", best_fitness)
print("=======================================")
self.population.Epoch()
else:
# play all of the population at the same time
for generation in range(self.generations):
# retrieve genome list and build the players list.
genome_list = NEAT.GetGenomeList(self.population)
players = list()
for genome in genome_list:
net = NEAT.NeuralNetwork()
genome.BuildPhenotype(net)
players.append(Dino_player_neat(net))
# start game and retrieve fitness list.
the_game = Dino_NEAT(players, generation)
fitness = the_game.on_execute()
if fitness is None:
print("Training stopped.")
break
# assign each genome to its corresponding fitness.
best_fitness = 0
for i in range(len(fitness)):
genome = genome_list[i]
if best_fitness < fitness[i]:
best_fitness = fitness[i]
genome.SetFitness(fitness[i])
# print best fitness and advance to the next generation
print("generation", generation, ":", best_fitness)
self.population.Epoch()
def evolve():
g = NEAT.Genome(0, 3, 0, 1, False, NEAT.ActivationFunction.UNSIGNED_SIGMOID,
NEAT.ActivationFunction.UNSIGNED_SIGMOID, 0, params, 0)
pop = NEAT.Population(g, params, True, 1.0, 1)
pop.RNG.Seed(int(time.clock()*100))
generations = 0
for generation in range(1000):
genome_list = NEAT.GetGenomeList(pop)
fitness_list = []
for genome in genome_list:
fitness_list.append(evaluate(genome))
NEAT.ZipFitness(genome_list, fitness_list)
pop.Epoch()
generations = generation
best = max(fitness_list)
bestG = pop.GetBestGenome()
plot_nn(bestG)
plt.pause(0.001)
plt.ion()
plt.show(block=False)
print("Mejor fitness [",generation,"]: ",best)
if best > 15.9:
break
return generations
def getbest(i):
g = NEAT.Genome(0, 3, 0, 1, False, NEAT.ActivationFunction.UNSIGNED_SIGMOID,
NEAT.ActivationFunction.UNSIGNED_SIGMOID, 0, params, 0)
pop = NEAT.Population(g, params, True, 1.0, i)
# pop.RNG.Seed(int(time.clock()*100))
pop.RNG.Seed(1234)
generations = 0
for generation in range(max_generations):
genome_list = NEAT.GetGenomeList(pop)
fitness_list = EvaluateGenomeList_Serial(genome_list, evaluate, display=False)
# fitness_list = EvaluateGenomeList_Parallel(genome_list, evaluate, display=False)
NEAT.ZipFitness(genome_list, fitness_list)
pop.Epoch()
generations = generation
best = max(fitness_list)
if best > 15.0:
break
net = NEAT.NeuralNetwork()
pop.GetBestGenome().BuildPhenotype(net)
# img = NEAT.viz.Draw(net)
# cv2.imshow("current best", img)
# cv2.waitKey(1)
return generations, net.NumHiddenNeurons(), net.NumConnections()
def evaluate_obj_functions(obj_function, generation):
"""
The function to perform evaluation of the objective functions population
Arguments:
obj_function: The population of objective functions
generation: The current generation of evolution
"""
obj_func_coeffs = []
n_items_list = []
# evaluate objective function genomes and collect novelty items
obj_func_genomes = NEAT.GetGenomeList(obj_function.population)
for genome in obj_func_genomes:
n_item = evaluate_individ_obj_function(genome=genome,
generation=generation)
n_items_list.append(n_item)
obj_func_coeffs.append(n_item.data)
# evaluate collected novelty items and set genomes fitness scores
max_fitness = 0
for i, genome in enumerate(obj_func_genomes):
fitness = obj_function.archive.evaluate_novelty_score(
item=n_items_list[i], n_items_list=n_items_list)
genome.SetFitness(fitness)
max_fitness = max(max_fitness, fitness)
return obj_func_coeffs, max_fitness
def getbest(i):
g = NEAT.Genome(0, substrate.GetMinCPPNInputs(), 0,
substrate.GetMinCPPNOutputs(), False,
NEAT.ActivationFunction.TANH, NEAT.ActivationFunction.TANH,
0, params)
pop = NEAT.Population(g, params, True, 1.0, i)
pop.RNG.Seed(i)
for generation in range(2000):
genome_list = NEAT.GetGenomeList(pop)
fitnesses = NEAT.EvaluateGenomeList_Serial(genome_list,
evaluate,
display=False)
[
genome.SetFitness(fitness)
for genome, fitness in zip(genome_list, fitnesses)
]
best = max([x.GetLeader().GetFitness() for x in pop.Species])
pop.Epoch()
generations = generation
if best > 15.0:
break
return generations
def single_generation(self):
"""
Single generation of evaluation for a population on an environment.
