def genetic():
ITERATIONS = 30
POP_SIZE = 40
CROSSOVER_RATE = 0.5
MUTATION_RATE = 0.01
TOURNAMENT_SIZE = 6
task = read('generatedTaskFile.csv')
print("Here it goes...")
population = init_population(task.numberOfObjects, POP_SIZE, task)
i = 0
while i < ITERATIONS:
print("New iteration")
j = 0
new_population = Population(task)
new_population.listOfIndividuals = []
while j < POP_SIZE:
parent1 = tournament(population, TOURNAMENT_SIZE, task)
parent2 = tournament(population, TOURNAMENT_SIZE, task)
child = Individual()
child.itemsTaken = crossover(parent1, parent2, CROSSOVER_RATE)
mutate(child, MUTATION_RATE)
new_population.addIndividualToPopulation(child)
j += 1
population = new_population
i += 1
return population.best()
def crossover_onepoint(p1, p2, mu, crossover_rate, passes):
c1 = copy.deepcopy(p1)
c2 = copy.deepcopy(p2)
# alpha=0.3
count = 0
point = np.random.randint(0, 32)
for list1, list2 in zip(p1['table'], p2['table']):
if (count >= point):
c1['table'][count][1] = list2[1]
c2['table'][count][1] = list1[1]
count = count + 1
if (count >= 32):
break
c1m = mutate.mutate(c1, mu)
c2m = mutate.mutate(c2, mu)
fs.finalState(c1m, passes)
fit.calculate_fitness(c1m)
fs.finalState(c1m, passes)
fit.calculate_fitness(c1m)
return c1m, c2m
def test_mutate_substitutions():
an_instance = genome.Genome()
an_instance.expand(n=10)
a_sequence = copy_genome(an_instance.sequence_A)
mutate.mutate(an_instance, {'singles': 50, 'expansions': 0, 'deletions': 0})
set_A = set(an_instance.sequence_A)
set_B = set(a_sequence)
assert len(set_A.intersection(set_B)) < len(a_sequence) # overlap
def test_mutate_expansions():
an_instance = genome.Genome()
an_instance.expand(n=10)
a_sequence = copy_genome(an_instance.sequence_A)
mutate.mutate(an_instance, {
'singles': 0.0,
'expansions': 50.0,
'deletions': 0.0}
)
assert len(an_instance.sequence_A) > len(a_sequence)
def nextGen(self):
self.save()
totalAvgFit = 0
remaining = 0
avgPopFit = 0
for spec in self.populationSpecies:
remaining += spec.removeHalf()
tmp = spec.calcAvgFitness()
totalAvgFit += tmp
avgPopFit += (tmp * len(spec.subpopulation))
children = []
#self.removeStale()
self.removeWeak()
print()
print("Generation", self.generationNumber)
print("Max Fitness:", self.maxFitness)
print("Innovation Number:", self.globalInnovationNumber)
print("No. of Species:", len(self.populationSpecies))
print("Avg pop fitness:", avgPopFit / remaining)
print("Total Population:", self.index - 1)
for spec in self.populationSpecies:
n = math.floor(
spec.avgFitness / totalAvgFit) * self.numberOfIndividuals - 1
for i in range(n):
ch = spec.getChild()
if ch:
self.globalInnovationNumber, ch = mutate.mutate(
ch, self.globalInnovationNumber)
children.append(ch)
spec.removeAllExceptOne()
while self.numberOfIndividuals > len(children) + len(
self.populationSpecies):
spec = self.populationSpecies[random.randrange(
0, len(self.populationSpecies))]
ch = spec.getChild()
if ch:
self.globalInnovationNumber, ch = mutate.mutate(
ch, self.globalInnovationNumber)
children.append(ch)
M = self.numberOfIndividuals - len(self.populationSpecies)
for i in range(0, M):
self.addChromosome(children[i])
print("Children produced:", M)
print()
self.generationNumber += 1
self.index = 0
self.maxFitness = 0
def next_generation(self, mutation_setup):
""" Swap current population with new individuals
"""
# matching individuals
# sorted by fitness
self.individuals.sort(key=lambda x: x.fitness)
for i, m_individual in enumerate(self.individuals):
if m_individual.paired_with != -1:
continue # skip already matched individuals
# for now: match next one
m_try = i + 1 if i < len(self.individuals) else i - randint(1, 3)
# safechecks
if m_try < 0 or m_try >= len(self.individuals):
continue
if m_try == i:
continue
if self.individuals[m_try].paired_with != -1:
continue
# assign mate
m_individual.paired_with = m_try
self.individuals[m_try].paired_with = i
