class QNET():
  def batch_train(self, batch_size=64):
        """Implement Double DQN Algorithm, batch training"""
        if self.exp.get_num() < self.exp.get_min():
            #The number of experiences is not enough for batch training
            return

        # get a batch of experiences
        state, action, reward, next_state, done = self.exp.get_batch(batch_size)
        state = state.reshape(batch_size, self.in_units)
        next_state = next_state.reshape(batch_size, self.in_units)

        # get actions by Q-network
        qnet_q_values = self.session.run(self.q, feed_dict={self.x:next_state})
        qnet_actions = np.argmax(qnet_q_values, axis=1)

        # calculate estimated Q-values with qnet_actions by using Target-network
        tnet_q_values = self.session.run(self.tnet.q, feed_dict={self.tnet.x:next_state})
        tnet_q = [np.take(tnet_q_values[i], qnet_actions[i]) for i in range(batch_size)]

        # Update Q-values of Q-network
        qnet_update_q = [r+0.95*q if not d else r for r, q, d in zip(reward, tnet_q, done)]

        # optimization
        indices=[[i,action[i]] for i in range(batch_size)]
        feed_dict={self.x:state, self.target:qnet_update_q, self.selected_idx:indices}
        self.session.run(self.train_opt, feed_dict)