def train(self, training_games_to_play = 500):
training_progress_log = rysy.Log(self.network_path + "progress_training.log")
if self.agent_type == "curiosity":
icm_training_progress_log = rysy.Log(self.network_path + "icm_progress_training.log")
self.training_games_to_play = training_games_to_play
#process training
while self.env.get_games_count() < self.training_games_to_play:
result = self.agent.main()
if result != 0:
print("ERROR : agent returned ", result, "\n\n\n\n")
return result
#print training progress %, and score, every 256th iterations
if self.env.get_iterations()%256 == 0:
str_progress = str(self.env.get_iterations()) + " "
str_progress+= str(self.env.get_games_count()) + " "
str_progress+= str(self.agent.get_epsilon_start()) + " "
str_progress+= str(self.env.get_score()) + " "
str_progress+= str(self.env.get_active_env_id()) + " "
score = self.env.get_envs_score()
games = self.env.get_envs_games_count()
for i in range(0, len(score)):
str_progress+= str(games[i]) + " "
str_progress+= str(score[i]) + " "
str_progress+= "\n"
training_progress_log.put_string(str_progress)
if self.agent_type == "curiosity":
str_icm_progress = str(self.env.get_iterations()) + " "
str_icm_progress+= str(self.env.get_games_count()) + " "
str_icm_progress+= str(self.agent.get_icm_result().inverse_loss) + " "
str_icm_progress+= str(self.agent.get_icm_result().forward_loss) + " "
str_icm_progress+= str(self.agent.get_icm_result().inverse_classification_success) + " "
str_icm_progress+= "\n"
icm_training_progress_log.put_string(str_icm_progress)
print("done = ", self.env.get_games_count()*100.0/self.training_games_to_play, "%", " eps = ", self.agent.get_epsilon_start(), " iterations = ", self.env.get_iterations(), " score = ", self.env.get_score(), " active_env = ", self.env.get_active_env_id())
if self.env.get_iterations()%50000 == 0:
print("SAVING network")
self.agent.save(self.network_path)
self.agent.save(self.network_path)
return 0
def train(self, training_games_to_play=500):
training_progress_log = rysy.Log(self.agent_config_path +
"progress_training.log")
self.training_games_to_play = training_games_to_play
#process training
while self.env.get_games_count() < self.training_games_to_play:
self.agent.main()
#print training progress %, ane score, every 256th iterations
if self.env.get_iterations() % 256 == 0:
str_progress = str(self.env.get_iterations()) + " "
str_progress += str(self.env.get_games_count()) + " "
str_progress += str(self.env.get_score()) + " "
str_progress += "\n"
training_progress_log.put_string(str_progress)
print(
"done = ",
self.env.get_games_count() * 100.0 /
self.training_games_to_play, "%", " iterations = ",
self.env.get_iterations(), " score = ",
self.env.get_score())
if self.env.get_iterations() % 50000 == 0:
print("SAVING network")
self.agent.save(self.agent_config_path)
self.agent.save(self.agent_config_path)
def test(self, testing_games_to_play=100):
testing_progress_log = rysy.Log(self.network_path +
"progress_testing.log")
self.agent.load(self.network_path)
#reset score
self.env.reset_score()
#choose only the best action
self.agent.run_best_enable()
#process testing games
while self.env.get_games_count(
) < testing_games_to_play + self.training_games_to_play:
self.agent.main()
if self.env.get_iterations() % 256 == 0:
str_progress = str(self.env.get_iterations()) + " "
str_progress += str(self.env.get_games_count() -
+self.training_games_to_play) + " "
str_progress += str(self.agent.get_epsilon_training()) + " "
str_progress += str(self.env.get_score()) + " "
str_progress += "\n"
testing_progress_log.put_string(str_progress)
print("TESTING SCORE =", self.env.get_score())
return self.env.get_score()
def train(self, training_games_to_play=500):
training_progress_log = rysy.Log(self.network_path +
"progress_training.log")
icm_training_progress_log = rysy.Log(self.network_path +
"icm_progress_training.log")
self.training_games_to_play = training_games_to_play
#process training
while self.env.get_games_count() < self.training_games_to_play:
self.agent.main()
#print training progress %, ane score, every 256th iterations
if self.env.get_iterations() % 256 == 0:
str_progress = str(self.env.get_iterations()) + " "
str_progress += str(self.env.get_games_count()) + " "
str_progress += str(self.agent.get_epsilon_training()) + " "
str_progress += str(self.env.get_score()) + " "
str_progress += "\n"
training_progress_log.put_string(str_progress)
print(
"done = ",
self.env.get_games_count() * 100.0 /
self.training_games_to_play, "%", " eps = ",
self.agent.get_epsilon_training(), " iterations = ",
self.env.get_iterations(), " score = ",
self.env.get_score())
if self.env.get_iterations() % 256 == 0:
str_icm_progress = str(self.env.get_iterations()) + " "
str_icm_progress += str(self.env.get_games_count()) + " "
str_icm_progress += str(
self.agent.get_icm_result().inverse_loss) + " "
str_icm_progress += str(
self.agent.get_icm_result().forward_loss) + " "
str_icm_progress += str(self.agent.get_icm_result().