# mutations
for i, _ in enumerate(self.individuals):
mutate(self.individuals[i].genome, mutation_setup)
# crossing overs
for i, _ in enumerate(self.individuals):
if self.individuals[i].paired_with == -1:
continue
crossover(self.individuals[i].genome, events=1)
# offspring
new_individuals = []
for i, o_individual in enumerate(self.individuals):
if o_individual.paired_with == -1:
continue # skip lone individuals
# probability of offspring dictated by fitness
# algorithm: 100% for every full 1, probability
# 0.1=10% for all other
offspring_to_make = o_individual.fitness
while offspring_to_make > 0.01:
if offspring_to_make < 1:
if randint(0, 100) > offspring_to_make * 100:
break # under 1 -> no more offspring
offspring_to_make -= 1
new_individuals.append(
Individual(genome=Genome(
cod_seqA=o_individual.genome.haploid(),
cod_seqB=self.individuals[
o_individual.paired_with].genome.haploid(),
), ))
# finish line
self.individuals = new_individuals
def SetupNextGen(gl: Global, evals: List[Evaluation], scores: List[Evaluation],
args):
gl.individuals = selection.reproduce(evals, gl.nIndividuals)
scores.sort(reverse=True)
if not args.trainingmode:
Simulate(gl, scores[0].individual)
print("BestScore: " + str(scores[0].score))
sumVal = 0
for ev in scores:
sumVal += ev.score
print("Average Score: " + str(sumVal / gl.nIndividuals))
gl.bestIndi = scores[0].individual
for indi in gl.individuals:
mutate.mutate(indi, gl, 2, 0.1, 0.1, 0.2, 0.6)
def get_best_from_population(population):
result = np.empty(ITERATIONS)
for i in range(ITERATIONS):
new_population = Population()
while new_population.population_size < POPULATION_SIZE:
parent1 = tournament(population, TOURNAMENT_SIZE)
parent2 = tournament(population, TOURNAMENT_SIZE)
child = crossover(parent1, parent2, CROSSOVER_RATE)
mutate(child, MUTATION_RATE)
new_population.add_individual(child)
print('.', end='')
population = new_population
result[i] = population.get_best_individual().evaluate()
print('-', end='')
return result
def runGeneticAlgorithim(populationsize, generations):
firstGen = ab.genPool(populationsize, int(sys.argv[5]))
for gen in range(generations):
if gen == 0:
lastGen = ab.breed(firstGen, gen, float(sys.argv[4]))
for x in range(len(lastGen)):
lastGen[x] = mt.mutate(lastGen[x])
else:
for x in range(len(lastGen)):
lastGen[x] = mt.mutate(lastGen[x])
lastGen = ab.breed(lastGen, gen, float(sys.argv[4]))
for aptamer in lastGen:
if hp.hairpin(aptamer[1]) != 0:
lastGen.remove(aptamer)
return [firstGen, lastGen]
def evolve_minion(worker_file, gen, rank, output_dir):
t_start = time.time()
worker_ID, seed, pop_size, num_return, randSeed, advice, BD_table, biases, configs = load_worker_file(worker_file)
output_dir, fitness_direction, population = init_minion(configs, randSeed, seed, pop_size)
for p in range(pop_size):
mutate.mutate(configs, population[p].net, biases=biases)
pressurize.pressurize(configs, population[p], None, advice) # None: don't write instances to file
population = sort_popn(population, fitness_direction)
write_out_worker(output_dir + "/to_master/" + str(gen) + "/" + str(rank), population, num_return)
report_timing(t_start, rank, gen, output_dir)
return len(population[0].net.nodes())
def reproduce(population, fitness, mutation_rate):
"""
Generates next generation of population by probabilistically choosing mating pool based on fitness, then
probabilistically reproducing molecules in the mating pool and randomly mutating the children
:param population: list of RDKit molecules
:param fitness: probability distribution of same length as population, returned by pop_fitness
:param mutation_rate: hyperparameter determining the likelihood of mutations occuring
:return: new_population
"""
mating_pool = []
for i in range(len(population)):
mating_pool.append(np.random.choice(population, p=fitness))
new_population = []
for n in range(len(population)):
