d9f1874e3489edc915228dcfdae64d5294034464,rllib/agents/pg/tests/test_pg.py,TestPG,test_pg_loss_functions,#TestPG#,34

Before Change

                expected_logp = expected_logp.numpy()
            expected_loss = -np.mean(
                expected_logp *
                (train_batch[Postprocessing.ADVANTAGES] if fw != "torch" else
                 train_batch[Postprocessing.ADVANTAGES].numpy()))
            check(results, expected_loss, decimals=4)

After Change

            adv = train_batch[Postprocessing.ADVANTAGES]
            if sess:
                expected_logp = sess.run(expected_logp)
            elif fw == "torch":
                expected_logp = expected_logp.detach().cpu().numpy()
                adv = adv.detach().cpu().numpy()
            else:
                expected_logp = expected_logp.numpy()
            expected_loss = -np.mean(expected_logp * adv)

Instances