def train_real(self, data_loc=None):
from utils.text_process import code_to_text
from utils.text_process import get_tokenlized
wi_dict, iw_dict = self.init_real_trainng(data_loc)
self.init_real_metric()
def get_real_test_file(dict=iw_dict):
with open(self.generator_file, 'r') as file:
codes = get_tokenlized(self.generator_file)
with open(self.test_file, 'w') as outfile:
outfile.write(code_to_text(codes=codes, dictionary=dict))
self.sess.run(tf.compat.v1.global_variables_initializer())
self.log = open(self.log_file, 'w')
generate_samples(self.sess, self.generator, self.batch_size, self.generate_num, self.generator_file)
self.gen_data_loader.create_batches(self.oracle_file)
print('Pre-training Generator...')
for epoch in range(self.pre_epoch_num):
start = time()
loss = pre_train_epoch(self.sess, self.generator, self.gen_data_loader)
end = time()
self.add_epoch()
if epoch % 5 == 0:
generate_samples(self.sess, self.generator, self.batch_size, self.generate_num, self.generator_file)
get_real_test_file()
self.evaluate()
print('Pre-training Discriminator...')
self.reset_epoch()
for epoch in range(self.pre_epoch_num):
self.train_discriminator()
self.reset_epoch()
print('Adversarial Training...')
self.reward = Reward()
for epoch in range(self.adversarial_epoch_num):
start = time()
for index in range(1):
samples = self.generator.generate(self.sess)
rewards = self.reward.get_reward(self.sess, samples, 16, self.discriminator)
feed = {
self.generator.x: samples,
self.generator.rewards: rewards
}
_ = self.sess.run(self.generator.g_updates, feed_dict=feed)
end = time()
self.add_epoch()
if epoch % 5 == 0 or epoch == self.adversarial_epoch_num - 1:
generate_samples(self.sess, self.generator, self.batch_size, self.generate_num, self.generator_file)
get_real_test_file()
self.evaluate()
for _ in range(15):
self.train_discriminator()
def train_oracle(self):
self.init_oracle_trainng()
self.init_metric()
self.sess.run(tf.global_variables_initializer())
self.pre_epoch_num = 80
self.adversarial_epoch_num = 100
self.log = open('experiment-log-maligan-basic.csv', 'w')
generate_samples(self.sess, self.oracle, self.batch_size,
self.generate_num, self.oracle_file)
generate_samples(self.sess, self.generator, self.batch_size,
self.generate_num, self.generator_file)
self.gen_data_loader.create_batches(self.oracle_file)
self.oracle_data_loader.create_batches(self.generator_file)
# rollout = Reward(generator, update_rate)
print('start pre-train generator:')
for epoch in range(self.pre_epoch_num):
start = time()
loss = pre_train_epoch(self.sess, self.generator,
self.gen_data_loader)
end = time()
print('epoch:' + str(epoch) + '\t time:' + str(end - start))
self.add_epoch()
if epoch % 5 == 0:
self.evaluate()
print('start pre-train discriminator:')
self.reset_epoch()
for epoch in range(self.pre_epoch_num):
print('epoch:' + str(epoch))
self.train_discriminator()
self.reset_epoch()
print('adversarial training:')
self.reward = Reward()
for epoch in range(self.adversarial_epoch_num):
# print('epoch:' + str(epoch))
start = time()
for index in range(50):
samples = self.generator.generate(self.sess)
rewards = self.reward.get_reward(self.sess, samples, 16,
self.discriminator)
feed = {
self.generator.x: samples,
self.generator.rewards: rewards
}
_ = self.sess.run(self.generator.g_updates, feed_dict=feed)
end = time()
self.add_epoch()
print('epoch:' + str(epoch) + '\t time:' + str(end - start))
if epoch % 5 == 0 or epoch == self.adversarial_epoch_num - 1:
self.evaluate()
for _ in range(15):
self.train_discriminator()
def train_oracle(self):
self.init_oracle_trainng()
self.init_metric()
self.sess.run(tf.compat.v1.global_variables_initializer())
self.log = open(self.log_file, 'w')
generate_samples(self.sess, self.oracle, self.batch_size, self.generate_num, self.oracle_file)
generate_samples(self.sess, self.generator, self.batch_size, self.generate_num, self.generator_file)
self.gen_data_loader.create_batches(self.oracle_file)
self.oracle_data_loader.create_batches(self.generator_file)
print('Pre-training Generator...')
for epoch in range(self.pre_epoch_num):
start = time()
loss = pre_train_epoch(self.sess, self.generator, self.gen_data_loader)
end = time()
self.add_epoch()
if epoch % 5 == 0:
self.evaluate()
print('Pre-training Discriminator...')
self.reset_epoch()
for epoch in range(self.pre_epoch_num):
self.train_discriminator()
self.reset_epoch()
print('Adversarial Training...')