parent_A = random.choice(mating_pool)
parent_B = random.choice(mating_pool)
# print Chem.MolToSmiles(parent_A),Chem.MolToSmiles(parent_B)
new_child = co.crossover(parent_A, parent_B)
# print new_child
if new_child is not None:
new_child = mu.mutate(new_child, mutation_rate)
# print("after mutation",new_child)
if new_child is not None:
new_population.append(new_child)
return new_population
def initializePopulation(self):
for i in range(self.numberOfIndividuals):
temp = deepcopy(chromosome())
#for j in range(100):
self.globalInnovationNumber, temp = mutate.mutate(
temp, self.globalInnovationNumber)
self.addChromosome(temp)
def crossover(p1, p2, mu, crossover_rate):
c1 = copy.deepcopy(p1)
c2 = copy.deepcopy(p2)
# alpha=0.3
count = 0
for l1, l2 in zip(p1, p2):
if (np.random.uniform(0, 1) <= crossover_rate):
c1[count] = l2
c2[count] = l1
count = count + 1
c1m = mutate.mutate(c1, mu)
c2m = mutate.mutate(c2, mu)
return c1m, c2m
def genetic(f):
N = 1000
x = 2
select_p = 0.5
mutate_chance = 0.01
inputs = list(range(6))
outputs = [f(x) for x in inputs]
population = [generate() for i in range(N)]
#results = [interpret(code)(x) for code in population]
#pprint(results)
while True:
scores = [evaluate(code, inputs, outputs) for code in population]
size = len(scores)
selected_indices = list(sorted(range(size), key=lambda t : scores[t]))[:int(select_p * size)]
selected = [population[i] for i in selected_indices]
selected_scores = [scores[i] for i in selected_indices]
if 0.0 in selected_scores:
print('Selected:')
pprint(selected[0])
break
crossover(selected)
mutated = [mutate(code) if random.random() < mutate_chance else code for code in selected]
population = [generate() for i in range(N)]
def genetic(parameters):
train_losses, test_losses = [], []
training_object_count, feature_count, training_features, training_labels, test_features, test_labels\
= prepare_data.read_from_file(filename)
num_generations = parameters['num_generations']
num_chromosomes = parameters['num_chromosomes']
num_parents = int(num_chromosomes * parameters['parent_proportion'])
num_children = num_chromosomes - num_parents
num_genes = feature_count
chromosomes_shape = (num_chromosomes, num_genes + 1)
chromosomes = np.random.uniform(low=-parameters['random_gene_amplitude'],
high=parameters['random_gene_amplitude'],
size=chromosomes_shape)
for generation in range(num_generations):
parents = choose_parents(chromosomes, training_features,
training_labels, num_parents,
parameters['regularization_strength'])
children = reproduce(parents, num_children, num_parents, num_genes + 1)
children = mutate(children, parameters['mutation_amplitude'])
chromosomes = np.concatenate((parents, children), axis=0)
train_loss = nrmse(predict(training_features, parents[0]),
training_labels)
train_losses.append(train_loss)
test_losses.append(
nrmse(predict(test_features, parents[0]), test_labels))
print('Train NRMSE: {}'.format(train_loss))
return train_losses, test_losses, parameters
def geneticVRP(self):
cus_num = len(self.dis_cus_cus)
zonglines = [] #用于存储历代所有种群
lines = [[0 for i in range(cus_num)] for j in range(self.popSize)]
R = lines[0] # 一个随机个体
R2 = R
R_len = [0 for i in range(self.popSize)] # 用来存储每条路径长度
t = 1
while t <= self.NGen:
print "第 %d 次迭代：" % (t)
#生成随机初始路径
for i in range(self.popSize):
line = random.sample(range(1, cus_num + 1), cus_num)
lines[i] = line
#交叉
farm = lines
farm2 = farm
for i in range(0, self.popSize, 2):
if random.random() < self.cxPb and i < self.popSize:
route1 = farm[i]
route2 = farm[i + 1]
route1, route2 = intercross.intercross(route1, route2)
farm[i] = route1
farm[i + 1] = route2
#变异
for i in range(0, self.popSize):
if random.random() <= self.mutPb:
farm[i] = mutate.mutate(farm[i])
#群体的选择和更新
FARM = [[0 for i in range(cus_num)] for j in range(self.popSize)]
fitness_value, Cost, R = fitness.fitness(
self.dis_cus_cus, self.dis_sta_cus, self.min_dis_depot,
