e70bdb2d089ae283781c45b8d97963823a984baa,ch10/00_pong_pg.py,,,#,51

Before Change

            grad_vars = 0.0
            grad_count = 0
            for p in net.parameters():
                grad_max = max(grad_max, p.grad.abs().max().data.cpu().numpy()[0])
                grad_means += (p.grad ** 2).mean().sqrt().data.cpu().numpy()[0]
                grad_vars += torch.var(p.grad).data.cpu().numpy()[0]
                grad_count += 1

After Change

            if USE_MEAN_BASELINE:
                batch_scales.append(exp.reward - baseline)
            else:
                batch_scales.append(exp.reward)

            // handle new rewards
            new_rewards = exp_source.pop_total_rewards()
            if new_rewards:

Instances