def test_evolution(self):
p = Population(10, 3, 1)
entire_population = []
for k, v in p.population.items():
if k != 's0':
entire_population = entire_population + v
pass
p.evolve_species(p.population.get('s0'), entire_population, 2)
class NEATRunner(BaseTrainer):
def __init__(
self,
config: Config,
evaluator: GymEvaluator,
reporters: Optional[Sequence[BaseReporter]] = None,
num_workers: Optional[int] = multiprocessing.cpu_count(),
):
self._evaluator = evaluator
self._population = Population(config)
reporters = reporters or []
for reporter in reporters:
self._population.add_reporter(reporter)
self._num_workers = num_workers
def _train(self, num_frames: Optional[int],
stop_time: Optional[int]) -> DefaultGenome:
if self._num_workers is None:
func = lambda g, c: self._evaluate_population_fitness(
g, c, num_frames)
else:
parallel = ParallelEvaluator(
num_workers=self._num_workers,
evaluator=self._evaluator,
max_num_frames=num_frames or float('inf'),
)
func = parallel.evaluate
if stop_time is not None:
# it may not be 100% reliable but it's the best we can achieve without writing a custom
# parallel executor
func = _timeout_func(func, time(), stop_time)
try:
return self._population.run(
fitness_function=func,
n=float('inf'),
)
except TimeoutError:
return self._population.best_genome
def _evaluate_population_fitness(
self,
genomes: Sequence[Tuple[int, DefaultGenome]],
config: Config,
max_num_frames: int,
):
if self._evaluator.num_frames >= max_num_frames:
raise TimeoutError()
for _, genome in genomes:
genome.fitness, num_frames = self._evaluator.evaluate(
genome, config)
self._evaluator.num_frames += num_frames
def __init__(
self,
config: Config,
evaluator: GymEvaluator,
reporters: Optional[Sequence[BaseReporter]] = None,
num_workers: Optional[int] = multiprocessing.cpu_count(),
):
self._evaluator = evaluator
self._population = Population(config)
reporters = reporters or []
for reporter in reporters:
self._population.add_reporter(reporter)
self._num_workers = num_workers
def test_evolution(self):
p = Population(10, 3, 1)
status = p.get_status()
for s in status.keys():
output = []
for i in range(status.get(s, 0)):
output.append(p.run(s, i, [1, 2, 3]))
p.set_score(s, i, 1)
p.evolve()
def learn_cartpole(config_path):
env = gym.make('CartPole-v0')
env._max_episode_steps = max_evaluation_steps
# Load configuration.
config = Config(StateMachineGenome, DefaultReproduction,
DefaultSpeciesSet, DefaultStagnation,
config_path)
# Create the population, which is the top-level object for a NEAT run.
p = Population(config)
# Add a stdout reporter to show progress in the terminal.
p.add_reporter(StdOutReporter(True))
stats = StatisticsReporter()
p.add_reporter(stats)
# Run for up for the given number of generations
f = lambda genomes, config: eval_genomes(genomes, config, env=env)
winner = p.run(f, num_generations)
input("Press Enter to continue...")
