bb400593d5aeccdccf5f827c1e30b0ea10774da7,chainerrl/agents/categorical_double_dqn.py,CategoricalDoubleDQN,_compute_target_values,#CategoricalDoubleDQN#Any#,20

Before Change

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

        with chainer.using_config("train", False), state_kept(self.q_function):
            next_qout = self.q_function(batch_next_state)

        target_next_qout = self.target_q_function(batch_next_state)

After Change

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

        if self.recurrent:
            target_next_qout, _ = self.target_model.n_step_forward(
                batch_next_state, exp_batch["next_recurrent_state"],
                output_mode="concat")
            next_qout, _ = self.model.n_step_forward(
                batch_next_state, exp_batch["next_recurrent_state"],
                output_mode="concat")
        else:
            target_next_qout = self.target_model(batch_next_state)
            next_qout = self.model(batch_next_state)

        next_q_max = target_next_qout.evaluate_actions(
            next_qout.greedy_actions)

        batch_size = batch_rewards.shape[0]

Instances