self.min_dis_station, self.time_win, self.tolerate_time_win,
self.demand, farm)
rank = [
index
for index, value in sorted(list(enumerate(fitness_value)),
key=lambda x: x[1])
]
for i in range(0, self.popSize):
FARM[i] = farm[rank[i]]
fitness_value_s = sorted(fitness_value)
zonglines.append(fitness_value_s[0])
R = FARM[0] #更新最短路径
t += 1
print "最佳路径为："
for i in range(0, len(FARM[0])):
print "%d," % (FARM[0][i]),
return zonglines, R
def __hill_climbing(self, n_hill_climb):
for point in self.points:
for nn in range(n_hill_climb):
new_point = mutate(point)
d = Point.pareto_dominance(new_point, point)
if d > 0:
self.remove_value(point)
self.add(new_point)
break
def _run(self):
signal.signal(signal.SIGINT, signal.SIG_IGN)
while not self._quit.is_set():
try: d = self.dna_q.get_nowait()
except(mpq.Empty): continue
self.critters = self.clean(self.critters, self.max)
try: pid = critter.hashtxt(dna.dna2txt(d))
except(KeyError,TypeError): continue
if random.random() < .5:
#if False:
try: mutate.mutate(d)
except(TypeError,KeyError): pass
try:
c = critter.Critter(d)
c.run() # run has to precede append, otherwise not threadsafe
self.critters.append(c)
if pid != c.my_id:
self.logging_q.put((self.name, pid, c.my_id))
except: pass
def runGeneticAlgorithim(aptamerData, generations):
firstGen = ab.getAptamers(aptamerData)
model = load_model(str(sys.argv[5])) #saved trained keras model
for gen in range(generations):
if gen == 0:
lastGen = ab.breed(firstGen, gen, model, float(sys.argv[4]))
print("finished breeding")
for x in range(len(lastGen)):
lastGen[x] = mt.mutate(lastGen[x], model)
print("finished mutating")
else:
for x in range(len(lastGen)):
lastGen[x] = mt.mutate(lastGen[x], model)
print("finished mutating")
lastGen = ab.breed(lastGen, gen, model, float(sys.argv[4]))
print("finished breeding")
for aptamer in lastGen:
if hp.hairpin(aptamer[1]) == True:
lastGen.remove(aptamer)
print("Finished gen " + str(gen + 1))
return [firstGen, lastGen]
def reproduce(mating_pool, population_size, mutation_rate):
new_population = []
while len(new_population) < population_size:
parent_A = random.choice(mating_pool)
parent_B = random.choice(mating_pool)
new_child = co.crossover(parent_A, parent_B)
if new_child != None:
new_child = mu.mutate(new_child, mutation_rate)
if new_child != None:
new_population.append(new_child)
return new_population
def reproduce(mating_pool, population_size, mutation_rate):
new_population = []
while len(new_population) < population_size:
parent_A = random.choice(mating_pool)
parent_B = random.choice(mating_pool)
new_child = co.crossover(parent_A, parent_B)
if new_child != None:
mutated_child = mu.mutate(new_child, mutation_rate)
if mutated_child != None:
#print(','.join([Chem.MolToSmiles(mutated_child),Chem.MolToSmiles(new_child),Chem.MolToSmiles(parent_A),Chem.MolToSmiles(parent_B)]))
new_population.append(mutated_child)
return new_population
def reproduce(mating_pool, mutation_rate):
"""
Args:
mating_pool: list of RDKit Mol
mutation_rate: rate of mutation
Returns:
"""
parent_a = random.choice(mating_pool)
parent_b = random.choice(mating_pool)
new_child = co.crossover(parent_a, parent_b)
if new_child is not None:
new_child = mu.mutate(new_child, mutation_rate)
return new_child
def pso(dm,population_size=40):
df=dm.copy(deep=True)
X=df.iloc[:,:-1]
X=np.array(X,dtype=float)
X_=preprocessing(X)
X=X_.copy()
y=df.iloc[:,len(df.columns)-1]
y=np.array(y,dtype=int)
population_size=40
population=population_initialisation(len(df.columns)-1,population_size=40,no_population_slots=4)
fitness=np.zeros(population_size,dtype=float)
historical_fitness=np.zeros(population_size,dtype=float)
historical_position=population.copy()
global_fitness=999999
global_position=np.zeros(len(df.columns)-1,dtype=int)
for i in range(0,population_size):
historical_fitness[i]=999999
k=0
while(k!=100):
print('\n',k+1,'th generation in process...')
