def converging(ip, population, n_population, n_gene, search_idx,
search_region):
n_children = 10
children = np.empty((n_children, n_gene + 1))
for i in range(n_children):
ip[2:] = np.random.choice(n_population, n_gene, replace=False)
children[i, :] = xover(population[ip, :], n_gene)
family = np.empty((n_children + 2, n_gene + 1))
family[:n_children, :] = children
family[-2, :] = population[ip[0], :]
family[-1, :] = population[ip[1], :]
family = family[np.argsort(family[:, -1]), :]
# Best, either of parents
population[ip[0], :] = family[0, :]
# Random
population[ip[1], :] = \
family[np.random.randint(low=1, high=n_children+2, dtype=np.int), :]
if np.isinf(population[ip[1], -1]):
population[ip[1], -1] = objective(population[ip[1], :n_gene],
search_idx, search_region)
population = population[np.argsort(population[:, -1]), :]
return ip, population
def get_new_child(parents, n_gene, search_idx, search_region):
MAXITER = 100
for _ in range(MAXITER):
child = undx(parents, n_gene)
if 0. <= np.min(child[:n_gene]) and np.max(child[:n_gene]) <= 1.:
break
else:
child[:n_gene] = np.clip(child[:n_gene], 0., 1.)
child[-1] = objective(child[:n_gene], search_idx, search_region)
return child
def get_initial_population(nth_paramset, n_population, n_gene, search_idx,
search_region):
population = np.full((n_population, n_gene + 1), np.inf)
with open('./out/%d/initpop.log' % (nth_paramset), mode='w') as f:
f.write('Generating the initial population. . .\n')
for i in range(n_population):
while np.isinf(population[i, -1]) or np.isnan(population[i, -1]):
population[i, :n_gene] = np.random.rand(n_gene)
population[i, -1] = objective(population[i, :n_gene], search_idx,
search_region)
with open('./out/%d/initpop.log' % (nth_paramset), mode='a') as f:
f.write('%d/%d\n' % (i + 1, n_population))
population = population[np.argsort(population[:, -1]), :]
return population
def get_initial_population_continue(nth_paramset, n_population, n_gene,
search_idx, search_region, p0_bounds):
best_generation = np.load('./out/%d/generation.npy' % (nth_paramset))
best_indiv = np.load('./out/%d/fit_param%d.npy' %
(nth_paramset, int(best_generation)))
population = np.full((n_population, n_gene + 1), np.inf)
with open('./out/%d/initpop.log' % (nth_paramset), mode='w') as f:
f.write('Generating the initial population. . .\n')
for i in range(n_population):
while np.isinf(population[i, -1]) or np.isnan(population[i, -1]):
population[i, :n_gene] = encode_bestindiv2randgene(
best_indiv, search_region, p0_bounds)
population[i, :n_gene] = np.clip(population[i, :n_gene], 0., 1.)
population[i, -1] = objective(population[i, :n_gene], search_idx,
search_region)
with open('./out/%d/initpop.log' % (nth_paramset), mode='a') as f:
f.write('%d/%d\n' % (i + 1, n_population))
population = population[np.argsort(population[:, -1]), :]
return population
def ga_v1_continue(nth_paramset, max_generation, n_population, n_children,
n_gene, allowable_error, search_idx, search_region,
p0_bounds):
count_num = np.load('./out/%d/count_num.npy' % (nth_paramset))
best_generation = np.load('./out/%d/generation.npy' % (nth_paramset))
best_indiv = np.load('./out/%d/fit_param%d.npy' %
(nth_paramset, int(best_generation)))
best_fitness = objective((np.log10(best_indiv) - search_region[0, :]) /
(search_region[1, :] - search_region[0, :]),
search_idx, search_region)
population = get_initial_population_continue(nth_paramset, n_population,
n_gene, search_idx,
search_region, p0_bounds)
if best_fitness < population[0, -1]:
population[0, :n_gene] = (
(np.log10(best_indiv) - search_region[0, :]) /
(search_region[1, :] - search_region[0, :]))
population[0, -1] = best_fitness
else:
best_indiv = decode_gene2variable(population[0, :n_gene],
search_region)
best_fitness = population[0, -1]
np.save(
'./out/%d/fit_param%d.npy' % (nth_paramset, int(count_num) + 1),
best_indiv)
with open('./out/%d/out.log' % (nth_paramset), mode='a') as f:
f.write('\n----------------------------------------\n\n' +
'Generation%d: Best Fitness = %e\n' %
(int(count_num) + 1, best_fitness))
if population[0, -1] <= allowable_error:
best_indiv = decode_gene2variable(population[0, :n_gene],
search_region)
best_fitness = population[0, -1]
return best_indiv, best_fitness
generation = 1
while generation < max_generation:
