a997c9720844894b9119be2d6ea8dd6fa057c143,samples/dqn_expreplay.py,,,#,16

Before Change

                q_vals = q_vals.cuda()
            l = loss_fn(model(states), q_vals)
            losses.append(l.data[0])
            mean_q.append(q_vals.mean().data[0])
            l.backward()
            optimizer.step()

After Change

            q_vals.append(train_q)
        return torch.from_numpy(np.array(states, dtype=np.float32)), torch.stack(q_vals)

    reward_sma = utils.SMAQueue(run.getint("stop", "mean_games", fallback=100))

    for idx in range(10000):
        exp_replay.populate(run.getint("exp_buffer", "populate"))

Instances