:return: performance measure (i.e. fitness).
"""
# Retrieve a list of all genomes in the population
genome_list = mneat.GetGenomeList(self.alg.pop)
# Main Population Evaluation Loop
for current_genome in genome_list:
# Evaluate the current genome
fitness = self.evaluate_agent(current_genome)
# Reset the current genome's fitness
current_genome.SetFitness(fitness)
# Call a new Epoch - runs mutation and crossover, creating offspring
self.alg.pop.Epoch()
def getbest(i):
g = NEAT.Genome(0, substrate.GetMinCPPNInputs(), 2,
substrate.GetMinCPPNOutputs(), False,
NEAT.ActivationFunction.TANH, NEAT.ActivationFunction.TANH,
0, params)
pop = NEAT.Population(g, params, True, 1.0, i)
pop.RNG.Seed(i)
for generation in range(2000):
genome_list = NEAT.GetGenomeList(pop)
# if sys.platform == 'linux':
# fitnesses = EvaluateGenomeList_Parallel(genome_list, evaluate, display=False)
# else:
fitnesses = EvaluateGenomeList_Serial(genome_list,
evaluate,
display=False)
[
genome.SetFitness(fitness)
for genome, fitness in zip(genome_list, fitnesses)
]
print('Gen: %d Best: %3.5f' % (generation, max(fitnesses)))
best = max(fitnesses)
pop.Epoch()
generations = generation
if best > 15.0:
break
return generations
def test_multi_neat(self):
params = NEAT.Parameters()
params.PopulationSize = 100
genome = NEAT.Genome(
0, # ID
3, # number of inputs. Note: always add one extra input, for bias
0, # number of hidden nodes
2, # number of outputs
False, # FS_NEAT; auto-determine an appropriate set of inputs for the evolved networks
NEAT.ActivationFunction.UNSIGNED_SIGMOID, # OutputActType
NEAT.ActivationFunction.UNSIGNED_SIGMOID, # HiddenActType
0, # SeedType
params # Parameters
)
seed = 42
pop = NEAT.Population(
genome,
params,
True, # whether the population should be randomized
1.0, # how much the population should be randomized,
seed)
for generation in range(3):
# retrieve a list of all genomes in the population
genome_list = NEAT.GetGenomeList(pop)
# apply the evaluation function to all genomes
for genome in genome_list:
fitness = self.evaluate(genome)
genome.SetFitness(fitness)
# advance to the next generation
pop.Epoch()
def getbest(i):
g = NEAT.Genome(0, 3, 0, 1, False,
NEAT.ActivationFunction.UNSIGNED_SIGMOID,
NEAT.ActivationFunction.UNSIGNED_SIGMOID, 0, params)
pop = NEAT.Population(g, params, True, 1.0, i)
pop.RNG.Seed(i)
generations = 0
for generation in range(1000):
genome_list = NEAT.GetGenomeList(pop)
fitness_list = NEAT.EvaluateGenomeList_Serial(genome_list,
evaluate,
display=False)
NEAT.ZipFitness(genome_list, fitness_list)
best = max([x.GetLeader().GetFitness() for x in pop.Species])
pop.Epoch()
generations = generation
if best > 15.0:
break
return generations
def getbest():
g = NEAT.Genome(0, 3, 0, 1, False,
NEAT.ActivationFunction.UNSIGNED_SIGMOID,
NEAT.ActivationFunction.UNSIGNED_SIGMOID, 0, params)
pop = NEAT.Population(g, params, True, 1.0)
generations = 0
for generation in range(1000):
genome_list = NEAT.GetGenomeList(pop)
fitness_list = NEAT.EvaluateGenomeList_Serial(genome_list,
evaluate,
display=False)
NEAT.ZipFitness(genome_list, fitness_list)
best = max([x.GetLeader().GetFitness() for x in pop.Species])
# print 'Best fitness:', best, 'Species:', len(pop.Species)
# test
net = NEAT.NeuralNetwork()
pop.Species[0].GetLeader().BuildPhenotype(net)
img = np.zeros((250, 250, 3), dtype=np.uint8)
img += 10
NEAT.DrawPhenotype(img, (0, 0, 250, 250), net)
cv2.imshow("nn_win", img)
cv2.waitKey(1)
pop.Epoch()
# print "Generation:", generation
generations = generation
if best > 15.5:
break
return generations
def run_experiment(config_file, trial_id, n_generations, out_dir, view_results=False, save_results=True):
"""
The function to run the experiment against hyper-parameters
defined in the provided configuration file.
The winner genome will be rendered as a graph as well as the
important statistics of neuroevolution process execution.