net = StateMachineNetwork.create(winner, config)
eval_network(net, env, True)
def test_another_crossover(self):
g1 = Genome(3, 1)
g2 = Genome(3, 1)
node_i = 4
conn_i = 0
for generation in range(10):
conn_i = g1.mutate_connections(conn_i)
node_i, conn_i = g1.mutate_nodes(node_i, conn_i)
conn_i = g2.mutate_connections(conn_i)
node_i, conn_i = g2.mutate_nodes(node_i, conn_i)
a1, a2 = Population.align_genome(g1, g2)
g1 = Population.crossover(a1, a2, g1.get_input_nodes(), g1.get_output_nodes())
g2 = Population.crossover(a1, a2, g1.get_input_nodes(), g1.get_output_nodes())
Printer(g1).print()
def test_align_genome(self):
g1 = Genome(3, 1)
g1.create_node(Node.TYPE_HIDDEN, 5)
g1.connect_nodes_by_id(1, 4, 1)
g1.connect_nodes_by_id(2, 4, 2)
g1.connect_nodes_by_id(3, 4, 3)
g1.connect_nodes_by_id(2, 5, 4)
g1.connect_nodes_by_id(5, 4, 5)
g1.connect_nodes_by_id(1, 5, 8)
g2 = Genome(3, 1)
g2.create_node(Node.TYPE_HIDDEN, 5)
g2.create_node(Node.TYPE_HIDDEN, 6)
g2.connect_nodes_by_id(1, 4, 1)
g2.connect_nodes_by_id(2, 4, 2)
g2.connect_nodes_by_id(3, 4, 3)
g2.connect_nodes_by_id(2, 5, 4)
g2.connect_nodes_by_id(5, 4, 5)
g2.connect_nodes_by_id(5, 6, 6)
g2.connect_nodes_by_id(6, 4, 7)
g2.connect_nodes_by_id(3, 5, 9)
g2.connect_nodes_by_id(1, 6, 10)
# p = Population(2, 3, 1)
# p.population['s0'] = [g1, g2]
a1, a2 = Population.align_genome(g1, g2)
self.assertEqual(1, a1[0].get_innovation())
self.assertEqual(2, a1[1].get_innovation())
self.assertEqual(3, a1[2].get_innovation())
self.assertEqual(4, a1[3].get_innovation())
self.assertEqual(5, a1[4].get_innovation())
self.assertEqual(None, a1[5])
self.assertEqual(None, a1[6])
self.assertEqual(8, a1[7].get_innovation())
self.assertEqual(None, a1[8])
self.assertEqual(None, a1[9])
self.assertEqual(1, a2[0].get_innovation())
self.assertEqual(2, a2[1].get_innovation())
self.assertEqual(3, a2[2].get_innovation())
self.assertEqual(4, a2[3].get_innovation())
self.assertEqual(5, a2[4].get_innovation())
self.assertEqual(6, a2[5].get_innovation())
self.assertEqual(7, a2[6].get_innovation())
self.assertEqual(None, a2[7])
self.assertEqual(9, a2[8].get_innovation())
self.assertEqual(10, a2[9].get_innovation())
def play():
env = gym.make('CartPole-v1')
ob = env.reset()
p = Population.load(FULLNAME)
winner = p.get_winner()
while True:
action = winner.run(ob)
ob, reward, done, info = env.step(np.argmax(action))
env.render()
if done:
ob = env.reset()
def test_flow(self):
p = Population(10, 3, 1)
for _ in range(100):
status = p.get_status()
for s in status.keys():
output = []
for i in range(status.get(s, 0)):
out = p.run(s, i, [1, 2, 3])[0]
output.append(out)
p.set_score(s, i, random.randrange(1, 10))
print(s, output)
p.evolve()
# print a sample
pr = Printer(p.population[next(iter(p.population))][0])
pr.print()
def play():
env = gym.make('LunarLanderContinuous-v2')
ob = env.reset()
p = Population.load(FULLNAME)
winner = p.get_winner()
while True:
action = action_final_activation(winner.run(ob))
print('Action: {}'.format(action))
ob, reward, done, info = env.step(action)
env.render()
if done:
ob = env.reset()
def test_crossover(self):
g1 = Genome(3, 1)
g1.create_node(Node.TYPE_HIDDEN, 5)
g1.connect_nodes_by_id(1, 4, 1)
g1.connect_nodes_by_id(2, 4, 2)
g1.connect_nodes_by_id(3, 4, 3)
g1.connect_nodes_by_id(2, 5, 4)
g1.connect_nodes_by_id(5, 4, 5)
g1.connect_nodes_by_id(1, 5, 8)
g2 = Genome(3, 1)
g2.create_node(Node.TYPE_HIDDEN, 5)
g2.create_node(Node.TYPE_HIDDEN, 6)
g2.connect_nodes_by_id(1, 4, 1)
g2.connect_nodes_by_id(2, 4, 2)
g2.connect_nodes_by_id(3, 4, 3)
g2.connect_nodes_by_id(2, 5, 4)
g2.connect_nodes_by_id(5, 4, 5)
g2.connect_nodes_by_id(5, 6, 6)
g2.connect_nodes_by_id(6, 4, 7)
g2.connect_nodes_by_id(3, 5, 9)
g2.connect_nodes_by_id(1, 6, 10)
p = Population(2, 3, 1)
p.population[0] = g1
