return observation
def test(self, dataset, batch_size=1000):
loss, acc = self.sess.run([self.loss, self.accuracy],
feed_dict=self.load_feed_dict(
data=dataset, batch_size=batch_size))
print('\naccuracy on test set: ', acc)
print('\nloss on test set: ', loss)
return loss, acc
if __name__ == "__main__":
from tool_func import *
log_dir = './logs'
data_dir = './datas/50div10gmm10-1.csv'
dataset_0 = load_datasets(data_dir)
env = Learner(
log_dir=log_dir,
i_data=dataset_0['train'],
j_data=dataset_0['train'],
feature_size=50,
)
k = env.reset()
# print(env.reset())
action = 2
b = env.step(action=action, trans_penalty=0.1)
k1 = env.reset()
k2 = env.step(action=action, trans_penalty=0.1)
print('00000000000000')
with open(file_name, 'a') as file:
file.write(
'\n=============================================\n==========================================\n'
)
break
if __name__ == "__main__":
# import os
# cuda = raw_input("CUDA DEVICES: ")
# os.environ["CUDA_VISIBLE_DEVICES"] = cuda
# log_dir = './ckpt'
data = [load_datasets('./datas/50div10gmm10-1.csv', tst_frac=0), load_datasets('./datas/50div10gmm10-2.csv', tst_frac=0),\
load_datasets('./datas/50div10gmm10-all.csv', tst_frac=0.8) ]
with tf.variable_scope('DQN_with_prioritized_replay'):
RL_prio = DQNPrioritizedReplay(
n_actions=6, #trans weights
n_features=6,
learning_rate=0.01, #RL learning rate
reward_decay=0.98, #reward decay --the infuluence of reward
e_greedy=0.95, #max e-greedy
e_greedy_increment=0.005,
replace_target_iter=20,
memory_size=800,
batch_size=20,
output_graph=False,
prioritized=False,
def test(self, batch_size=100):
tst_loss, total_acc, scalar_acc = self.sess.run(
[self.loss, self.accuracy, self.test_accuracy],
feed_dict=self.load_feed_dict(data=self.reward_tst_data,
batch_size=batch_size))
self.writer.add_summary(scalar_acc, self.learn_step_counter)
print('\naccuracy on test set: ', total_acc)
print('\nloss on test set: ', tst_loss)
return tst_loss, total_acc
def destroy(self):
self.writer.close()
self.sess.close()
if __name__ == "__main__":
log_dir = './5.1_Double_DQN/logs'
data_dir = './5.1_Double_DQN/datas/origin50div50gmm10.csv'
train_data, test_data, dataset_0, dataset_1, dataset_2 = load_datasets(
data_dir)
env = Learner(log_dir=log_dir, i_data=dataset_0, j_data=dataset_1)
env.reset()
for i in range(5):
env.step(action=0)
env.step(action=4)
print('00000000000000')
env.reset()
env.step(action=0)
env.step(action=0)
file_name = './logs/records-f.txt'
# acc = 0.9888
# with open(file_name , 'a') as file:
# file.write(str(acc))
log_dir = './logs/gmm/'
# data_dir_1 = './datas/10in15_50div50_100-0.csv'
# data_dir_2 = './datas/10in15_50div50_100-1.csv'
# data_dir_3 = './datas/10in15_50div50_100-2.csv'
# data_dir_4 = './datas/10in15_50div50_100-3.csv'
# data_all ='./datas/d-all.csv'
# data = [load_datasets(data_dir_1, tst_frac=0.1), load_datasets(data_dir_2, tst_frac=0.1), load_datasets(data_dir_3, tst_frac=0.1), \
# load_datasets(data_dir_4, tst_frac=0.1), load_datasets(data_all, tst_frac=0.1)]
# print(data[1])
data = [
load_datasets('./datas/f-1.csv', tst_frac=0.1),
load_datasets('./datas/f-2.csv', tst_frac=0.1)
]
with tf.variable_scope('DQN_with_prioritized_replay'):
RL_prio = DQNPrioritizedReplay(
n_actions=ACTION_SPACE,
n_features=5,
learning_rate=0.005,
reward_decay=0.9,
e_greedy=0.99,
replace_target_iter=20,
memory_size=10000,
batch_size=20,
e_greedy_increment=0.005,
output_graph=False,
if __name__ == "__main__":
# import os
# cuda = raw_input("CUDA DEVICES: ")
# os.environ["CUDA_VISIBLE_DEVICES"] = cuda
file_name = './logs/records-e-0.txt'
# acc = 0.9888
# with open(file_name , 'a') as file:
# file.write(str(acc))
log_dir = './logs/gmm/'
data_dir_1 = './datas/f-1.csv'
# data_dir_2 = '../datas/e-1.csv'
# data_dir_3 = '../datas/e-2.csv'
# data_dir_4 = '../datas/e-3.csv'
# data_all ='../datas/e-all.csv'
data = [load_datasets(data_dir_1, tst_frac=0.2), load_datasets(data_dir_1, tst_frac=0.2)]
# print(data[1])
env = Learner(
log_dir=log_dir,
i_data=data[0]['train'],
j_data=data[0]['train'],
tst_data = data[0]['tst'],
learning_steps = 500,
learning_steps_max = 50000,
hidden_layer_size=10,
feature_size=3,
)
env.reset()