Arguments:
config_file: The path to the file with experiment
configuration
trial_id: The ID of current trial
n_generations: The number of evolutionary generations
out_dir: The directory to save intermediate results.
view_results: The flag to control if intermediate results should be displayed after each trial
save_results: The flag to control whether intermediate results should be saved after each trial.
Returns:
The tuple (solution_found, generation, complexity, best_genome_fitness) that has flag indicating whether
solution was found, the generation when solution was found, the complextity of best genome, and the fitness
of best genome.
"""
g = NEAT.Genome(0, 4+1, 0, 1+1, False, NEAT.ActivationFunction.TANH,
NEAT.ActivationFunction.TANH, 0, params, 0)
pop = NEAT.Population(g, params, True, 1.0, trial_id)
# set random seed
seed = int(time.time())
pop.RNG.Seed(seed)
generations = 0
solved = False
best_trial_fitness = 0
best_trial_complexity = 0
for generation in range(n_generations):
genome_list = NEAT.GetGenomeList(pop)
fitness_list = EvaluateGenomeList_Serial(genome_list, evaluate, display=view_results)
NEAT.ZipFitness(genome_list, fitness_list)
generations = generation
best = max(genome_list, key=get_fitness)
best_fitness = best.GetFitness()
complexity = best.NumNeurons() + best.NumLinks()
solved = best_fitness >= cart.MAX_FITNESS # Changed to correspond limit used with other tested libraries
if solved:
best_trial_fitness = best_fitness
best_trial_complexity = complexity
print("Trial: %2d\tgeneration: %d\tfitness: %f\tcomplexity: %d\tseed: %d" %
(trial_id, generations, best_trial_fitness, complexity, seed))
break
# check if best fitness in this generation is better than current maximum
if best_fitness > best_trial_fitness:
best_trial_complexity = complexity
best_trial_fitness = best_fitness
# move to the next epoch
pop.Epoch()
if not solved:
print("Trial: %2d\tFAILED\t\tfitness: %f\tcomplexity: %d\tseed: %d" %
(trial_id, best_trial_fitness, best_trial_complexity, seed))
return solved, generations, best_trial_complexity, best_trial_fitness
def run_neat():
generation = 0
while True:
generation += 1
for genome in NEAT.GetGenomeList(population):
fitness = train(genome, generation=generation)
genome.SetFitness(fitness)
if validate != None:
population.Epoch()
for genome in NEAT.GetGenomeList(population):
fitness = train(genome, generation=generation)
genome.SetFitness(fitness)
current_best = pickle.dumps(population.GetBestGenome())
else:
current_best = pickle.dumps(population.GetBestGenome())
population.Epoch()
yield generation, current_best
def sparseness(genome):
distances = []
for g in NEAT.GetGenomeList(pop):
d = genome.behavior.distance_to(g.behavior)
distances.append(d)
# get the distances from the archive as well
for ab in archive:
distances.append(genome.behavior.distance_to(ab))
distances = sorted(distances)
sp = np.mean(distances[1:ns_K + 1])
return sp
def evolve():
global generations
global global_best
global rrse_list
global mae_list
global rows
global cols
# print("LOO Validation:")
g = NEAT.Genome(0, (cols - 1), (3), 1, False,
NEAT.ActivationFunction.LINEAR,
NEAT.ActivationFunction.LINEAR, 1, params, 1)
for test_idx in range(rows):
pop = NEAT.Population(g, params, True, 1.0, 0)
pop.RNG.Seed(int(time.clock() * 100))
generations = 0
global_best = -99999999
no_improvement = 0
# Run for a maximum of N generations
while no_improvement < 7 and generations < 100: #TODO: make max gens into variable and set via command line
# Reset the population if this path does not seem promising
if (generations > 7 and global_best < -150):
pop = NEAT.Population(g, params, True, 1.0, 0)
pop.RNG.Seed(int(time.clock() * 100))
