e37a3673f57936713fea57581962937ce096e1a0,reagent/evaluation/seq2reward_evaluator.py,Seq2RewardEvaluator,evaluate,#Seq2RewardEvaluator#Any#,21

Before Change

        detached_loss = loss.cpu().detach().item()

        if self.trainer.view_q_value:
            q_values = (
                get_Q(
                    self.trainer.seq2reward_network, eval_batch, self.trainer.all_permut
                )
                .cpu()
                .mean(0)
                .tolist()
            )
        else:
            q_values = [0] * len(self.trainer.params.action_names)

        self.reward_net.train(reward_net_prev_mode)

After Change

            minlength=len(self.trainer.params.action_names),
        )
        // normalize
        action_distribution = (
            action_distribution.float() / torch.sum(action_distribution)
        ).tolist()

        self.reward_net.train(reward_net_prev_mode)
        return (detached_loss, q_values, action_distribution)

Instances