dqn_pendulum.py 文件源码

def run(self, train=True, movie=False, enableLog=False):

        self.env.reset(0,0)

        self.reset_seq()
        total_reward=0

        for i in range(300):
            # ???state????????????
            old_seq = self.seq.copy()

            # ?????????????
            action = self.agent.get_action(old_seq,train)

            # ??????????
            self.env.update_state(action)
            reward=self.env.get_reward()
            total_reward +=reward

            # ???????state???????????
            state = self.env.get_state()
            self.push_seq(state)
            new_seq = self.seq.copy()

            # ??????????????????
            self.agent.experience_local(old_seq, action, reward, new_seq)

            if enableLog:
                self.log.append(np.hstack([old_seq[0], action, reward]))

            # ????????????
            if movie:
                display.clear_output(wait=True)
                display.display(self.env.get_svg())
                time.sleep(0.01)


        # ????????????????????????????
        self.agent.experience_global(total_reward)

        if train:
            # ??????????????????
            self.agent.update_model(old_seq, action, reward, new_seq)
            self.agent.reduce_epsilon()

        if enableLog:
            return total_reward,self.log

        return total_reward