for i in range(0,population_size):
fitness[i]=svm_fitness(X,y,population[i])
if(str(fitness[i])!=str(historical_fitness[i])):
f=[fitness[i],historical_fitness[i]]
if(f.index(min(f))==0):
historical_fitness[i]=(fitness[i]).copy()
historical_position[i]=(population[i]).copy()
if(str(historical_fitness[i])!=str(global_fitness)):
f=[historical_fitness[i],global_fitness]
if(f.index(min(f))==0):
global_fitness=(historical_fitness[i]).copy()
global_position=(population[i]).copy()
for i in range(0,population_size):
population[i]=(tpmbx(population[i],global_position,historical_position[i])).copy()
population[i]=(mutate(population[i])).copy()
k=k+1
features_best=list(*(np.where(global_position==1)))
acc=svm_fitness(X,y,global_position,1,1)
return [features_best,acc]
def crossover(p1, p2, mu, crossover_rate, passes):
c1 = copy.deepcopy(p1)
c2 = copy.deepcopy(p2)
# alpha=0.3
count = 0
for list1, list2 in zip(p1['table'], p2['table']):
if (np.random.uniform(0, 1) <= crossover_rate):
c1['table'][count][1] = list2[1]
c2['table'][count][1] = list1[1]
count = count + 1
c1m = mutate.mutate(c1, mu)
c2m = mutate.mutate(c2, mu)
fs.finalState(c1m, passes)
fit.calculate_fitness(c1m)
fs.finalState(c2m, passes)
fit.calculate_fitness(c2m)
return c1m, c2m
def reproduce(mating_pool, population_size, mutation_rate):
new_population = []
for n in range(population_size):
parent_A = random.choice(mating_pool)
parent_B = random.choice(mating_pool)
#print Chem.MolToSmiles(parent_A),Chem.MolToSmiles(parent_B)
new_child = co.crossover(parent_A, parent_B)
#print new_child
if new_child != None:
new_child = mu.mutate(new_child, mutation_rate)
#print "after mutation",new_child
if new_child != None:
new_population.append(new_child)
return new_population
def test_mutation(self):
room_ids = list(ROOM_INFO_DICT_1)
chromosome = CHROMOSOME_1
copy_chromosome = copy.deepcopy(chromosome)
self.assertEqual(chromosome, copy_chromosome)
chromosome = mutator.mutate(chromosome, room_ids, 1)
self.assertNotEqual(chromosome, copy_chromosome)
for gene in chromosome.values():
gene_len = len(gene)
timeslot_len = len(gene[1])
self.assertTrue(gene_len == 3) # 3 things in gene
self.assertTrue(timeslot_len == 2) # 2 elements in timeslot
self.assertTrue(gene[1][0] >= 0) # first timeslot 0 or more
# second timeslot less than max timeslot
self.assertTrue(gene[1][1] <= MAX_TIMESLOT)
self.assertTrue(gene[1][0] < gene[1][1]) # ending after begining
def traverse(node):
"""DFS visit nodes from tree"""
if node.is_root():
sequence = Tree369.rand_seq(L)
node.set_sequence(sequence)
else:
branch_length = node.get_parent().get_height(
) - node.get_height()
parent_sequence = node.get_parent().get_sequence()
mutated_sequence = mutate(parent_sequence, branch_length, mu)
node.set_sequence(mutated_sequence)
seq_list.append((int(node.get_label()), node.get_sequence()))
if node.is_leaf():
return
traverse(node.get_children()[0])
traverse(node.get_children()[1])
def reproduce(mating_pool, population_size, mutation_rate):
new_population = []
count = 0
while len(new_population) < population_size:
parent_A = random.choice(mating_pool)
parent_B = random.choice(mating_pool)
#print Chem.MolToSmiles(parent_A),Chem.MolToSmiles(parent_B)
new_child = co.crossover(parent_A, parent_B)
#print new_child
if new_child != None:
new_child = mu.mutate(new_child, mutation_rate)
#print "after mutation",new_child
if new_child != None:
mol = co.mol_OK(new_child)
if mol != True:
count += 1
else:
new_population.append(new_child)