p.population[1] = g2
a1, a2 = Population.align_genome(g1, g2)
g3 = Population.crossover(a1, a2, g1.get_input_nodes(), g1.get_output_nodes())
self.assertEqual(10, len(g3.connections))
self.assertEqual(6, len(g3.nodes))
self.assertEqual(1, g3.select_connection_by_innovation(1).get_innovation())
self.assertEqual(2, g3.select_connection_by_innovation(2).get_innovation())
self.assertEqual(3, g3.select_connection_by_innovation(3).get_innovation())
self.assertEqual(4, g3.select_connection_by_innovation(4).get_innovation())
self.assertEqual(5, g3.select_connection_by_innovation(5).get_innovation())
self.assertEqual(6, g3.select_connection_by_innovation(6).get_innovation())
self.assertEqual(7, g3.select_connection_by_innovation(7).get_innovation())
self.assertEqual(8, g3.select_connection_by_innovation(8).get_innovation())
self.assertEqual(9, g3.select_connection_by_innovation(9).get_innovation())
self.assertEqual(10, g3.select_connection_by_innovation(10).get_innovation())
self.assertEqual((2, 3), Population.calculate_excess_disjoint(g1, g2))
self.assertEqual(5, species_distance(g1, g2))
def _epoch(self):
if self.current_generation == 0:
self.population = Population(self.configuration)
else:
# Create the next generation from the current generation.
self.population.population = self.population.reproduction.reproduce(
self.population.config, self.population.species,
self.population.config.pop_size, self.population.generation)
# Check for complete extinction.
if not self.population.species.species:
self.population.reporters.complete_extinction()
# If requested by the user, create a completely new population,
# otherwise raise an exception.
if self.population.config.reset_on_extinction:
self.population.population = self.population.reproduction.create_new(
self.population.config.genome_type,
self.population.config.genome_config,
self.population.config.pop_size)
else:
raise CompleteExtinctionException()
# Evaluate all genomes using the user-provided function.
self._eval_genomes(list(iteritems(self.population.population)),
self.population.config)
# Gather and report statistics.
best = None
for g in itervalues(self.population.population):
if best is None or g.fitness > best.fitness:
best = g
# Track the best genome ever seen.
if self.population.best_genome is None or best.fitness > self.population.best_genome.fitness:
self.population.best_genome = best
self.champion = self.population.best_genome
# Divide the new population into species.
self.population.species.speciate(self.population.config,
self.population.population,
self.population.generation)
return self.stopping_criterion.evaluate(self)
def print_population():
g = Population.load(FULLNAME).get_winner()
Printer(g).print()
def ttt_test(config):
gens = config['max_generations']
pop = None
start_genome = None
#//Hold records for each run
evals = [ 0 for i in range(neatconfig.num_runs) ]
genes = [ 0 for i in range(neatconfig.num_runs) ]
nodes = [ 0 for i in range(neatconfig.num_runs) ]
# Made as lists for pass-by-reference
winnernum = [ 0 ]
winnergenes = [ 0 ]
winnernodes = [ 0 ]
#//For averaging
totalevals = 0
totalgenes = 0
totalnodes = 0
dprint(DEBUG_INFO, "START TTT TEST")
if config['seed_with_start_genome']:
gene_filename = neatconfig.genedir + "/" + config['start_genome_file']
iFile = open(gene_filename, "r")
if not iFile:
dprint(DEBUG_ERROR, "Unable to open starting genome file %s." % (gene_filename, ))
return pop
dprint(DEBUG_INFO, "Opened start genome file %s." % (gene_filename, ))
#//Read in the start Genome
dprint(DEBUG_INFO, "Reading in the start genome")
line = iFile.readline()
words = line.strip().split()
curword = words[0]
gid = int(words[1])
if curword != "genomestart":
dprint(DEBUG_ERROR, "Bad starting genome file %s." % (gene_filename, ))
return pop