generations = 0
global_best = -99999999
no_improvement = 0
genome_list = NEAT.GetGenomeList(pop)
fitness_list = []
for genome in genome_list:
fitness_list.append(evaluate(genome, test_idx))
NEAT.ZipFitness(genome_list, fitness_list)
pop.Epoch()
generations += 1
best = max(fitness_list)
#print("[ROW:", test_idx, "] ", -global_best, " (", no_improvement, " g. of no improvement)")
if best > global_best:
no_improvement = 0
global_best = best
else:
no_improvement += 1
#print("LOO test error (RRSE):")
#print(rrse_list[test_idx])
#print("LOO test error (MAE):")
#print(mae_list[test_idx])
print(rrse_list)
print(mae_list)
avg_rrse = np.mean(rrse_list)
avg_mae = np.mean(mae_list)
return [avg_rrse, avg_mae]
def getbest(run):
g = NEAT.Genome(0, substrate.GetMinCPPNInputs(), 0,
substrate.GetMinCPPNOutputs(), False,
NEAT.ActivationFunction.TANH, NEAT.ActivationFunction.TANH,
0, params)
pop = NEAT.Population(g, params, True, 1.0, run)
for generation in range(1000):
# Evaluate genomes
genome_list = NEAT.GetGenomeList(pop)
fitnesses = EvaluateGenomeList_Serial(genome_list,
evaluate_xor,
display=False)
[
genome.SetFitness(fitness)
for genome, fitness in zip(genome_list, fitnesses)
]
print('Gen: %d Best: %3.5f' % (generation, max(fitnesses)))
# Print best fitness
# print("---------------------------")
# print("Generation: ", generation)
# print("max ", max([x.GetLeader().GetFitness() for x in pop.Species]))
# Visualize best network's Genome
net = NEAT.NeuralNetwork()
pop.Species[0].GetLeader().BuildPhenotype(net)
img = np.zeros((500, 500, 3), dtype=np.uint8)
img += 10
NEAT.DrawPhenotype(img, (0, 0, 500, 500), net)
cv2.imshow("CPPN", img)
# Visualize best network's Pheotype
net = NEAT.NeuralNetwork()
pop.Species[0].GetLeader().BuildESHyperNEATPhenotype(
net, substrate, params)
img = np.zeros((500, 500, 3), dtype=np.uint8)
img += 10
NEAT.DrawPhenotype(img, (0, 0, 500, 500), net, substrate=True)
cv2.imshow("NN", img)
cv2.waitKey(1)
if max(fitnesses) > 15.0:
break
# Epoch
generations = generation
pop.Epoch()
return generations
def optimize_neat(config, n_inputs, n_hidden, n_outputs, out_dir):
print('Starting Optimization')
params = NEAT.Parameters()
params.PopulationSize = 60
params.OldAgeTreshold = 10
params.SpeciesMaxStagnation = 20
params.AllowLoops = False
for i in range(config['n_morphogens']):
addTrait(params, 'K%i' % i, (.03, .08))
addTrait(params, 'F%i' % i, (.01, .06))
addTrait(params, 'diffU%i' % i, (.005, .02))
addTrait(params, 'diffV%i' % i, (.0025, .01))
######################## Create NEAT objects ###############################
fs_neat = False
seed_type = 0
out_type = NEAT.ActivationFunction.UNSIGNED_SIGMOID
hidden_type = NEAT.ActivationFunction.UNSIGNED_SIGMOID
genome_prototye = NEAT.Genome(0, n_inputs, n_hidden, n_outputs, fs_neat, \
out_type, hidden_type, seed_type, params, 0)
rand_seed = int(time.time())
pop = NEAT.Population(genome_prototye, params, True, 1.0, rand_seed)
######################## Main evolution loop ###############################
top_fitness = 0 # Fitness function is defined in [0, 1]
top_grid = None
for generation in range(config['generations']):
print('Starting generation', generation)
genomes = NEAT.GetGenomeList(pop)
fitness_list = [simulate_genome(g, config)[0] for g in genomes]
NEAT.ZipFitness(genomes, fitness_list)
max_fitness = max(fitness_list)
print('Generation complete')
print('Max fitness', max_fitness)
print('Mean fitness', np.mean(fitness_list))
if max_fitness > top_fitness:
print('New best fitness')
best_genome = genomes[fitness_list.index(max_fitness)]
_, best_grid = simulate_genome(best_genome, config)
top_fitness = max_fitness
top_grid = best_grid
