def update_Q_table(self, observation):
index_1 = state2index_(L2=self.state[-3], O=self.state[-4])
index_2 = state2index_(L2=self.state[-1], O=self.state[-2])
if observation == 'B' or observation == 'S':
reward = 1
else:
reward = 0
self.Q_table[index_1][self.action] += self.learning_rate \
* (reward + self.gamma * np.max(self.Q_table[index_2]) - self.Q_table[index_1][self.action])
def select_action(self, observation):
# 0 is wait, 1 is transmit
###########################################
index = state2index_(L2=self.queue, O=observation)
self.epsilon *= self.epsilon_decay
self.epsilon = max(self.epsilon_min, self.epsilon)
if self.epsilon > np.random.uniform():
self.action = round(np.random.uniform())
else:
self.action = np.argmax(self.Q_table[index])
if sum(self.queue) == 0:
self.action = 0