def train_agent(config, agent):
torch.manual_seed(config.SEED)
env = NeodroidGymWrapper(
BinaryActionEncodingWrapper(
environment_name=config.ENVIRONMENT_NAME,
connect_to_running=config.CONNECT_TO_RUNNING,
))
env.seed(config.SEED)
agent.build(env)
listener = add_early_stopping_key_combination(agent.stop_procedure)
if listener:
listener.start()
try:
(
trained_model,
running_signals,
running_lengths,
*training_statistics,
) = agent.train(env,
env,
rollouts=config.ROLLOUTS,
render=config.RENDER_ENVIRONMENT)
except ValueError as e:
running_signals = None
running_lengths = None
trained_model = None
raise e
finally:
if listener:
listener.stop()
draugr.save_statistic(
running_signals,
stat_name="running_signals",
config_name=C.CONFIG_NAME,
project_name=C.PROJECT_NAME,
directory=C.LOG_DIRECTORY,
)
draugr.save_statistic(
running_lengths,
stat_name="running_lengths",
directory=C.LOG_DIRECTORY,
config_name=C.CONFIG_NAME,
project_name=C.PROJECT_NAME,
)
U.save_model(trained_model, **config)
env.close()
def train_agent(config, agent):
neo.seed(config.SEED)
torch.manual_seed(config.SEED)
env = neo(
environment_name=config.ENVIRONMENT_NAME,
connect_to_running=config.CONNECT_TO_RUNNING,
)
env.seed(config.SEED)
agent.build(env)
listener = add_early_stopping_key_combination(agent.stop_procedure)
if listener:
listener.start()
try:
(
trained_model,
running_signals,
running_lengths,
*training_statistics,
) = agent.train(env, config.ROLLOUTS, render=config.RENDER_ENVIRONMENT)
finally:
if listener:
listener.stop()
draugr.save_statistic(
running_signals,
stat_name="running_signals",
config_name=C.CONFIG_NAME,
project_name=C.PROJECT_NAME,
directory=C.LOG_DIRECTORY,
)
draugr.save_statistic(
running_lengths,
stat_name="running_lengths",
directory=C.LOG_DIRECTORY,
config_name=C.CONFIG_NAME,
project_name=C.PROJECT_NAME,
)
U.save_model(trained_model, **config)
env.close()