np.save(out_dir + '/grid_%i' % generation, best_grid)
best_genome.Save(out_dir + '/genome_%i' % generation)
pop.Epoch()
print()
def simGeneration(): genomeList = NEAT.GetGenomeList(pop) #make graph graph = mp.Map(20, 20) for g in genomeList: net = NEAT.NeuralNetwork() g.BuildPhenotype(net) fitness = evaluate(net, graph) g.SetFitness(fitness) pop.Epoch()
def evaluate_solutions(robot, obj_func_coeffs, generation):
best_robot_genome = None
solution_found = False
distances = []
n_items_list = []
# evaluate robot genomes against maze simulation
robot_genomes = NEAT.GetGenomeList(robot.population)
for genome in robot_genomes:
found, distance, n_item = evaluate_individual_solution(
genome=genome, generation=generation, robot=robot)
# store returned values
distances.append(distance)
n_items_list.append(n_item)
if found:
best_robot_genome = genome
solution_found = True
# evaluate novelty scores of robot genomes and calculate fitness
max_fitness = 0
best_coeffs = None
best_distance = 1000
best_novelty = 0
for i, n_item in enumerate(n_items_list):
novelty = robot.archive.evaluate_novelty_score(
item=n_item, n_items_list=n_items_list)
# The sanity check
assert robot_genomes[i].GetID() == n_item.genomeId
# calculate fitness
fitness, coeffs = evaluate_solution_fitness(distances[i], novelty,
obj_func_coeffs)
robot_genomes[i].SetFitness(fitness)
if not solution_found:
# find the best genome in population
if max_fitness < fitness:
max_fitness = fitness
best_robot_genome = robot_genomes[i]
best_coeffs = coeffs
best_distance = distances[i]
best_novelty = novelty
elif best_robot_genome.GetID() == n_item.genomeId:
# store fitness of winner solution
max_fitness = fitness
best_coeffs = coeffs
best_distance = distances[i]
best_novelty = novelty
return best_robot_genome, solution_found, max_fitness, distances, best_coeffs, best_distance, best_novelty
def getbest(run):
g = NEAT.Genome(0, 7, 1, True, NEAT.ActivationFunction.SIGNED_SIGMOID,
NEAT.ActivationFunction.SIGNED_SIGMOID, params)
pop = NEAT.Population(g, params, True, 1.0, run)
for generation in range(1000):
#Evaluate genomes
genome_list = NEAT.GetGenomeList(pop)
fitnesses = NEAT.EvaluateGenomeList_Serial(genome_list,
evaluate_xor,
display=False)
[
genome.SetFitness(fitness)
for genome, fitness in zip(genome_list, fitnesses)
]
# Print best fitness
#print("---------------------------")
#print("Generation: ", generation)
#print("max ", max([x.GetLeader().GetFitness() for x in pop.Species]))
# Visualize best network's Genome
'''
net = NEAT.NeuralNetwork()
pop.Species[0].GetLeader().BuildPhenotype(net)
img = np.zeros((500, 500, 3), dtype=np.uint8)
img += 10
NEAT.DrawPhenotype(img, (0, 0, 500, 500), net )
cv2.imshow("CPPN", img)
# Visualize best network's Pheotype
net = NEAT.NeuralNetwork()
pop.Species[0].GetLeader().Build_ES_Phenotype(net, substrate, params)
img = np.zeros((500, 500, 3), dtype=np.uint8)
img += 10
Utilities.DrawPhenotype(img, (0, 0, 500, 500), net, substrate=True )
cv2.imshow("NN", img)
cv2.waitKey(1)
'''
if max([x.GetLeader().GetFitness() for x in pop.Species]) > 15.0:
break
# Epoch
generations = generation
pop.Epoch()
return generations
def getbest():
g = NEAT.Genome(0, substrate.GetMinCPPNInputs(), 0,
substrate.GetMinCPPNOutputs(), False,
NEAT.ActivationFunction.SIGNED_GAUSS,
NEAT.ActivationFunction.SIGNED_GAUSS, 0, params)
pop = NEAT.Population(g, params, True, 1.0)
for generation in range(1000):
genome_list = NEAT.GetGenomeList(pop)
# fitnesses = NEAT.EvaluateGenomeList_Parallel(genome_list, evaluate)
fitnesses = NEAT.EvaluateGenomeList_Serial(genome_list,
evaluate,
display=False)
[
genome.SetFitness(fitness)
for genome, fitness in zip(genome_list, fitnesses)
]
