f09a705de11b1b6ade4a7353f238e8f379d15ce5,ch09/04_pong_pg.py,,,#,47

Before Change

            m_grad_mean.append(grad_means / grad_count)

            if train_step_idx % 10 == 0:
                writer.add_scalar("baseline", np.mean(m_baseline), step_idx)
                writer.add_scalar("batch_scales", np.mean(m_batch_scales), step_idx)
                writer.add_scalar("loss_entropy", np.mean(m_loss_entropy), step_idx)
                writer.add_scalar("loss_policy", np.mean(m_loss_policy), step_idx)
                writer.add_scalar("loss_total", np.mean(m_loss_total), step_idx)

After Change

            batch_actions_t = torch.LongTensor(batch_actions)
            bs = np.array(batch_scales, dtype=np.float32)
            bs -= bs.mean()
            if abs(bs.std()) > 1e-5:
                bs /= bs.std()

            batch_scale_v = Variable(torch.from_numpy(bs))
            if args.cuda:
                states_v = states_v.cuda()
                batch_actions_t = batch_actions_t.cuda()

Instances