def make_replay()

in rlmeta/agents/ppo/ppo_agent.py [0:0]

• 22 lines of code
• 5 McCabe index (conditional complexity)

    def make_replay(self) -> List[NestedTensor]:
        v = 0.0
        gae = 0.0
        ret = []
        for cur in reversed(self.trajectory):
            reward = cur.pop("reward")
            v_ = v
            v = cur["v"]
            if self.reward_rescaling:
                v = self.reward_rescaler.recover(v)
            delta = reward + self.gamma * v_ - v
            gae = delta + self.gamma * self.gae_lambda * gae
            cur["gae"] = gae
            ret.append(gae + v)

        if self.reward_rescaling:
            ret = data_utils.stack_tensors(ret)
            self.reward_rescaler.update(ret)
            ret = self.reward_rescaler.rescale(ret)
            ret = ret.unbind()
        for cur, r in zip(self.trajectory, reversed(ret)):
            cur["return"] = r

        return self.trajectory