8acf099847ebf73ad8cdae1341d0f768dbe1c094,ch09/04_pong_pg.py,,,#,28

Before Change

            if train_step_idx % 100 == 0:
                writer.add_scalar("baseline", baseline, step_idx)
                writer.add_scalar("batch_scales", np.mean(batch_scales), step_idx)
                writer.add_scalar("loss_entropy", entropy_loss_v.data.cpu().numpy()[0], step_idx)
                writer.add_scalar("loss_policy", loss_policy_v.data.cpu().numpy()[0], step_idx)
                writer.add_scalar("loss_total", loss_v.data.cpu().numpy()[0], step_idx)

After Change

            entropy_loss_v = ENTROPY_BETA * (prob_v * log_prob_v).sum()
            loss_v = loss_policy_v + entropy_loss_v

            m_baseline.append(baseline)
            m_batch_scales.append(np.mean(batch_scales))
            m_loss_entropy.append(entropy_loss_v.data.cpu().numpy()[0])
            m_loss_policy.append(loss_policy_v.data.cpu().numpy()[0])
            m_loss_total.append(loss_v.data.cpu().numpy()[0])

Instances