population = mgg_alternation(population, n_population, n_children,
n_gene, search_idx, search_region)
best_indiv = decode_gene2variable(population[0, :n_gene],
search_region)
if population[0, -1] < best_fitness:
np.save(
'./out/%d/fit_param%d.npy' %
(nth_paramset, generation + int(count_num) + 1), best_indiv)
np.save('./out/%d/generation.npy' % (nth_paramset),
generation + int(count_num) + 1)
np.save('./out/%d/best_fitness' % (nth_paramset), best_fitness)
best_fitness = population[0, -1]
np.save('./out/%d/count_num.npy' % (nth_paramset),
generation + int(count_num) + 1)
with open('./out/%d/out.log' % (nth_paramset), mode='a') as f:
f.write('Generation%d: Best Fitness = %e\n' %
(generation + int(count_num) + 1, best_fitness))
if population[0, -1] <= allowable_error:
best_indiv = decode_gene2variable(population[0, :n_gene],
search_region)
best_fitness = population[0, -1]
return best_indiv, best_fitness
generation += 1
best_indiv = decode_gene2variable(population[0, :n_gene], search_region)
best_fitness = population[0, -1]
return best_indiv, best_fitness
def ga_v2_continue(nth_paramset, max_generation, n_population, n_children,
n_gene, allowable_error, search_idx, search_region,
p0_bounds):
if n_population < n_gene + 2:
raise ValueError('n_population must be larger than %d' % (n_gene + 2))
n_iter = 1
n0 = np.empty(2 * n_population)
count_num = np.load('./out/%d/count_num.npy' % (nth_paramset))
best_generation = np.load('./out/%d/generation.npy' % (nth_paramset))
best_indiv = np.load('./out/%d/fit_param%d.npy' %
(nth_paramset, int(best_generation)))
best_fitness = objective((np.log10(best_indiv) - search_region[0, :]) /
(search_region[1, :] - search_region[0, :]),
search_idx, search_region)
population = get_initial_population_continue(nth_paramset, n_population,
n_gene, search_idx,
search_region, p0_bounds)
if best_fitness < population[0, -1]:
population[0, :n_gene] = (
(np.log10(best_indiv) - search_region[0, :]) /
(search_region[1, :] - search_region[0, :]))
population[0, -1] = best_fitness
else:
best_indiv = decode_gene2variable(population[0, :n_gene],
search_region)
best_fitness = population[0, -1]
np.save(
'./out/%d/fit_param%d.npy' % (nth_paramset, int(count_num) + 1),
best_indiv)
with open('./out/%d/out.log' % (nth_paramset), mode='a') as f:
f.write('\n----------------------------------------\n\n' +
'Generation%d: Best Fitness = %e\n' %
(int(count_num) + 1, best_fitness))
n0[0] = population[0, -1]
if population[0, -1] <= allowable_error:
best_indiv = decode_gene2variable(population[0, :n_gene],
search_region)
best_fitness = population[0, -1]
return best_indiv, best_fitness
generation = 1
while generation < max_generation:
ip = np.random.choice(n_population, n_gene + 2, replace=False)
ip, population = converging(ip, population, n_population, n_gene,
search_idx, search_region)
ip, population = local_search(ip, population, n_population, n_children,
n_gene, search_idx, search_region)
for _ in range(n_iter - 1):
ip = np.random.choice(n_population, n_gene + 2, replace=False)
ip, population = converging(ip, population, n_population, n_gene,
search_idx, search_region)
if generation % len(n0) == 0:
n0 = np.empty_like(n0)
n0[0] = population[0, -1]
elif generation % len(n0) == len(n0) - 1:
n0[-1] = population[0, -1]
if n0[0] == n0[-1]:
n_iter *= 2
else:
n_iter = 1
else:
n0[generation % len(n0)] = population[0, -1]
best_indiv = decode_gene2variable(population[0, :n_gene],
search_region)
if population[0, -1] < best_fitness:
np.save('./out/%d/generation.npy' % (nth_paramset),
generation + int(count_num) + 1)
np.save(
'./out/%d/fit_param%d.npy' %
(nth_paramset, generation + int(count_num) + 1), best_indiv)
np.save('./out/%d/best_fitness' % (nth_paramset), best_fitness)
best_fitness = population[0, -1]
np.save('./out/%d/count_num.npy' % (nth_paramset),
generation + int(count_num) + 1)
with open('./out/%d/out.log' % (nth_paramset), mode='a') as f:
f.write('Generation%d: Best Fitness = %e\n' %
(generation + int(count_num) + 1, best_fitness))
if population[0, -1] <= allowable_error:
best_indiv = decode_gene2variable(population[0, :n_gene],
search_region)
best_fitness = population[0, -1]
return best_indiv, best_fitness
generation += 1
best_indiv = decode_gene2variable(population[0, :n_gene], search_region)
best_fitness = population[0, -1]
return best_indiv, best_fitness