def get_trajectory(self, last_pred=None):
"""Get trajectory"""
# Need copy, when run with explore time > 1,
# if not, will clear trajectory before sent.
# trajectory = message(self.trajectory.copy())
trajectory = message(deepcopy(self.trajectory))
set_msg_info(trajectory, agent_id=self.id)
return trajectory
def get_trajectory(self, last_pred=None):
for _data_key in ("cur_state", "logit", "action"):
self.trajectory[_data_key] = np.asarray(self.trajectory[_data_key])
self.trajectory["action"].astype(np.int32)
# self.trajectory["cur_state"].astype(np.int32)
# print(self.trajectory)
trajectory = message(self.trajectory)
set_msg_info(trajectory, agent_id=self.id)
return trajectory
def transfer_to_broker(self):
"""Send train data to learner."""
while True:
data = self.recv_agent.recv()
info_cmd = get_msg_info(data, "cmd")
new_cmd = info_cmd + self.learner_postfix
set_msg_info(data,
broker_id=self.broker_id,
explorer_id=self.explorer_id,
cmd=new_cmd)
self.send_broker.send(data)
def transfer_to_broker(self):
"""Send train data to learner."""
while True:
data = self.recv_agent.recv()
info_cmd = get_msg_info(data, "cmd")
# print("info_cmd in explorer: ", info_cmd, data)
data_type = "buf_reduce" if info_cmd == "buf_reduce" else "data"
set_msg_info(data,
broker_id=self.broker_id,
explorer_id=self.explorer_id)
self.send_broker.send(data, data_type=data_type)
def get_trajectory(self, last_pred=None):
for env_id in range(self.vector_env_size):
for _data_key in ("cur_state", "logit", "action", "reward", "done",
"info"):
self.trajectory[_data_key].extend(
self.sample_vector[env_id][_data_key])
# merge data into env_num * seq_len
for _data_key in self.trajectory:
self.trajectory[_data_key] = np.stack(self.trajectory[_data_key])
self.trajectory["action"].astype(np.int32)
trajectory = message(self.trajectory.copy())
set_msg_info(trajectory, agent_id=self.id)
return trajectory
def predict(self):
"""Predict action."""
while True:
start_t0 = time()
data = self.request_q.recv()
state = get_msg_data(data)
self._stats.obs_wait_time += time() - start_t0
start_t1 = time()
with self.lock:
action = self.alg.predict(state)
self._stats.inference_time += time() - start_t1
set_msg_info(data, cmd="predict_reply")
set_msg_data(data, action)
# logging.debug("msg to explore: ", data)
self.reply_q.send(data)
self._stats.iters += 1
if self._stats.iters > self._report_period:
_report = self._stats.get()
self.stats_deliver.send(_report, block=True)
def get_trajectory(self):
trajectory = message(self.trajectory.copy())
set_msg_info(trajectory, agent_id=self.id)
return trajectory
def get_trajectory(self):
transition = self.batch.data.transition_data
transition.update(self._info.copy()) # record win rate within train
trajectory = message(transition)
set_msg_info(trajectory, agent_id=self.id)
return trajectory