else:
count += 1
else:
count += 1
return new_population, count
def nearly_equal(word, word2):
if word2 in mutate.mutate(word):
return True
def test_mutate_int(self):
for _ in range(10):
x = mutate.mutate(0)
self.assertIn(x, [-1, 1])
def runModel():
lexs = [sampleLexicon(), sampleLexicon()]
scores = [posteriorScore(lexs[0], corpus), posteriorScore(lexs[1], corpus)]
names = ['sticky', 'MH']
numAccepted = [0, 0]
interval = 50
print "alpha:", params.alpha
print "gamma:", params.gamma
print "kappa:", params.kappa
print "num_iterations:", num_iterations
print "Initial score:", scores[0]
print "Initial lexicon:"
utils.printLex(lexs[0])
bestLex = lexs[0]
bestScore = scores[0]
icons = ['\\', '|', '/', '-']
iconIdx = 0
lastPrinted = bestLex
sys.stdout.write(icons[iconIdx])
for i in range(num_iterations):
iconIdx = (iconIdx+1)%4
sys.stdout.write('\b')
sys.stdout.write(icons[iconIdx])
sys.stdout.flush()
if i%300==0:
lexs[0] = bestLex
scores[0] = bestScore
if i%interval==0:
sys.stdout.write('\b')
print "Reached iteration\t", i
print "\tsticky \tscore:", scores[0], "\tlen:", lexs[0].getLen(), "\tnum accepted:", numAccepted[0], "/", interval
print "\tMH \tscore:", scores[1], "\tlen:", lexs[1].getLen(), "\tnum accepted:", numAccepted[1], "/", interval
numAccepted = [0, 0]
if i%250==0 and cmp(lastPrinted, bestLex) != 0:
print "Current best lexicon (score = %d, len = %d):"%(bestScore, bestLex.getLen())
print "F-score: %f, precision: %f, recall: %f"%utils.fScore(bestLex)
utils.printLex(bestLex)
lastPrinted = bestLex
for j in range(len(lexs)):
lex = lexs[j]
score = scores[j]
temp = 6000
newLex = mutate(lex)
newScore = posteriorScore(newLex, corpus)
prob = np.exp(newScore - score) if newScore<score else 1 # Scores are in log space
sys.stdout.write('\b')
if bool(Bernoulli(probs=prob).sample()):
numAccepted[j] += 1
lexs[j] = newLex
scores[j] = newScore
if newScore > bestScore:
diff = newScore-bestScore
bestLex = newLex
bestScore = newScore
sys.stdout.write('\b')
print "New high score!\t", bestScore
print "\tdiscovered by:\t", names[j], "\tlen:", bestLex.getLen(), "\tscore diff\t", diff, "\titeration:",i
print "\tF-score: %f, precision: %f, recall: %f"%utils.fScore(bestLex)
if i>100:
utils.printLex(bestLex)
if j != 0:
lexs[0] = bestLex
scores[0] = bestScore
return (bestLex, bestScore)
def test_mutation_rate(self):
mutate.mutation_rate = 0.0
for _ in range(10):
x = mutate.mutate(0)
self.assertEqual(x, 0)
mutate.mutation_rate = 1.0
def mutate(self):
"""
Returns a Colicin with an id shifted by one
"""
new_id = mutate(self.id)
return Colicin(new_id)
def evolve_minion(worker_file, gen, rank, output_dir):
t_start = time.time()
with open(str(worker_file), 'rb') as file:
worker_ID, seed, worker_gens, pop_size, num_return, randSeed, curr_gen, configs = pickle.load(
file)
file.close()
survive_fraction = float(configs['worker_percent_survive']) / 100
num_survive = math.ceil(survive_fraction * pop_size)
output_dir = configs['output_directory'].replace(
"v4nu_minknap_1X_both_reverse/", '')
#output_dir += str(worker_ID)
max_gen = int(configs['max_generations'])
control = configs['control']
if (control == "None"): control = None
fitness_direction = str(configs['fitness_direction'])
node_edge_ratio = float(configs['edge_to_node_ratio'])
random.seed(randSeed)
population = gen_population_from_seed(seed, pop_size)
start_size = len(seed.net.nodes())
pressurize_time = 0
mutate_time = 0
for g in range(worker_gens):
gen_percent = float(curr_gen / max_gen)