best = max([x.GetLeader().GetFitness() for x in pop.Species])
# print 'Best fitness:', best
# test
net = NEAT.NeuralNetwork()
pop.Species[0].GetLeader().BuildPhenotype(net)
img = np.zeros((250, 250, 3), dtype=np.uint8)
img += 10
NEAT.DrawPhenotype(img, (0, 0, 250, 250), net)
cv2.imshow("CPPN", img)
net = NEAT.NeuralNetwork()
pop.Species[0].GetLeader().BuildHyperNEATPhenotype(net, substrate)
img = np.zeros((250, 250, 3), dtype=np.uint8)
img += 10
NEAT.DrawPhenotype(img, (0, 0, 250, 250), net, substrate=True)
cv2.imshow("NN", img)
cv2.waitKey(1)
pop.Epoch()
# print "Generation:", generation
generations = generation
if best > 15.5:
break
return generations
def getbest(i):
g = NEAT.Genome(0, 3, 0, 1, False, NEAT.ActivationFunction.UNSIGNED_SIGMOID,
NEAT.ActivationFunction.UNSIGNED_SIGMOID, 0, params)
pop = NEAT.Population(g, params, True, 1.0, i)
pop.RNG.Seed(int(time.clock()*100))
generations = 0
for generation in range(1000):
genome_list = NEAT.GetGenomeList(pop)
fitness_list = EvaluateGenomeList_Serial(genome_list, evaluate, display=False)
NEAT.ZipFitness(genome_list, fitness_list)
pop.Epoch()
generations = generation
best = max(fitness_list)
if best > 15.0:
break
return generations
def getbest():
g = NEAT.Genome(0, 7, 1, False, NEAT.ActivationFunction.SIGNED_SIGMOID,
NEAT.ActivationFunction.SIGNED_SIGMOID, params)
pop = NEAT.Population(g, params, True, 1.0)
for generation in range(2000):
genome_list = NEAT.GetGenomeList(pop)
# fitnesses = NEAT.EvaluateGenomeList_Parallel(genome_list, evaluate)
fitnesses = NEAT.EvaluateGenomeList_Serial(genome_list,
evaluate_xor,
display=True)
[
genome.SetFitness(fitness)
for genome, fitness in zip(genome_list, fitnesses)
]
best = max([x.GetLeader().GetFitness() for x in pop.Species])
net = NEAT.NeuralNetwork()
pop.Species[0].GetLeader().BuildPhenotype(net)
img = np.zeros((500, 500, 3), dtype=np.uint8)
img += 10
NEAT.DrawPhenotype(img, (0, 0, 500, 500), net)
cv2.imshow("CPPN", img)
net = NEAT.NeuralNetwork()
pop.Species[0].GetLeader().Build_ES_Phenotype(net, substrate, params)
img = np.zeros((500, 500, 3), dtype=np.uint8)
img += 10
utilities.DrawPhenotype(img, (0, 0, 500, 500), net, substrate=True)
cv2.imshow("NN", img)
cv2.waitKey(1)
generations = generation
if best > 15.0:
break
pop.Epoch()
return generations
def objective_driven(seed):
i = seed
g = NEAT.Genome(0, 6, 0, 4, False, NEAT.ActivationFunction.SIGNED_SIGMOID,
NEAT.ActivationFunction.SIGNED_SIGMOID, 0, params)
pop = NEAT.Population(g, params, True, 1.0, i)
#pop.RNG.Seed(i)
generations = 0
for generation in range(250):
genome_list = NEAT.GetGenomeList(pop)
fitness_list = NEAT.EvaluateGenomeList_Serial(genome_list,
evaluate,
display=False)
fitness_list = [k[0] for k in fitness_list]
NEAT.ZipFitness(genome_list, fitness_list)
best_fits = [x.GetLeader().GetFitness() for x in pop.Species]
best = max(best_fits)
idx = best_fits.index(best)
print best, pop.Species[idx].GetLeader().GetFitness()
imgs, res = evaluate(pop.Species[idx].GetLeader(),
debug=True,
save="gen%d.ply" % generation)
plt.ion()
plt.clf()
subfig = 1
t_imgs = len(imgs)
for img in imgs:
plt.subplot(t_imgs, 1, subfig)
plt.title("Confidence: %0.2f%%" %
(res[subfig - 1, target_class] * 100.0))
plt.imshow(img)
subfig += 1
plt.draw()
plt.pause(0.1)
plt.savefig("out%d.png" % generation)
pop.Epoch()
generations = generation
return generations
def main():
# for as many as maxGeneration
for generation in range(50):
# retrieve a list of all genomes in the population
genome_list = NEAT.GetGenomeList(pop)
# apply the evaluation function to all genomes
sum = 0
for genome in genome_list:
fitness = evaluate(genome)
genome.SetFitness(fitness)
sum += fitness
avg = sum / float(len(genome_list))
