def select_action(self,
observation: types.ObservationType) -> types.ActionData:
if np.random.uniform() < self.epsilon:
return types.ActionData(action=np.random.choice(self.action_space))
distribution = self.policy.get_distribution(observation)
return types.ActionData(
action=np.random.choice(self.action_space, p=distribution))
def select_action(self,
observation: types.ObservationType) -> types.ActionData:
self._memory_buffer.append(observation)
if np.random.uniform() < self.epsilon:
return types.ActionData(action=np.random.choice(self.action_space))
observation = np.asarray(list(self._memory_buffer))
distribution = self.policy.get_distribution(observation)
return types.ActionData(action=np.argmax(distribution))
def select_action(self,
observation: types.ObservationType) -> types.ActionData:
distribution = self.policy.get_distribution(observation)
noise = np.random.normal(loc=0.,
scale=self.sigma,
size=len(self.action_space))
distribution += noise
return types.ActionData(action=np.argmax(distribution))