4e008e79147043207a5d0032aa771d7f811af0e1,chainerrl/agents/dqn.py,DQN,_compute_target_values,#DQN#Any#Any#,251

Before Change

    def _compute_target_values(self, experiences, gamma):

        batch_next_state = self._batch_states(
            [elem["next_state"] for elem in experiences])

        target_next_qout = self.target_model(batch_next_state, test=True)
        next_q_max = target_next_qout.max
        next_q_max.creator = None

        batch_rewards = self.xp.asarray(
            [elem["reward"] for elem in experiences], dtype=np.float32)

        batch_terminal = self.xp.asarray(
            [elem["is_state_terminal"] for elem in experiences],
            dtype=np.float32)

        return batch_rewards + self.gamma * (1.0 - batch_terminal) * next_q_max

    def _compute_y_and_t(self, experiences, gamma):

After Change

    def _compute_target_values(self, exp_batch, gamma):

        batch_next_state = exp_batch["next_state"]

        target_next_qout = self.target_model(batch_next_state, test=True)
        next_q_max = target_next_qout.max
        next_q_max.creator = None

        batch_rewards = exp_batch["reward"]
        batch_terminal = exp_batch["is_state_terminal"]

        return batch_rewards + self.gamma * (1.0 - batch_terminal) * next_q_max

    def _compute_y_and_t(self, exp_batch, gamma):

Instances