4296a765125fff6491892a1bb70fb32ac516dae6,ch15/01_train_a2c.py,,,#,54

Before Change

                adv_v = vals_ref_v.unsqueeze(dim=-1) - value_v.detach()
                log_prob_v = adv_v * calc_logprob(mu_v, var_v, actions_v)
                loss_policy_v = -log_prob_v.mean()
                entropy_loss_v = ENTROPY_BETA * (-(torch.log(2*math.pi*var_v) + 1)/2).mean()

                loss_v = loss_policy_v + entropy_loss_v + loss_value_v
                loss_v.backward()
                optimizer.step()

After Change

    exp_source = ptan.experience.ExperienceSourceFirstLast(envs, agent, GAMMA, steps_count=REWARD_STEPS)

    opt_act = optim.Adam(net_act.parameters(), lr=LEARNING_RATE_ACTOR)
    opt_crt = optim.Adam(net_crt.parameters(), lr=LEARNING_RATE_CRITIC)

    batch = []
    best_reward = None
    with ptan.common.utils.RewardTracker(writer) as tracker:

Instances