def _publish_reward_topic(self, reward, episode_number=1):
"""
This function publishes the given reward in the reward topic for
easy access from ROS infrastructure.
:param reward:
:param episode_number:
:return:
"""
reward_msg = RLExperimentInfo()
reward_msg.episode_number = episode_number
reward_msg.episode_reward = reward
self.reward_pub.publish(reward_msg)
def _publish_reward_topic(self, reward, steps, episode_number=1):
"""
This function publishes the given reward in the reward topic for
easy access from ROS infrastructure.
:param reward:
:param episode_number:
:return:
"""
reward_msg = RLExperimentInfo()
reward_msg.episode_number = episode_number
reward_msg.episode_reward = reward
self.reward_pub.publish(reward_msg)
self.reward_list.append(reward)
self.episode_list.append(episode_number)
self.step_list.append(steps)
liste = str(reward) + ";" + str(episode_number) + ";" + str(steps) + "\n"
with open(self.csv_name + '.csv', 'a') as csv:
csv.write(str(liste))
def run(self, agent, num_episodes, do_train=False):
scores = deque(maxlen=100)
for episode in range(num_episodes):
state = self.env.reset().reshape(
1, self.env.observation_space.shape[0])
total_reward = 0
for t in range(self.max_timesteps):
action = agent.select_action(state, do_train)
next_state, reward, done, _ = self.env.step(action)
next_state = next_state.reshape(
1, self.env.observation_space.shape[0])
reward = self.calc_reward(state, action, reward, next_state,
done)
if do_train:
agent.record(state, action, reward, next_state, done)
total_reward += reward
state = next_state
if done:
reward_msg = RLExperimentInfo()
reward_msg.episode_number = episode
reward_msg.episode_reward = total_reward
self.reward_pub.publish(reward_msg)
break
scores.append(total_reward)
mean_score = np.mean(scores)
if do_train:
agent.replay()
print(
'[Episode {}] - Mean survival time over last 100 episodes was {} ticks.'
.format(episode, mean_score))
done_msg = Bool()
done_msg.data = True
self.done_pub.publish(done_msg)
def _publish_reward_topic(self, reward, episode_number=1):
reward_msg = RLExperimentInfo()
reward_msg.episode_number = episode_number
reward_msg.episode_reward = reward
self.reward_pub.publish(reward_msg)
if next_state_[j] < 0:
next_state_[j] = 0
elif next_state_[j] > env.shape[j] - 1:
next_state_[j] = env.shape[j] - 1
print("Go into state: ", next_state_, " from state: ", state_,
" by action: ", stringify(action))
# Make the algorithm learn based on the results
qlearn.learn(tuple(state_), action, reward, tuple(next_state_))
if not (done):
state = next_state
else:
rospy.loginfo("DONE")
last_time_steps = np.append(last_time_steps, [int(t + 1)])
reward_msg = RLExperimentInfo()
reward_msg.episode_number = x
reward_msg.episode_reward = cumulated_reward
reward_pub.publish(reward_msg)
break
m, s = divmod(int(time.time() - start_time), 60)
h, m = divmod(m, 60)
rospy.loginfo(
("EP: " + str(x + 1) + " - [alpha: " +
str(round(qlearn.alpha, 2)) + " - gamma: " +
str(round(qlearn.gamma, 2)) + " - epsilon: " +
str(round(qlearn.epsilon, 2)) + "] - Reward: " +
str(cumulated_reward) + " Time: %d:%02d:%02d" % (h, m, s)))
done_msg = Bool()
done_msg.data = True