if (g != 0):
for p in range(num_survive, pop_size):
population[p] = population[p % num_survive].copy()
#assert (population[p] != population[p%num_survive])
#assert (population[p].net != population[p % num_survive].net)
for p in range(pop_size):
t0 = ptime()
mutate.mutate(configs, population[p].net, gen_percent,
node_edge_ratio)
t1 = ptime()
mutate_time += t1 - t0
if (control == None):
pressure_results = pressurize.pressurize(
configs, population[p].net, None
) # false: don't track node fitness, None: don't write instances to file
population[p].fitness_parts[0], population[p].fitness_parts[
1], population[p].fitness_parts[2] = pressure_results[
0], pressure_results[1], pressure_results[2]
else:
util.cluster_print(
output_dir, "ERROR in minion(): unknown control config: " +
str(control))
old_popn = population
population = fitness.eval_fitness(old_popn, fitness_direction)
del old_popn
#debug(population,worker_ID, output_dir)
curr_gen += 1
write_out_worker(output_dir + "/to_master/" + str(gen) + "/" + str(rank),
population, num_return)
# some output, probably outdated
if (worker_ID == 0):
orig_dir = configs['output_directory']
end_size = len(population[0].net.nodes())
growth = end_size - start_size
output.minion_csv(orig_dir, pressurize_time, growth, end_size)
#debug(population, worker_ID, output_dir)
#if (worker_ID==0): util.cluster_print(output_dir,"Pressurizing took " + str(pressurize_time) + " secs, while mutate took " + str(mutate_time) + " secs.")
t_end = time.time()
time_elapsed = t_end - t_start
if (rank == 1 or rank == 32 or rank == 63):
util.cluster_print(
output_dir, "Worker #" + str(rank) + " finishing after " +
str(time_elapsed) + " seconds")
aa = 0 # a counter
max_score = 0 # keep track of the max score achieved by a strain
counter = 0 # another counter
# Do While: stop condition is when a circuit that fulfills the truth table is found
while (temp_string[0][0][0] < (100*len(tt_out[0])*len(tt_in)+2-100)):
for xx in range (0, keep_top_X_strains):
bit_string[xx] = copy.deepcopy(temp_string[xx]) # Copy "elite" strains to bit_string
for xx in range (keep_top_X_strains, crossover_X_strains+keep_top_X_strains):
bit_string[xx] = copy.deepcopy(temp_string[0]) # Copy "elite" strains to bit_string
for xx in range (keep_top_X_strains, crossover_X_strains+keep_top_X_strains):
mt.crossover(bit_string[xx][1], temp_string[xx][1], rows, crossover_percent_elite) # Crossover
for xx in range(1, crossover_X_strains+keep_top_X_strains):
mt.mutate(bit_string[xx][1], num_of_bits_to_mutate) # Mutation of strains (excluding the best strain)
## Simulate each strain and assign a fitness
for xx in range(0, num_of_strains):
bit_string[xx][0][0] = cs.sim(bit_string[xx][1],tt_in,tt_out,rows,cells_per_row, bits_per_cell)
# Because I use python's "sort" function, I added a random number to the start of each strain.
# This avoids sorting the strains by fitness, and then the input of the first logic gate (bias)
bit_string[xx][0][1] = rnd.randint(0,900)
# Sort strains with best (highest) fitness first
bit_string.sort(reverse=True)
temp_string = []
for xx in range (0, crossover_X_strains+keep_top_X_strains+1):
temp_string.append(bit_string[xx]) #Copy the best XX strains to temp_string
def mutate(self):
"""
Returns an immunity instance shifted by one
(Binding range unchanged)
"""
return Immunity(mutate(self.id), self.binding_range)
def nearlyEqual(s1, s2):
mutated_list = mutate.mutate(s2)
return s1 in mutated_list