inverse_classification_success) + " "
str_icm_progress += "\n"
icm_training_progress_log.put_string(str_icm_progress)
if self.env.get_iterations() % 50000 == 0:
print("SAVING network")
self.agent.save(self.network_path)
self.agent.save(self.network_path)
def __init__(self):
env.Env.__init__(self)
self.width = 4
self.height = 4
self.depth = 1
self.time = 1
self.actions_count = 4
self.observation_init()
self.reset()
self.gui = gl_gui.GLVisualisation()
self.max_value = 0.0
self.max_score_log = rysy.Log("2048_max_score.log")
def test(self, testing_games_to_play = 100):
testing_progress_log = rysy.Log(self.network_path + "progress_testing.log")
self.agent.load(self.network_path + "trained/")
#reset score
self.env.reset_score()
#choose only the best action
self.agent.run_best_enable()
#process testing games
while self.env.get_games_count() < testing_games_to_play + self.training_games_to_play:
result = self.agent.main()
if result != 0:
print("ERROR : agent returned ", result, "\n\n\n\n")
return result
if self.env.get_iterations()%256 == 0:
str_progress = str(self.env.get_iterations()) + " "
str_progress+= str(self.env.get_games_count() - self.training_games_to_play) + " "
str_progress+= str(self.agent.get_epsilon_start()) + " "
str_progress+= str(self.env.get_score()) + " "
str_progress+= str(self.env.get_active_env_id()) + " "
score = self.env.get_envs_score()
games = self.env.get_envs_games_count()
for i in range(0, len(score)):
str_progress+= str(games[i]) + " "
str_progress+= str(score[i]) + " "
str_progress+= "\n"
testing_progress_log.put_string(str_progress)
print("TESTING SCORE =", env.get_score())
return env.get_score()
def test(self, log_filename_prefix, testing_games_to_play=100):
self.agent.load(self.agent_config_path)
#choose only the best action
self.agent.run_best_enable()
score = []
game_id = 0
#process testing games
while self.env.get_games_count(
) < testing_games_to_play + self.training_games_to_play:
self.agent.main()
if self.env.get_games_count() != game_id:
game_id = self.env.get_games_count()
score.append(self.env.get_score())
self.env.reset_score()
print(score)
mean_score = numpy.mean(score)
std = numpy.std(score)
result = "games count : " + str(len(score)) + "\n"
result += "mean score : " + str(mean_score) + "\n"
result += "std score : " + str(std) + "\n"
result += "games : " + "\n"
for i in range(0, len(score)):
result += str(score[i]) + "\n"
testing_progress_log = rysy.Log(self.agent_config_path +
log_filename_prefix +
"result_testing.log")
testing_progress_log.put_string(result)
epsilon_decay = 0.99999
#init DQN agent
agent = libs.libs_agent.agent_dqn.DQNAgent(env, network_path + "network_config.json", gamma, replay_buffer_size, epsilon_training, epsilon_testing, epsilon_decay)
'''
agent.load(network_path + "trained/")
agent.run_best_enable()
while True:
agent.main()
env._print()
'''
training_progress_log = rysy.Log(network_path + "progress_training.log")
testing_progress_log = rysy.Log(network_path + "progress_testing.log")
#process training
total_games_to_play = 20000
while env.get_games_count() < total_games_to_play:
agent.main()
#print training progress %, ane score, every 256th iterations
if verbose:
if env.get_iterations()%256 == 0:
env._print()
env.render()
if env.get_iterations()%256 == 0: