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

Before Change

        new_logits_v = net(states_v)
        new_prob_v = F.softmax(new_logits_v, dim=1)
        kl_div_v = -((new_prob_v / prob_v).log() * prob_v).sum(dim=1).mean()
        writer.add_scalar("kl", kl_div_v.data.cpu().numpy()[0], step_idx)

        grad_max = 0.0
        grad_means = 0.0

After Change

        // calc KL-div
        new_logits_v = net(states_v)
        new_prob_v = F.softmax(new_logits_v, dim=1)
        kl_div_v = -((new_prob_v / prob_v).log() * prob_v).sum(dim=1).mean()
        writer.add_scalar("kl", kl_div_v.item(), step_idx)

        grad_max = 0.0
        grad_means = 0.0

Instances