dprint(DEBUG_INFO, "Reading in Genome id %d." % (gid,))
start_genome = Genome()
start_genome.SetFromFile(gid, iFile)
iFile.close()
dprint(DEBUG_INFO, "Verifying start_genome")
if not start_genome.verify():
dprint(DEBUG_ERROR, "genome.verify() failed:", start_genome.deep_string())
#Complete a number of runs
for expcount in range(neatconfig.num_runs):
if config['seed_with_start_genome']:
#//Spawn the Population
dprint(DEBUG_INFO, "Spawning Population off Genome")
pop = Population()
pop.SetFromGenome(start_genome, neatconfig.pop_size)
dprint(DEBUG_INFO, "Verifying Spawned Population[%d]" % (expcount, ))
if not pop.verify():
dprint(DEBUG_ERROR, "Population[%d] verification failed" % (expcount, ))
elif config['seed_with_previous_population']:
population_filename = neatconfig.genedir + "/" + config['previous_population_file']
dprint(DEBUG_INFO, "Reading Population from %s" % (population_filename, ))
pop = Population()
pop.SetFromFilename(population_filename)
dprint(DEBUG_INFO, "Verifying Start Population[%d]" % (expcount, ))
if not pop.verify():
dprint(DEBUG_ERROR, "Population[%d] verification failed" % (expcount, ))
#// evolve up to gens generations
gen = 1
while gen <= gens:
dprint(DEBUG_INFO, "Evaluating Spawned Population[%d] Epoch[%d]" % (expcount, gen))
# if not pop.verify():
# dprint(DEBUG_ERROR, "Population[%d] Epoch[%d] verification failed" % (expcount, gen))
#print "Epoch", gen
generation_filename = neatconfig.generationdir + "/tttgen_%d" % (gen,)
# Evaluate one generation, checking for a successful end
#//Check for success
if ttt_epoch(pop, gen, generation_filename, winnernum, winnergenes, winnernodes):
evals[expcount] = neatconfig.pop_size * (gen-1) + winnernum[0]
genes[expcount] = winnergenes[0]
nodes[expcount] = winnernodes[0]
break
# in case we want to change after run has started
config = ttt_read_config(neatconfig.configdir + '/ttt.config')
gens = config['max_generations']
gen += 1
# end of generation loop
if g_found_optimal:
break
# end of num_runs loop
#//Average and print stats
dprint(DEBUG_INFO, "Nodes: ")
for expcount in range(neatconfig.num_runs):
dprint(DEBUG_INFO, nodes[expcount])
totalnodes += nodes[expcount]
dprint(DEBUG_INFO, "Genes: ")
for expcount in range(neatconfig.num_runs):
dprint(DEBUG_INFO, genes[expcount])
totalgenes += genes[expcount]
dprint(DEBUG_INFO, "Evals ")
samples = 0
for expcount in range(neatconfig.num_runs):
dprint(DEBUG_INFO, evals[expcount])
if evals[expcount] > 0:
totalevals += evals[expcount]
samples += 1
if samples > 0:
avgnodes = float(totalnodes)/samples
avggenes = float(totalgenes)/samples
avgevals = float(totalevals)/samples
else:
avgnodes = 0
avggenes = 0
avgevals = 0
dprint(DEBUG_INFO, "Failures:", (neatconfig.num_runs - samples), "out of", neatconfig.num_runs, "runs")
dprint(DEBUG_INFO, "Average Nodes:", avgnodes)
dprint(DEBUG_INFO, "Average Genes:", avggenes)
dprint(DEBUG_INFO, "Average Evals:", avgevals)
return pop
def xor_test(gens):
gene_filename = neatconfig.genedir + "/xorstartgenes"
pop = None
start_genome = None
#//Hold records for each run
evals = [ 0 for i in range(neatconfig.num_runs) ]
genes = [ 0 for i in range(neatconfig.num_runs) ]
nodes = [ 0 for i in range(neatconfig.num_runs) ]
# Made as lists for pass-by-reference
winnernum = [ 0 ]
winnergenes = [ 0 ]
winnernodes = [ 0 ]
#//For averaging
totalevals = 0
totalgenes = 0
totalnodes = 0
iFile = open(gene_filename, "r")
if not iFile:
dprint(DEBUG_ERROR, "Unable to open starting genome file %s." % (gene_filename, ))
return pop
print "START XOR TEST"
#//Read in the start Genome
print "Reading in the start genome"
line = iFile.readline()
words = line.strip().split()
curword = words[0]
gid = int(words[1])
if curword != "genomestart":
print "Bad starting genome file %s." % (gene_filename, )
return pop
