72313bc158d57e1f36e2c17848338a3079982def,ch16/01_cartpole_es.py,,train_step,#,69

Before Change

    opt.zero_grad()
    for p, p_update in zip(net.parameters(), weighted_noise):
        update = p_update / (len(batch_reward) * NOISE_STD)
        p.grad = Variable(update)
//        p.data += LEARNING_RATE * update
//        m_updates.append(torch.norm(update))
    opt.step()
    writer.add_scalar("update_l2", np.mean(m_updates), step_idx)

After Change

    for p, p_update in zip(net.parameters(), weighted_noise):
        update = p_update / (len(batch_reward) * NOISE_STD)
        p.data += LEARNING_RATE * update
        m_updates.append(torch.norm(update))
    writer.add_scalar("update_l2", np.mean(m_updates), step_idx)


if __name__ == "__main__":

Instances