def train(self):
self.initParams()
ga = GA(self)
population = []
for i in range(self.MAX_ITER):
population = ga.selection(ga.population)
parents = ga.select_parents(population)
children = ga.crossover(parents)
newPopulation = ga.mutation(children)
population = newPopulation
best_chromosome, best_fit = ga.get_best_chromosome(population)
W1, W2 = ga.chromosome_to_matrix(best_chromosome)
self.setAllParams(W1, W2)
print(best_fit)
print("{}".format(204 * "-"))
print("|{:^19s} | {:^54} | {:^123}|".format(
"OPTIMAL", "MANUFACTURER", "RETAILER"))
print("{}".format(204 * "-"))
print('[{:^3s}] | {:^12s} | {:^8s} | {:^6s} | {:^8s} | {:^10s}'.format(
'STT', '#NP', 'A', 'C', 'DEMAND', 'NP_M'), end='')
for i in range(len(config.M)):
print('| {:^5s}'.format('n' + str(i+1)), end='')
for i in range(len(config.K)):
idx = str(i+1)
print('| {:^8s} | {:^12s}'.format('a' + idx, "NP_b" + idx), end='')
print("|\n{}".format(204 * "-"))
for i in range(generation_count):
ga.evaluation()
ga.selection()
ga.crossover()
ga.mutation()
if i % 1 == 0:
print('[{:>3d}] | '.format(i), end='')
ga.show_optimal()
manufacture = ga.get_best_mf()
e_A = manufacture.e_A
A = manufacture.A
sum_e_bi = 0
for i in range(len(config.e_a)):
sum_e_bi += manufacture.retailers[i].e_a
sys.exit()
for i in range(len(config.e_a)):
_cnt_generations = 100
fitness_avg = list()
fitness_max = list()
diversity = list()
valid_candidate = list()
gen = GA(population_cnt=_population_cnt,
rate_crossover=_rate_crossover,
rate_mutation=_rate_mutation,
sample_candidate=_sample_candidate)
print("Cross: " + str(_rate_crossover) + "- Mutation: " + str(_rate_mutation) +
"- tourSize: " + str(_tournament_size) + "- tourRate: " +
str(_tournament_win_rate))
for generation in range(_cnt_generations):
gen.mutate()
gen.crossover(strategy=_strategy)
gen.evaluate(calc_diversity=_calc_diversity)
gen.selection(tournament_win_rate=_tournament_win_rate,
tournament_size=_tournament_size)
if gen.generation % 10 == 0:
print("Gen: " + str(gen.generation) + "- Fitness_avg: " +
str(gen.fitness_avg) + "- Fitness_best: " +
str(gen.best_candidate.get_fitness()) + "- Diversity: " +
str(gen.diversity))
fitness_avg.append(gen.fitness_avg)
fitness_max.append(gen.best_candidate.get_fitness())
diversity.append(gen.diversity)
if (gen.best_candidate.get_fitness() > 1.0):
valid_candidate.append(deepcopy(gen.best_candidate))
for best in valid_candidate:
print("Candidate Fitness " + str(best.get_fitness()))
hidden_shape=int(np.sqrt(traindata)),
kernel='gaussian')
model[2] = RBFN(input_shape=dimension,
hidden_shape=int(np.sqrt(traindata)),
kernel='gaussian')
resetmodel(x, y)
max_iter = 100
ga = GA(pop_size=100,
dimension=dimension,
lower_bound=lower_bound,
upper_bound=upper_bound)
ga.init_Population()
for i in range(max_iter):
updatemodel(ga.pop)
ga.crossover(ga.pc)
ga.mutation(ga.pm)
ga.pop = np.unique(ga.pop, axis=0)
for j in range(0, 3):
temp = model[j].predict(ga.pop)
if j == 0:
fit_value = temp
else:
fit_value = fit_value + temp
fit_value = fit_value.reshape((len(ga.pop), 1))
ga.selection(fit_value)
resetmodel(x, y)
optimum = ga.first[-1]
endtime = time.perf_counter()
print('Optimal solution :', optimum)
#Calculate fitness for each population_vector in population
fitness_train, fitness_test = fitness(population_vectors)
train_error.append(np.min(fitness_train))
test_error.append(np.min(fitness_test))
print('Train Error:', train_error[-1])
print('Test Error:', test_error[-1])
#Select best parents
parents = ga.select_mating_pool(population=population_vectors,
fitness=fitness_train.copy(),
mode='min',
num_parents=num_parents_mating)
print(parents.shape[0], 'New parents generated...')
#Crossover parents
offsprings = ga.crossover(parents=parents,
num_offsprings=sol_per_pop - num_parents_mating)
print(offsprings.shape[0], 'New offsprings produced...')
#Mutate offsprings
mutated_offsprings = ga.mutate(offsprings,
mutation_percent=mutation_percent)
print(mutated_offsprings.shape[0], 'offsprings mutated')
#Create new population
population_vectors[:parents.shape[0], :] = parents
population_vectors[parents.shape[0]:, :] = mutated_offsprings
print('New population created')
#Update the result of this generation in report files
f = open('train_error.pkl', 'wb')
pickle.dump(train_error, f)
