e4f051b6cce414997a97b896276563c4e361d0b8,ch09/04_cartpole_pg.py,,,#,35

Before Change

        writer.add_scalar("entropy", entropy_v.data.cpu().numpy()[0], 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)
        writer.add_scalar("grad_l2", grad_means / grad_count, step_idx)
        writer.add_scalar("grad_max", grad_max, step_idx)

After Change

        if len(batch_states) < BATCH_SIZE:
            continue

        states_v = torch.FloatTensor(batch_states)
        batch_actions_t = torch.LongTensor(batch_actions)
        batch_scale_v = torch.FloatTensor(batch_scales)

        optimizer.zero_grad()

Instances