def __init__(self,
args,
observation_size,
action_size,
network_type,
task_queue,
result_queue,
worker_id,
name_scope='planning_worker'):
# the multiprocessing initialization
multiprocessing.Process.__init__(self)
self.args = args
self._name_scope = name_scope
self._worker_id = worker_id
self._network_type = network_type
self._npr = np.random.RandomState(args.seed + self._worker_id)
self._observation_size = observation_size
self._action_size = action_size
self._task_queue = task_queue
self._result_queue = result_queue
logger.info('Worker {} online'.format(self._worker_id))
self._base_dir = init_path.get_base_dir()
def __init__(self,
args,
observation_size,
action_size,
network_type,
task_queue,
result_queue,
worker_id,
name_scope='planning_worker'):
# the base agent
super(worker,
self).__init__(args, observation_size, action_size, network_type,
task_queue, result_queue, worker_id, name_scope)
self._base_dir = init_path.get_base_dir()
def __init__(self,
args,
observation_size,
action_size,
network_type,
task_queue,
result_queue,
worker_id,
name_scope='mbmf_worker'):
# the base agent
super(worker,
self).__init__(args, observation_size, action_size, network_type,
task_queue, result_queue, worker_id, name_scope)
self._base_dir = init_path.get_base_dir()
# build the environments
self._build_env()
def __init__(self,
args,
observation_size,
action_size,
network_type,
task_queue,
result_queue,
worker_id,
name_scope='planning_worker'):
# the base agent
super(worker,
self).__init__(args, observation_size, action_size, network_type,
task_queue, result_queue, worker_id, name_scope)
self._base_dir = init_path.get_base_dir()
self._alpha = args.cem_learning_rate
self._num_iters = args.cem_num_iters
self._elites_fraction = args.cem_elites_fraction
def __init__(self, sess, summary_name, enable=True, summary_dir=None):
# the interface we need
self.summary = None
self.sess = sess
self.enable = enable
if not self.enable: # the summary handler is disabled
return
if summary_dir is None:
self.path = os.path.join(
init_path.get_base_dir(), 'summary'
)
else:
self.path = os.path.join(summary_dir, 'summary')
self.path = os.path.abspath(self.path)
if not os.path.exists(self.path):
os.makedirs(self.path)
self.path = os.path.join(self.path, summary_name)
self.train_writer = tf.summary.FileWriter(self.path, self.sess.graph)
logger.info(
'summary write initialized, writing to {}'.format(self.path))