self.reward = Reward()
for epoch in range(self.adversarial_epoch_num):
start = time()
for index in range(50):
samples = self.generator.generate(self.sess)
rewards = self.reward.get_reward(self.sess, samples, 16, self.discriminator)
feed = {
self.generator.x: samples,
self.generator.rewards: rewards
}
_ = self.sess.run(self.generator.g_updates, feed_dict=feed)
end = time()
self.add_epoch()
if epoch % 5 == 0 or epoch == self.adversarial_epoch_num - 1:
self.evaluate()
for _ in range(15):
self.train_discriminator()
self.log.close()
def train_cfg(self):
import json
from utils.text_process import get_tokenlized
from utils.text_process import code_to_text
cfg_grammar = """
S -> S PLUS x | S SUB x | S PROD x | S DIV x | x | '(' S ')'
PLUS -> '+'
SUB -> '-'
PROD -> '*'
DIV -> '/'
x -> 'x' | 'y'
"""
wi_dict_loc, iw_dict_loc = self.init_cfg_training(cfg_grammar)
with open(iw_dict_loc, 'r') as file:
iw_dict = json.load(file)
def get_cfg_test_file(dict=iw_dict):
with open(self.generator_file, 'r') as file:
codes = get_tokenlized(self.generator_file)
with open(self.test_file, 'w') as outfile:
outfile.write(code_to_text(codes=codes, dictionary=dict))
self.init_cfg_metric(grammar=cfg_grammar)
self.sess.run(tf.global_variables_initializer())
self.pre_epoch_num = 80
self.adversarial_epoch_num = 100
self.log = open('experiment-log-maliganbasic-cfg.csv', 'w')
# generate_samples(self.sess, self.oracle, self.batch_size, self.generate_num, self.oracle_file)
generate_samples(self.sess, self.generator, self.batch_size, self.generate_num, self.generator_file)
self.gen_data_loader.create_batches(self.oracle_file)
self.oracle_data_loader.create_batches(self.generator_file)
print('start pre-train generator:')
for epoch in range(self.pre_epoch_num):
start = time()
loss = pre_train_epoch(self.sess, self.generator, self.gen_data_loader)
end = time()
print('epoch:' + str(self.epoch) + '\t time:' + str(end - start))
self.add_epoch()
if epoch % 5 == 0:
generate_samples(self.sess, self.generator, self.batch_size, self.generate_num, self.generator_file)
get_cfg_test_file()
self.evaluate()
print('start pre-train discriminator:')
self.reset_epoch()
for epoch in range(self.pre_epoch_num * 3):
print('epoch:' + str(epoch))
self.train_discriminator()
self.reset_epoch()
print('adversarial training:')
self.reward = Reward()
for epoch in range(self.adversarial_epoch_num):
# print('epoch:' + str(epoch))
start = time()
for index in range(1):
samples = self.generator.generate(self.sess)
rewards = self.reward.get_reward(self.sess, samples, 16, self.discriminator)
feed = {
self.generator.x: samples,
self.generator.rewards: rewards
}
_ = self.sess.run(self.generator.g_updates, feed_dict=feed)
end = time()
self.add_epoch()
print('epoch:' + str(self.epoch) + '\t time:' + str(end - start))
if epoch % 5 == 0 or epoch == self.adversarial_epoch_num - 1:
generate_samples(self.sess, self.generator, self.batch_size, self.generate_num, self.generator_file)
get_cfg_test_file()
self.evaluate()
for _ in range(15):
self.train_discriminator()
return
def train_real(self, data_loc=None):
from utils.text_process import code_to_text
from utils.text_process import get_tokenlized
wi_dict, iw_dict = self.init_real_trainng(data_loc)
self.init_real_metric()
def get_real_test_file(dict=iw_dict):
with open(self.generator_file, 'r') as file:
codes = get_tokenlized(self.generator_file)
with open(self.test_file, 'w') as outfile:
outfile.write(code_to_text(codes=codes, dictionary=dict))
self.sess.run(tf.global_variables_initializer())
self.pre_epoch_num = 40
self.adversarial_epoch_num = 50
self.log = open(
'experiment-log-maligan-real_' + self.start_time + str('.csv'),
'w', 1)
#self.log = open('experiment-log-maligan-real.csv', 'w')
generate_samples(self.sess, self.generator, self.batch_size,
self.generate_num, self.generator_file)
self.gen_data_loader.create_batches(self.oracle_file)
print('start pre-train generator:')
for epoch in range(self.pre_epoch_num):
start = time()
loss = pre_train_epoch(self.sess, self.generator,
self.gen_data_loader)
end = time()
print('epoch:' + str(self.epoch) + '\t time:' + str(end - start))
self.add_epoch()
if epoch % 5 == 0:
generate_samples(self.sess, self.generator, self.batch_size,
self.generate_num, self.generator_file)
get_real_test_file()
self.evaluate()
print('start pre-train discriminator:')
self.reset_epoch()
for epoch in range(self.pre_epoch_num):
print('epoch:' + str(epoch))
self.train_discriminator()
self.reset_epoch()
print('adversarial training:')
self.reward = Reward()
for epoch in range(self.adversarial_epoch_num):
# print('epoch:' + str(epoch))
start = time()
for index in range(1):
samples = self.generator.generate(self.sess)
rewards = self.reward.get_reward(self.sess, samples, 16,
self.discriminator)
feed = {
self.generator.x: samples,
self.generator.rewards: rewards
}
_ = self.sess.run(self.generator.g_updates, feed_dict=feed)
end = time()
self.add_epoch()
print('epoch:' + str(self.epoch) + '\t time:' + str(end - start))
if epoch % 5 == 0 or epoch == self.adversarial_epoch_num - 1:
generate_samples(self.sess, self.generator, self.batch_size,
self.generate_num, self.generator_file)
get_real_test_file()
self.evaluate()
for _ in range(15):
self.train_discriminator()
#Generate final samples
generate_samples(self.sess, self.generator, self.batch_size,
self.generate_final_num, self.generator_final_file)
with open(self.generator_final_file, 'r') as file:
codes = get_tokenlized(self.generator_final_file)
with open(self.test_final_file, 'w') as outfile:
outfile.write(code_to_text(codes=codes, dictionary=iw_dict))