def _try_save(self, agent: AgentABC):
agent_logs = agent_logger.flush(self.agent_logger_cid)
mean_test_reward = np.mean(agent_logs[0]["test_episode_total_reward"])
if mean_test_reward > self.best_score or not self.save_best_only:
if mean_test_reward > self.best_score:
self.best_score = mean_test_reward
self._save_agent(agent, mean_test_reward)
def on_iteration_end(self, agent: AgentABC):
agent_logs = agent_logger.flush(self.agent_logger_cid)
mean_test_reward = np.mean(agent_logs[0]["test_episode_total_reward"])
break_flag = mean_test_reward >= self.target_reward
if break_flag and self.verbose:
print(
"Early stopping in iteration_number %s. "
"Achieved mean raw reward of %.4f (target was %.4f)" %
(agent.iteration_count, mean_test_reward, self.target_reward))
return break_flag
def on_iteration_end(self, agent: AgentABC):
agent_logs = agent_logger.flush(self.agent_logger_cid)
mean_test_reward = np.mean(agent_logs[0]["test_episode_total_reward"])
mean_test_episode_length = np.mean(
agent_logs[0]["test_episode_length"])
history_summary = (mean_test_reward, mean_test_episode_length)
if self.on_screen:
self._log_on_screen(agent.iteration_count, history_summary)
if self.to_file:
self._log_to_file(agent.iteration_count, history_summary)
def on_iteration_end(self, agent: AgentABC):
recent_agent_logs = agent_logger.flush(self.agent_logger_cid)
recent_nn_logs = nn_logger.flush(self.nn_logger_cid)
summaries = self._calculate_summaries(recent_agent_logs,
recent_nn_logs)
if self.on_screen:
self._log_on_screen(agent.iteration_count, *summaries)
if self.to_file:
self._log_to_file(agent.iteration_count, *summaries)