99abcc6e9b57f441999ce10dbc31ca1bed79c356,ch15/04_train_ppo.py,,,#,55

Before Change

                surr_obj_v = adv_v * torch.exp(logprob_pi_v - logprob_old_pi_v)
                clipped_surr_v = torch.clamp(surr_obj_v, 1.0 - PPO_EPS, 1.0 + PPO_EPS)
                loss_policy_v = -torch.min(surr_obj_v, clipped_surr_v).mean()
                entropy_loss_v = ENTROPY_BETA * (-(torch.log(2*math.pi*var_v) + 1)/2).mean()

                loss_v = loss_policy_v + entropy_loss_v + loss_value_v
                loss_v.backward()
                optimizer.step()

                tb_tracker.track("advantage", adv_v, step_idx)
                tb_tracker.track("values", value_v, step_idx)
                tb_tracker.track("batch_rewards", vals_ref_v, step_idx)
                tb_tracker.track("loss_entropy", entropy_loss_v, step_idx)
                tb_tracker.track("loss_policy", loss_policy_v, step_idx)
                tb_tracker.track("loss_value", loss_value_v, step_idx)
                tb_tracker.track("loss_total", loss_v, step_idx)

After Change

    test_env = gym.make(ENV_ID)

    net_act = model.ModelActor(envs[0].observation_space.shape[0], envs[0].action_space.shape[0])
    net_crt = model.ModelCritic(envs[0].observation_space.shape[0])
    if args.cuda:
        net_act.cuda()
        net_crt.cuda()

Instances