d5b0cd8e7960c247bb7c5b7c832358f8831780fb,ch15/03_train_trpo.py,,,#,90

Before Change

            traj_states = [t[0].state for t in trajectory]
            traj_actions = [t[0].action for t in trajectory]
            traj_states_v = Variable(torch.from_numpy(np.array(traj_states, dtype=np.float32)))
            traj_actions_v = Variable(torch.from_numpy(np.array(traj_actions, dtype=np.float32)))
            if args.cuda:
                traj_states_v = traj_states_v.cuda()
                traj_actions_v = traj_actions_v.cuda()

            traj_adv_v, traj_ref_v = calc_adv_ref(trajectory, net_crt, traj_states_v, cuda=args.cuda)
            mu_v = net_act(traj_states_v)
            old_logprob_v = calc_logprob(mu_v, net_act.logstd, traj_actions_v)

After Change

            traj_states = [t[0].state for t in trajectory]
            traj_actions = [t[0].action for t in trajectory]
            traj_states_v = torch.FloatTensor(traj_states).to(device)
            traj_actions_v = torch.FloatTensor(traj_actions).to(device)
            traj_adv_v, traj_ref_v = calc_adv_ref(trajectory, net_crt, traj_states_v, device=device)
            mu_v = net_act(traj_states_v)
            old_logprob_v = calc_logprob(mu_v, net_act.logstd, traj_actions_v)

Instances