def make_replay()

in rlmeta/agents/dqn/apex_dqn_agent.py [0:0]

• 24 lines of code
• 4 McCabe index (conditional complexity)

    def make_replay(self) -> Optional[List[NestedTensor]]:
        trajectory_len = len(self.trajectory)
        if trajectory_len <= self.multi_step:
            return None

        replay = []
        append = replay.append
        for i in range(0, trajectory_len - self.multi_step):
            cur = self.trajectory[i]
            nxt = self.trajectory[i + self.multi_step]
            obs = cur["obs"]
            act = cur["action"]
            next_obs = nxt["obs"]
            done = nxt["done"]
            reward = 0.0
            for j in range(self.multi_step):
                reward += (self.gamma**j) * self.trajectory[i + j]["reward"]
            append({
                "obs": obs,
                "action": act,
                "reward": torch.tensor([reward]),
                "next_obs": next_obs,
                "done": torch.tensor([done]),
            })

        return replay