print avg
# at this point we may output some information regarding the progress of evolution, best fitness, etc.
# it's also the place to put any code that tracks the progress and saves the best genome or the entire
# population. We skip all of this in the tutorial.
# advance to the next generation
pop.Epoch()
def getbest(i):
g = NEAT.Genome(0, substrate.GetMinCPPNInputs(), 0,
substrate.GetMinCPPNOutputs(), False,
NEAT.ActivationFunction.TANH, NEAT.ActivationFunction.TANH,
0, params, 0)
pop = NEAT.Population(g, params, True, 1.0, i)
pop.RNG.Seed(i)
for generation in range(max_generations):
genome_list = NEAT.GetGenomeList(pop)
# if sys.platform == 'linux':
# fitnesses = EvaluateGenomeList_Parallel(genome_list, evaluate, display=False)
# else:
fitnesses = EvaluateGenomeList_Serial(genome_list,
evaluate,
display=False)
[
genome.SetFitness(fitness)
for genome, fitness in zip(genome_list, fitnesses)
]
net = NEAT.NeuralNetwork()
pop.GetBestGenome().BuildPhenotype(net)
complexity = "complexity ({}, {})".format(net.NumHiddenNeurons(),
net.NumConnections())
print('Gen: %d/%d Best: %3.5f. %s' %
(generation, max_generations - 1, max(fitnesses), complexity))
best = max(fitnesses)
pop.Epoch()
generations = generation
if best > 15.0:
break
return generations
def evolve_neat(params, generations, out_dir, run_id, pool):
pop = create_initial_population(params)
max_ever = None
t = time.time()
for generation in range(generations):
print(run_id, 'Starting generation', generation)
genomes = NEAT.GetGenomeList(pop)
if pool:
data = [(g, g.GetGenomeTraits(), params) for g in genomes]
fitnesses = pool.starmap(evaluate, data)
else:
fitnesses = [
evaluate(g, g.GetGenomeTraits(), params) for g in genomes
]
NEAT.ZipFitness(genomes, fitnesses)
maxf, meanf = max(fitnesses), sum(fitnesses) / float(len(fitnesses))
runtime = time.time() - t
t = time.time()
print()
print('Generation %i ran in %f, %f per coral' % \
(generation, runtime, runtime/len(genomes)))
print('Max fitness:', maxf, 'Mean fitness:', meanf)
if max_ever is None or maxf > max_ever:
best = pop.GetBestGenome()
max_ever = maxf
print('New best fitness.', best.NumNeurons(), best.NumLinks())
coral = simulate_and_save(best, params, out_dir, generation, maxf,
meanf)[0]
pop.Epoch()
print('#' * 80)
print('Run Complete.')
def evolve():
g = NEAT.Genome(0, 2, 0, 1, False, NEAT.ActivationFunction.LINEAR,
NEAT.ActivationFunction.LINEAR, 0, params, 0)
pop = NEAT.Population(g, params, True, 1.0, 1)
pop.RNG.Seed(int(time.clock()*100))
generations = 0
for generation in range(50):
genome_list = NEAT.GetGenomeList(pop)
fitness_list = []
for genome in genome_list:
fitness_list.append(evaluate(genome))
NEAT.ZipFitness(genome_list, fitness_list)
pop.Epoch()
generations = generation
best = -max(fitness_list)
bestG = pop.GetBestGenome()
plot_nn(bestG)
plt.pause(0.01)
plt.ion()
plt.show(block=False)
print("Mejor fitness [",generation,"]: ",best)
if best < 0.01:
break
testNet = NEAT.NeuralNetwork()
bestG.BuildPhenotype(testNet)
for i in range(10):
testNet.Flush()
testNet.Input(np.array([float(100+2*i), 1]))
for _ in range(2):
testNet.Activate()
o = testNet.Output()
print(100+2*i,"/ 2 = ",o[0])
return generations
def run_experiment(params, trial_id, n_generations, out_dir=None, view_results=False, save_results=True):
g = NEAT.Genome(0, 3, 0, 1, False, NEAT.ActivationFunction.UNSIGNED_SIGMOID,
NEAT.ActivationFunction.UNSIGNED_SIGMOID, 0, params, 0)
pop = NEAT.Population(g, params, True, 1.0, trial_id)
# set random seed
seed = int(time.time())
pop.RNG.Seed(seed)
generations = 0
solved = False
max_fitness = 0
complexity = 0
for generation in range(n_generations):
genome_list = NEAT.GetGenomeList(pop)
fitness_list = EvaluateGenomeList_Serial(genome_list, evaluate, display=view_results)
NEAT.ZipFitness(genome_list, fitness_list)
generations = generation
best = max(genome_list, key=get_fitness)
best_fitness = best.GetFitness()
complexity = best.NumNeurons() + best.NumLinks()
solved = best_fitness > 15.5 # Changed to correspond limit used with other tested libraries
if solved:
max_fitness = best_fitness
print("Trial: %2d\tgeneration: %d\tfitness: %f\tcomplexity: %d\tseed: %d" % (trial_id, generations, max_fitness, complexity, seed))
break
# check if best fitness in this generation is better than current maximum
max_fitness = max(best_fitness, max_fitness)
# move to the next epoch
pop.Epoch()
if not solved:
print("Trial: %2d\tFAILED\t\tfitness: %f\tcomplexity: %d\tseed: %d" % (trial_id, max_fitness, complexity, seed))
return solved, generations, complexity, max_fitness
def evolutionary_run(gens, pop_size, output_path, run_num, numofjoints):
""" Conduct an evolutionary run using the snake and muscle model.
