def start(self, observation):
state = observation.intArray[0]
action = self.egreedy(state)
return_action = Action()
return_action.intArray = [action]
self.lastAction = copy.deepcopy(return_action)
self.lastObservation = copy.deepcopy(observation)
return return_action
def step(self, reward, observation):
# Generate random action, 0 or 1
int_action = self.rand_generator.randint(0, 1)
return_action = Action()
return_action.intArray = [int_action]
last_action = copy.deepcopy(return_action)
last_observation = copy.deepcopy(observation)
return return_action
def choose_action(self):
pi_s = self._learner.policy(self._laststate)
lastaction = pi_s.policy_action
pi_s = lastaction.policy(self._laststate)
self._lastaction = lastaction
assert self._lastaction
# print(lastaction)
return_action = Action()
return_action.intArray = self._lastaction.tolist()
return return_action
def step(self, reward, observation):
state = observation.intArray[0]
last_state = self.lastObservation.intArray[0]
last_action = self.lastAction.intArray[0]
action = self.egreedy(state)
Q_sa = self.value_function[last_state][last_action]
Q_sprime_aprime = self.value_function[state][action]
new_Q_sa = Q_sa + self.sarsa_stepsize * (reward + self.sarsa_gamma * Q_sprime_aprime - Q_sa)
if not self.policyFrozen:
self.value_function[last_state][last_action] = new_Q_sa
return_action = Action()
return_action.intArray = [action]
self.lastAction = copy.deepcopy(return_action)
self.lastObservation = copy.deepcopy(observation)
return return_action
def get_Action(self):
return Action.from_AbstractType(self.get_AbstractType())