def run(self):
event = self._sim.get_next_event()
if isinstance(event, EpisodeStartEvent):
log.event("Episode Start")
observation = self._sim.gym_episode_start(params(event))
event.initial_state = state(observation)
elif isinstance(event, SimulateEvent):
log.event("Simulate")
obs, rwd, done, _ = self._sim.gym_simulate(action(event))
event.state = state(obs)
event.reward = reward(rwd)
event.terminal = terminal(done)
self._episode_reward += rwd
self._sim.render()
elif isinstance(event, EpisodeFinishEvent):
log.event("Episode Finish")
print("Episode {} reward: {}".format(
self._episode_count, self._episode_reward))
self._episode_count += 1
self._episode_reward = 0
elif isinstance(event, FinishedEvent):
log.event("Finished")
return False
elif event is None:
return False
return True
def episode_start(self, parameters=None):
""" called at the start of every episode. should
reset the simulation and return the initial state
"""
log.info('Episode {} Starting'.format(self.episode_count))
self.model.reset()
return star.state(self.model.state)
def simulate(self, brain_action):
action = star.action(brain_action)
self.model.step(action)
terminal = star.terminal(self.model.state)
reward = star.reward(self.model.state, terminal)
brain_state = star.state(self.model.state)
if terminal:
log.info(f'Episode {self.episode_count}: '
f'iterations={self.iteration_count:-3.0f} reward={self.episode_reward:-3.1f}')
return (brain_state, reward, terminal)
def simulate(self, brain_action):
""" run a single step of the simulation.
if the simulation has reached a terminal state, mark it as such.
"""
action = star.action(brain_action)
self.model.step(action)
if self.iteration_count >= 200:
terminal = True
else:
terminal = star.terminal(self.model.state)
reward = star.reward(self.model.state, terminal)
brain_state = star.state(self.model.state)
return (brain_state, reward, terminal)
def episode_start(self, parameters=None):
self.model.reset()
return star.state(self.model.state)
def episode_start(self, parameters=None):
theta = np.random.uniform(-2 * RAD_PER_DEG, 2 * RAD_PER_DEG) # +/- 2 degree sd in radians
self.model.reset(theta=theta)
return star.state(self.model.state)