print "Reading in Genome id %d." % (gid,)
start_genome = Genome()
start_genome.SetFromFile(gid, iFile)
iFile.close()
print "Verifying start_genome"
if not start_genome.verify():
dprint(DEBUG_ERROR, "genome.verify() failed:", start_genome.deep_string())
#Complete a number of runs
for expcount in range(neatconfig.num_runs):
#//Spawn the Population
print "Spawning Population off Genome2"
pop = Population()
pop.SetFromGenome(start_genome, neatconfig.pop_size)
print "Verifying Spawned Population[%d]" % (expcount, )
if not pop.verify():
dprint(DEBUG_ERROR, "Population[%d] verification failed" % (expcount, ))
#// evolve up to gens generations
for gen in range(1, gens+1):
print "Verifying Spawned Population[%d] Epoch[%d]" % (expcount, gen)
if not pop.verify():
dprint(DEBUG_ERROR, "Population[%d] Epoch[%d] verification failed" % (expcount, gen))
#print "Epoch", gen
generation_filename = neatconfig.generationdir + "/gen_%d" % (gen,)
# Evaluate one generation, checking for a successful end
#//Check for success
if xor_epoch(pop, gen, generation_filename, winnernum, winnergenes, winnernodes):
evals[expcount] = neatconfig.pop_size * (gen-1) + winnernum[0]
genes[expcount] = winnergenes[0]
nodes[expcount] = winnernodes[0]
break
# end of generation loop
if g_found_optimal:
break
# end of num_runs loop
#//Average and print stats
print "Nodes: "
for expcount in range(neatconfig.num_runs):
print nodes[expcount]
totalnodes += nodes[expcount]
print "Genes: "
for expcount in range(neatconfig.num_runs):
print genes[expcount]
totalgenes += genes[expcount]
print "Evals "
samples = 0
for expcount in range(neatconfig.num_runs):
print evals[expcount]
if evals[expcount] > 0:
totalevals += evals[expcount]
samples += 1
if samples > 0:
avgnodes = float(totalnodes)/samples
avggenes = float(totalgenes)/samples
avgevals = float(totalevals)/samples
else:
avgnodes = 0
avggenes = 0
avgevals = 0
print "Failures:", (neatconfig.num_runs - samples), "out of", neatconfig.num_runs, "runs"
print "Average Nodes:", avgnodes
print "Average Genes:", avggenes
print "Average Evals:", avgevals
return pop
def test_empty_crossover(self):
g1 = Genome(3, 1)
g2 = Genome(3, 1)
a1, a2 = Population.align_genome(g1, g2)
c = Population.crossover(a1, a2, g1.get_input_nodes(), g1.get_output_nodes())
self.assertGreaterEqual(4, len(c.nodes))
def __init__(self, network, populationsize):
self.population = Population(populationsize, network)
self.network = network
class Trainer:
def __init__(self, network, populationsize):
self.population = Population(populationsize, network)
self.network = network
#The fitness function rewards based on how accurate the output is
#Optimal fitness is reached if input[0] is substracted from input[1]
#Iterations is the amount of training cycles
#Mutationrate is the chance that offspring is mutated
#Elitism is the amount of genomes that are put into the next generation without change
def train(self, iterations, mutationrate, elitism):
oldFitnessSum = 0
newFitnessSum = 0
for i in range(iterations):
#EVALUATION
for member in self.population.population:
self.network.loadDNA(member)
#repeat network evalutation 5 times for accuracy
for _ in range(5):
#two random inputs
inputs = []
inputs.append(uniform(-5, 10))
inputs.append(uniform(-10, 20))
self.network.setInputs(inputs)
result = self.network.run()[0]
expectation = inputs[0] - inputs[1]
#perfect fitness if there is no difference
#we also square the difference because we want some dope ass inequality
member.fitness += 100 - pow(abs(result - expectation), 2)
#PRINT GENERATION PROGRESS
if (i % 2 == 0):
oldFitnessSum = sum(dna.fitness
for dna in self.population.population)
else:
newFitnessSum = sum(dna.fitness
for dna in self.population.population)
if (oldFitnessSum != 0 and newFitnessSum != 0):
