class ModelTrainer(object):
def __init__(self, sim, predict=False):
self._sim = sim
self.episode_count = 0
self.reset_iteration_metrics()
self.star = Star(predict)
self.star.logger = sim.results_logger
if log_iterations is True:
self.logged_observations = self.star.define_logged_observations()
self.logged_observations = self.update_logged_observations(
self.logged_observations)
bonsai_tools.log_observations_columns(self.star.logger,
self.logged_observations)
def episode_start(self, event):
if getattr(self._sim, 'sim_id', -1) == -1:
self.sim_id = self._sim._impl._sim_id
#if self.sim_id != -1:
# print('SimID', self.sim_id)
self.start_episode()
self.episode_count += 1
# Check https://docs.bons.ai/references/library-reference.html#event-class for SDK event class documentation from Product
event.initial_state = self.star.get_state()
event.terminal = self.star.get_terminal(event.initial_state)
event.reward = 0 #the initial reward is an arbitrary value since there are no actions taken by BRAIN in initial state
if log_iterations is True:
self.logged_observations = self.star.define_logged_observations()
self.logged_observations = self.update_logged_observations(
self.logged_observations)
bonsai_tools.log_iteration(self.star.logger,
self.logged_observations)
def run(self):
event = self._sim.get_next_event()
if isinstance(event, EpisodeStartEvent):
log.event("Episode Start Train")
self.episode_start(event)
# Receive the action from the BRAIN as event.action, run the simulation one step and return the state, action, and reward to the BRAIN.
elif isinstance(event, SimulateEvent):
log.event("Simulate")
self.iteration_count += 1
self.action = event.action
self.star.set_action(self.action)
event.state = self.star.get_state()
event.terminal = self.star.get_terminal(event.state)
event.reward = self.star.get_reward(event.state, event.terminal)
#print(event.state)
self.reward = event.reward
self.terminal = event.terminal
self.episode_reward += event.reward
self.logged_observations = self.star.define_logged_observations()
self.logged_observations = self.update_logged_observations(
self.logged_observations)
if log_iterations is True:
bonsai_tools.log_iteration(self.star.logger,
self.logged_observations)
else:
bonsai_tools.print_progress(self.logged_observations)
# The episode is terminal. Finish the episode.
elif isinstance(event, EpisodeFinishEvent):
log.event("Episode Finish")
print(
"episode count: {}, iteration count: {}, episode reward: {:6.2f}"
.format(self.episode_count, self.iteration_count,
self.episode_reward))
elif isinstance(event, FinishedEvent):
log.event("Finished")
return False
elif event is None:
return False
return True
def start_episode(self, config=None):
self.star.simulator_reset_config()
self.reset_iteration_metrics()
def reset_iteration_metrics(self):
"""Executed once every start of episode
"""
self.reward = 0
self.terminal = False
self.episode_reward = 0.0
self.iteration_count = 0
self._cpu_pc = psutil.cpu_percent()
self._vmem = psutil.virtual_memory().percent
def update_logged_observations(self, logged_observations):
self._cpu_pc = psutil.cpu_percent()
self._vmem = psutil.virtual_memory().percent
updated_observations = {
'episode_count': self.episode_count,
'iteration_count': self.iteration_count,
'terminal': self.terminal,
'reward': self.reward,
'episode_reward': self.episode_reward,
'cpu_pc': self._cpu_pc,
'vmem': self._vmem
}
#updated_observations.update(self.action)
updated_observations.update(logged_observations)
return updated_observations
class ModelTrainer(object):
def __init__(self, sim, predict=False):
self._sim = sim
self.episode_count = 0
self.reset_iteration_metrics()
print('model type that is being sent to the star is:', modeltype)
time.sleep(1)
self.star = Star(predict,modeltype=modeltype)
self.star.logger = sim.results_logger
if log_iterations is True:
self.logged_observations = self.star.define_logged_observations()
self.logged_observations = self.update_logged_observations(self.logged_observations)
bonsai_tools.log_observations_columns(self.star.logger, self.logged_observations)
def episode_start(self,event):
if getattr(self._sim, 'sim_id', -1) == -1:
self.sim_id = self._sim._impl._sim_id
#if self.sim_id != -1:
# print('SimID', self.sim_id)
self.start_episode()
self.episode_count += 1
event.initial_state = self.star.get_state()
event.terminal = self.star.get_terminal(event.initial_state)
event.reward = 0 #the initial reward is an arbitrary value since there are no actions taken by BRAIN in initial state
if log_iterations is True:
self.logged_observations = self.star.define_logged_observations()
self.logged_observations = self.update_logged_observations(self.logged_observations)
bonsai_tools.log_iteration(self.star.logger, self.logged_observations)
def run(self):
event = self._sim.get_next_event()
if isinstance(event, EpisodeStartEvent):
log.event("Episode Start Train")
self.episode_start(event)
# Receive the action from the BRAIN as event.action, run the simulation one step and return the state, action, and reward to the BRAIN.
elif isinstance(event, SimulateEvent):
log.event("Simulate")
self.iteration_count += 1
self.action = event.action
self.star.set_action(self.action)
event.state = self.star.get_state()
event.terminal = self.star.get_terminal(event.state)
event.reward = self.star.get_reward(event.state, event.terminal)
#print(event.state)
self.reward = event.reward
self.terminal = event.terminal
self.episode_reward += event.reward
self.logged_observations = self.star.define_logged_observations()
self.logged_observations = self.update_logged_observations(self.logged_observations)
if log_iterations is True:
bonsai_tools.log_iteration(self.star.logger, self.logged_observations)
else:
bonsai_tools.print_progress(self.logged_observations)
# The episode is terminal. Finish the episode.
elif isinstance(event, EpisodeFinishEvent):
log.event("Episode Finish")
print("episode count: {}, iteration count: {}, episode reward: {:6.2f}".format(
self.episode_count, self.iteration_count, self.episode_reward))
elif isinstance(event, FinishedEvent):
log.event("Finished")
return False
elif event is None:
return False
return True
def start_episode(self, config=None):
self.star.reset()
self.reset_iteration_metrics()
def reset_iteration_metrics(self):
"""Executed once every start of episode
"""
self.reward = 0
self.terminal = False
self.episode_reward = 0.0
self.iteration_count = 0
def update_logged_observations(self, logged_observations):
updated_observations = {
'episode_count':self.episode_count,
'iteration_count':self.iteration_count,
'terminal':self.terminal,
'reward':self.reward,
'episode_reward':self.episode_reward
}
#updated_observations.update(self.action)
updated_observations.update(logged_observations)
return updated_observations