"""
params = NEAT.Parameters()
params.CompatTreshold = 5.0
params.CompatTresholdModifier = 0.3
params.YoungAgeTreshold = 15
params.SpeciesMaxStagnation = 1000
params.OldAgeTreshold = 35
params.MinSpecies = 1
params.MaxSpecies = 25
params.RouletteWheelSelection = False
params.RecurrentProb = 0.25
params.OverallMutationRate = 0.33
params.MutateWeightsProb = 0.90
params.WeightMutationMaxPower = 1.0
params.WeightReplacementMaxPower = 5.0
params.MutateWeightsSevereProb = 0.5
params.WeightMutationRate = 0.75
params.MaxWeight = 20
params.MutateAddNeuronProb = 0.4
params.MutateAddLinkProb = 0.4
params.MutateRemLinkProb = 0.05
params.CrossoverRate = 0.4
assert pop_size >= 0, "wrong population size argument! pop_size: %d" % pop_size
params.PopulationSize = pop_size
# worm has only one dof (turning around y-axis) per joint
num_outputs = numofjoints
# the inputs for the ANN are the 7 current joint positions and the amplitude of a sine wave as well as a bias
num_inputs = numofjoints + 1 + 1
# Initialize the population
# Genome(ID, NumInputs, NumHidden, NumOutputs, ActivationFunction?, Output activation function, Hidden layer acitvation function, seed, params)
genome = NEAT.Genome(0, num_inputs, 0, num_outputs, False,
NEAT.ActivationFunction.SIGNED_SIGMOID,
NEAT.ActivationFunction.SIGNED_SIGMOID, 0, params)
# Population(Genome, params, randomizedweights?, randomrange)
population = NEAT.Population(genome, params, True, 1.0)
genome_list = NEAT.GetGenomeList(population)
morph_pop = MorphGenomes(pop_size)
for ind in genome_list:
morph_pop.addIndividual(ind.GetID())
morph_genomes = morph_pop.getGenomes()
# ANN genome and morphology genome are zipped together
# Zip the two genome components together for use in the parallel call.
zip_args = [(ind, morph_genomes[ind.GetID()]) for ind in genome_list]
mnlog.write_population_statistics_headers(output_path + str(run_num) +
"_fitnesses.dat")
# Setup multiprocessing
cores = mpc.cpu_count()
#cores = 1
pool = mpc.Pool(initializer=initProcess,
initargs=(
GlobalVarWorkaround.args,
GlobalVarWorkaround.man,
GlobalVarWorkaround.worm,
),
processes=cores)
assert gens >= 0, "wrong number of generations as argument! gens: %d" % gens
for gen in xrange(gens):
print gen
#fitnesses = map(evaluate_individual,zip_args) # serial execution
fitnesses = pool.map(evaluate_individual, zip_args)
replace_nanfitnesses(fitnesses)
for g, f in zip(genome_list, fitnesses):
g.SetFitness(f)
print("Generation " + str(gen) + "\t: " + str(max(fitnesses)))
# Write the best performing individual to a file.
mnlog.write_best_individual(
output_path + "best_individuals/Evo_NEAT_run_" + str(run_num) +
"_best_gen_" + str(gen) + ".dat",
genome_list[fitnesses.index(max(fitnesses))])
morph_pop.logGenome(
genome_list[fitnesses.index(max(fitnesses))].GetID(),
"Evo_NEAT_run_" + str(run_num) + "_best_gen_" + str(gen),
output_path)
# Write information about the best individual we wrote.
with open(
output_path + "/" + str(run_num) +
"_best_individuals_logging.dat", "a") as f:
f.write("Generation: "+str(gen)+" Individual is: "+str(genome_list[fitnesses.index(max(fitnesses))].GetID())+\
" Fitness is: "+str(max(fitnesses))+"\n")
# Log the progress of the entire population.
mnlog.write_population_statistics(
output_path + str(run_num) + "_fitnesses.dat", genome_list,
fitnesses, gen)
# Log the final population for later evaluation.
if gen == gens - 1:
population.Save(output_path + "run_" + str(run_num) +
"_population_generation_" + str(gen) + ".dat")
# Create the next generation
population.Epoch()
genome_list = NEAT.GetGenomeList(population)
morph_pop.NextGen()
zip_args = []
for ind in genome_list:
# PID .. parent ID
pid1 = ind.GetPID1()
# if pid2 is negative it means that no crossover happend!
pid2 = ind.GetPID2()
gid = ind.GetID()
# Handle Crossover
morph_pop.Crossover(gid, pid1, pid2)
# Handle Mutation
morph_pop.MutatePopulation()
# Zip the arguments for calling the evolution function.
morph_genomes = morph_pop.getGenomes()
zip_args = [(ind, morph_genomes[ind.GetID()]) for ind in genome_list]
action = np.argmax(out)
observation, reward, done, info = env.step(action)
if done: break
f += reward
avg_reward += f
return avg_reward
try:
for generation in range(100):
for i_episode, genome in enumerate(NEAT.GetGenomeList(pop)):
net = NEAT.NeuralNetwork()
genome.BuildPhenotype(net)
avg_reward = 0
for trial in range(trials):
avg_reward += do_trial()
avg_reward /= trials
#print(avg_reward)
genome.SetFitness(1000000 + avg_reward)