print("Gen Nr: " + str(i))
print("Difference between Generation Fitness Sum: " +
str(round(newFitnessSum - oldFitnessSum, 2)))
percentage = (newFitnessSum -
oldFitnessSum) / oldFitnessSum * 100
print("Percentage difference between generation: " +
str(round(percentage, 2)) + "%")
print("New Fitness Sum: " + str(round(newFitnessSum, 2)))
print(
"Highest Fitness: " +
str(max(dna.fitness
for dna in self.population.population)))
#CREATE NEW POPULATION
#best genomes at beginning of list, so we can include them easily (elitism)
self.population.population.sort(key=lambda x: x.fitness,
reverse=True)
#dont make a new generation on last iteration, so we can
#take the population as-is and evaluate it one more time in main.py
if (i < iterations - 1):
population2 = []
#add 5 best genomes unchanged
for elitismIndex in range(elitism):
self.population.population[elitismIndex].fitness = 0
population2.append(
self.population.population[elitismIndex])
for _ in range(len(self.population.population) - elitism):
parent1 = self.population.rouletteSelect()
parent2 = self.population.rouletteSelect()
child = self.population.crossover(parent1, parent2)
if (uniform(0, 1) < mutationrate):
child = self.population.mutate(child)
child.fitness = 0
population2.append(child)
self.population.population = population2
return self.population
import gym
import numpy as np
from neat import Population, Printer
env = gym.make('MountainCar-v0')
done = False
strain_count = 10
passing_score = 500
p = Population(1000, 2, 3)
target_reward = 0.5
max_reward = -999999
winner = None
max_position = -1.2
min_position = 0.6
while True:
status = p.get_status()
for s in status.keys():
output = []
for i in range(status.get(s, 0)):
ob = env.reset()
reward_sum = 200
while True:
action = p.run(s, i, ob)
ob, reward, done, info = env.step(np.argmax(action))
max_position = np.max([max_position, ob[0]])
min_position = np.min([min_position, ob[0]])
reward_sum = reward_sum + reward
if done:
def test_status(self):
p = Population(10, 3, 1)
status = p.get_status()
self.assertEqual({'s0': 10}, status)
def test_init(self):
p = Population(10, 3, 1)
self.assertEqual(10, len(p.population.get('s0')))
self.assertEqual(5, p.node_innovation)
self.assertEqual(1, p.conn_innovation)
def train():
env = gym.make('CartPole-v1')
try:
p = Population.load(FULLNAME)
print('Existing state loaded')
except FileNotFoundError as e:
print('Creating new state')
p = Population(1000, env.observation_space.shape[0],
env.action_space.n)
max_reward = 0
while True:
try:
status = p.get_status()
for s in status.keys():
for i in range(status.get(s, 0)):
ob = env.reset()
reward_sum = 0
while True:
action = p.run(s, i, ob)
ob, reward, done, info = env.step(np.argmax(action))
reward_sum = reward_sum + reward
if done:
break
p.set_score(s, i, reward_sum)
max_reward = np.max([reward_sum, max_reward])
print(p.generation, max_reward, p.population.keys())
p.evolve()
except KeyboardInterrupt as e:
try:
print('\nsaving before exit')
p.save(FULLNAME)
sys.exit('Bye!')
except RuntimeError as e:
print('error saving: {}'.format(str(e)))
import sys
sys.path.insert(0, 'H:\\Documents\\Coding_projects')
sys.path.insert(1, 'H:\\Documents\\Coding_projects\\neat')
from neat import Population
from neat import Network
from neat import DataConf
import csv
import pygame
pop = Population(1000, 3, 1, False)
nets = pop.get_pop()
iterations = 100
pygame.init()
win = pygame.display.set_mode((500,500))
data_handler = DataConf(win)
with open('H:\\Documents\\Coding_projects\\xor\\data\\scores.csv', mode='w') as csv_file:
csv_file = csv.writer(csv_file)
csv_file.writerow(["Score"])
for i in range(iterations):
fitness_tot = []
for n in nets:
answer = n.feed_forward([0, 0, 1])[0]
n.fitness += abs(0 - answer)
answer = n.feed_forward([1, 0, 1])[0]
